<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD Journal Publishing with OASIS Tables v3.0 20080202//EN" "journalpub-oasis3.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:oasis="http://docs.oasis-open.org/ns/oasis-exchange/table" xml:lang="en" dtd-version="3.0">
  <front>
    <journal-meta><journal-id journal-id-type="publisher">ESSD</journal-id><journal-title-group>
    <journal-title>Earth System Science Data</journal-title>
    <abbrev-journal-title abbrev-type="publisher">ESSD</abbrev-journal-title><abbrev-journal-title abbrev-type="nlm-ta">Earth Syst. Sci. Data</abbrev-journal-title>
  </journal-title-group><issn pub-type="epub">1866-3516</issn><publisher>
    <publisher-name>Copernicus Publications</publisher-name>
    <publisher-loc>Göttingen, Germany</publisher-loc>
  </publisher></journal-meta>
    <article-meta>
      <article-id pub-id-type="doi">10.5194/essd-13-777-2021</article-id><title-group><article-title>OceanSODA-ETHZ: a global gridded data set of the surface ocean carbonate system for seasonal to decadal studies of ocean acidification</article-title><alt-title>Surface ocean acidification data set</alt-title>
      </title-group><?xmltex \runningtitle{Surface ocean acidification data set}?><?xmltex \runningauthor{L.~Gregor and N.~Gruber}?>
      <contrib-group>
        <contrib contrib-type="author" corresp="yes">
          <name><surname>Gregor</surname><given-names>Luke</given-names></name>
          <email>luke.gregor@usys.ethz.ch</email>
        <ext-link>https://orcid.org/0000-0001-6071-1857</ext-link></contrib>
        <contrib contrib-type="author" corresp="no">
          <name><surname>Gruber</surname><given-names>Nicolas</given-names></name>
          
        <ext-link>https://orcid.org/0000-0002-2085-2310</ext-link></contrib>
        <aff id="aff1"><institution>Environmental Physics, Institute of Biogeochemistry and Pollutant Dynamics,<?xmltex \hack{\break}?> ETH Zurich, 8092 Zurich, Switzerland</institution>
        </aff>
      </contrib-group>
      <author-notes><corresp id="corr1">Luke Gregor (luke.gregor@usys.ethz.ch)</corresp></author-notes><pub-date><day>2</day><month>March</month><year>2021</year></pub-date>
      
      <volume>13</volume>
      <issue>2</issue>
      <fpage>777</fpage><lpage>808</lpage>
      <history>
        <date date-type="received"><day>4</day><month>October</month><year>2020</year></date>
           <date date-type="rev-request"><day>22</day><month>October</month><year>2020</year></date>
           <date date-type="rev-recd"><day>8</day><month>January</month><year>2021</year></date>
           <date date-type="accepted"><day>11</day><month>January</month><year>2021</year></date>
      </history>
      <permissions>
        <copyright-statement>Copyright: © 2021 Luke Gregor</copyright-statement>
        <copyright-year>2021</copyright-year>
      <license license-type="open-access"><license-p>This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit <ext-link ext-link-type="uri" xlink:href="https://creativecommons.org/licenses/by/4.0/">https://creativecommons.org/licenses/by/4.0/</ext-link></license-p></license></permissions><self-uri xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021.html">This article is available from https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021.html</self-uri><self-uri xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021.pdf">The full text article is available as a PDF file from https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021.pdf</self-uri>
      <abstract><title>Abstract</title>
    <p id="d1e88">Ocean acidification has profoundly altered the ocean's carbonate chemistry since preindustrial times, with potentially serious consequences for marine life. Yet, no long-term, global observation-based data set exists that allows us to study changes in ocean acidification for all carbonate system parameters over the last few decades. Here, we fill this gap and present a methodologically consistent global data set of all relevant surface ocean parameters, i.e., dissolved inorganic carbon (DIC), total alkalinity (TA), partial pressure of CO<inline-formula><mml:math id="M1" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M2" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>), pH, and the saturation state with respect to mineral CaCO<inline-formula><mml:math id="M3" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M4" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>) at a monthly resolution over the period 1985 through 2018 at a spatial resolution of <inline-formula><mml:math id="M5" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula>. This data set, named OceanSODA-ETHZ, was created by extrapolating in time and space the surface ocean observations of <inline-formula><mml:math id="M6" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (from the Surface Ocean CO<inline-formula><mml:math id="M7" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Atlas, SOCAT) and total alkalinity (TA; from the Global Ocean Data Analysis Project, GLODAP) using the newly developed Geospatial Random Cluster Ensemble Regression (GRaCER) method (code available at <ext-link xlink:href="https://doi.org/10.5281/zenodo.4455354" ext-link-type="DOI">10.5281/zenodo.4455354</ext-link>, <xref ref-type="bibr" rid="bib1.bibx43" id="altparen.1"/>). This method is based on a two-step (cluster-regression) approach but extends it by considering an ensemble of such cluster regressions, leading to improved robustness. Surface ocean DIC, pH, and <inline-formula><mml:math id="M8" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> were then computed from the globally mapped <inline-formula><mml:math id="M9" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA using the thermodynamic equations of the carbonate system. For the open ocean, the cluster-regression method estimates <inline-formula><mml:math id="M10" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA with global near-zero biases and root mean squared errors of 12 <inline-formula><mml:math id="M11" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm and 13 <inline-formula><mml:math id="M12" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M13" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, respectively. Taking into account also the measurement and representation errors, the total uncertainty increases to 14 <inline-formula><mml:math id="M14" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm and 21 <inline-formula><mml:math id="M15" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M16" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, respectively. We assess the fidelity of the computed parameters by comparing them to direct observations from GLODAP, finding surface ocean pH and DIC global biases of near zero, as well as root mean squared errors of 0.023 and 16 <inline-formula><mml:math id="M17" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M18" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, respectively. These uncertainties are very comparable to those expected by propagating the total uncertainty from <inline-formula><mml:math id="M19" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA through the thermodynamic computations, indicating a robust and conservative assessment of the uncertainties. We illustrate the potential of this new data set by analyzing the climatological mean seasonal cycles of the different parameters of the surface ocean carbonate system, highlighting their commonalities and differences. Further, this data set provides a novel constraint on the global- and basin-scale trends in ocean acidification for all parameters. Concretely, we find for the period 1990 through 2018 global mean trends of 8.6 <inline-formula><mml:math id="M20" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1 <inline-formula><mml:math id="M21" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M22" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> per decade for DIC, <inline-formula><mml:math id="M23" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.016 <inline-formula><mml:math id="M24" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.000 per decade for pH, 16.5 <inline-formula><mml:math id="M25" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1 <inline-formula><mml:math id="M26" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm  per decade for <inline-formula><mml:math id="M27" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M28" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, and <inline-formula><mml:math id="M29" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.07 <inline-formula><mml:math id="M30" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.00 per decade for <inline-formula><mml:math id="M31" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. The OceanSODA-ETHZ data can be downloaded from <ext-link xlink:href="https://doi.org/10.25921/m5wx-ja34" ext-link-type="DOI">10.25921/m5wx-ja34</ext-link> <xref ref-type="bibr" rid="bib1.bibx44" id="paren.2"/>.</p>
  </abstract>
    </article-meta>
  </front>
<body>
      

<?pagebreak page778?><sec id="Ch1.S1" sec-type="intro">
  <label>1</label><title>Introduction</title>
      <p id="d1e417">The oceans have taken up roughly one-quarter of the anthropogenic CO<inline-formula><mml:math id="M32" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> that has been released into the atmosphere since the start of the industrial era <xref ref-type="bibr" rid="bib1.bibx96 bib1.bibx50" id="paren.3"/>, modulating the increase in atmospheric CO<inline-formula><mml:math id="M33" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> substantially.
However, this buffering of anthropogenic climate change by the ocean comes with a substantial cost, i.e., ocean acidification <xref ref-type="bibr" rid="bib1.bibx26" id="paren.4"/>.
The uptake of anthropogenic CO<inline-formula><mml:math id="M34" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> over the last 150 years has made the surface ocean more acidic with a decrease in global mean pH from <inline-formula><mml:math id="M35" display="inline"><mml:mo>∼</mml:mo></mml:math></inline-formula> 8.2 around 1850 to <inline-formula><mml:math id="M36" display="inline"><mml:mo>∼</mml:mo></mml:math></inline-formula> 8.1 today <xref ref-type="bibr" rid="bib1.bibx33 bib1.bibx55" id="paren.5"/>.
This decrease in pH equates to a <inline-formula><mml:math id="M37" display="inline"><mml:mo>∼</mml:mo></mml:math></inline-formula> 30 % increase in the concentration of H<inline-formula><mml:math id="M38" display="inline"><mml:msup><mml:mi/><mml:mo>+</mml:mo></mml:msup></mml:math></inline-formula> ions. Some of the anthropogenic CO<inline-formula><mml:math id="M39" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> taken up from the atmosphere remains in the seawater as dissolved CO<inline-formula><mml:math id="M40" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, thus increasing its partial pressure (<inline-formula><mml:math id="M41" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M42" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>).
In fact, surface ocean <inline-formula><mml:math id="M43" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M44" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> tends to track the increase in atmospheric <inline-formula><mml:math id="M45" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M46" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> rather closely (e.g., <xref ref-type="bibr" rid="bib1.bibx4" id="altparen.6"/>) owing to the <inline-formula><mml:math id="M47" display="inline"><mml:mo>∼</mml:mo></mml:math></inline-formula> 1-year timescale for the equilibration of CO<inline-formula><mml:math id="M48" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> across the air–sea interface <xref ref-type="bibr" rid="bib1.bibx97" id="paren.7"/>, which is smaller than the decadal timescale increase in atmospheric CO<inline-formula><mml:math id="M49" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx35" id="paren.8"/>.
While some of the added CO<inline-formula><mml:math id="M50" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> stays as CO<inline-formula><mml:math id="M51" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, the majority of it is titrated away by the ocean's carbonate ion <xref ref-type="bibr" rid="bib1.bibx97" id="paren.9"/>, leading to a substantial reduction in its concentration.
This reduces the saturation state (<inline-formula><mml:math id="M52" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>) with regard to the mineral calcium carbonate (CaCO<inline-formula><mml:math id="M53" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula>), where an <inline-formula><mml:math id="M54" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> of <inline-formula><mml:math id="M55" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> leads to dissolution of CaCO<inline-formula><mml:math id="M56" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula>.</p>
      <p id="d1e654">These chemical changes, collectively described as ocean acidification, will have a profound impact on marine organisms, especially those that form shells made of CaCO<inline-formula><mml:math id="M57" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx86 bib1.bibx29 bib1.bibx26 bib1.bibx5 bib1.bibx27" id="paren.10"/>.
Calcifying organisms living in high latitudes and subtropical and tropical upwelling regions, with their naturally low <inline-formula><mml:math id="M58" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> and pH, may be particularly vulnerable, as these regions will be among the first to cross critical saturation thresholds <xref ref-type="bibr" rid="bib1.bibx86 bib1.bibx100 bib1.bibx49 bib1.bibx34 bib1.bibx30 bib1.bibx51 bib1.bibx82" id="paren.11"/>.
However, marine organisms may be susceptible to changes even where <inline-formula><mml:math id="M59" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> <inline-formula><mml:math id="M60" display="inline"><mml:mrow><mml:mo>&gt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> due to a shift in energetic requirements for shell formation <xref ref-type="bibr" rid="bib1.bibx86 bib1.bibx91" id="paren.12"/>.
For example, it is well known that corals start to decrease their calcification already at saturation states well above 3 <xref ref-type="bibr" rid="bib1.bibx37" id="paren.13"/>.
Ocean acidification will thus have a significant economic impact on fisheries and tourism through the impact on shellfish and corals, respectively <xref ref-type="bibr" rid="bib1.bibx20 bib1.bibx27" id="paren.14"/>.</p>
      <p id="d1e706">At the global scale, most of what we know about the progression of ocean acidification in the recent decades has come from either models <xref ref-type="bibr" rid="bib1.bibx8 bib1.bibx60" id="paren.15"/> or from the combination of model-based trends with observation-based climatologies <xref ref-type="bibr" rid="bib1.bibx33 bib1.bibx55" id="paren.16"/>.
A notable exception is the large number of studies that have analyzed the trends and variability of surface ocean <inline-formula><mml:math id="M61" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M62" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (e.g., <xref ref-type="bibr" rid="bib1.bibx62 bib1.bibx65 bib1.bibx94 bib1.bibx22 bib1.bibx46" id="altparen.17"/>) and the effort of <xref ref-type="bibr" rid="bib1.bibx69" id="text.18"/> and <xref ref-type="bibr" rid="bib1.bibx108" id="text.19"/> to analyze long-term trends in pH and <inline-formula><mml:math id="M63" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> respectively.
But these studies remained limited to one single parameter. At the local-to-regional scale, a number of long-term time series have provided excellent insights into the processes and trends of ocean acidification across all carbonate system parameters (e.g., <xref ref-type="bibr" rid="bib1.bibx4" id="altparen.20"/>), but no global comprehensive view of the historical development of ocean acidification based on observations exists. This is largely a consequence of the limited observations, although observational efforts have increased substantially in the recent decades through efforts such as GOA-ON (Global Ocean Acidification Observing Network, <xref ref-type="bibr" rid="bib1.bibx107" id="altparen.21"/>).
The OceanSODA (Satellite Oceanographic Datasets for Acidification) project (<uri>https://esa-oceansoda.org</uri>, last access: 12 September 2020), which this study forms part of, aims to close this gap by linking satellite observations with in situ observations of the marine carbonate system.</p>
      <p id="d1e757">In line with the goal of the OceanSODA project, we aim to develop a global, observation-based data set documenting the progression of ocean acidification over the recent decades.
Such a data set will be crucial to put the current trends of ocean acidification into the context of the changes over the last few decades.
By also describing the level of variability in ocean acidification around the long-term trend, it will also help to better understand the challenges that marine organisms are facing.
Additionally, it will permit us to explore in much more detail how ocean acidification has unfolded regionally and potentially deviated from the simple model of it being dependent on the rise in atmospheric CO<inline-formula><mml:math id="M64" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>.</p>
      <p id="d1e770">The well-measurable parameters of the marine carbonate system are dissolved inorganic carbon (DIC), total alkalinity (TA), pH, and the partial pressure of carbon dioxide (<inline-formula><mml:math id="M65" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M66" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>).
Very few measurement programs measure all of these parameters concurrently.
In fact, the vast majority of the observational programs measure only one parameter, with <inline-formula><mml:math id="M67" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M68" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> being the most often measured one, followed by DIC, TA, and pH <xref ref-type="bibr" rid="bib1.bibx2 bib1.bibx85" id="paren.22"/>.
Since two parameters are sufficient to fully describe the marine carbonate system, any combination of two will permit us to fully reconstruct the entire carbonate system.
But not all combination are equally suited, given the uncertainties in the measurements, the uncertainties in the coefficients of the carbonate chemistry, and the spatiotemporal coverage vis-à-vis the variability of these parameters.</p>
      <?pagebreak page779?><p id="d1e808">We use here the pair <inline-formula><mml:math id="M69" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M70" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA as the basis for our reconstruction for two reasons.
First, these are the best observed parameters relative to their spatiotemporal variability, permitting us to develop better predictive models for the global surface ocean than possible for, e.g., DIC and pH.
Second, detailed assessments of the internal consistency of the oceanic carbonate system have shown that <inline-formula><mml:math id="M71" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M72" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA are a well-suited pair to estimate pH, owing to the reliability of the measurements and the predictive accuracy <xref ref-type="bibr" rid="bib1.bibx7 bib1.bibx2 bib1.bibx92" id="paren.23"/>.
This is not the case if DIC was used instead of TA.
Our choice is supported by <xref ref-type="bibr" rid="bib1.bibx102" id="text.24"/>, who developed the first seasonal climatology of all surface ocean carbonate system parameters using the same pair.</p>
      <p id="d1e850">Measurements of <inline-formula><mml:math id="M73" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M74" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> are abundant compared to the other variables, due to a well-established and robust underway sampling protocol that allows instruments to also be installed on non-scientific vessels under the Voluntary Observing Ship (VOS) program <xref ref-type="bibr" rid="bib1.bibx2 bib1.bibx90" id="paren.25"/>.
High-quality <inline-formula><mml:math id="M75" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M76" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> data are also easily accessible thanks to the Surface Ocean CO<inline-formula><mml:math id="M77" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Atlas (SOCAT) that consolidates underway <inline-formula><mml:math id="M78" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M79" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> observations and ensures the quality of observations <xref ref-type="bibr" rid="bib1.bibx2" id="paren.26"/>.
Total alkalinity is not as widely measured as <inline-formula><mml:math id="M80" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M81" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> due to the fact that measurements are made discretely with bottle samples <xref ref-type="bibr" rid="bib1.bibx24" id="paren.27"/>.
But, fortunately, TA is highly correlated with salinity on a global scale (<inline-formula><mml:math id="M82" display="inline"><mml:mrow><mml:mi>r</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.96</mml:mn></mml:mrow></mml:math></inline-formula>), making it a suitable variable for prediction with a <inline-formula><mml:math id="M83" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">10</mml:mn></mml:mrow></mml:math></inline-formula> % error of the observed range <xref ref-type="bibr" rid="bib1.bibx72 bib1.bibx85 bib1.bibx13" id="paren.28"/>.
Further, the accessibility of TA measurements is made possible through the continued efforts of the Global Ocean Data Analysis Project <xref ref-type="bibr" rid="bib1.bibx85" id="paren.29"><named-content content-type="pre">GLODAP;</named-content></xref>.
We discarded the option to use DIC instead of TA, even though DIC is slightly more often sampled than TA.
This decision is based on the fact that DIC is more variable than TA, and also its correlation with salinity is much lower.
As a result, it is difficult to develop predictive models for DIC that are as accurate and precise as those for TA. Oceanic pH is also not an option, since historically it has been measured far less often than the other parameters.
This is changing, since progress with reference materials and new sensors have permitted a tremendous increase in the number of pH measurements in recent years, largely benefiting from deployments of biogeochemical Argo floats <xref ref-type="bibr" rid="bib1.bibx19" id="paren.30"/>.</p>
      <p id="d1e970">The actual spatial and temporal coverage for any of these parameters is very low.
Even for <inline-formula><mml:math id="M84" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M85" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, i.e., the parameter with the densest coverage, only about 1.4 % of the global surface ocean has been sampled in any given month over the past 30 years <xref ref-type="bibr" rid="bib1.bibx2" id="paren.31"/>.
Thus, the global-scale reconstruction of the progression of ocean acidification requires a very substantial interpolation and extrapolation effort.
Advances in remote sensing <xref ref-type="bibr" rid="bib1.bibx61" id="paren.32"/>, and the increasing power and usability of machine learning techniques, have permitted us to address this challenge, leading to a proliferation of such efforts.
However, they vary greatly between the different parameters of the marine carbonate system.</p>
      <p id="d1e995">By far the most established efforts are those that interpolate and extrapolate the ocean <inline-formula><mml:math id="M86" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M87" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> observations, as demonstrated by the intercomparison project by <xref ref-type="bibr" rid="bib1.bibx95" id="text.33"/>.
Feed-forward neural networks (FFNNs) have become one of the favored tools <xref ref-type="bibr" rid="bib1.bibx62 bib1.bibx114 bib1.bibx22" id="paren.34"/>, but other statistical and machine learning methods, such as Bayesian regression and tree-based regression, have also been used with similar success <xref ref-type="bibr" rid="bib1.bibx94 bib1.bibx46" id="paren.35"/>.
However, the specific implementation of the methods is what sets the assortment of methods apart.
For example, the SOM-FFN method of <xref ref-type="bibr" rid="bib1.bibx62" id="text.36"/> and the CSIR-ML6 method of <xref ref-type="bibr" rid="bib1.bibx46" id="text.37"/> (among others) first cluster the data based on a certain set of climatological predictors and then perform a regression on <inline-formula><mml:math id="M88" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M89" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> for each resulting cluster.
An alternate approach, used by both the CMEMS-FFNNv2 <xref ref-type="bibr" rid="bib1.bibx22" id="paren.38"/> and NIES-FNN <xref ref-type="bibr" rid="bib1.bibx114" id="paren.39"/> methods, is to include the positional coordinates, without the need for subsetting the data by clustering.
Despite the differences in implementation and regression algorithms, the majority of methods achieve for the open ocean a root mean squared error (RMSE) of roughly 18 <inline-formula><mml:math id="M90" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm when compared with SOCAT <xref ref-type="bibr" rid="bib1.bibx46" id="paren.40"/>.
However, each of these methods has its strengths and weaknesses; for example, the SOM-FFN and CSIR-ML6 methods are able to generalize estimates in data-sparse regions due to information sharing within a cluster, but the methods suffer from discrete boundaries where clusters meet <xref ref-type="bibr" rid="bib1.bibx46" id="paren.41"/>.
These discrete boundaries may introduce artifacts when applied to certain questions.  This is also the case, for example, in the blended open-ocean–coastal-ocean product of <xref ref-type="bibr" rid="bib1.bibx66" id="text.42"/>, where the authors combined the open-ocean estimate of <xref ref-type="bibr" rid="bib1.bibx65" id="text.43"/> with the coastal product of <xref ref-type="bibr" rid="bib1.bibx68" id="text.44"/>.</p>
      <p id="d1e1076">The extrapolation of TA onto a global grid is also well established <xref ref-type="bibr" rid="bib1.bibx48 bib1.bibx80 bib1.bibx72 bib1.bibx102 bib1.bibx39 bib1.bibx17 bib1.bibx6" id="paren.45"/>.
The highly linear relationship between salinity and TA means that linear regressions have been able for quite some time to estimate TA with adequate accuracy.
For example, <xref ref-type="bibr" rid="bib1.bibx48" id="text.46"/> developed a globally applicable multi-linear regression (MLR) model involving salinity and the conservative tracer PO (PO <inline-formula><mml:math id="M91" display="inline"><mml:mo>=</mml:mo></mml:math></inline-formula> O<inline-formula><mml:math id="M92" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> <inline-formula><mml:math id="M93" display="inline"><mml:mo>+</mml:mo></mml:math></inline-formula> 170 <inline-formula><mml:math id="M94" display="inline"><mml:mo>⋅</mml:mo></mml:math></inline-formula> PO<inline-formula><mml:math id="M95" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>, <xref ref-type="bibr" rid="bib1.bibx12" id="altparen.47"/>) and achieved a global RMSE of 11 <inline-formula><mml:math id="M96" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M97" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
<xref ref-type="bibr" rid="bib1.bibx72" id="text.48"/> also used a MLR approach but differentiated it regionally using salinity, temperature, and spatial coordinates as independent variables. The same approach was followed by <xref ref-type="bibr" rid="bib1.bibx104" id="text.49"/>.
More recently, more nuanced and non-linear regression approaches have improved upon the MLR approaches <xref ref-type="bibr" rid="bib1.bibx98 bib1.bibx16 bib1.bibx13 bib1.bibx6" id="paren.50"/>. For example, the locally interpolated alkalinity regression (LIARv2) still makes use of linear regression but interpolates the regression coefficients spatially from a fixed set of trained regression nodes located at every fifth point <xref ref-type="bibr" rid="bib1.bibx15" id="paren.51"/>.
<xref ref-type="bibr" rid="bib1.bibx98" id="text.52"/> used a self-organizing map approach coupled with a local linear optimizer (called SOMLO) and achieved a global RMSE of 9 <inline-formula><mml:math id="M98" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M99" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
A similar RMSE was achieved by <xref ref-type="bibr" rid="bib1.bibx13" id="text.53"/> using a neural network approach (NNGv2). These low RMSE levels were also achieved by these studies avoiding the nearshore and coastal<?pagebreak page780?> environments, where variability in the surface ocean carbonate system is much higher than in the open ocean <xref ref-type="bibr" rid="bib1.bibx68" id="paren.54"/>. In addition to these global regressions, several regionally specific regressions were developed (see Table 1 in <xref ref-type="bibr" rid="bib1.bibx61" id="altparen.55"/>).</p>
      <p id="d1e1195">In comparison, very few efforts attempted to interpolate and extrapolate DIC.
<xref ref-type="bibr" rid="bib1.bibx71" id="text.56"/> were the first to produce a global map of DIC using a regression methodology; however, their application employed a regional multi-linear regression model similar to that used later to map TA.
But their application was limited to the generation of a seasonal climatology.
It was not until <xref ref-type="bibr" rid="bib1.bibx98" id="text.57"/> when the first global reconstruction of the temporal progression of DIC over multiple years was published.
They used the same SOMLO method as they had used for TA, creating global maps of DIC with a RMSE of 11 <inline-formula><mml:math id="M100" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M101" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
More recently, <xref ref-type="bibr" rid="bib1.bibx59" id="text.58"/> used the SOM-FFN method of <xref ref-type="bibr" rid="bib1.bibx62" id="text.59"/> to reconstruct DIC throughout the upper water column on a monthly basis, but they limited their discussion to the mean seasonal cycle.</p>
      <p id="d1e1231">Here, to map TA and <inline-formula><mml:math id="M102" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M103" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> globally, including the coastal ocean, the Arctic, and the Mediterranean, we use a newly developed two-step cluster-regression approach that is similar in design to the SOM-FFN method <xref ref-type="bibr" rid="bib1.bibx62 bib1.bibx65" id="paren.60"/> but extend it by using an ensemble of such cluster regressions.
This method, referred to as Geospatial Random Cluster Ensemble Regression (GRaCER), increases the robustness of the estimates considerably.
It also removes the boundary problems inherent in all methods that use fixed regional boundaries.
We apply the same methodology to TA and <inline-formula><mml:math id="M104" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M105" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, resulting in methodologically consistent global estimates of the two parameters, from which DIC, pH, and <inline-formula><mml:math id="M106" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> can then be computed using the well-established thermodynamic models of the seawater carbonate system.
These latter estimates can then be compared against the many available DIC and pH measurements, providing a large set of independent data to assess the fidelity of our estimates.
This also requires a good understanding of the different sources of uncertainties – including those emanating from sampling and measurement, from the statistical modeling, and from the lack of representativeness, i.e., the fact that a local measurement is not representative for the large pixel (<inline-formula><mml:math id="M107" display="inline"><mml:mrow><mml:mn mathvariant="normal">100</mml:mn><mml:mo>×</mml:mo><mml:mn mathvariant="normal">100</mml:mn></mml:mrow></mml:math></inline-formula> km and 1 month) – that one models in our regressions.</p>
      <p id="d1e1289">The rest of the paper describes the data and methods used to calculate this data set for ocean acidification.
The uncertainties of the predictions are assessed, followed by the presentation of the data with a focus on the seasonal cycle. Last, we discuss the implications of the uncertainties for the use of the derived marine carbonate system.</p>
</sec>
<sec id="Ch1.S2">
  <label>2</label><title>Methods</title>
      <p id="d1e1300">To reconstruct the global progression of all parameters of the surface ocean carbonate system over the last three decades (1985 through 2018), we follow the three steps depicted by the flow diagram in Fig. <xref ref-type="fig" rid="Ch1.F1"/>. First, we develop a statistical model for the measured TA and <inline-formula><mml:math id="M108" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M109" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> using the newly developed GRaCER method. This method itself consists of two steps, i.e., a cluster step, where the target variables are clustered regionally, and a regression step, where for each cluster a regression is evaluated. These two steps are repeated multiple times, creating an ensemble of models; second, we map these two quantities globally and over time using this ensemble of statistical models and global observations of the predictor variables; third and last, we use a thermodynamic model of the seawater carbonate system to compute the remaining parameters of the surface ocean carbonate system, namely DIC, pH, and <inline-formula><mml:math id="M110" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. Along the way, we extensively evaluate and test each step with independent observations. We refer to the data set with the evaluated and complete marine carbonate system as OceanSODA-ETHZ.</p>
      <p id="d1e1328">Next, we describe the concept of the GRaCER method and then detail its implementation for <inline-formula><mml:math id="M111" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M112" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA. This is followed by a description of the numerous types of data employed and how they were prepared. Lastly, we demonstrate how we used a thermodynamic model to derive the remaining parameters of the marine carbonate system.</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F1" specific-use="star"><?xmltex \currentcnt{1}?><?xmltex \def\figurename{Figure}?><label>Figure 1</label><caption><p id="d1e1349">Schematic flow diagram showing the three steps required to reconstruct the surface ocean carbonate system. In the first step (yellow hexagons), the GRaCER (Geospatial Random Cluster Ensemble regression) method is used to develop statistical models for the observed TA (left) and <inline-formula><mml:math id="M113" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M114" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (right) fields. In the second step (orange rectangles), these statistical models are used to extrapolate these two parameters over time and space using ancillary observations, primarily stemming from satellite observations. In the third step (red oval), the interpolated and extrapolated TA and <inline-formula><mml:math id="M115" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M116" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fields are then used to compute the remaining parameters of the surface ocean carbonate system, namely DIC, pH, and the saturation state of seawater with regard to mineral CaCO<inline-formula><mml:math id="M117" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula>, <inline-formula><mml:math id="M118" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. The output of steps two and three is the OceanSODA-ETHZ product. Also shown are the various data sets and data flows used in this study. The different lines indicate whether data are used for training (solid lines), testing (dashed lines), or output with an estimate of uncertainty, where independent test data are shown with gray dashed lines. The gridded/satellite data are summarized in Table <xref ref-type="table" rid="Ch1.T1"/>. Independent test data are shown by the purple box. pyCO2SYS is the software used to solve the marine carbonate system and propagate uncertainties <xref ref-type="bibr" rid="bib1.bibx53" id="paren.61"/>.</p></caption>
        <?xmltex \igopts{width=341.433071pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f01.png"/>

      </fig>

      <?xmltex \floatpos{t}?><fig id="Ch1.F2" specific-use="star"><?xmltex \currentcnt{2}?><?xmltex \def\figurename{Figure}?><label>Figure 2</label><caption><p id="d1e1415">A schematic showing the steps used in the GRaCER method for a single month. Panels <bold>(a)</bold>–<bold>(c)</bold> show a subset of the clusters of a single ensemble member <bold>(h)</bold>, with the adjacent scatter plots <bold>(d)</bold>–<bold>(f)</bold> showing the training data for each cluster and the linear regression models for that cluster (with toy data). Panels <bold>(g)</bold>–<bold>(i)</bold> show the ensemble member estimates for a subset of three members for <inline-formula><mml:math id="M119" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>. Panel <bold>(j)</bold> shows the ensemble mean for all ensemble members, which includes ensemble members not shown in panels <bold>(g)</bold>–<bold>(i)</bold>.</p></caption>
        <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f02.png"/>

      </fig>

<sec id="Ch1.S2.SS1">
  <label>2.1</label><title>GRaCER algorithm</title>
      <p id="d1e1476">The GRaCER algorithm builds conceptually on a series of cluster-regression algorithms that have been successfully used for the interpolation and extrapolation of surface ocean <inline-formula><mml:math id="M120" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M121" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx98 bib1.bibx62 bib1.bibx65 bib1.bibx54 bib1.bibx46" id="paren.62"/>. The main advantage of such a two-step approach is that the first clustering step organizes the variability regionally and temporally. This greatly enhances then the fidelity of the second step, i.e., the regression, as the size of the regression problem is reduced from the global domain to smaller, more homogeneous regions. A second advantage is that this clustering brings together regions with similar seasonality and similar co-variability with potential predictors, irrespective of the number of observations. The regression step explains the variability within each region over time and space dimensions, including interannual variability. Further, the clustering permits the regression to transfer information from spatially distant but geochemically similar regions, making the interpolation and extrapolation more robust in data-poor regions. The main innovation of the GRaCER algorithm relative to the previously used two-step approaches is its use of ensembles of cluster regressions, i.e., the generation of a whole series of clusters and corresponding regressions, which overcomes the boundary problems that are inherent in all two-step approaches.</p>
      <?pagebreak page782?><p id="d1e1498">For the clustering step, we use monthly climatological data of <inline-formula><mml:math id="M122" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M123" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA and related parameters (Fig. <xref ref-type="fig" rid="Ch1.F2"/>a–c) to determine the main patterns of variability of the target variable and its co-variability with potential predictor variables. We opted for a clustering on climatological data rather than on the actual monthly data in order to clearly focus on the clustering step's role to isolate primarily regions with the same seasonal cycle. The alternative approach, i.e., to cluster on the monthly data, is also more prone to errors since the climatological distributions are better known than their month-to-month variations. Finally, clustering on monthly data would also take away signals from the regression step, which is actually better suited to capture the smaller level variations associated with the interannual variability and trends. The mini-batch <inline-formula><mml:math id="M124" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-means implementation in the Python scikit-learn package is used to perform the clustering due to its computational efficiency and scalability with large data sets <xref ref-type="bibr" rid="bib1.bibx89" id="paren.63"/>.
A user-defined number of cluster centers are initiated, where cluster centers represent the mean of the points in a cluster. The <inline-formula><mml:math id="M125" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-means<inline-formula><mml:math id="M126" display="inline"><mml:mrow><mml:mo>+</mml:mo><mml:mo>+</mml:mo></mml:mrow></mml:math></inline-formula> algorithm is used to initiate the cluster centers; it randomly selects the location of the first cluster center and then iteratively selects a best-guess location for the remaining cluster centers. Thereafter, the algorithm minimizes the distance between cluster centers and data points in the variable space.
Once the clusters have been defined for the climatological domain, the co-located training data are assigned to the monthly clusters.</p>
      <p id="d1e1547">The regression is then performed individually for each of the clusters (Fig. <xref ref-type="fig" rid="Ch1.F2"/>d–f). The GRaCER method does not use a prescribed regression method – rather the appropriate algorithm for the particular use case is implemented. Importantly, the algorithm must be able to scale appropriately to the size of the problem. For example, the training data set for TA is <inline-formula><mml:math id="M127" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:mo>/</mml:mo><mml:mn mathvariant="normal">20</mml:mn></mml:mrow></mml:math></inline-formula>th of the size of the <inline-formula><mml:math id="M128" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M129" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> training data set; thus a more computationally expensive method can be used to predict TA.</p>
      <p id="d1e1580">The ensemble members are created by performing the cluster-regression step multiple times. Creating an ensemble is possible due to the fact that each clustering instance is slightly different (Fig. <xref ref-type="fig" rid="Ch1.F2"/>g–i). In practice, the spatial distribution of the clusters is similar; i.e., there is consistency in the typology of the clusters, particularly in regions where clusters are well defined, such as in the subtropical gyres and in the tropical eastern Pacific. However, there are regions that belong to different clusters; i.e., there is slight variance in the typology between ensemble members. The differences are due to the random initialization of the first cluster center in the <inline-formula><mml:math id="M130" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-means clustering step and the fact that clustering variables for some regions have weak gradients in spatial auto-correlation resulting in weak association with a cluster.
In practice, this means that the locations of cluster boundaries vary between ensemble members; thus the ensemble mean does not have discrete boundaries (Fig. <xref ref-type="fig" rid="Ch1.F2"/>j).</p>
</sec>
<sec id="Ch1.S2.SS2">
  <label>2.2</label><title>Algorithm implementation</title>
<sec id="Ch1.S2.SS2.SSS1">
  <label>2.2.1</label><title>Total alkalinity</title>
      <p id="d1e1609">For the estimation of TA, we employ the support vector regression (SVR) method with 12 clusters and 16 ensemble members. The clustering is performed on climatological mean TA, sea-surface salinity (SSS), sea-surface temperature (SST), and nitrate (<inline-formula><mml:math id="M131" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">NO</mml:mi></mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:msubsup></mml:mrow></mml:math></inline-formula>; Table <xref ref-type="table" rid="Ch1.T1"/> and Sect. <xref ref-type="sec" rid="Ch1.S2.SS3"/> below). The optimal variables on which clustering should be performed were selected by assessing the regression scores of each combination of variables following the methodology of <xref ref-type="bibr" rid="bib1.bibx46" id="text.64"/>.
All data are standardized to the mean (<inline-formula><mml:math id="M132" display="inline"><mml:mi mathvariant="italic">μ</mml:mi></mml:math></inline-formula>) and standard deviation (<inline-formula><mml:math id="M133" display="inline"><mml:mi mathvariant="italic">σ</mml:mi></mml:math></inline-formula>) prior to clustering (<inline-formula><mml:math id="M134" display="inline"><mml:mstyle displaystyle="false"><mml:mfrac style="text"><mml:mrow><mml:mo>(</mml:mo><mml:mi>x</mml:mi><mml:mo>-</mml:mo><mml:mi mathvariant="italic">μ</mml:mi><mml:mo>)</mml:mo></mml:mrow><mml:mi mathvariant="italic">σ</mml:mi></mml:mfrac></mml:mstyle></mml:math></inline-formula>), after which TA is given 3 times the weight of the other variables.</p>
      <p id="d1e1668">A similar exhaustive search was used for determining the number of clusters. The number of ensemble members was chosen by the number above which there is no longer an increase in performance, analogous to the number of trees in a random forest.  Test data are a subset of years spaced 3 years apart starting in 1985. We ensure that the models are not overfitted by selecting hyper-parameters using <inline-formula><mml:math id="M135" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-fold cross validation (further details are in Sect. <xref ref-type="sec" rid="App1.Ch1.S1.SS3"/>).</p>
      <?pagebreak page783?><p id="d1e1680">To regress and map TA, we use  SSS, SST, silicic acid (H<inline-formula><mml:math id="M136" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M137" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>), and <inline-formula><mml:math id="M138" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup><mml:mo>=</mml:mo><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">NO</mml:mi></mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:msubsup><mml:mo>-</mml:mo><mml:mn mathvariant="normal">16</mml:mn><mml:mo>⋅</mml:mo><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">PO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx47" id="paren.65"><named-content content-type="pre">nitrogen excess relative to phosphate in terms of the Redfield N : P ratio simplified from</named-content></xref> as predictors.
Our choice of SSS and SST as predictors is easily justified by these two variables accounting for the majority of TA variability <xref ref-type="bibr" rid="bib1.bibx72 bib1.bibx16" id="paren.66"/>.
The addition of H<inline-formula><mml:math id="M139" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M140" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula> and <inline-formula><mml:math id="M141" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> as predictors is to account for seasonal changes in primary production that has an impact on TA <xref ref-type="bibr" rid="bib1.bibx113 bib1.bibx16" id="paren.67"/>.
Further, <inline-formula><mml:math id="M142" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> expresses the zonal differences between and within the large ocean basins better than using simply <inline-formula><mml:math id="M143" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">NO</mml:mi></mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:msubsup></mml:mrow></mml:math></inline-formula> or <inline-formula><mml:math id="M144" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">PO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula> – an important consideration, since coordinates (i.e., latitude and longitude) are not included in our set of predictors.</p>
</sec>
<sec id="Ch1.S2.SS2.SSS2">
  <label>2.2.2</label><?xmltex \opttitle{Partial pressure of CO${}_{2}$}?><title>Partial pressure of CO<inline-formula><mml:math id="M145" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></title>
      <p id="d1e1840">For the estimation of <inline-formula><mml:math id="M146" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, we use two regression methods, i.e., GBDTs (gradient boosted decision trees) and FFNNv2 (feed-forward neural network). These are implemented with 21 clusters and 16 ensemble members (eight each).
The number of clusters is at the upper end of the range compared with the number of clusters used by the MPI-SOMFFN or CSIR-ML6 methods.
However, testing has shown that additional clusters are required to account for the additional complexity by our inclusion of data from the coastal, Arctic, and Mediterranean seas.</p>
      <p id="d1e1857">Clustering is performed on climatological values of <inline-formula><mml:math id="M147" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M148" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, SST, mixed-layer depth (MLD), and chlorophyll <inline-formula><mml:math id="M149" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>, with additional weighting given to <inline-formula><mml:math id="M150" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M151" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>.
As with TA, all variables are standardized prior to clustering with <inline-formula><mml:math id="M152" display="inline"><mml:mstyle displaystyle="false"><mml:mfrac style="text"><mml:mrow><mml:mo>(</mml:mo><mml:mi>x</mml:mi><mml:mo>-</mml:mo><mml:mi mathvariant="italic">μ</mml:mi><mml:mo>)</mml:mo></mml:mrow><mml:mi mathvariant="italic">σ</mml:mi></mml:mfrac></mml:mstyle></mml:math></inline-formula>, after which <inline-formula><mml:math id="M153" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is multiplied by 3 to give it stronger weighting.</p>
      <p id="d1e1934">Details of the regression method and of the hyper-parameter selection are given in Sect. <xref ref-type="sec" rid="App1.Ch1.S1.SS3"/>.
Test data are selected as every fifth year starting in 1985, and validation data for early stopping are selected using the same approach starting in 1987, where the latter are used for early stopping to reduce overfitting and keep model complexity within bounds.</p>
      <p id="d1e1940">The regression and mapping is performed with the following variables as predictors: SST, SSS, the logarithm of chlorophyll <inline-formula><mml:math id="M154" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>, the logarithm of mixed-layer depth, the meridional and zonal components of the surface winds, the sine and cosine of <inline-formula><mml:math id="M155" display="inline"><mml:mstyle displaystyle="false"><mml:mfrac style="text"><mml:mrow><mml:mtext>day of year</mml:mtext><mml:mo>⋅</mml:mo><mml:mi mathvariant="italic">π</mml:mi></mml:mrow><mml:mrow><mml:mn mathvariant="normal">365</mml:mn><mml:mo>⋅</mml:mo><mml:mn mathvariant="normal">180</mml:mn></mml:mrow></mml:mfrac></mml:mstyle></mml:math></inline-formula>, and the atmospheric dry-air mixing ratio (<inline-formula><mml:math id="M156" display="inline"><mml:mi>x</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M157" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>).
These predictors are the same as used by <xref ref-type="bibr" rid="bib1.bibx46" id="text.68"/>, and various combinations of these methods have been used by previous approaches <xref ref-type="bibr" rid="bib1.bibx63 bib1.bibx22" id="paren.69"/>.</p>
      <p id="d1e1994">It is important to note that the predictors are proxies for the spatiotemporal changes in <inline-formula><mml:math id="M158" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M159" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and do not necessarily explain the physical mechanism by which changes in <inline-formula><mml:math id="M160" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M161" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> are driven.
For example, an increase in sea-surface temperature in the subtropics results in an increase in <inline-formula><mml:math id="M162" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M163" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> as shown by <xref ref-type="bibr" rid="bib1.bibx104" id="text.70"/> and <xref ref-type="bibr" rid="bib1.bibx73" id="text.71"/>.
In contrast, surface warming in the Southern Ocean can be a proxy for stratification that reduces outcropping of high-CO<inline-formula><mml:math id="M164" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> waters <xref ref-type="bibr" rid="bib1.bibx64 bib1.bibx45" id="paren.72"/>.
Similarly, changes in SSS and MLD also capture the distribution and processes that drive changes in surface <inline-formula><mml:math id="M165" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, such as stratification and mixing.
However, the climatological MLD product used here does not capture interannual variability in stratification and mixing.
We thus include the two surface wind components that, along with SST, are a proxy for wind-driven mixing and upwelling.
Chlorophyll is also an important driver of <inline-formula><mml:math id="M166" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M167" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> on a local scale, particularly in the high-latitude regions where high primary productivity results in rapid uptake of <inline-formula><mml:math id="M168" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M169" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx1 bib1.bibx45" id="paren.73"/>.
Lastly, <inline-formula><mml:math id="M170" display="inline"><mml:mi>x</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M171" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> is included to account for the close tracking of oceanic <inline-formula><mml:math id="M172" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M173" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> to atmospheric CO<inline-formula><mml:math id="M174" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> concentrations <xref ref-type="bibr" rid="bib1.bibx4" id="paren.74"/>.</p>
</sec>
</sec>
<sec id="Ch1.S2.SS3">
  <label>2.3</label><title>Data</title>
      <p id="d1e2168">Data are used to develop the two-step GRaCER model, i.e., clustering and regression, and to evaluate the estimates. Table <xref ref-type="table" rid="Ch1.T1"/> provides an overview of all data employed and the purposes for which they are used, and Table <xref ref-type="table" rid="Ch1.T2"/> shows the corresponding source of the data. We describe each data set by parameter and use.</p>

<?xmltex \floatpos{t}?><table-wrap id="Ch1.T1" specific-use="star"><?xmltex \currentcnt{1}?><label>Table 1</label><caption><p id="d1e2178">Variables used as the clustering features and predictor variables for regression. Details about these data are given in the text. Note that clustering features are all resampled to monthly climatologies. <inline-formula><mml:math id="M175" display="inline"><mml:msup><mml:mi/><mml:mi mathvariant="normal">a</mml:mi></mml:msup></mml:math></inline-formula> TA<inline-formula><mml:math id="M176" display="inline"><mml:msup><mml:mi/><mml:mi mathvariant="normal">map</mml:mi></mml:msup></mml:math></inline-formula> is the exception where a quasi-annual mean is used. The first column for regression shows the target variable. All machine learning models use the same variables to train and predict the final estimates with the exception of <inline-formula><mml:math id="M177" display="inline"><mml:msup><mml:mi/><mml:mi mathvariant="normal">b</mml:mi></mml:msup></mml:math></inline-formula> SSS, where ungridded GLODAP data are used to train and SODA salinity is used to predict. All other references to SSS refer to the gridded SODA product. </p></caption><oasis:table frame="topbot"><oasis:tgroup cols="2">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="left"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Clustering</oasis:entry>
         <oasis:entry colname="col2">Clustering features (monthly climatology)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">TA</oasis:entry>
         <oasis:entry colname="col2">TA<inline-formula><mml:math id="M178" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:msup><mml:mi mathvariant="normal">map</mml:mi><mml:mi mathvariant="normal">a</mml:mi></mml:msup></mml:mrow></mml:msup></mml:math></inline-formula>, SSS, SST, <inline-formula><mml:math id="M179" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
       <?xmltex \interline{[5.690551pt]}?></oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"><inline-formula><mml:math id="M180" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M181" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M182" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M183" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">map</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula>, SST, Chl <inline-formula><mml:math id="M184" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>, MLD<inline-formula><mml:math id="M185" display="inline"><mml:msub><mml:mi/><mml:mi mathvariant="normal">clim</mml:mi></mml:msub></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Regression and mapping</oasis:entry>
         <oasis:entry colname="col2">Predictors (monthly)</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">TA<inline-formula><mml:math id="M186" display="inline"><mml:msup><mml:mi/><mml:mi mathvariant="normal">GLODAP</mml:mi></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col2">SSS<inline-formula><mml:math id="M187" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mi mathvariant="normal">SODA</mml:mi><mml:mrow><mml:msup><mml:mi mathvariant="normal">GLODAP</mml:mi><mml:mi mathvariant="normal">b</mml:mi></mml:msup></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula>, SST, H<inline-formula><mml:math id="M188" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M189" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>, <inline-formula><mml:math id="M190" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
       <?xmltex \interline{[5.690551pt]}?></oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><inline-formula><mml:math id="M191" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M192" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M193" display="inline"><mml:mi>x</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M194" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">atm</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula>, SST, SSS, Chl <inline-formula><mml:math id="M195" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>, MLD<inline-formula><mml:math id="M196" display="inline"><mml:msub><mml:mi/><mml:mi mathvariant="normal">clim</mml:mi></mml:msub></mml:math></inline-formula>, <inline-formula><mml:math id="M197" display="inline"><mml:mi>u</mml:mi></mml:math></inline-formula> wind, <inline-formula><mml:math id="M198" display="inline"><mml:mi>v</mml:mi></mml:math></inline-formula> wind</oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

<?xmltex \floatpos{t}?><table-wrap id="Ch1.T2" specific-use="star"><?xmltex \currentcnt{2}?><label>Table 2</label><caption><p id="d1e2467">Data sources used in this study.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="4">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="left"/>
     <oasis:colspec colnum="3" colname="col3" align="left"/>
     <oasis:colspec colnum="4" colname="col4" align="left"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Product name</oasis:entry>
         <oasis:entry colname="col2">Variable</oasis:entry>
         <oasis:entry colname="col3">Abbreviation</oasis:entry>
         <oasis:entry colname="col4">Reference</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">SOCAT</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M199" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M200" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M201" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M202" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx2" id="text.75"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">MPI-SOMFFN</oasis:entry>
         <oasis:entry colname="col2">Gap-filled <inline-formula><mml:math id="M203" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M204" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M205" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M206" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">map</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx65" id="text.76"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Jena-MLS</oasis:entry>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx94" id="text.77"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">LSCE-FFNN</oasis:entry>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx22" id="text.78"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">LDEO</oasis:entry>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx106" id="text.79"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">GLODAP</oasis:entry>
         <oasis:entry colname="col2">Total alkalinity (in situ)</oasis:entry>
         <oasis:entry colname="col3">TA</oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx84" id="text.80"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">Total alkalinity (mapped)</oasis:entry>
         <oasis:entry colname="col3">TA<inline-formula><mml:math id="M207" display="inline"><mml:msup><mml:mi/><mml:mi mathvariant="normal">map</mml:mi></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx70" id="text.81"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">OSTIA</oasis:entry>
         <oasis:entry colname="col2">Sea-surface temperature</oasis:entry>
         <oasis:entry colname="col3">SST</oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx41" id="text.82"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">Sea-ice fraction</oasis:entry>
         <oasis:entry colname="col3">ICE</oasis:entry>
         <oasis:entry colname="col4"/>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">SODA v3.4.2</oasis:entry>
         <oasis:entry colname="col2">Salinity</oasis:entry>
         <oasis:entry colname="col3">SSS</oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx17" id="text.83"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">Mixed-layer depth</oasis:entry>
         <oasis:entry colname="col3">MLD</oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx52" id="text.84"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">ERA5</oasis:entry>
         <oasis:entry colname="col2">Sea-level pressure</oasis:entry>
         <oasis:entry colname="col3">Pres</oasis:entry>
         <oasis:entry colname="col4">Hersbach et al. (2018)</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M208" display="inline"><mml:mi>U</mml:mi></mml:math></inline-formula> component of wind</oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M209" display="inline"><mml:mi>U</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"/>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M210" display="inline"><mml:mi>V</mml:mi></mml:math></inline-formula> component of wind</oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M211" display="inline"><mml:mi>V</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"/>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">NOAA: ATM</oasis:entry>
         <oasis:entry colname="col2">Mole fraction of CO<inline-formula><mml:math id="M212" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">atm</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M213" display="inline"><mml:mrow><mml:mi>x</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">atm</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx25" id="text.85"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">GlobColour</oasis:entry>
         <oasis:entry colname="col2">Chlorophyll <inline-formula><mml:math id="M214" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3">Chl <inline-formula><mml:math id="M215" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx76" id="text.86"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">WOA</oasis:entry>
         <oasis:entry colname="col2">Phosphate</oasis:entry>
         <oasis:entry colname="col3">PO<inline-formula><mml:math id="M216" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">
                    <xref ref-type="bibr" rid="bib1.bibx10" id="text.87"/>
                  </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">Nitrate</oasis:entry>
         <oasis:entry colname="col3">NO<inline-formula><mml:math id="M217" display="inline"><mml:mrow><mml:msubsup><mml:mi/><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">Silicic acid</oasis:entry>
         <oasis:entry colname="col3">H<inline-formula><mml:math id="M218" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M219" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"/>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

<sec id="Ch1.S2.SS3.SSS1">
  <label>2.3.1</label><title>Data for clustering</title>
      <p id="d1e2994">For the clustering of TA, we used the mapped product of total alkalinity (TA<inline-formula><mml:math id="M220" display="inline"><mml:msub><mml:mi/><mml:mrow class="chem"><mml:mi mathvariant="normal">map</mml:mi></mml:mrow></mml:msub></mml:math></inline-formula>) from the GLODAPv2 <xref ref-type="bibr" rid="bib1.bibx70" id="paren.88"/>. This product represents a quasi-annual mean as it was generated without consideration of the seasonal cycle.
We repeat this quasi-annual mean TA to create a monthly data set over which clustering can be performed.
We thus assume that the spatial variability of TA is larger than the seasonal variability.
This is backed by <xref ref-type="bibr" rid="bib1.bibx102" id="text.89"/> and <xref ref-type="bibr" rid="bib1.bibx13" id="text.90"/>, who found that the seasonal variability of TA for the majority of the ocean was more than a factor of 10 smaller than the spatial variability.</p>
      <p id="d1e3016">For the clustering step of <inline-formula><mml:math id="M221" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, we use four data-based products resampled and gridded to a monthly by <inline-formula><mml:math id="M222" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> resolution (<inline-formula><mml:math id="M223" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mrow class="chem"><mml:mi mathvariant="normal">map</mml:mi></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula>), namely LDEO by <xref ref-type="bibr" rid="bib1.bibx106" id="text.91"/>, MPI-SOMFFN by <xref ref-type="bibr" rid="bib1.bibx65" id="text.92"/>, Jena-MLS by <xref ref-type="bibr" rid="bib1.bibx94" id="text.93"/>, and CMEMS-FFNNv2 by <xref ref-type="bibr" rid="bib1.bibx22" id="text.94"/>.
It may seem tautological to use other machine learning estimates, but these data are just used to create regional clusters; i.e., they are not used in the regression step.
Relative to previous two-step approaches <xref ref-type="bibr" rid="bib1.bibx65 bib1.bibx22" id="paren.95"/>, which used just the LDEO product, we expanded on this by including three more estimates.
In doing so, we make the implicit assumption that this ensemble of estimates is a better representation of the <inline-formula><mml:math id="M224" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> monthly climatology than the LDEO climatology alone.</p>
      <p id="d1e3100">SSS is from the Simple Ocean Data Assimilation (SODA) analysis <xref ref-type="bibr" rid="bib1.bibx17" id="paren.96"/> and SST from the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) product <xref ref-type="bibr" rid="bib1.bibx40 bib1.bibx41" id="paren.97"/>.
<inline-formula><mml:math id="M225" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> is calculated using monthly climatologies of <inline-formula><mml:math id="M226" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">NO</mml:mi></mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:msubsup></mml:mrow></mml:math></inline-formula> and <inline-formula><mml:math id="M227" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">PO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula> from the World Ocean Atlas updated in 2018 <xref ref-type="bibr" rid="bib1.bibx10" id="paren.98"/>.
We use the monthly climatology of density-based mixed-layer depth (MLD) from <xref ref-type="bibr" rid="bib1.bibx52" id="text.99"/> that is estimated from Argo float profiles.
The MLD data product merges monthly estimates of MLD from multiple years but is averaged into a climatology due to the paucity of data on an annual scale.
A two-dimensional moving average filter is applied to the MLD to interpolate missing data and remove the noise introduced by interannual and sub-monthly variability.</p>
      <?pagebreak page784?><p id="d1e3158"><?xmltex \hack{\newpage}?>Chlorophyll <inline-formula><mml:math id="M228" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> (Chl <inline-formula><mml:math id="M229" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>) is from the GlobColour project where a monthly climatology is calculated for the period from 1998 through 2018 <xref ref-type="bibr" rid="bib1.bibx76" id="paren.100"/>.
The missing data in the high latitudes during winter are filled with a 0.3 mg m<inline-formula><mml:math id="M230" display="inline"><mml:msup><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msup></mml:math></inline-formula>, which is roughly the 20th percentile of global chlorophyll <inline-formula><mml:math id="M231" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula>.
Lastly, we take the log transformation (base 10) of Chl <inline-formula><mml:math id="M232" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> to convert the log-like distribution of Chl <inline-formula><mml:math id="M233" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> to a normal distribution for improved performance in the gradient descent algorithm of the FFNNv2.</p>
</sec>
<sec id="Ch1.S2.SS3.SSS2">
  <label>2.3.2</label><title>Data for regression and mapping</title>
      <?pagebreak page785?><p id="d1e3218">For the regression step of TA, the bottle measurements from the GLODAP v2 product are used as the target variable <xref ref-type="bibr" rid="bib1.bibx85" id="paren.101"/>.
Following <xref ref-type="bibr" rid="bib1.bibx72" id="text.102"/>, we select data shallower than 20 m in latitudes lower than 30<inline-formula><mml:math id="M234" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> and shallower than 30 m at higher latitudes. We do not exclude measurements taken in nearshore and coastal environments as was previously done.
The quality of TA measurements was historically not as rigorous as the SOCAT <inline-formula><mml:math id="M235" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> data due to the lack of reference standards before the mid-1990s <xref ref-type="bibr" rid="bib1.bibx7" id="paren.103"/>.
However, most of the biases in the cruises were corrected based on calibration to deep samples, where it is assumed that interannual TA variability is negligible relative to the magnitude of the bias <xref ref-type="bibr" rid="bib1.bibx85" id="paren.104"/>.
These bias corrections amount to <inline-formula><mml:math id="M236" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 5 <inline-formula><mml:math id="M237" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M238" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> on average.</p>
      <p id="d1e3284">For the regression of <inline-formula><mml:math id="M239" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, we use SOCAT v2019 where only data with a SOCAT cruise quality flag of A to D and a World Ocean Circulation Experiment (WOCE) quality flag of 2 are used. As is the case for TA, we do not exclude data from coastal and nearshore environments.
The fugacity of <inline-formula><mml:math id="M240" display="inline"><mml:mrow><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (<inline-formula><mml:math id="M241" display="inline"><mml:mrow><mml:mi>f</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>) reported in SOCAT v2019 is converted to <inline-formula><mml:math id="M242" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> using
              <disp-formula id="Ch1.E1" content-type="numbered"><label>1</label><mml:math id="M243" display="block"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub><mml:mo>=</mml:mo><mml:mi>f</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub><mml:mo>⋅</mml:mo><mml:mi>exp⁡</mml:mi><mml:mfenced close=")" open="("><mml:mrow><mml:msubsup><mml:mi>P</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow><mml:mi mathvariant="normal">surf</mml:mi></mml:msubsup><mml:mo>⋅</mml:mo><mml:mstyle displaystyle="true"><mml:mfrac style="display"><mml:mrow><mml:mi>B</mml:mi><mml:mo>+</mml:mo><mml:mn mathvariant="normal">2</mml:mn><mml:mo>⋅</mml:mo><mml:mi mathvariant="italic">δ</mml:mi></mml:mrow><mml:mrow><mml:mi>R</mml:mi><mml:mo>⋅</mml:mo><mml:msup><mml:mi>T</mml:mi><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msup></mml:mrow></mml:mfrac></mml:mstyle></mml:mrow></mml:mfenced><mml:mo>,</mml:mo></mml:mrow></mml:math></disp-formula>
            where <inline-formula><mml:math id="M244" display="inline"><mml:mi>P</mml:mi></mml:math></inline-formula> is atmospheric pressure at sea level from the ERA5 reanalysis product <xref ref-type="bibr" rid="bib1.bibx21" id="paren.105"/>.
<inline-formula><mml:math id="M245" display="inline"><mml:mi>B</mml:mi></mml:math></inline-formula> and <inline-formula><mml:math id="M246" display="inline"><mml:mi mathvariant="italic">δ</mml:mi></mml:math></inline-formula> are virial coefficients, <inline-formula><mml:math id="M247" display="inline"><mml:mi>R</mml:mi></mml:math></inline-formula> is the gas constant, and <inline-formula><mml:math id="M248" display="inline"><mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msup></mml:mrow></mml:math></inline-formula> is the ship intake temperature in <inline-formula><mml:math id="M249" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula>C <xref ref-type="bibr" rid="bib1.bibx24" id="paren.106"/>.
In exploratory work for this study, we tested predicting <inline-formula><mml:math id="M250" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub><mml:mo>=</mml:mo><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">atm</mml:mi></mml:msubsup><mml:mo>-</mml:mo><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> instead of just <inline-formula><mml:math id="M251" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> but found that this did not produce credible results; for a more in-depth discussion see Sect. <xref ref-type="sec" rid="App1.Ch1.S1.SS2"/>.</p>
      <p id="d1e3516">The discrete measurements of <inline-formula><mml:math id="M252" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA are resampled on a monthly grid (January 1985 through December 2018) with a spatial resolution of <inline-formula><mml:math id="M253" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> to match the predictors used in the mapping step.</p>
      <p id="d1e3553">Finally, outliers are removed from gridded <inline-formula><mml:math id="M254" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula> using the methods described in Sect. <xref ref-type="sec" rid="App1.Ch1.S1.SS1"/>.
In total, 2425 points are removed from the gridded <inline-formula><mml:math id="M255" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula> using these outlier removal approaches, equivalent to 0.85 % of the original gridded data.</p>
      <p id="d1e3591">We use sea-surface temperature from OSTIA for both TA and <inline-formula><mml:math id="M256" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> regression <xref ref-type="bibr" rid="bib1.bibx41" id="paren.107"/>.
The TA model is trained using in situ salinity from GLODAP v2, but salinity from SODA v4.3.2 is used for the mapping step <xref ref-type="bibr" rid="bib1.bibx85 bib1.bibx17" id="paren.108"/>.
The <inline-formula><mml:math id="M257" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mo>*</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> and H<inline-formula><mml:math id="M258" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M259" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula> are the same as used in the clustering step but are repeated for the number of years.
Similarly, the mixed-layer depth climatology described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS3.SSS1"/> is repeated for each year.
We use the global mean of the mole fraction of CO<inline-formula><mml:math id="M260" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> for the marine boundary layer (<inline-formula><mml:math id="M261" display="inline"><mml:mrow><mml:mi>x</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">mbl</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula>) as a predictor in the regression, as the correction for water vapor pressure may otherwise introduce co-variance with other predictors (i.e., SST, SSS).
Missing data in the monthly GlobColour chlorophyll <inline-formula><mml:math id="M262" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> product are filled with climatological data described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS3.SSS1"/>.
The meridional and zonal components of the surface winds are averaged from the hourly output from the ERA5 reanalysis <xref ref-type="bibr" rid="bib1.bibx21" id="paren.109"/>.</p>
</sec>
<sec id="Ch1.S2.SS3.SSS3">
  <label>2.3.3</label><title>Evaluation variables</title>
      <p id="d1e3692">The machine learning estimates of TA, <inline-formula><mml:math id="M263" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, and the computed DIC and pH are evaluated against data that are not used in the training or mapping step.</p>
      <p id="d1e3709">For DIC and pH, we use the directly measured data from GLODAP v2.2019 <xref ref-type="bibr" rid="bib1.bibx85" id="paren.110"/>.
<xref ref-type="bibr" rid="bib1.bibx7" id="text.111"/> report a measurement error of <inline-formula><mml:math id="M264" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 5 <inline-formula><mml:math id="M265" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi><mml:mi mathvariant="normal">mol</mml:mi></mml:mrow></mml:math></inline-formula> kg<inline-formula><mml:math id="M266" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for GLODAP DIC in an inter-laboratory comparison.
<xref ref-type="bibr" rid="bib1.bibx85" id="text.112"/> estimate the measurement error for pH to be 0.01.
To be consistent with TA, we select data shallower than 20 m in latitudes lower than 30<inline-formula><mml:math id="M267" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> and shallower than 30 m at higher latitudes and resampled the data to monthly by <inline-formula><mml:math id="M268" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula>.</p>
      <p id="d1e3780">Three long-term time series stations are used to provide direct independent comparisons for DIC and TA, namely the Hawaii Ocean Time-series at 22.57<inline-formula><mml:math id="M269" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 158<inline-formula><mml:math id="M270" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W (HOT, <xref ref-type="bibr" rid="bib1.bibx28" id="altparen.113"/>); the Bermuda Atlantic Time-series Study  at 32<inline-formula><mml:math id="M271" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 64<inline-formula><mml:math id="M272" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W (BATS, <xref ref-type="bibr" rid="bib1.bibx3" id="altparen.114"/>); and the Irminger station in the high northern Atlantic (64.3<inline-formula><mml:math id="M273" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 28<inline-formula><mml:math id="M274" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W, <xref ref-type="bibr" rid="bib1.bibx83" id="altparen.115"/>, only for DIC).
The accuracy for these measurements is reported to be below 2 <inline-formula><mml:math id="M275" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M276" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for DIC and <inline-formula><mml:math id="M277" display="inline"><mml:mo>∼</mml:mo></mml:math></inline-formula> 4 <inline-formula><mml:math id="M278" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M279" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for TA for all stations.
We use the same depth constraints for the long-term stations as for GLODAP, explained in the paragraph above. <inline-formula><mml:math id="M280" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is also calculated from DIC and TA for HOT and BATS to provide an additional constraint.</p>
      <p id="d1e3909">Data present in the Lamont-Doherty Earth Observatory <inline-formula><mml:math id="M281" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> data set but not in SOCAT are used to independently compare <inline-formula><mml:math id="M282" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx103" id="paren.116"/>.
<xref ref-type="bibr" rid="bib1.bibx103" id="text.117"/> report an error estimate of <inline-formula><mml:math id="M283" display="inline"><mml:mrow><mml:mo>±</mml:mo><mml:mn mathvariant="normal">2.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow></mml:mrow></mml:math></inline-formula>, but it must be added that some of the data unique to LDEO may be excluded from SOCAT due to stricter quality control criteria for the latter; thus errors for the LDEO data are expected to be larger <xref ref-type="bibr" rid="bib1.bibx2" id="paren.118"/>.</p>
      <p id="d1e3967">Finally, we include Argo float measurements of pH from the Southern Ocean Carbon and Climate Observations and Modeling project (SOCCOM) <xref ref-type="bibr" rid="bib1.bibx57 bib1.bibx112" id="paren.119"/>.
<xref ref-type="bibr" rid="bib1.bibx56" id="text.120"/> report a mean uncertainty of <inline-formula><mml:math id="M284" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.019 for pH for the entire water column, though this is likely higher for the upper 30 m as the authors report lower errors for estimates below 50 m.</p>
</sec>
</sec>
<sec id="Ch1.S2.SS4">
  <label>2.4</label><?xmltex \opttitle{Computation of DIC, pH, and $\Omega$}?><title>Computation of DIC, pH, and <inline-formula><mml:math id="M285" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula></title>
      <p id="d1e3999">The remaining parameters of the marine carbonate system, i.e., DIC, pH, and <inline-formula><mml:math id="M286" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>, are computed using the Python version of CO2SYS <xref ref-type="bibr" rid="bib1.bibx53" id="paren.121"/> originally developed by <xref ref-type="bibr" rid="bib1.bibx74" id="text.122"/>.
In addition to <inline-formula><mml:math id="M287" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA, CO2SYS requires the input of sea-surface temperature, sea-surface salinity, pressure (assumed 0 dbar at the surface), <inline-formula><mml:math id="M288" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">PO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula>, and H<inline-formula><mml:math id="M289" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M290" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>.
We use the same data sources described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS3.SSS2"/> and Table <xref ref-type="table" rid="Ch1.T1"/>.
Climatologies of H<inline-formula><mml:math id="M291" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula>SiO<inline-formula><mml:math id="M292" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:math></inline-formula> and <inline-formula><mml:math id="M293" display="inline"><mml:mrow><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">PO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:mrow><mml:mn mathvariant="normal">3</mml:mn><mml:mo>-</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula> are repeated for each year, thus assuming no interannual variability. The impact of this assumption is minimal (<inline-formula><mml:math id="M294" display="inline"><mml:mo lspace="0mm">≪</mml:mo></mml:math></inline-formula> 2 <inline-formula><mml:math id="M295" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M296" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>), as can be shown by varying these nutrients over seasonal cycle.
The dissociation constants by <xref ref-type="bibr" rid="bib1.bibx23" id="text.123"/> for <inline-formula><mml:math id="M297" display="inline"><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mrow><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">HSO</mml:mi></mml:mrow><mml:mn mathvariant="normal">4</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula> and the total boron–salinity relationship by <xref ref-type="bibr" rid="bib1.bibx109" id="text.124"/> were used, as recommended by <xref ref-type="bibr" rid="bib1.bibx87" id="text.125"/> and <xref ref-type="bibr" rid="bib1.bibx92" id="text.126"/>.
For further details<?pagebreak page786?> on the calculation and the full description of the marine carbonate system, see <xref ref-type="bibr" rid="bib1.bibx24" id="text.127"/>.</p>
      <p id="d1e4164">An important consideration in these calculations is the internal consistency of the marine carbonate system, i.e., the error due to uncertainties in the equations and coefficients that describe the marine carbonate system. <xref ref-type="bibr" rid="bib1.bibx92" id="text.128"/> pointed out that the <inline-formula><mml:math id="M298" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>–TA pair has the lowest error in the calculation of pH (0.003 <inline-formula><mml:math id="M299" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.008 pH units) using the dissociation constants by <xref ref-type="bibr" rid="bib1.bibx79" id="text.129"/> as refitted by Dickson and Millero (1987). However, using the same pair and the same dissociation constants resulted in an estimate of <inline-formula><mml:math id="M300" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> with respect to aragonite that is very different from that computed using the DIC–TA pair. But since <xref ref-type="bibr" rid="bib1.bibx92" id="text.130"/> lacked direct measurements of <inline-formula><mml:math id="M301" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>, it remains unclear which pair is actually better for <inline-formula><mml:math id="M302" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. We cannot resolve this here but need to acknowledge that this inconsistency adds some additional uncertainty to our computed <inline-formula><mml:math id="M303" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> values.</p>
</sec>
</sec>
<sec id="Ch1.S3">
  <label>3</label><title>Uncertainty assessment</title>
      <p id="d1e4235">Any application of our data product requires a firm understanding of the errors and uncertainties associated with each of the reported parameters of the surface ocean carbonate system. We first discuss the errors and uncertainties associated with the statistically modeled quantities TA and <inline-formula><mml:math id="M304" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and then those with the computed parameters DIC, pH, and <inline-formula><mml:math id="M305" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. Then, we will compare these propagated uncertainties with the uncertainty of the computed DIC and pH by comparing these values with in situ observations. This provides a strong check on our ability to establish a full uncertainty budget. Here, we use the term “uncertainty” to characterize the range of values within which the true value is asserted to lie with some level of confidence. The term “error” is used in two ways: first, as a process that leads to deviations between the measurement and the true value and, second, as an estimate quantifiable against a known value. For the purpose of our analysis here, we consider the training data sets for TA and <inline-formula><mml:math id="M306" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M307" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> as such known values; i.e., we assume that these observations are unbiased. This can be justified on the basis of their having undergone extensive secondary quality control.</p>
<sec id="Ch1.S3.SS1">
  <label>3.1</label><?xmltex \opttitle{Sources of errors for TA and $p$CO${}_{2}$}?><title>Sources of errors for TA and <inline-formula><mml:math id="M308" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M309" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></title>
      <p id="d1e4298">We identify three sources of errors that contribute to the total uncertainty for <inline-formula><mml:math id="M310" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M311" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA, namely the uncertainty stemming from the measurement (<inline-formula><mml:math id="M312" display="inline"><mml:mi>M</mml:mi></mml:math></inline-formula>), representation (<inline-formula><mml:math id="M313" display="inline"><mml:mi>R</mml:mi></mml:math></inline-formula>), and prediction (<inline-formula><mml:math id="M314" display="inline"><mml:mi>P</mml:mi></mml:math></inline-formula>) errors. Assuming independence of the three error sources, the total uncertainty (<inline-formula><mml:math id="M315" display="inline"><mml:mi>E</mml:mi></mml:math></inline-formula>) for the TA and <inline-formula><mml:math id="M316" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates can thus be expressed as the root of the squared sum of the uncertainties from the three error sources:
            <disp-formula id="Ch1.E2" content-type="numbered"><label>2</label><mml:math id="M317" display="block"><mml:mrow><mml:mi>E</mml:mi><mml:mo>=</mml:mo><mml:msqrt><mml:mrow><mml:msup><mml:mi>M</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:msup><mml:mi>R</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:msup><mml:mi>P</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow></mml:msqrt><mml:mo>.</mml:mo></mml:mrow></mml:math></disp-formula></p>
      <p id="d1e4392"><?xmltex \hack{\newpage}?>The measurement error reflects the combination of potential biases (systematic errors) from sampling and measurement as well as random errors associated with sampling and the imprecise nature of the measurement system. Since both TA and <inline-formula><mml:math id="M318" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M319" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> are being measured against certified reference materials and have undergone extensive secondary quality control, we assume that they have no systematic error, i.e., that their bias is zero. We also assume the sampling error to be small, so that the uncertainty <inline-formula><mml:math id="M320" display="inline"><mml:mi>M</mml:mi></mml:math></inline-formula> associated with the measurement error can be well approximated by the precision of the employed measurement methodology <xref ref-type="bibr" rid="bib1.bibx24" id="paren.131"/>.</p>
      <p id="d1e4422">The representation error, <inline-formula><mml:math id="M321" display="inline"><mml:mi>R</mml:mi></mml:math></inline-formula>, is a result of the fact that we develop our statistical model (GRaCER) on a grid that is in many places coarser in time and space than the typical scales of variability of TA and <inline-formula><mml:math id="M322" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M323" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>. As a result, any given observation may not be representative for the <inline-formula><mml:math id="M324" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> monthly grid cell used as a basis for our regression, leading to a bias in the estimated mean relative to the true spatial and temporal mean. This problem is particularly severe if the number of samples within any grid cell is low and the spatiotemporal variability is high. This is often the case. For example, more than 90 % of the data in the monthly gridded <inline-formula><mml:math id="M325" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M326" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> product are based on a single day of sampling within the month. The situation is more dire for TA, for which there are 10-fold fewer observations than <inline-formula><mml:math id="M327" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M328" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>. Since we are lacking full knowledge of the spatial and temporal variability of TA and <inline-formula><mml:math id="M329" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M330" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, we cannot fully quantify the representation error. Instead, we approximate it using the few regions where we have sufficient observations or then using closely related parameters for which we have more observations. For simplicity, we make the assumption that the uncertainty associated with this error is, on a global average, normally distributed with a bias of 0.</p>
      <p id="d1e4518">The uncertainty associated with the prediction error, <inline-formula><mml:math id="M331" display="inline"><mml:mi>P</mml:mi></mml:math></inline-formula>, is determined by the test scores from the evaluation of the statistical model vis-à-vis the independent test data. The test scores describe the error incurred in the prediction of the subset of data that are not used in the training step. This error includes also the propagated uncertainty associated with the predictor variables.</p>
      <p id="d1e4528">We summarize these uncertainties with mean biases (<inline-formula><mml:math id="M332" display="inline"><mml:mstyle displaystyle="false"><mml:mfrac style="text"><mml:mrow><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mo>(</mml:mo><mml:msub><mml:mover accent="true"><mml:mi>y</mml:mi><mml:mo mathvariant="normal" stretchy="false">^</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>-</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>)</mml:mo></mml:mrow><mml:mi>N</mml:mi></mml:mfrac></mml:mstyle></mml:math></inline-formula>) and root mean squared error (RMSE, <inline-formula><mml:math id="M333" display="inline"><mml:msqrt><mml:mstyle displaystyle="false"><mml:mfrac style="text"><mml:mrow><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mo>(</mml:mo><mml:msub><mml:mover accent="true"><mml:mi>y</mml:mi><mml:mo mathvariant="normal" stretchy="false">^</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>-</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msup><mml:mo>)</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow><mml:mi>N</mml:mi></mml:mfrac></mml:mstyle></mml:msqrt></mml:math></inline-formula>), where <inline-formula><mml:math id="M334" display="inline"><mml:mi>y</mml:mi></mml:math></inline-formula> is the target value, <inline-formula><mml:math id="M335" display="inline"><mml:mover accent="true"><mml:mi>y</mml:mi><mml:mo mathvariant="normal" stretchy="false">^</mml:mo></mml:mover></mml:math></inline-formula> is the predicted value, and <inline-formula><mml:math id="M336" display="inline"><mml:mi>N</mml:mi></mml:math></inline-formula> is the number of observations. We separate the coastal and open-ocean regions using the COastal Segmentation and related CATchments (COSCATs) mask <xref ref-type="bibr" rid="bib1.bibx67" id="paren.132"/> in order to reflect their very different levels of spatiotemporal variability.</p>

<?xmltex \floatpos{t}?><table-wrap id="Ch1.T3" specific-use="star"><?xmltex \currentcnt{3}?><label>Table 3</label><caption><p id="d1e4648">Summary of the uncertainties of total alkalinity and <inline-formula><mml:math id="M337" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> from the different error sources (see Table <xref ref-type="disp-formula" rid="Ch1.E2"/>) separately evaluated for the open ocean and for coastal regions (defined by the COSCATs regions, <xref ref-type="bibr" rid="bib1.bibx67" id="altparen.133"/>). See text for details on how the different sources were quantified.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="5">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="right" colsep="1"/>
     <oasis:colspec colnum="4" colname="col4" align="right"/>
     <oasis:colspec colnum="5" colname="col5" align="right"/>
     <oasis:thead>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry rowsep="1" namest="col2" nameend="col3" align="center" colsep="1">Alkalinity (<inline-formula><mml:math id="M338" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M339" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) </oasis:entry>
         <oasis:entry rowsep="1" namest="col4" nameend="col5" align="center"><inline-formula><mml:math id="M340" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M341" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M342" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm) </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Uncertainty</oasis:entry>
         <oasis:entry colname="col2">Open</oasis:entry>
         <oasis:entry colname="col3">Coastal</oasis:entry>
         <oasis:entry colname="col4">Open</oasis:entry>
         <oasis:entry colname="col5">Coastal</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">ocean</oasis:entry>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4">ocean</oasis:entry>
         <oasis:entry colname="col5"/>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">Measurement</oasis:entry>
         <oasis:entry namest="col2" nameend="col3" align="center">5 (<inline-formula><mml:math id="M343" display="inline"><mml:mo lspace="0mm">≤</mml:mo></mml:math></inline-formula> 10) </oasis:entry>
         <oasis:entry namest="col4" nameend="col5" align="center">2 (<inline-formula><mml:math id="M344" display="inline"><mml:mo lspace="0mm">≤</mml:mo></mml:math></inline-formula> 5) </oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Representation</oasis:entry>
         <oasis:entry colname="col2">16</oasis:entry>
         <oasis:entry colname="col3">34</oasis:entry>
         <oasis:entry colname="col4">7</oasis:entry>
         <oasis:entry colname="col5">17</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Prediction</oasis:entry>
         <oasis:entry colname="col2">13</oasis:entry>
         <oasis:entry colname="col3">28</oasis:entry>
         <oasis:entry colname="col4">12</oasis:entry>
         <oasis:entry colname="col5">27</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Total</oasis:entry>
         <oasis:entry colname="col2">21</oasis:entry>
         <oasis:entry colname="col3">45</oasis:entry>
         <oasis:entry colname="col4">14</oasis:entry>
         <oasis:entry colname="col5">32</oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

<sec id="Ch1.S3.SS1.SSS1">
  <label>3.1.1</label><title>Uncertainty for total alkalinity</title>
      <?pagebreak page787?><p id="d1e4866">We adopt an uncertainty <inline-formula><mml:math id="M345" display="inline"><mml:mi>M</mml:mi></mml:math></inline-formula> associated with the measurement error of TA of <inline-formula><mml:math id="M346" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 5 <inline-formula><mml:math id="M347" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M348" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> based on the laboratory intercomparison by <xref ref-type="bibr" rid="bib1.bibx7" id="text.134"/>.
This is only half the accuracy of <inline-formula><mml:math id="M349" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 10 <inline-formula><mml:math id="M350" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M351" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> or 0.5 % reported by GLODAPv2 <xref ref-type="bibr" rid="bib1.bibx85" id="paren.135"/>.
We consider this to be an overly conservative estimate, since <xref ref-type="bibr" rid="bib1.bibx7" id="text.136"/> pointed out that the majority of the laboratories involved in the round-robin exercise achieved an accuracy of better than <inline-formula><mml:math id="M352" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 5 <inline-formula><mml:math id="M353" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M354" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
We thus opted for this lower value that is more representative of the majority of the data.</p>
      <p id="d1e4968">Owing to the sparseness of the TA observations, we cannot estimate the uncertainty <inline-formula><mml:math id="M355" display="inline"><mml:mi>R</mml:mi></mml:math></inline-formula> associated with the representation error directly.
Instead, we use the high correlation between TA and salinity.
This permits us to determine the representation error for TA indirectly from an estimate of the representation error of sea-surface salinity.
Concretely, we compare the test RMSE of TA predicted with GLODAPv2's in situ salinity with the RMSE of TA predicted with the satellite-based SODA salinity (see Table <xref ref-type="table" rid="Ch1.T1"/>).
Since the latter salinity is supposed to reflect the true time–space average over each grid cell, the difference between these two salinities is a direct estimate of the uncertainty associated with the representation error for salinity.
Consequently, the difference in TA from these two estimates is an estimate of the uncertainty associated with the representation error for TA.
The resulting estimates for the open and coastal ocean are summarized in Table <xref ref-type="table" rid="Ch1.T3"/>.</p>
      <p id="d1e4982">The uncertainty <inline-formula><mml:math id="M356" display="inline"><mml:mi>P</mml:mi></mml:math></inline-formula> associated with the prediction error is based on the model's RMSE score calculated from test data and is listed in Table <xref ref-type="table" rid="Ch1.T3"/> and Fig. <xref ref-type="fig" rid="Ch1.F3"/>a.
The global mean prediction error for the open ocean amounts to 13 <inline-formula><mml:math id="M357" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M358" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, with some regional differences. The prediction error is more than twice this number in the coastal regions, i.e., 28 <inline-formula><mml:math id="M359" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M360" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> (coastal regions are defined by the COSCATs regions, <xref ref-type="bibr" rid="bib1.bibx67" id="altparen.137"/>).
We find especially high prediction errors, for example, in the highly dynamic Amazon outflow region or the Gulf of Maine in the northwestern Atlantic.
However, in such regions, one can expect that part of the high prediction error is actually stemming from a representation error, as we are not using directly co-measured variables when we train our regression model.
In Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>, we show the spatially and climatologically mapped test errors for TA in Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/> using the GRaCER approach.</p>
      <p id="d1e5044">While the global bias of the TA product of OceanSODA-ETHZ is near zero (0.5 <inline-formula><mml:math id="M361" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M362" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>), confirming our assumption about the unbiased nature of our prediction error, this is not the case regionally. For example, OceanSODA-ETHZ tends to consistently overestimate TA in the southeastern Atlantic and underestimate TA in the southern Indian Ocean (Fig. <xref ref-type="fig" rid="Ch1.F3"/>b). A seasonal breakdown of the biases into DJF and JJA reveals that the winter period of each hemisphere has biases in the high latitudes, though the paucity of data means that we can place less weight on this finding.</p>

      <?xmltex \floatpos{p}?><fig id="Ch1.F3" specific-use="star"><?xmltex \currentcnt{3}?><?xmltex \def\figurename{Figure}?><label>Figure 3</label><caption><p id="d1e5072">Test metrics for total alkalinity <bold>(a, b)</bold> and <inline-formula><mml:math id="M363" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> <bold>(c, d)</bold>. The left-hand column <bold>(a, d)</bold> shows the root mean squared error (RMSE) compared with the target data. Similarly, the middle column <bold>(b, e)</bold> shows bias compared to the respective training data sets (GLODAP v2.2019 and SOCAT v2019). The right-hand column shows the zonally averaged biases for June, July, and August (JJA, blue) and December, January, and February (DJF, orange). A 2D spatial convolution was first applied to the <inline-formula><mml:math id="M364" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> pixels to make regional patterns in the biases and RMSE clearer, and data were then aggregated into <inline-formula><mml:math id="M365" display="inline"><mml:mrow><mml:mn mathvariant="normal">4</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">4</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> pixels for clearer visualization.</p></caption>
            <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f03.png"/>

          </fig>

      <p id="d1e5148">A good check on the model prediction error is provided by comparing the estimated TA against  independent observations.
To this end, we use data from the Hawaii Ocean Time-series (HOT), the Bermuda Atlantic Time-series Study (BATS), and the Irminger station shown in Fig. <xref ref-type="fig" rid="Ch1.F4"/>a, b, e, f, i, and j and Table <xref ref-type="table" rid="Ch1.T4"/>.
For the period 1990–2018, the bias for BATS is 3 <inline-formula><mml:math id="M366" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M367" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> and for HOT <inline-formula><mml:math id="M368" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>2 <inline-formula><mml:math id="M369" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M370" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, indicating that the method captures the overall structure and variability of TA well at these subtropical stations.
Further, the mean seasonal cycle is relatively well represented at HOT and BATS, being within 1 standard deviation of the interannual variability when averaged as a climatology (Fig. <xref ref-type="fig" rid="Ch1.F4"/>b, f).
However, the results are not as good for the Irminger station in the Atlantic high latitudes (<inline-formula><mml:math id="M371" display="inline"><mml:mo lspace="0mm">∼</mml:mo></mml:math></inline-formula> 65<inline-formula><mml:math id="M372" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N), where OceanSODA-ETHZ has a large negative bias (<inline-formula><mml:math id="M373" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>10 <inline-formula><mml:math id="M374" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M375" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) when compared to TA computed from the observed <inline-formula><mml:math id="M376" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and DIC.
OceanSODA-ETHZ also overestimates the weak seasonal cycle of TA at the Irminger station, contributing to the large bias that is particularly strong from December to May.
The RMSE at Irminger station is 15 <inline-formula><mml:math id="M377" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M378" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>, less than 5 <inline-formula><mml:math id="M379" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M380" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> larger than the RMSE for HOT and BATS stations (10 <inline-formula><mml:math id="M381" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M382" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> respectively) owing to the small interannual and seasonal amplitude at Irminger.
The RMSEs are thus smaller than the mean prediction error (13 <inline-formula><mml:math id="M383" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M384" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) at the subtropical stations yet exceed this mean estimate at the high-latitude station.</p>

      <?xmltex \floatpos{p}?><fig id="Ch1.F4" specific-use="star"><?xmltex \currentcnt{4}?><?xmltex \def\figurename{Figure}?><label>Figure 4</label><caption><p id="d1e5346">A comparison of a subset of measurements from long-term observation stations (gray) with predicted total alkalinity (TA) <bold>(a, b, e, f, i, j)</bold> and partial pressure of CO<inline-formula><mml:math id="M385" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M386" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>) <bold>(c, d, g ,h, k, l)</bold>. The top row <bold>(a–d)</bold> shows data for the Bermuda Ocean Time-series Study (BATS), the middle row <bold>(e–h)</bold> shows data for the Hawaii Ocean Time-series (HOT),  and the bottom row <bold>(i, l)</bold> shows data for the Irminger station. The narrow panels show the average of the seasonal climatology for the time series. The gray shading shows the standard deviation of the observations for the period 1990 to 2018, while the orange/blue lines show the average estimate. TA for the Irminger station is calculated from <inline-formula><mml:math id="M387" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and DIC, and <inline-formula><mml:math id="M388" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is calculated for BATS and HOT using DIC and TA, as described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS4"/>.</p></caption>
            <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f04.png"/>

          </fig>

<?xmltex \floatpos{t}?><table-wrap id="Ch1.T4" specific-use="star"><?xmltex \currentcnt{4}?><label>Table 4</label><caption><p id="d1e5427">Comparison of training and independent data sources with various methods for the open-ocean region using the COSCATs coastal mask by <xref ref-type="bibr" rid="bib1.bibx67" id="text.138"/>. GLODAP refers to the GLODAP v2 2019 data <xref ref-type="bibr" rid="bib1.bibx85" id="paren.139"/>, HOT refers to the Hawaii Ocean Time-series <xref ref-type="bibr" rid="bib1.bibx28" id="paren.140"/>, BATS refers to Bermuda Atlantic Time-series Study <xref ref-type="bibr" rid="bib1.bibx3" id="paren.141"/>, SOCAT is the 2019 version of the Surface Ocean Carbon Atlas <xref ref-type="bibr" rid="bib1.bibx2" id="paren.142"/>, LDEO is the Lamont-Doherty Earth Observatory <inline-formula><mml:math id="M389" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> data set for points not present in SOCAT <xref ref-type="bibr" rid="bib1.bibx103" id="paren.143"/>, and SOCCOM is the pH measured by autonomous floats from the Southern Ocean Carbon and Climate Observations and Modeling project <xref ref-type="bibr" rid="bib1.bibx56" id="paren.144"/>. Statistical outliers were excluded in the calculation of LDEO RMSE. OS-ETHZ is the OceanSODA-ETHZ data from this study, NNGv2 is from <xref ref-type="bibr" rid="bib1.bibx13" id="text.145"/>, LIARv2 from <xref ref-type="bibr" rid="bib1.bibx16" id="text.146"/>, CMEMS-FFNNv2 from <xref ref-type="bibr" rid="bib1.bibx22" id="text.147"/>, and SOMFFN from <xref ref-type="bibr" rid="bib1.bibx65" id="text.148"/>. NNGv2 and LIARv2 predictions are made with SODA salinity and OSTIA sea-surface temperature, resulting in different estimates to the original publications <xref ref-type="bibr" rid="bib1.bibx13 bib1.bibx16" id="paren.149"/>. Note that the full data set is used for OceanSODA-ETHZ, unlike Table <xref ref-type="table" rid="Ch1.T3"/>, which presents the errors for test years.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="12">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="left"/>
     <oasis:colspec colnum="3" colname="col3" align="right"/>
     <oasis:colspec colnum="4" colname="col4" align="right"/>
     <oasis:colspec colnum="5" colname="col5" align="right" colsep="1"/>
     <oasis:colspec colnum="6" colname="col6" align="right"/>
     <oasis:colspec colnum="7" colname="col7" align="right" colsep="1"/>
     <oasis:colspec colnum="8" colname="col8" align="right"/>
     <oasis:colspec colnum="9" colname="col9" align="right"/>
     <oasis:colspec colnum="10" colname="col10" align="right" colsep="1"/>
     <oasis:colspec colnum="11" colname="col11" align="right"/>
     <oasis:colspec colnum="12" colname="col12" align="right"/>
     <oasis:thead>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"/>
         <oasis:entry rowsep="1" namest="col3" nameend="col5" align="center" colsep="1">TA (<inline-formula><mml:math id="M390" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M391" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) </oasis:entry>
         <oasis:entry rowsep="1" namest="col6" nameend="col7" align="center" colsep="1"><inline-formula><mml:math id="M392" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M393" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M394" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm) </oasis:entry>
         <oasis:entry rowsep="1" namest="col8" nameend="col10" align="center" colsep="1">DIC (<inline-formula><mml:math id="M395" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M396" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) </oasis:entry>
         <oasis:entry rowsep="1" namest="col11" nameend="col12" align="center">pH </oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3">GLODAP</oasis:entry>
         <oasis:entry colname="col4">HOT</oasis:entry>
         <oasis:entry colname="col5">BATS</oasis:entry>
         <oasis:entry colname="col6">SOCAT</oasis:entry>
         <oasis:entry colname="col7">LDEO</oasis:entry>
         <oasis:entry colname="col8">GLODAP</oasis:entry>
         <oasis:entry colname="col9">HOT</oasis:entry>
         <oasis:entry colname="col10">BATS</oasis:entry>
         <oasis:entry colname="col11">GLODAP</oasis:entry>
         <oasis:entry colname="col12">SOCCOM</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">Bias</oasis:entry>
         <oasis:entry colname="col2">this study</oasis:entry>
         <oasis:entry colname="col3">0.5</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M397" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>2.1</oasis:entry>
         <oasis:entry colname="col5">2.6</oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M398" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.4</oasis:entry>
         <oasis:entry colname="col7">0.1</oasis:entry>
         <oasis:entry colname="col8">0.5</oasis:entry>
         <oasis:entry colname="col9"><inline-formula><mml:math id="M399" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>1.0</oasis:entry>
         <oasis:entry colname="col10">0.4</oasis:entry>
         <oasis:entry colname="col11"><inline-formula><mml:math id="M400" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.001</oasis:entry>
         <oasis:entry colname="col12">0.009</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">LIAR <inline-formula><mml:math id="M401" display="inline"><mml:mo>+</mml:mo></mml:math></inline-formula> FFNN</oasis:entry>
         <oasis:entry colname="col3">0.3</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M402" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>3.1</oasis:entry>
         <oasis:entry colname="col5">0.2</oasis:entry>
         <oasis:entry colname="col6">0.5</oasis:entry>
         <oasis:entry colname="col7">0.6</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11">0.001</oasis:entry>
         <oasis:entry colname="col12">0.013</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">NNGv2</oasis:entry>
         <oasis:entry colname="col3">1.2</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M403" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>3.2</oasis:entry>
         <oasis:entry colname="col5">4.3</oasis:entry>
         <oasis:entry colname="col6"/>
         <oasis:entry colname="col7"/>
         <oasis:entry colname="col8">2.3</oasis:entry>
         <oasis:entry colname="col9">2.2</oasis:entry>
         <oasis:entry colname="col10"><inline-formula><mml:math id="M404" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.4</oasis:entry>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">SOMFFN</oasis:entry>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4"/>
         <oasis:entry colname="col5"/>
         <oasis:entry colname="col6">0.4</oasis:entry>
         <oasis:entry colname="col7">0.4</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">RMSE</oasis:entry>
         <oasis:entry colname="col2">this study</oasis:entry>
         <oasis:entry colname="col3">17.5</oasis:entry>
         <oasis:entry colname="col4">9.5</oasis:entry>
         <oasis:entry colname="col5">10.1</oasis:entry>
         <oasis:entry colname="col6">11.1</oasis:entry>
         <oasis:entry colname="col7">19.9</oasis:entry>
         <oasis:entry colname="col8">16.3</oasis:entry>
         <oasis:entry colname="col9">8.7</oasis:entry>
         <oasis:entry colname="col10">9.1</oasis:entry>
         <oasis:entry colname="col11">0.024</oasis:entry>
         <oasis:entry colname="col12">0.036</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">LIAR <inline-formula><mml:math id="M405" display="inline"><mml:mo>+</mml:mo></mml:math></inline-formula> FFNN</oasis:entry>
         <oasis:entry colname="col3">18.0</oasis:entry>
         <oasis:entry colname="col4">8.8</oasis:entry>
         <oasis:entry colname="col5">8.8</oasis:entry>
         <oasis:entry colname="col6">13.1</oasis:entry>
         <oasis:entry colname="col7">19.6</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11">0.023</oasis:entry>
         <oasis:entry colname="col12">0.037</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">NNGv2</oasis:entry>
         <oasis:entry colname="col3">16.2</oasis:entry>
         <oasis:entry colname="col4">6.7</oasis:entry>
         <oasis:entry colname="col5">10.4</oasis:entry>
         <oasis:entry colname="col6"/>
         <oasis:entry colname="col7"/>
         <oasis:entry colname="col8">23.1</oasis:entry>
         <oasis:entry colname="col9">9.5</oasis:entry>
         <oasis:entry colname="col10">15.2</oasis:entry>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">SOMFFN</oasis:entry>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4"/>
         <oasis:entry colname="col5"/>
         <oasis:entry colname="col6">11.7</oasis:entry>
         <oasis:entry colname="col7">21.4</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><inline-formula><mml:math id="M406" display="inline"><mml:mrow><mml:msup><mml:mi>r</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col2">this study</oasis:entry>
         <oasis:entry colname="col3">0.91</oasis:entry>
         <oasis:entry colname="col4">0.58</oasis:entry>
         <oasis:entry colname="col5">0.13</oasis:entry>
         <oasis:entry colname="col6">0.82</oasis:entry>
         <oasis:entry colname="col7">0.45</oasis:entry>
         <oasis:entry colname="col8">0.93</oasis:entry>
         <oasis:entry colname="col9">0.77</oasis:entry>
         <oasis:entry colname="col10">0.76</oasis:entry>
         <oasis:entry colname="col11">0.67</oasis:entry>
         <oasis:entry colname="col12">0.047</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">LIAR <inline-formula><mml:math id="M407" display="inline"><mml:mo>+</mml:mo></mml:math></inline-formula> FFNN</oasis:entry>
         <oasis:entry colname="col3">0.91</oasis:entry>
         <oasis:entry colname="col4">0.6</oasis:entry>
         <oasis:entry colname="col5">0.21</oasis:entry>
         <oasis:entry colname="col6">0.78</oasis:entry>
         <oasis:entry colname="col7">0.44</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11">0.67</oasis:entry>
         <oasis:entry colname="col12"><inline-formula><mml:math id="M408" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.043</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">NNGv2</oasis:entry>
         <oasis:entry colname="col3">0.93</oasis:entry>
         <oasis:entry colname="col4">0.75</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M409" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.1</oasis:entry>
         <oasis:entry colname="col6"/>
         <oasis:entry colname="col7"/>
         <oasis:entry colname="col8">0.82</oasis:entry>
         <oasis:entry colname="col9">0.71</oasis:entry>
         <oasis:entry colname="col10">0.38</oasis:entry>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">SOMFFN</oasis:entry>
         <oasis:entry colname="col3"/>
         <oasis:entry colname="col4"/>
         <oasis:entry colname="col5"/>
         <oasis:entry colname="col6">0.82</oasis:entry>
         <oasis:entry colname="col7">0.49</oasis:entry>
         <oasis:entry colname="col8"/>
         <oasis:entry colname="col9"/>
         <oasis:entry colname="col10"/>
         <oasis:entry colname="col11"/>
         <oasis:entry colname="col12"/>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

</sec>
<?pagebreak page789?><sec id="Ch1.S3.SS1.SSS2">
  <label>3.1.2</label><?xmltex \opttitle{Uncertainty for $p$CO${}_{2}$}?><title>Uncertainty for <inline-formula><mml:math id="M410" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M411" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula></title>
      <p id="d1e6191">For the uncertainty <inline-formula><mml:math id="M412" display="inline"><mml:mi>M</mml:mi></mml:math></inline-formula> associated with the measurement error of <inline-formula><mml:math id="M413" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M414" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, we adopt a value of <inline-formula><mml:math id="M415" display="inline"><mml:mrow><mml:mo>±</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M416" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm. This reflects the fact that 80 % of the data we have used from SOCAT (flags A and B) have a precision better than that number and an accuracy of similar magnitude. The remaining data we used (SOCAT flags C and D) have a measurement precision and accuracy of less than <inline-formula><mml:math id="M417" display="inline"><mml:mrow><mml:mo>±</mml:mo><mml:mn mathvariant="normal">5</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M418" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm.</p>
      <p id="d1e6254">We estimate the uncertainty <inline-formula><mml:math id="M419" display="inline"><mml:mi>R</mml:mi></mml:math></inline-formula> associated with the representation error of <inline-formula><mml:math id="M420" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M421" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> on the basis of a spatiotemporal gradient analysis.
To this end, we compare the <inline-formula><mml:math id="M422" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M423" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> in our regular grid that has a resolution of <inline-formula><mml:math id="M424" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> by 1 month, with the <inline-formula><mml:math id="M425" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M426" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> binned to a grid with twice this resolution, i.e., <inline-formula><mml:math id="M427" display="inline"><mml:mrow><mml:mn mathvariant="normal">0.5</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">0.5</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> by 15 d.
In regions with high spatiotemporal coverage, the difference in the average of adjacent grid cells represents the potential change that can occur within the coarser <inline-formula><mml:math id="M428" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> by a 1-month grid cell.
The spatial and temporal gradients are calculated separately, and we take the average of these two elements.
Using this analysis, we estimate a representation error of our <inline-formula><mml:math id="M429" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M430" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> estimates of 7  and 17 <inline-formula><mml:math id="M431" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for the open and coastal ocean respectively (Table <xref ref-type="table" rid="Ch1.T3"/>).</p>
      <p id="d1e6400">From the RMSE of our test data, we estimate an uncertainty <inline-formula><mml:math id="M432" display="inline"><mml:mi>P</mml:mi></mml:math></inline-formula> associated with the  prediction error of <inline-formula><mml:math id="M433" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M434" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> of 12 <inline-formula><mml:math id="M435" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for the open ocean and 28 <inline-formula><mml:math id="M436" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for the coastal ocean (Table <xref ref-type="table" rid="Ch1.T3"/>). Within the open ocean (Fig. <xref ref-type="fig" rid="Ch1.F3"/>e), the eastern tropical Pacific has the highest RMSE, but this is also the region with the highest variance in the observations. The strong horizontal gradients in the region increase the errors, particularly at cluster boundaries. The high RMSE for the coastal region stems primarily from coastal Antarctica as well as some coastal regions in the higher latitudes of the Northern Hemisphere. The former is due to large uncertainties in <inline-formula><mml:math id="M437" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> during the summer months when retreating ice and ensuing rapid net primary production result in large gradients <xref ref-type="bibr" rid="bib1.bibx1" id="paren.150"/>.
The climatologically mapped errors for <inline-formula><mml:math id="M438" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> are shown in Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>b and d, created by mapping the test errors to the clusters and averaged over the ensemble.</p>
      <p id="d1e6481">The comparison between the regression estimated and observed <inline-formula><mml:math id="M439" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M440" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> reveals regional biases (Fig. <xref ref-type="fig" rid="Ch1.F3"/>e), despite the global bias being close to zero (<inline-formula><mml:math id="M441" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>0.37 <inline-formula><mml:math id="M442" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm). Some of the highest biases are found, again, in the eastern tropical Pacific, where strong horizontal gradients drive the observed juxtaposed biases. The large negative biases in winter in the Southern Ocean are likely driven by the paucity of data in this region <xref ref-type="bibr" rid="bib1.bibx46 bib1.bibx42 bib1.bibx14" id="paren.151"/>.</p>
      <?pagebreak page790?><p id="d1e6521">The time series comparisons show that the seasonal cycle is well represented at BATS and HOT with <inline-formula><mml:math id="M443" display="inline"><mml:mrow><mml:msup><mml:mi>r</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula> scores of 0.89 and 0.82 respectively (Fig. <xref ref-type="fig" rid="Ch1.F4"/>d, h).
Low biases (<inline-formula><mml:math id="M444" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M445" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm in absolute terms) further demonstrate that <inline-formula><mml:math id="M446" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates are reliable in the subtropics.
The seasonal cycle is also well captured at the Irminger station in the high latitudes, but a lower <inline-formula><mml:math id="M447" display="inline"><mml:mrow><mml:msup><mml:mi>r</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula> score and larger bias (<inline-formula><mml:math id="M448" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>8.0 <inline-formula><mml:math id="M449" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M450" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) allude to the dampened amplitude of the seasonal cycle particularly in the winter months (Fig. <xref ref-type="fig" rid="Ch1.F4"/>).</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F5" specific-use="star"><?xmltex \currentcnt{5}?><?xmltex \def\figurename{Figure}?><label>Figure 5</label><caption><p id="d1e6612">Root mean squared error (RMSE – <bold>a</bold>, <bold>d</bold>) and biases <bold>(b, c, e, f)</bold> for dissolved inorganic carbon (DIC, <bold>a, b, c</bold>) and pH <bold>(d, e, f)</bold> compared with in situ GLODAP v2.2019 data. The two subplots in the rightmost column compare the zonally averaged bias for JJA and DJF. Data were processed for plotting as described in Sect. 4.</p></caption>
            <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f05.png"/>

          </fig>

      <?xmltex \floatpos{t}?><fig id="Ch1.F6"><?xmltex \currentcnt{6}?><?xmltex \def\figurename{Figure}?><label>Figure 6</label><caption><p id="d1e6638">A comparison of observations from long-term observation stations (gray) with predicted dissolved inorganic carbon (DIC). The top row <bold>(a, b)</bold> shows data for the Bermuda Ocean Time-series Study (BATS), the middle row <bold>(c, d)</bold> shows data for the Hawaii Ocean Time-series (HOT), and the bottom row <bold>(e, f)</bold> shows data for the Irminger station. The narrow panels on the right show the average of the seasonal climatology for the time series, where the gray shading shows the standard deviation of the observations for the period 1990 to 2018, while the green line shows the average estimate. </p></caption>
            <?xmltex \igopts{width=241.848425pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f06.png"/>

          </fig>

</sec>
</sec>
<sec id="Ch1.S3.SS2">
  <label>3.2</label><title>Uncertainties of the calculated parameters</title>
      <p id="d1e6665">We determine the uncertainties of the calculated parameters in two ways. First, we propagate the uncertainties of <inline-formula><mml:math id="M451" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M452" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and TA through pyCO2SYS <xref ref-type="bibr" rid="bib1.bibx88 bib1.bibx53" id="paren.152"/> onto the computed parameters DIC, pH, and <inline-formula><mml:math id="M453" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>. This yields an expected uncertainty that we refer to as a bottom-up total uncertainty estimate. Second, we obtain a top-down uncertainty estimate by comparing the calculated DIC and pH with independent measurements (see Table <xref ref-type="table" rid="Ch1.T4"/>). We first describe the top-down estimates and then the bottom-up estimates.
For the top-down estimate, we use the GLODAP DIC and pH data as our independent test of the method's performance. This is because these data are not used in any way in our estimation.</p>
      <p id="d1e6696">In the global mean, the computed DIC in OceanSODA-ETHZ has a very low bias compared with in situ GLODAP measurements (0.5 <inline-formula><mml:math id="M454" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M455" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) (Table <xref ref-type="table" rid="Ch1.T4"/>).
Spatially, the DIC biases reveal a more nuanced picture, with large positive biases in the western equatorial Pacific and negative biases in the western equatorial Atlantic.
The bias in the western equatorial Atlantic matches the negative bias in the <inline-formula><mml:math id="M456" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> in the same region; however, the source of the DIC bias in the Pacific is not clear from the <inline-formula><mml:math id="M457" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA test data biases.
The global mean of the top-down uncertainty estimate for DIC (16.3 <inline-formula><mml:math id="M458" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol<inline-formula><mml:math id="M459" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for the open ocean) is then perhaps a better reflection of the uncertainty as positive and negative values do not cancel each other out.</p>
      <p id="d1e6770">It is interesting to point out that the computed DIC in OceanSODA-ETHZ compares very favorably to directly estimated DIC products, such as that provided by NNGv2. Our uncertainty associated with the prediction error for DIC of 16.3 <inline-formula><mml:math id="M460" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol<inline-formula><mml:math id="M461" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> is substantially better than theirs (23.1 <inline-formula><mml:math id="M462" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M463" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) across all independent data.</p>
      <p id="d1e6813">The  comparison of the DIC time series data (BATS, HOT, and Irminger stations) supports the findings of the global top-down estimates (Fig. <xref ref-type="fig" rid="Ch1.F6"/>).
The biases are relatively low for HOT and BATS (<inline-formula><mml:math id="M464" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>1.0 and 0.4 <inline-formula><mml:math id="M465" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M466" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> respectively), but the bias is much larger at Irminger station (<inline-formula><mml:math id="M467" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>6.9 <inline-formula><mml:math id="M468" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M469" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>), which is at a much higher latitude (64<inline-formula><mml:math id="M470" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N) compared to HOT and BATS in the subtropics (<inline-formula><mml:math id="M471" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">35</mml:mn></mml:mrow></mml:math></inline-formula><inline-formula><mml:math id="M472" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N).
However, this bias is not reflected in the zonal average of the seasonal biases (Fig. <xref ref-type="fig" rid="Ch1.F5"/>c).
Similarly, the RMSE is also larger at Irminger station (14 <inline-formula><mml:math id="M473" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M474" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) compared to HOT and BATS (<inline-formula><mml:math id="M475" display="inline"><mml:mo lspace="0mm">∼</mml:mo></mml:math></inline-formula> 9 <inline-formula><mml:math id="M476" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M477" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) in the subtropics.
Despite these differences in the top-down error, the same amount of variability is represented by the OceanSODA DIC for all three stations (<inline-formula><mml:math id="M478" display="inline"><mml:mo lspace="0mm">∼</mml:mo></mml:math></inline-formula> 0.76) owing to the larger seasonal cycle at Irminger station.</p>
      <p id="d1e6958">The pH comparison with the GLODAP pH measurements shows that OceanSODA-ETHZ has a negligible bias (0.001).
As with DIC, regional biases in pH are larger than the global average, with the coastal and high-latitude oceans contributing significantly to the regional biases. The RMSE of pH with respect to GLODAP is also low (0.024) but is slightly outperformed by the RMSE of pH calculated with  LIARv2 TA and FFNNv2 <inline-formula><mml:math id="M479" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (made available in the FFNNv2 data set, 0.023). But given the uncertainty of GLODAP pH (0.01), the difference is negligible.</p>
      <p id="d1e6975">For the bottom-up estimate, we propagate the total uncertainty of <inline-formula><mml:math id="M480" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and only the uncertainties associated with the measurement and prediction errors for TA. We do this to avoid including the representation error twice in the bottom-up estimate, as we hypothesize that the representation error of TA is largely accounted for by the representation error of <inline-formula><mml:math id="M481" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and vice versa.
We choose to use the representation error of <inline-formula><mml:math id="M482" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> rather than TA as the larger number of samples gives us greater confidence in the estimate.
Moreover, we feel that the assumptions we make in the estimate of the TA representation error are larger than those for <inline-formula><mml:math id="M483" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, thus further justifying our choice in using only the representation error of <inline-formula><mml:math id="M484" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F7"><?xmltex \currentcnt{7}?><?xmltex \def\figurename{Figure}?><label>Figure 7</label><caption><p id="d1e7050">A comparison of propagated uncertainties with independent uncertainties as an assertion of the validity of uncertainty estimates. The map <bold>(a)</bold> shows the separation between coastal and open ocean, panel <bold>(b)</bold> shows the error contributions in the open ocean, and panel <bold>(c)</bold> shows those in the coastal ocean. The total uncertainty has been broken into the three different components. Note that the values represented by the bar plots are not equivalent to values in Table <xref ref-type="table" rid="Ch1.T3"/> as the latter shows <inline-formula><mml:math id="M485" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA total uncertainties for test data only, while the bar charts show total uncertainties for all data; further, the breakdown of the error contributions is proportional to the contribution of the sum of the squares (see Eq. <xref ref-type="disp-formula" rid="Ch1.E2"/>).</p></caption>
          <?xmltex \igopts{width=241.848425pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f07.png"/>

        </fig>

      <p id="d1e7087">The comparison of top-down vs. bottom-up uncertainty estimates for open and coastal oceans is shown in Fig. <xref ref-type="fig" rid="Ch1.F7"/>, with values of the total uncertainties also shown.
The top-down and bottom-up uncertainty estimates for DIC are relatively accurate, with the estimates being within 5 % of each other in the open ocean and 9 % in the coastal ocean.
However, the estimates are not as coherent for pH where the bottom-up error is 23 % smaller than the top-down uncertainty in the open ocean.
The difference is even bigger in the coastal ocean where the bottom-up uncertainty is 31 % smaller than the top-down uncertainty.</p>
</sec>
</sec>
<sec id="Ch1.S4">
  <label>4</label><title>Results</title>
<sec id="Ch1.S4.SS1">
  <label>4.1</label><title>Comparison with other climatologies</title>
      <p id="d1e7108">A spatial comparison between OceanSODA-ETHZ and existing products might reveal potential biases in our product if the bias is present in all comparisons.
We compare TA against LIARv2 and NNGv2 <xref ref-type="bibr" rid="bib1.bibx16 bib1.bibx13" id="paren.153"/> by first taking the difference for each month and then calculating the average for these differences.
The same approach is used to compare <inline-formula><mml:math id="M486" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> with SOMFFN and FFNNv2 <xref ref-type="bibr" rid="bib1.bibx65 bib1.bibx22" id="paren.154"/>.</p>
      <p id="d1e7131">The differences in TA between OceanSODA-ETHZ and NNGv2 and LIARv2 are on the same order of magnitude in the open ocean as the prediction error (13 <inline-formula><mml:math id="M487" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M488" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) but are slightly larger on average for NNGv2 than for LIARv2 (Fig. <xref ref-type="fig" rid="Ch1.F8"/>a, b).
There is also some agreement in the spatial pattern of the distribution, particularly in data-sparse parts of the Pacific Ocean and Indian Ocean.
The larger differences may stem from the fact that both LIARv2 and NNGv2 are more constrained by spatial coordinates than GRaCER.
NNGv2 uses latitude and longitude as predictors, while the LIARv2 approach interpolates the linear regression coefficients for every 5<inline-formula><mml:math id="M489" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> grid cell<?pagebreak page791?> <xref ref-type="bibr" rid="bib1.bibx13 bib1.bibx16" id="paren.155"/>.
The divergence in data-sparse regions is thus not surprising.
Although, it must be emphasized that this comparison serves more as a sanity check than as a ground-truthing exercise.</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F8" specific-use="star"><?xmltex \currentcnt{8}?><?xmltex \def\figurename{Figure}?><label>Figure 8</label><caption><p id="d1e7170">A comparison of the mean differences between TA <bold>(a, b)</bold> and <inline-formula><mml:math id="M490" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> <bold>(c, d)</bold> for OceanSODA-ETHZ and other published methods: <bold>(a)</bold> LIARv2, <bold>(b)</bold> NNGv2, <bold>(c)</bold> CMEMS-FFNNv2, and <bold>(d)</bold> MPI-SOMFFN. The markers in the North Pacific show the locations used in the comparison of the climatology in Fig. <xref ref-type="fig" rid="Ch1.F11"/>.</p></caption>
          <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f08.png"/>

        </fig>

      <p id="d1e7215">The differences between OceanSODA-ETHZ <inline-formula><mml:math id="M491" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and FFNNv2 and SOMFFN are smaller and not as spatially coherent compared to those for TA.
The differences are marginally larger for the OceanSODA-FFNNv2 than for OceanSODA-SOMFFN (Fig. <xref ref-type="fig" rid="Ch1.F8"/>c, d).
This is consistent with the data in Table <xref ref-type="table" rid="Ch1.T4"/>, where the metrics for the SOMFFN are very similar to OceanSODA-ETHZ.
The smaller difference between the latter should not come as a surprise as the GRaCER approach is built on the two-step cluster-regression approach of the SOMFFN, while the FFNNv2 approach includes spatial coordinates <xref ref-type="bibr" rid="bib1.bibx65 bib1.bibx22" id="paren.156"/>.
In general, the dissimilarity between the differences is encouraging as it indicates that OceanSODA-ETHZ is not consistently biased relative to SOMFFN and FFNNv2.</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F9" specific-use="star"><?xmltex \currentcnt{9}?><?xmltex \def\figurename{Figure}?><label>Figure 9</label><caption><p id="d1e7241">A basin–mean comparison of OceanSODA-ETHZ <inline-formula><mml:math id="M492" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> with four gap-filling methods: MPI-SOMFFN, JENA-MLS, CMEMS-FFNNv2, and CSIR-ML6. The thin lines show the differences compared to the individual methods, while the thick line shows the mean difference across the four methods. We do not show the Arctic Ocean as OceanSODA-ETHZ covers only 23 % of the region. The vertical dashed line in each figure marks the year 1990, where estimates prior to this period show biases in panels <bold>(b)</bold> and <bold>(c)</bold>.</p></caption>
          <?xmltex \igopts{width=369.885827pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f09.png"/>

        </fig>

      <p id="d1e7270">We also show the temporal evolution of the basin–mean differences between OceanSODA-ETHZ <inline-formula><mml:math id="M493" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and other gap-filling methods (Fig. <xref ref-type="fig" rid="Ch1.F9"/>).
In the Atlantic (Fig. <xref ref-type="fig" rid="Ch1.F9"/>a), OceanSODA-ETHZ <inline-formula><mml:math id="M494" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is <inline-formula><mml:math id="M495" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M496" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm lower than the mean of the other gap-filling methods for the period 1990 to 2008.
Thereafter, the difference is <inline-formula><mml:math id="M497" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M498" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm.
In the Indian Ocean, our <inline-formula><mml:math id="M499" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates have a persistent negative difference of <inline-formula><mml:math id="M500" display="inline"><mml:mrow><mml:mo>∼</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M501" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm (Fig. <xref ref-type="fig" rid="Ch1.F9"/>c).
The comparison in the Pacific (Fig. <xref ref-type="fig" rid="Ch1.F9"/>b) is the most consistent with the other methods, with a slight positive difference in the beginning of the period (pre-1990).
The OceanSODA-ETHZ estimates of <inline-formula><mml:math id="M502" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> in the Southern Ocean (Fig. <xref ref-type="fig" rid="Ch1.F9"/>d) have a large positive difference prior to 1990 – up to 6 <inline-formula><mml:math id="M503" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for one of the ensemble members.
This difference quickly diminishes and is near zero by 1990.
There is also a negative difference later in the period (2004 to 2015); however, the ensemble spread over this period is large.</p>
      <p id="d1e7403">The comparison with other methods illustrates that while gap-filling methods are converging on a global scale, there are regional differences. Further, large differences in <inline-formula><mml:math id="M504" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> between methods prior to 1990 indicate high uncertainty for this period.</p>
</sec>
<sec id="Ch1.S4.SS2">
  <label>4.2</label><title>Seasonal climatologies</title>
      <p id="d1e7428">The climatological mean spatial distribution of TA, <inline-formula><mml:math id="M505" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, DIC, pH, and <inline-formula><mml:math id="M506" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula>, obtained by averaging the estimates from 1985 through 2018 reveals a very rich and diverse pattern of variability with commonalities and differences (Fig. <xref ref-type="fig" rid="Ch1.F10"/>).
The climatological maps are accompanied by Hovmöller diagrams that show the zonal average of the seasonal cycle for each of the variables (Fig. <xref ref-type="fig" rid="Ch1.F11"/>).
We also show climatological time series for each of the variables at high-latitude (55<inline-formula><mml:math id="M507" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 170<inline-formula><mml:math id="M508" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W), midlatitude (30<inline-formula><mml:math id="M509" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 170<inline-formula><mml:math id="M510" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W), and low-latitude (10<inline-formula><mml:math id="M511" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 170<inline-formula><mml:math id="M512" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> W) locations.</p>

      <?xmltex \floatpos{p}?><fig id="Ch1.F10" specific-use="star"><?xmltex \currentcnt{10}?><?xmltex \def\figurename{Figure}?><label>Figure 10</label><caption><p id="d1e7513">Mean maps of the GRaCER-based estimates for the period 1985–2018 for <bold>(a)</bold> total alkalinity and <bold>(c)</bold> <inline-formula><mml:math id="M513" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M514" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, as well as those of the computed variables, <bold>(b)</bold> dissolved inorganic carbon, <bold>(d)</bold> pH, <bold>(e)</bold> <inline-formula><mml:math id="M515" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mi mathvariant="normal">calc</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> (saturation state with regard to calcite), and <bold>(f)</bold> <inline-formula><mml:math id="M516" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mi mathvariant="normal">arag</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> (saturation state with regard to aragonite). The three black markers in each plot show the locations chosen for the seasonal analysis in Fig. <xref ref-type="fig" rid="Ch1.F11"/>f–j.</p></caption>
          <?xmltex \igopts{width=369.885827pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f10.png"/>

        </fig>

      <p id="d1e7581">Total alkalinity shows the largest differences between basins, with the mean alkalinity being much higher in the saltier Atlantic than in the Pacific and Indian basins <xref ref-type="bibr" rid="bib1.bibx9" id="paren.157"/> (Fig. <xref ref-type="fig" rid="Ch1.F10"/>a). The spatial variability of TA on a global scale exceeds the variability of the seasonal cycle (Fig. <xref ref-type="fig" rid="Ch1.F11"/>a, e).
Seasonal variability of TA is on the order of 20 <inline-formula><mml:math id="M517" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M518" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>  at the chosen mid- and high-latitude locations, while the latitudinal gradient is as large as 150 <inline-formula><mml:math id="M519" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M520" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
However, much of the TA seasonality is driven by seasonal changes in salinity due to precipitation and ice melt in the respective regions.</p>
      <p id="d1e7633">Dissolved inorganic carbon is more homogeneous across the basins but has a much larger meridional gradient than TA, amounting to more than 150 <inline-formula><mml:math id="M521" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M522" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> (Fig. <xref ref-type="fig" rid="Ch1.F10"/>b). This<?pagebreak page792?> meridional gradient is seasonally substantially more modified than is the case for TA, particularly in the high latitudes of the Northern Hemisphere where the seasonal cycle is as large as 100 <inline-formula><mml:math id="M523" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M524" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> (Fig. <xref ref-type="fig" rid="Ch1.F11"/>b, g). The larger seasonal cycle in DIC is due to the carbon uptake by springtime phytoplankton blooms and stratification during the warmer seasons <xref ref-type="bibr" rid="bib1.bibx99" id="paren.158"/>.
The magnitude of the springtime blooms is dampened by iron limitation in the Southern Ocean, visible by a smaller seasonal cycle amplitude (<inline-formula><mml:math id="M525" display="inline"><mml:mo lspace="0mm">∼</mml:mo></mml:math></inline-formula> 40 <inline-formula><mml:math id="M526" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M527" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>) <xref ref-type="bibr" rid="bib1.bibx110 bib1.bibx101" id="paren.159"/>. However, the background DIC concentration in the Southern Ocean is much larger due to upwelling of DIC-rich circumpolar deep waters driven by the persistent westerlies south of 50<inline-formula><mml:math id="M528" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> S <xref ref-type="bibr" rid="bib1.bibx77" id="paren.160"/>.</p>
      <p id="d1e7727">The spatial distribution and seasonal cycles of <inline-formula><mml:math id="M529" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and pH (Figs. <xref ref-type="fig" rid="Ch1.F10"/>c, d  and <xref ref-type="fig" rid="Ch1.F11"/>c, d, h, i) are strongly negatively correlated due to the inverse stoichiometric relationship between dissolved aqueous CO<inline-formula><mml:math id="M530" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and [H<inline-formula><mml:math id="M531" display="inline"><mml:msup><mml:mi/><mml:mo>+</mml:mo></mml:msup></mml:math></inline-formula>] <xref ref-type="bibr" rid="bib1.bibx24" id="paren.161"/>. The reduction in DIC is concomitant with the reduction in <inline-formula><mml:math id="M532" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> in the high latitudes due to the biological uptake of [CO<inline-formula><mml:math id="M533" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>] (Fig. <xref ref-type="fig" rid="Ch1.F11"/>g, h).
However, this relationship does not hold true in the midlatitudes, where the slight decrease in DIC is contrasted by a relatively strong increase in <inline-formula><mml:math id="M534" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.
This is due to the positive temperature dependence of <inline-formula><mml:math id="M535" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (the opposite is true for pH) <xref ref-type="bibr" rid="bib1.bibx104" id="paren.162"/>, which will be elaborated on in the discussion.</p>

      <?xmltex \floatpos{p}?><fig id="Ch1.F11" specific-use="star"><?xmltex \currentcnt{11}?><?xmltex \def\figurename{Figure}?><label>Figure 11</label><caption><p id="d1e7829">Hovmöller plots <bold>(a–e)</bold> showing the zonally averaged seasonal climatologies for <bold>(a)</bold> total alkalinity, <bold>(b)</bold> dissolved inorganic carbon, <bold>(c)</bold> <inline-formula><mml:math id="M536" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, <bold>(d)</bold> pH, and <bold>(e)</bold> aragonite saturation state (<inline-formula><mml:math id="M537" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mi mathvariant="normal">Ar</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula>). The second row of panels <bold>(f)</bold>–<bold>(j)</bold> shows the corresponding variables for a high-latitude (55<inline-formula><mml:math id="M538" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 180<inline-formula><mml:math id="M539" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> E, blue), midlatitude (30<inline-formula><mml:math id="M540" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 180<inline-formula><mml:math id="M541" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> E, orange), and low-latitude (10<inline-formula><mml:math id="M542" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> N, 180<inline-formula><mml:math id="M543" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> E, green) locations. The units for panels <bold>(f)</bold>–<bold>(j)</bold> correspond with the units in panels <bold>(a)</bold>–<bold>(e)</bold>.</p></caption>
          <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f11.png"/>

        </fig>

      <p id="d1e7956">The spatial distribution of <inline-formula><mml:math id="M544" display="inline"><mml:mi mathvariant="normal">Ω</mml:mi></mml:math></inline-formula> (Figs. <xref ref-type="fig" rid="Ch1.F10"/>e, f and <xref ref-type="fig" rid="Ch1.F11"/>e, j) strongly reflects the concentration of the carbonate ion, which can be well approximated by the difference between TA and DIC <xref ref-type="bibr" rid="bib1.bibx97" id="paren.163"/>.
Given that the seasonal cycle of TA is much weaker than DIC, the latter dominates the seasonal cycle of <inline-formula><mml:math id="M545" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">Ar</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula> (Fig. <xref ref-type="fig" rid="Ch1.F11"/>e, j).
This would also be true for <inline-formula><mml:math id="M546" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">Ca</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula>, which only differs from <inline-formula><mml:math id="M547" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">Ar</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula> in magnitude and not in distribution or seasonality (the latter is not shown).</p>
</sec>
<?pagebreak page793?><sec id="Ch1.S4.SS3">
  <label>4.3</label><title>Global- and basin-scale long-term trends</title>
      <p id="d1e8020">The OceanSODA-ETHZ data set can provide important novel constraints on the long-term trends in ocean acidification. We determine the long-term trends by a linear regression approach, restricting the period to 1990 through 2018, thus leaving out the 1980s, where the estimates are much more uncertain.</p>
      <?pagebreak page795?><p id="d1e8023">The global- and basin-scale trends for <inline-formula><mml:math id="M548" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> are remarkably similar (varying by only about 0.5 <inline-formula><mml:math id="M549" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow></mml:math></inline-formula> per decade around the global mean of 16.5 <inline-formula><mml:math id="M550" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow></mml:math></inline-formula> per decade) (Table <xref ref-type="table" rid="Ch1.T5"/>. The ocean trends are also very consistent with the  trends in atmospheric CO<inline-formula><mml:math id="M551" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> (<inline-formula><mml:math id="M552" display="inline"><mml:mo lspace="0mm">∼</mml:mo></mml:math></inline-formula> 18.6 <inline-formula><mml:math id="M553" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow></mml:math></inline-formula> per decade), reflecting the fact that the CO<inline-formula><mml:math id="M554" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> system in the surface ocean is following the increase in the atmosphere very closely globally as well as in all ocean basins. The atmospheric trend is slightly steeper than those in the ocean. This is as expected since the atmospheric CO<inline-formula><mml:math id="M555" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> increase forces the ocean, leading to a slight delay, a reflection of an increase in the air–sea disequilibrium over time (see, e.g., discussion in <xref ref-type="bibr" rid="bib1.bibx78" id="altparen.164"/>). This growth in the air–sea disequilibrium is the driving force behind the increase in the oceanic sink strength for anthropogenic CO<inline-formula><mml:math id="M556" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> over time <xref ref-type="bibr" rid="bib1.bibx48" id="paren.165"/>.</p>
      <p id="d1e8123">The basin-scale consistency holds true for pH as well (<inline-formula><mml:math id="M557" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>0.016 units per decade) and <inline-formula><mml:math id="M558" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">ar</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula> (<inline-formula><mml:math id="M559" display="inline"><mml:mo lspace="0mm">-</mml:mo></mml:math></inline-formula>0.07 units per decade), where global values are very similar to those averaged across the different ocean basins.  In contrast, there is more spatial variability in the DIC trends, but the overall magnitude of about 9 <inline-formula><mml:math id="M560" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M561" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> per decade largely represents the uptake of anthropogenic CO<inline-formula><mml:math id="M562" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> from the atmosphere. The spatial differences in the DIC trends are also mirrored in the spatial variations in the TA trends, with regions with higher DIC trends having higher TA trends. This makes sense in terms of positive trends in TA increasing the buffering capacity, hence permitting DIC to grow faster for the same increase in seawater  <inline-formula><mml:math id="M563" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>. The trends in TA themselves are driven almost entirely by salinity with a basin-scale correlation of 0.99 (see <xref ref-type="bibr" rid="bib1.bibx18" id="altparen.166"/>).</p>

<?xmltex \floatpos{t}?><table-wrap id="Ch1.T5" specific-use="star"><?xmltex \currentcnt{5}?><label>Table 5</label><caption><p id="d1e8203">Linear trends and their standard errors for OceanSODA-ETHZ variables for the period 1990 to 2018. All columns show increases per decade (decade<inline-formula><mml:math id="M564" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>). All trends in the table are significant (<inline-formula><mml:math id="M565" display="inline"><mml:mrow><mml:mi>P</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.05</mml:mn></mml:mrow></mml:math></inline-formula>). We exclude the Arctic as the OceanSODA-ETHZ product only covers 23 % of this region and may thus give spurious trends. The ocean basins are defined by the map shown in Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F16"/>.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="7">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="right"/>
     <oasis:colspec colnum="4" colname="col4" align="right"/>
     <oasis:colspec colnum="5" colname="col5" align="right"/>
     <oasis:colspec colnum="6" colname="col6" align="right"/>
     <oasis:colspec colnum="7" colname="col7" align="right"/>
     <oasis:thead>
       <oasis:row>
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2">TA</oasis:entry>
         <oasis:entry colname="col3">DIC</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M566" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">Ω</mml:mi><mml:mrow class="chem"><mml:mi mathvariant="normal">ar</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5">pH</oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M567" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col7"><inline-formula><mml:math id="M568" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mrow class="chem"><mml:mi mathvariant="normal">atm</mml:mi></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M569" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M570" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> decade<inline-formula><mml:math id="M571" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M572" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M573" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> decade<inline-formula><mml:math id="M574" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4">units decade<inline-formula><mml:math id="M575" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5">units decade<inline-formula><mml:math id="M576" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col6"> <inline-formula><mml:math id="M577" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm decade<inline-formula><mml:math id="M578" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col7"> <inline-formula><mml:math id="M579" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm decade<inline-formula><mml:math id="M580" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">Global</oasis:entry>
         <oasis:entry colname="col2">1.5 <inline-formula><mml:math id="M581" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col3">8.6 <inline-formula><mml:math id="M582" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M583" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.07 <inline-formula><mml:math id="M584" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.00</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M585" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.016 <inline-formula><mml:math id="M586" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.000</oasis:entry>
         <oasis:entry colname="col6">16.5 <inline-formula><mml:math id="M587" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col7">18.6 <inline-formula><mml:math id="M588" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Atlantic</oasis:entry>
         <oasis:entry colname="col2">3.1 <inline-formula><mml:math id="M589" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.2</oasis:entry>
         <oasis:entry colname="col3">10.0 <inline-formula><mml:math id="M590" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.4</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M591" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.07 <inline-formula><mml:math id="M592" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.00</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M593" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.016 <inline-formula><mml:math id="M594" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.000</oasis:entry>
         <oasis:entry colname="col6">16.7 <inline-formula><mml:math id="M595" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.2</oasis:entry>
         <oasis:entry colname="col7">18.8 <inline-formula><mml:math id="M596" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.2</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Pacific</oasis:entry>
         <oasis:entry colname="col2">0.5 <inline-formula><mml:math id="M597" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col3">8.1 <inline-formula><mml:math id="M598" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.3</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M599" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.07 <inline-formula><mml:math id="M600" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.00</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M601" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.016 <inline-formula><mml:math id="M602" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.000</oasis:entry>
         <oasis:entry colname="col6">16.7 <inline-formula><mml:math id="M603" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col7">18.7 <inline-formula><mml:math id="M604" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Indian</oasis:entry>
         <oasis:entry colname="col2">4.4 <inline-formula><mml:math id="M605" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.4</oasis:entry>
         <oasis:entry colname="col3">10.9 <inline-formula><mml:math id="M606" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.6</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M607" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.06 <inline-formula><mml:math id="M608" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.00</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M609" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.015 <inline-formula><mml:math id="M610" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.000</oasis:entry>
         <oasis:entry colname="col6">16.2 <inline-formula><mml:math id="M611" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.4</oasis:entry>
         <oasis:entry colname="col7">18.3 <inline-formula><mml:math id="M612" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Southern</oasis:entry>
         <oasis:entry colname="col2">0.5 <inline-formula><mml:math id="M613" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
         <oasis:entry colname="col3">7.0 <inline-formula><mml:math id="M614" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.6</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M615" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.06 <inline-formula><mml:math id="M616" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.01</oasis:entry>
         <oasis:entry colname="col5"><inline-formula><mml:math id="M617" display="inline"><mml:mo>-</mml:mo></mml:math></inline-formula>0.017 <inline-formula><mml:math id="M618" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.001</oasis:entry>
         <oasis:entry colname="col6">16.0 <inline-formula><mml:math id="M619" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.4</oasis:entry>
         <oasis:entry colname="col7">18.5 <inline-formula><mml:math id="M620" display="inline"><mml:mo>±</mml:mo></mml:math></inline-formula> 0.1</oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

</sec>
</sec>
<sec id="Ch1.S5">
  <label>5</label><title>Discussion</title>
<sec id="Ch1.S5.SS1">
  <label>5.1</label><title>Choosing the appropriate machine learning configuration</title>
      <p id="d1e8875">Here we consider two notable decisions that have a large impact on the final estimates: (1) the use of the ensemble approach and (2) the choice of regression algorithm.
For details on the minor choices, see Sect. <xref ref-type="sec" rid="App1.Ch1.S1.SS3"/> in the Appendix.</p>
      <p id="d1e8880">As previously motivated, we opt for the cluster-regression approach that is able to generalize estimates in sparse regions due to information sharing within a cluster.
However, cluster boundaries are often semi-discrete, resulting in artifactual boundaries.
This makes the output of cluster-regression approaches less suitable for studies where gradients over short time periods or distances are assessed, e.g., for the detection of extreme events.
Our approach removes these boundaries and improves the robustness of the estimates by eliminating, to a large extent, the sensitivity of the regression to the clustering algorithm.</p>
      <p id="d1e8883">The second major consideration is the choice of regression algorithm.
Our choice of different algorithms for TA and <inline-formula><mml:math id="M621" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> may seem peculiar. However, this decision was informed by the nature of the problems.
While regressing TA and <inline-formula><mml:math id="M622" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> are conceptually similar, the size of each data set and the distribution of training samples set them apart.
When gridded to a monthly by <inline-formula><mml:math id="M623" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> resolution, the number of data are <inline-formula><mml:math id="M624" display="inline"><mml:mrow><mml:mo>∼</mml:mo><mml:mn mathvariant="normal">300</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mn mathvariant="normal">000</mml:mn></mml:mrow></mml:math></inline-formula> for SOCAT <inline-formula><mml:math id="M625" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and <inline-formula><mml:math id="M626" display="inline"><mml:mrow><mml:mo>∼</mml:mo><mml:mn mathvariant="normal">16</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mn mathvariant="normal">000</mml:mn></mml:mrow></mml:math></inline-formula> for GLODAP TA – i.e., nearly 20 times more data for the former.
However, the strong linear correlation between TA and salinity (<inline-formula><mml:math id="M627" display="inline"><mml:mrow><mml:mi>r</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.96</mml:mn></mml:mrow></mml:math></inline-formula>) compensates for the poor sampling distribution – that is, as long as the regression method is able to extrapolate – a criterion that the support vector regression (SVR) method meets.</p>
      <p id="d1e8987"><inline-formula><mml:math id="M628" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is poorly correlated to any proxy, suggesting that the regression problem is more complex, thus requiring a method that is appropriately non-linear.
Gradient boosted decision trees (GBDTs) and feed-forward neural networks (FFNNs) meet this criterion.
But why not just use one of these approaches?
Work by <xref ref-type="bibr" rid="bib1.bibx46" id="text.167"/> found that an ensemble of methods (SVR, GBDTs, and FFNN) outperformed each individual member.
And while SVR performed well in <xref ref-type="bibr" rid="bib1.bibx46" id="text.168"/>, the method does not scale to larger problems, which GBDTs and FFNNv2 are capable of, leading to our choice of the latter two.
One critique of GBDTs in this application may be that, being a tree-based method, it is not able to extrapolate.
However, we feel that the cluster-regression approach combined with the large number of training data for <inline-formula><mml:math id="M629" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> compensates for this shortcoming.
GBDTs provide also provide useful diagnostics, such as feature importances, that, when combined with the GRaCER approach, provide useful information about the spatial and seasonal importance of the proxies.</p>
</sec>
<sec id="Ch1.S5.SS2">
  <label>5.2</label><title>Why are pH uncertainties less well constrained?</title>
      <p id="d1e9031">One of the novel contributions of this study is that we are able to assert the validity of our results by comparing the bottom-up (propagated) with the top-down uncertainties (in situ comparisons).
Using this approach, we show that the uncertainty estimates of DIC are remarkably well constrained, with the top-down estimate being within 5 % of the bottom-up uncertainty estimate for the open ocean (Fig. <xref ref-type="fig" rid="Ch1.F7"/>b).
The same statistics for pH yields a 23 % difference between the two budgets.
The question is thus, why are pH bottom-up and top-down uncertainty estimates much less consistent?</p>
      <p id="d1e9036">To assess this problem from the bottom-up perspective, we need to consider the uncertainties of <inline-formula><mml:math id="M630" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA.
But given that the DIC uncertainty budgets are well constrained, we can, with some certainty, rule out the bottom-up uncertainty estimate as the source for the larger mismatch in pH.</p>
      <?pagebreak page796?><p id="d1e9053">The source of the mismatch must thus be driven primarily by uncertainties in the top-down perspective, where it may be that the representation error of pH is larger than for DIC.
We rule out the measurement error as a contributor to the mismatch, as the bias of the measurements (provided accurate calibration to reference samples) should be normally distributed around zero.
Thus, the representation error is the most likely candidate, due to the temperature- and pressure-sensitive nature of pH compared to the conservative nature of DIC with respect to the same variables <xref ref-type="bibr" rid="bib1.bibx24" id="paren.169"/>.
This is important given that our surface pH data from GLODAP can be as deep as 30 m, while <inline-formula><mml:math id="M631" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is typically measured at the intake depth of most ships, which is typically at 5 m <xref ref-type="bibr" rid="bib1.bibx2" id="paren.170"/>.
A basic sensitivity study of the variability of pH in the surface layer shows that the median standard deviation of surface pH at a single station is 0.004 units (even when normalized to a standard temperature and depth).
The same approach for DIC yields a standard deviation of 2 <inline-formula><mml:math id="M632" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M633" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
This suggests that the vertical mismatch error of GLODAP alone already explains a good deal of the discrepancy in the uncertainty budgeting.
One might reduce this uncertainty by placing tighter constraints on the definition of surface pH, limiting the depths between 3  and 15 m, for example, but at the loss of valuable test samples.
The representation error (described in Sect. <xref ref-type="sec" rid="Ch1.S3.SS1"/>) could further explain the remaining disparity in the pH error budget.</p>
</sec>
<sec id="Ch1.S5.SS3">
  <label>5.3</label><title>Can we reduce the total uncertainty?</title>
      <p id="d1e9107">The last two decades have seen major improvements in the accuracy and precision of the TA and <inline-formula><mml:math id="M634" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> measurements, leading to substantial reductions in the measurement errors. The introduction of certified reference materials for TA and a standardized approach for measuring <inline-formula><mml:math id="M635" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> with reference gases means that the uncertainties associated with the measurements are low.</p>
      <p id="d1e9138">In contrast, the prediction uncertainty is the largest contributor to the total uncertainty for both DIC and TA, suggesting that this could be a fruitful avenue to pursue.
However, current literature suggests that this is unlikely.
<xref ref-type="bibr" rid="bib1.bibx46" id="text.171"/> showed that, within a selection of six gap-filling methods, all achieved similar accuracy scores when compared with independent data, upon which the authors suggested that we have hit a wall.</p>
      <p id="d1e9144">This leaves the representation error, which contributes a moderate fraction to the total <inline-formula><mml:math id="M636" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and TA uncertainties in the open ocean and even less in the coastal ocean.
A back-of-the-envelope calculation shows that increasing the resolution of <inline-formula><mml:math id="M637" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> 4-fold (from monthly by 1<inline-formula><mml:math id="M638" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> <inline-formula><mml:math id="M639" display="inline"><mml:mo>×</mml:mo></mml:math></inline-formula> 1<inline-formula><mml:math id="M640" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> to 8-daily by 0.25<inline-formula><mml:math id="M641" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula> <inline-formula><mml:math id="M642" display="inline"><mml:mo>×</mml:mo></mml:math></inline-formula> 0.25<inline-formula><mml:math id="M643" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula>) could decrease the representation error by 2.5 <inline-formula><mml:math id="M644" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm (35 %) for the open ocean and 3.2 <inline-formula><mml:math id="M645" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for the coastal ocean (20 %).
This is perhaps not as much as expected, but these small gains are larger than those that are currently being made with regard to the prediction errors <xref ref-type="bibr" rid="bib1.bibx46" id="paren.172"/>.
This is not applicable for TA, where ungridded values are already used to train GRaCER and where the selection of the predictor variables is likely  more important.</p>
      <p id="d1e9245">Why are these gains smaller than hoped?
It may be that our gradient approach for calculating the representation error breaks down as the resolution increases due to the decreasing number of adjacent grid points.
This is hardly surprising considering that 78 % of grid cells in the SOCAT v2019 monthly gridded product are represented by sampling on a single day that falls within that period <xref ref-type="bibr" rid="bib1.bibx2" id="paren.173"/>.
Another possibility is that decreasing both the spatial and temporal resolution exposes the sharp mesoscale gradients that are otherwise averaged over at larger resolutions <xref ref-type="bibr" rid="bib1.bibx93 bib1.bibx81" id="paren.174"/>.
With the currently available data, it seems as if the reduction to be made in the total uncertainty by increasing the resolution of the prediction data will be small.</p>

      <?xmltex \floatpos{t}?><fig id="Ch1.F12" specific-use="star"><?xmltex \currentcnt{12}?><?xmltex \def\figurename{Figure}?><label>Figure 12</label><caption><p id="d1e9257">Attribution of DIC, TA, and temperature to the seasonal cycle of <inline-formula><mml:math id="M646" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>. The top two figures show the seasonal cycle of <inline-formula><mml:math id="M647" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and the drivers thereof for <bold>(a)</bold> high latitudes, <bold>(b)</bold> midlatitudes, and <bold>(c)</bold> low latitudes. These locations are shown with the markers in the Pacific Ocean in panel <bold>(c)</bold>. The map <bold>(d)</bold> shows the dominant driver of the seasonal cycle for each region calculated as the value with the maximum seasonal amplitude. </p></caption>
          <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f12.png"/>

        </fig>

</sec>
<sec id="Ch1.S5.SS4">
  <label>5.4</label><?xmltex \opttitle{Regional sensitivity of $p{\protect\chem{CO}}_{2}$ to driver variables}?><title>Regional sensitivity of <inline-formula><mml:math id="M648" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> to driver variables</title>
      <p id="d1e9333">Here we demonstrate one of the possible ways in which the OceanSODA-ETHZ data can be used to gain further insight into the marine carbonate system.</p>
      <p id="d1e9336">We decompose and attribute the mean seasonal cycle variability of <inline-formula><mml:math id="M649" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> to its drivers, namely TA, DIC, temperature, and salinity.
Past studies using observation based products have been limited to a simpler thermal/non-thermal decomposition of <inline-formula><mml:math id="M650" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> due to the lack of DIC and TA <xref ref-type="bibr" rid="bib1.bibx105 bib1.bibx64" id="paren.175"/>.
This is thus the first time that a full decomposition of <inline-formula><mml:math id="M651" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> has been applied to observation-based data for a global domain.
The decomposition is performed with pyCO2SYS by keeping all but one of the drivers constant (to the average) and assessing the influence on <inline-formula><mml:math id="M652" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> <xref ref-type="bibr" rid="bib1.bibx53" id="paren.176"/>.
This is similar to previous studies that applied an analogous decomposition to ocean simulation output <xref ref-type="bibr" rid="bib1.bibx75 bib1.bibx31 bib1.bibx36" id="paren.177"/>.
An important note to make, considering that we intend that OceanSODA-ETHZ is used for ocean acidification studies, is that the decomposition of pH would result in virtually the same contribution of the drivers <xref ref-type="bibr" rid="bib1.bibx24" id="paren.178"/>.</p>
      <?pagebreak page797?><p id="d1e9408">The seasonal amplitude of <inline-formula><mml:math id="M653" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is driven predominantly by changes in DIC in the high and equatorial latitudes and by temperature in the midlatitudes (Fig. <xref ref-type="fig" rid="Ch1.F12"/>d).
At first glance, these results may seem similar to those that the simpler thermal decomposition might result in, requiring only temperature.
However, in regions where the seasonal amplitude of <inline-formula><mml:math id="M654" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is smaller, the importance of TA becomes more apparent.
For example, at the midlatitude station (Fig. <xref ref-type="fig" rid="Ch1.F12"/>b), TA and DIC synergistically act to dampen the impact of temperature on <inline-formula><mml:math id="M655" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.
Conversely, at the equatorial station (Fig. <xref ref-type="fig" rid="Ch1.F12"/>c) the effect of TA on <inline-formula><mml:math id="M656" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> opposes that of the more dominant DIC.
Further, there are regions in the tropics where TA is the dominant driver due to the weak seasonal cycle of both temperature and DIC.</p>
</sec>
<sec id="Ch1.S5.SS5">
  <label>5.5</label><title>Recommendations for use</title>
      <p id="d1e9483">In order to use the OceanSODA-ETHZ product in an optimal manner, it is important to be aware of its strengths and weaknesses.</p>
      <p id="d1e9486"><?xmltex \hack{\newpage}?>The primary use of the OceanSODA-ETHZ data set is to determine and assess the seasonality, the interannual variations, and trends of ocean acidification thanks to its containing all relevant parameters of the marine carbonate system <xref ref-type="bibr" rid="bib1.bibx64 bib1.bibx65 bib1.bibx45 bib1.bibx58" id="paren.179"/>.
However, users of the OceanSODA-ETHZ product should be aware of the fact that data prior to the 1990s should be treated with care due to the paucity of SOCAT <inline-formula><mml:math id="M657" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> training data during this period <xref ref-type="bibr" rid="bib1.bibx95 bib1.bibx111" id="paren.180"/>.
This was recently demonstrated by <xref ref-type="bibr" rid="bib1.bibx111" id="text.181"/>, who used an ensemble of various regression approaches to show that the spread of <inline-formula><mml:math id="M658" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates prior to the 1990s is large due to the paucity of data.
Similarly, <xref ref-type="bibr" rid="bib1.bibx46" id="text.182"/> showed that <inline-formula><mml:math id="M659" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates prior to 1990 tend to have a slightly positive bias.</p>
      <?pagebreak page798?><p id="d1e9545">The product is also very well suited for assessing models. Thanks to the spatially resolved estimates of uncertainty for TA and <inline-formula><mml:math id="M660" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>), one cannot only assess the model–observation mismatches, but also weigh them with the appropriate uncertainties.</p>
      <p id="d1e9564">A strength of the OceanSODA-ETHZ product is that it extends further into the coastal margin than most previous studies <xref ref-type="bibr" rid="bib1.bibx54 bib1.bibx65 bib1.bibx22" id="paren.183"/>.
This is achieved (i) by including coastal observations during the training and (ii) by using a larger number of clusters compared to other clustering approaches <xref ref-type="bibr" rid="bib1.bibx65 bib1.bibx111" id="paren.184"/>. This permits us to better separate open-ocean and coastal variability through the inclusion of suitable variables in the clustering step (e.g., Chl <inline-formula><mml:math id="M661" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> for <inline-formula><mml:math id="M662" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, and see Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F17"/> to see a representation of cluster boundaries).
This gives us confidence in the coastal estimates of the climatological seasonal cycle. Our product is therefore comparable to that of <xref ref-type="bibr" rid="bib1.bibx66" id="text.185"/>, who blended separate coastal and open-ocean <inline-formula><mml:math id="M663" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> products into a single climatological product with a monthly resolution <xref ref-type="bibr" rid="bib1.bibx65 bib1.bibx68" id="paren.186"/>.</p>
      <p id="d1e9618">The total uncertainties of our estimates in the coastal ocean are considerably larger compared to the open-ocean estimates (Fig. <xref ref-type="fig" rid="Ch1.F7"/>).
This reflects the much higher spatiotemporal variability of the physical and chemical environment in the coastal ocean, leading to much higher variations in the marine carbonate system  <xref ref-type="bibr" rid="bib1.bibx68" id="paren.187"/>.
Since our predictor variables are only partially reflecting this variability, a large portion of the high total uncertainty is due to a high representation error (Table <xref ref-type="table" rid="Ch1.T3"/>).
Increasing the resolution of the products may improve the total uncertainty of coastal estimates as done by <xref ref-type="bibr" rid="bib1.bibx68" id="text.188"/>.
Until we arrive at this point, the OceanSODA-ETHZ data should be used with care in the coastal ocean.
Further, we recommend that researchers interested in the investigation of interannual variability and trends in the coastal ocean using the OceanSODA-ETHZ product should also look a the underlying  in situ data to gain a better understanding of the variability, trends, and uncertainties for the coastal region of interest.</p>
</sec>
</sec>
<sec id="Ch1.S6">
  <label>6</label><title>Code and data availability</title>
      <p id="d1e9640">Software for the GRaCER framework is available on Zenodo <ext-link xlink:href="https://doi.org/10.5281/zenodo.4455354" ext-link-type="DOI">10.5281/zenodo.4455354</ext-link> <xref ref-type="bibr" rid="bib1.bibx43" id="paren.189"/>. The OceanSODA-ETHZ data set is available at <ext-link xlink:href="https://doi.org/10.25921/m5wx-ja34" ext-link-type="DOI">10.25921/m5wx-ja34</ext-link> <xref ref-type="bibr" rid="bib1.bibx44" id="paren.190"/>.</p>
</sec>
<sec id="Ch1.S7" sec-type="conclusions">
  <label>7</label><title>Summary</title>
      <p id="d1e9663">Our approach for estimating TA and <inline-formula><mml:math id="M664" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is an evolution of the cluster-regression approach: we create an ensemble of estimates by repeating the cluster-regression step multiple times, each with a different variation of clustering.
We call this approach the Geospatial Random Cluster Ensemble Regression (GRaCER).
The result is an estimate that is more robust with better generalization, and the output does not have the discrete cluster boundaries that single-member cluster-regression approaches have.</p>
      <p id="d1e9680">We find that our estimates of TA are within the ballpark of previous methods with a prediction error (root mean square error) of 13 <inline-formula><mml:math id="M665" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M666" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for open-ocean estimates, while biases are <inline-formula><mml:math id="M667" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M668" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M669" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula>.
Taking into consideration all sources of error (measurement and representation errors), the total uncertainty is 17 <inline-formula><mml:math id="M670" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M671" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> for TA.
The prediction error for <inline-formula><mml:math id="M672" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> in the open ocean is 12 <inline-formula><mml:math id="M673" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm, also with a bias of <inline-formula><mml:math id="M674" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M675" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm.
Including the measurement and representation errors for <inline-formula><mml:math id="M676" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> results in a total uncertainty of 14 <inline-formula><mml:math id="M677" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm for the open ocean.
We estimate the total uncertainty of DIC and pH to be 19 <inline-formula><mml:math id="M678" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>mol kg<inline-formula><mml:math id="M679" display="inline"><mml:msup><mml:mi/><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> and 0.022 units when compared with independent GLODAPv2 data for the open ocean.
Finally, we compare the aforementioned top-down uncertainty estimates of DIC and pH with the bottom-up uncertainty estimates that are calculated by propagating the TA and <inline-formula><mml:math id="M680" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> total uncertainty estimates through the marine carbonate system.
This budgeting approach shows that we have a good grasp on the uncertainties of DIC for both the open and coastal oceans.
However, pH uncertainties are not as well resolved, most likely due to a mismatch in the representativity of the measured pH.</p>
      <p id="d1e9851">We demonstrate a use case of the OceanSODA-ETHZ data set in which we decompose the seasonal variability of <inline-formula><mml:math id="M681" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> into four driver components of DIC, TA, temperature, and salinity.
We find that DIC is the dominant driver in the high and equatorial latitudes, while temperature contributes the majority of the signal in the subtropics.
Importantly, DIC and temperature are antagonistic drivers of <inline-formula><mml:math id="M682" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, while alkalinity always acts in opposition to the stronger of the two primary drivers. We also show the strong constraints OceanSODA-ETHZ can pose on the long-term trends in ocean acidification.</p>
      <p id="d1e9882">Finally, OceanSODA-ETHZ will be maintained and updated for future work.</p><?xmltex \hack{\clearpage}?>
</sec>

      
      </body>
    <back><app-group>

<?pagebreak page799?><app id="App1.Ch1.S1">
  <?xmltex \currentcnt{A}?><label>Appendix A</label><title>Supplement to the methods</title>
<sec id="App1.Ch1.S1.SS1">
  <label>A1</label><?xmltex \opttitle{$p{\protect\chem{CO}}_{2}$ outlier removal}?><title><inline-formula><mml:math id="M683" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> outlier removal</title>
      <p id="d1e9918">The first outlier removal method requires the <inline-formula><mml:math id="M684" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> to be adjusted from the ship intake temperature to the satellite SST as described by <xref ref-type="bibr" rid="bib1.bibx38" id="text.191"/>:
            <disp-formula id="App1.Ch1.S1.E3" content-type="numbered"><label>A1</label><mml:math id="M685" display="block"><mml:mtable columnspacing="1em" class="split" rowspacing="0.2ex" displaystyle="true" columnalign="right left"><mml:mtr><mml:mtd><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SST</mml:mi></mml:msubsup></mml:mrow></mml:mtd><mml:mtd><mml:mrow><mml:mo>=</mml:mo><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd/><mml:mtd><mml:mrow><mml:mo>×</mml:mo><mml:mi>exp⁡</mml:mi><mml:mfenced close=")" open="("><mml:mrow><mml:mn mathvariant="normal">0.0433</mml:mn><mml:mo>⋅</mml:mo><mml:mfenced close=")" open="("><mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mi mathvariant="normal">SST</mml:mi></mml:msup><mml:mo>-</mml:mo><mml:msup><mml:mi>T</mml:mi><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msup></mml:mrow></mml:mfenced></mml:mrow></mml:mfenced><mml:mo>,</mml:mo></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:math></disp-formula>
          where the ship intake depth varies due to inconsistent depth between vessels and the water column state (e.g., well stratified or mixed).
Here <inline-formula><mml:math id="M686" display="inline"><mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mi mathvariant="normal">SST</mml:mi></mml:msup></mml:mrow></mml:math></inline-formula> is the foundation temperature given by the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) product <xref ref-type="bibr" rid="bib1.bibx41" id="paren.192"/>.
The OSTIA product is matched to the ungridded <inline-formula><mml:math id="M687" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula> at daily by <inline-formula><mml:math id="M688" display="inline"><mml:mrow><mml:mn mathvariant="normal">0.25</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">0.25</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> resolution <xref ref-type="bibr" rid="bib1.bibx41" id="paren.193"/>.
The corrected <inline-formula><mml:math id="M689" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is then binned to monthly by <inline-formula><mml:math id="M690" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula> resolution without weighting.
Data are excluded where the absolute difference between <inline-formula><mml:math id="M691" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SST</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula> and <inline-formula><mml:math id="M692" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msubsup><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mi mathvariant="normal">SOCAT</mml:mi></mml:msubsup></mml:mrow></mml:math></inline-formula> is larger than 40 <inline-formula><mml:math id="M693" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">µ</mml:mi></mml:mrow></mml:math></inline-formula>atm.</p>
      <p id="d1e10130">Secondly, we exclude data that lie outside the expected ranges for the monthly climatology of <inline-formula><mml:math id="M694" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.
The expected ranges are defined using the interquartile range outlier detection method for each pixel in a given month with the following equation:

                <disp-formula specific-use="gather" content-type="numbered"><mml:math id="M695" display="block"><mml:mtable displaystyle="true"><mml:mlabeledtr id="App1.Ch1.S1.E4"><mml:mtd><mml:mtext>A2</mml:mtext></mml:mtd><mml:mtd><mml:mrow><mml:mstyle displaystyle="true" class="stylechange"/><mml:mi mathvariant="normal">IQR</mml:mi><mml:mo>=</mml:mo><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">3</mml:mn></mml:msub><mml:mo>-</mml:mo><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">1</mml:mn></mml:msub><mml:mo>,</mml:mo></mml:mrow></mml:mtd></mml:mlabeledtr><mml:mlabeledtr id="App1.Ch1.S1.E5"><mml:mtd><mml:mtext>A3</mml:mtext></mml:mtd><mml:mtd><mml:mrow><mml:mstyle displaystyle="true" class="stylechange"/><mml:mtext>lower limit</mml:mtext><mml:mo>=</mml:mo><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">1</mml:mn></mml:msub><mml:mo>-</mml:mo><mml:mi mathvariant="normal">IQR</mml:mi><mml:mo>⋅</mml:mo><mml:mn mathvariant="normal">1.5</mml:mn><mml:mo>,</mml:mo></mml:mrow></mml:mtd></mml:mlabeledtr><mml:mlabeledtr id="App1.Ch1.S1.E6"><mml:mtd><mml:mtext>A4</mml:mtext></mml:mtd><mml:mtd><mml:mrow><mml:mstyle displaystyle="true" class="stylechange"/><mml:mtext>upper limit</mml:mtext><mml:mo>=</mml:mo><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">3</mml:mn></mml:msub><mml:mo>+</mml:mo><mml:mi mathvariant="normal">IQR</mml:mi><mml:mo>⋅</mml:mo><mml:mn mathvariant="normal">1.5</mml:mn><mml:mo>,</mml:mo></mml:mrow></mml:mtd></mml:mlabeledtr></mml:mtable></mml:math></disp-formula>

            where <inline-formula><mml:math id="M696" display="inline"><mml:mrow><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">1</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and <inline-formula><mml:math id="M697" display="inline"><mml:mrow><mml:msub><mml:mi>Q</mml:mi><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> are the 25th and 75th percentiles, respectively.
This approach is only applied where there are enough data present for a particular month of the year.</p>
</sec>
<sec id="App1.Ch1.S1.SS2">
  <label>A2</label><?xmltex \opttitle{Target variable: $\Delta p{\protect\chem{CO}}_{2}$ vs. $p{\protect\chem{CO}}_{2}$}?><title>Target variable: <inline-formula><mml:math id="M698" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> vs. <inline-formula><mml:math id="M699" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula></title>
      <p id="d1e10290">In this study, one of the avenues that we explored was to predict <inline-formula><mml:math id="M700" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> instead of <inline-formula><mml:math id="M701" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.
The motivation for predicting <inline-formula><mml:math id="M702" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is that it might allow new measurements from recent years to add new information about the seasonal cycle in regions where sampling was previously seasonally biased, e.g., the SOCCOM float data <xref ref-type="bibr" rid="bib1.bibx42" id="paren.194"/>.
<xref ref-type="bibr" rid="bib1.bibx14" id="text.195"/> showed that including the new information about the seasonal cycle of <inline-formula><mml:math id="M703" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> in machine learning estimates resulted in stronger winter outgassing, but their results could only show this for the period in which SOCCOM float data are present.
One of the reasons that machine learning approaches are not able to propagate this information back through time is that the larger <inline-formula><mml:math id="M704" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is anchored by the <inline-formula><mml:math id="M705" display="inline"><mml:mrow><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> concentrations that are used as a proxy.
Atmospheric <inline-formula><mml:math id="M706" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is required as a predictor variable to capture the interannual signal of <inline-formula><mml:math id="M707" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.
Predicting <inline-formula><mml:math id="M708" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> might thus allow one to remove atmospheric <inline-formula><mml:math id="M709" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> as a driver because the interannual term trend of <inline-formula><mml:math id="M710" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is removed.</p>
      <p id="d1e10459">The results appeared promising, but on further investigation we found that the regressions that were trying to predict <inline-formula><mml:math id="M711" display="inline"><mml:mrow><mml:mi mathvariant="normal">Δ</mml:mi><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> were not able to represent the increasing strength of the sink.
Furthermore, we found that the interannual variability of <inline-formula><mml:math id="M712" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> was reduced compared to results that include atmospheric <inline-formula><mml:math id="M713" display="inline"><mml:mrow><mml:mi>x</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> as a driver.
Ultimately, we abandoned the approach.</p>
</sec>
<sec id="App1.Ch1.S1.SS3">
  <label>A3</label><title>Hyper-parameter selection for regression methods</title>
<sec id="App1.Ch1.S1.SS3.SSS1">
  <label>A3.1</label><title>Total alkalinity: support vector regression</title>
      <p id="d1e10521">Hyper-parameters for the support vector regression (SVR) were chosen on a per-cluster basis using grid search cross validation, where unshuffled <inline-formula><mml:math id="M714" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-fold cross validation with five splits was used.
The <inline-formula><mml:math id="M715" display="inline"><mml:mi mathvariant="italic">ν</mml:mi></mml:math></inline-formula>SVR variety of the algorithm from the scikit-learn package in Python was used.
The parameters <inline-formula><mml:math id="M716" display="inline"><mml:mi>C</mml:mi></mml:math></inline-formula>, <inline-formula><mml:math id="M717" display="inline"><mml:mi mathvariant="italic">γ</mml:mi></mml:math></inline-formula>, and <inline-formula><mml:math id="M718" display="inline"><mml:mi mathvariant="italic">ν</mml:mi></mml:math></inline-formula> were selected.</p>
</sec>
<sec id="App1.Ch1.S1.SS3.SSS2">
  <label>A3.2</label><?xmltex \opttitle{$p{\protect\chem{CO}}_{2}$: gradient boosted decision trees}?><title><inline-formula><mml:math id="M719" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>: gradient boosted decision trees</title>
      <p id="d1e10581">We used the LightGBM package to perform the gradient boosted regression with decision trees (GBDTs).
The GBDT algorithm was trained using early stopping, which determines the number of trees used in the model – typically one of the most important hyper-parameters.
Every fifth year from 1987 to 2019 was set aside as the validation data used in the early stopping.
The total number of leaves per tree and the minimum number of training points per terminal leaf were both set to <inline-formula><mml:math id="M720" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mn mathvariant="normal">0.5</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula>, where <inline-formula><mml:math id="M721" display="inline"><mml:mi>N</mml:mi></mml:math></inline-formula> is the number of training points in a given cluster.
The number of leaves per tree determines the size of the tree.
The difference with LightGBM compared to other packages, like XGBoost, is that trees are grown on a leaf-wise basis rather than a level-wise basis, where the depth of the tree would be a more important hyper-parameter.
The minimum number of training points per terminal leaf determines how many points are aggregated in an estimate – a small number could thus result in overfitting.
The value <inline-formula><mml:math id="M722" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mn mathvariant="normal">0.5</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula> was determined experimentally with a single ensemble member, where the optimal values were determined with <inline-formula><mml:math id="M723" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-fold cross validation.
The results were in the ballpark of <inline-formula><mml:math id="M724" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mn mathvariant="normal">0.5</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula>, showing relatively low sensitivity to changes in these hyper-parameters.
Further, the learning rate was set to 0.2, and L1 and L2 regularization were both set to 20.</p>
</sec>
<sec id="App1.Ch1.S1.SS3.SSS3">
  <label>A3.3</label><?xmltex \opttitle{$p{\protect\chem{CO}}_{2}$: feed-forward neural network}?><title><inline-formula><mml:math id="M725" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>: feed-forward neural network</title>
      <p id="d1e10653">Given that the problem of solving <inline-formula><mml:math id="M726" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> is not very complex (i.e., it is within the capability of a single layer neural network), the multi-layer perceptron regressor from the scikit-learn package was used.
The size of the hidden layer for each cluster was determined by shuffled <inline-formula><mml:math id="M727" display="inline"><mml:mi>K</mml:mi></mml:math></inline-formula>-fold cross validation with five splits.
The maximum number of weights in a hidden layer was set to <inline-formula><mml:math id="M728" display="inline"><mml:mrow><mml:msup><mml:mi>N</mml:mi><mml:mn mathvariant="normal">0.55</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula>.
Back propagation was performed using the Adam optimizer.
The learning rate of the optimizer was selected in the cross-validation process.
Early<?pagebreak page800?> stopping was used to speed up the training process and prevent overfitting where one random third of the data were used in early stopping.</p>

      <?xmltex \floatpos{t}?><fig id="App1.Ch1.S1.F13" specific-use="star"><?xmltex \currentcnt{A1}?><?xmltex \def\figurename{Figure}?><label>Figure A1</label><caption><p id="d1e10690">Time series of TA (orange) and <inline-formula><mml:math id="M729" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> (blue) estimates, with the respective training data sets (GLODAPv2 and SOCAT). The dashed lines show the number of training data (right axis). The vertical line shows the year 1990, before which OceanSODA-ETHZ <inline-formula><mml:math id="M730" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> estimates tend to be larger than other gap-filling estimates of <inline-formula><mml:math id="M731" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>.</p></caption>
            <?xmltex \igopts{width=341.433071pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f13.png"/>

          </fig>

</sec>
</sec>
<sec id="App1.Ch1.S1.SS4">
  <label>A4</label><title>GRaCER mapped cluster metrics</title>
      <p id="d1e10750">One of the advantages of using the GRaCER approach is that any metric can be mapped from the results to the appropriate clusters, resulting in an ensemble of metric scores.
The possible metrics that can be applied include bias, root mean squared error, and mean absolute error.
Further, these metrics can be applied to test data, meaning that the resulting scores can be based on test scores – that is, data that are unseen by the model during the training process, thus giving a true representation of the uncertainty.
Given that the clusters used in this study are climatological, we can get fully mapped climatological estimates of uncertainty.</p>
      <p id="d1e10753">The uncertainty of TA remains fairly constant between summer and winter, with the Amazon plume showing increased uncertainty in the Northern Hemisphere summer (Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>c).</p>
      <p id="d1e10758">The seasonal difference is larger for <inline-formula><mml:math id="M732" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> than for TA.
For example, uncertainty in the Southern Ocean is much larger in the Southern Hemisphere summer (Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>b, DJF) compared with winter (Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F14"/>d, JJA).</p>

      <?xmltex \floatpos{t}?><fig id="App1.Ch1.S1.F14" specific-use="star"><?xmltex \currentcnt{A2}?><?xmltex \def\figurename{Figure}?><label>Figure A2</label><caption><p id="d1e10782">The Huber test scores mapped to the ensemble clusters for total alkalinity (TA) and <inline-formula><mml:math id="M733" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>. The top row shows Huber scores averaged for December, January, and February (DJF) and the bottom row June, July, and August (JJA). The Huber score is a blend between root mean squared error (RMSE) and mean absolute error (MAE), where MAE is applied to values that are considered outliers. Only test data are used to calculate these climatological scores, meaning that the scores are based on GLODAP2 and SOCAT data for TA and <inline-formula><mml:math id="M734" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> respectively. </p></caption>
          <?xmltex \igopts{width=398.338583pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f14.png"/>

        </fig>

      <p id="d1e10819">Similarly, the spatial distribution of feature importances can be determined with the GRaCER approach when using gradient boosted decision trees as the regression method (Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F15"/>). Each ensemble member has a feature importance assigned to a cluster. When averaged over the ensemble members, a smoothed climatological estimate of feature importance can be estimated.</p>

      <?xmltex \floatpos{t}?><fig id="App1.Ch1.S1.F15" specific-use="star"><?xmltex \currentcnt{A3}?><?xmltex \def\figurename{Figure}?><label>Figure A3</label><caption><p id="d1e10826">Feature importances determined by gradient boosted decision trees for <inline-formula><mml:math id="M735" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> predictions. A subset of four proxies are shown for the months of June, July, and August. The feature importances allow one to make informed decisions about the inclusion or exclusion of proxy variables. Here, temperature <bold>(a)</bold> is one of the more important features, Chl <inline-formula><mml:math id="M736" display="inline"><mml:mi>a</mml:mi></mml:math></inline-formula> is most important in the high northern latitudes, and the <inline-formula><mml:math id="M737" display="inline"><mml:mi>U</mml:mi></mml:math></inline-formula> and <inline-formula><mml:math id="M738" display="inline"><mml:mi>V</mml:mi></mml:math></inline-formula> components of the winds are important along the coastal regions, particularly the eastern boundary upwelling systems. </p></caption>
          <?xmltex \igopts{width=341.433071pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f15.png"/>

        </fig>

      <?xmltex \floatpos{t}?><fig id="App1.Ch1.S1.F16" specific-use="star"><?xmltex \currentcnt{A4}?><?xmltex \def\figurename{Figure}?><label>Figure A4</label><caption><p id="d1e10875">Ocean basin boundaries used in Table <xref ref-type="table" rid="Ch1.T5"/> as used by the RECCAP2 project (<uri>https://reccap2-ocean.github.io/regions/</uri>, last access: 21 November 2020). The Southern Ocean and North Atlantic boundaries are defined by biome boundaries defined in <xref ref-type="bibr" rid="bib1.bibx32" id="text.196"/>.</p></caption>
          <?xmltex \igopts{width=341.433071pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f16.png"/>

        </fig>

<sec id="App1.Ch1.S1.SS4.SSS1">
  <label>A4.1</label><title>Cluster boundaries</title>
      <p id="d1e10899">The GRaCER method introduces the idea of using an ensemble of clusters, thus removing the variability that may be introduced in the clustering step.
The location of the clusters varies from ensemble member to ensemble member.
This creates a high-variance–low-bias scenario that is used by other ensemble methods such as random forests <xref ref-type="bibr" rid="bib1.bibx11" id="paren.197"/>.
The location of these boundaries can give information about the mean distribution of the clusters.
For example, locations where there are no cluster boundaries (<inline-formula><mml:math id="M739" display="inline"><mml:mrow><mml:mo>≤</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:math></inline-formula> in Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F17"/>)  indicate cluster centers that fall within the same cluster for the majority of time steps and ensemble members.
While regions where cluster boundaries occur very often (<inline-formula><mml:math id="M740" display="inline"><mml:mrow><mml:mo>&gt;</mml:mo><mml:mn mathvariant="normal">8</mml:mn></mml:mrow></mml:math></inline-formula> in Fig. <xref ref-type="fig" rid="App1.Ch1.S1.F17"/>) are indicative of regions where boundaries are found for most time steps and ensemble members.</p>

      <?xmltex \floatpos{t}?><fig id="App1.Ch1.S1.F17" specific-use="star"><?xmltex \currentcnt{A5}?><?xmltex \def\figurename{Figure}?><label>Figure A5</label><caption><p id="d1e10931">Map of the position of cluster boundaries across all ensemble members and months for <bold>(a)</bold> total alkalinity and <bold>(b)</bold> <inline-formula><mml:math id="M741" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:msub><mml:mrow class="chem"><mml:mi mathvariant="normal">CO</mml:mi></mml:mrow><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>. The white regions indicate locations that belong almost exclusively to the same cluster. Dark regions show where cluster boundaries are persistent.</p></caption>
            <?xmltex \igopts{width=341.433071pt}?><graphic xlink:href="https://essd.copernicus.org/articles/13/777/2021/essd-13-777-2021-f17.png"/>

          </fig>

<?xmltex \hack{\clearpage}?>
</sec>
</sec>
</app>
  </app-group><notes notes-type="authorcontribution"><title>Author contributions</title>

      <p id="d1e10968">LG and NG conceived of the study and developed the method. LG performed the analysis and testing of the method and wrote the paper with substantial input by NG.</p>
  </notes><notes notes-type="competinginterests"><title>Competing interests</title>

      <p id="d1e10974">The authors declare that they have no conflict of interest.</p>
  </notes><ack><title>Acknowledgements</title><p id="d1e10980">We are deeply indebted to the scientists who sampled, analyzed, and contributed to the global databases for ocean carbon data, namely the Surface Ocean CO<inline-formula><mml:math id="M742" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Atlas (SOCAT) and the Global Ocean Analysis Project (GLODAP). We also thank the funding agencies that made these efforts possible. SOCAT and GLODAP are international efforts, endorsed by the International Ocean Carbon Coordination Project (IOCCP), the Surface Ocean Lower Atmosphere Study (SOLAS), and the Integrated Marine Biosphere Research (IMBeR) program. We thank Meike Vogt, Fabio Benedetti, and Peter Land for their valuable input and discussions. We are indebted to Jamie Shutler for initiating and leading the OceanSODA project and for his substantial input to our quantification of the uncertainty budget.</p></ack><notes notes-type="financialsupport"><title>Financial support</title>

      <p id="d1e10994">This research has been supported by the European Space Agency (OceanSODA project, grant no. 4000112091/14/I-LG), the European Commission (COMFORT project, grant no. 820989), and the Horizon 2020 (4C project, grant no. 821003).</p>
  </notes><notes notes-type="reviewstatement"><title>Review statement</title>

      <p id="d1e11001">This paper was edited by Giuseppe M. R. Manzella and reviewed by two anonymous referees.</p>
  </notes><ref-list>
    <title>References</title>

      <ref id="bib1.bibx1"><?xmltex \def\ref@label{{Bakker et~al.(2008)}}?><label>Bakker et al.(2008)</label><?label Bakker2008?><mixed-citation>Bakker, D. C. E., Hoppema, M., Schröder, M., Geibert, W., and de Baar, H. J. W.: A rapid transition from ice covered CO<inline-formula><mml:math id="M743" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>–rich waters to a biologically mediated CO<inline-formula><mml:math id="M744" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> sink in the eastern Weddell Gyre, Biogeosciences, 5, 1373–1386, <ext-link xlink:href="https://doi.org/10.5194/bg-5-1373-2008" ext-link-type="DOI">10.5194/bg-5-1373-2008</ext-link>, 2008.</mixed-citation></ref>
      <ref id="bib1.bibx2"><?xmltex \def\ref@label{{Bakker et~al.(2016)}}?><label>Bakker et al.(2016)</label><?label Bakker2016?><mixed-citation>Bakker, D. C., Pfeil, B., Landa, C. S., Metzl, N., O'Brien, K. M., Olsen, A.,
Smith, K., Cosca, C., Harasawa, S., Jones, S. D., Nakaoka, S.-I., Nojiri, Y.,
Schuster, U., Steinhoff, T., Sweeney, C., Takahashi, T. T., Tilbrook, B.,
Wada, C., Wanninkhof, R. H., Alin, S. R., Balestrini, C. F., Barbero, L.,
Bates, N. R., Bianchi, A. A., Bonou, F., Boutin, J., Bozec, Y., Burger,
E. F., Cai, W.-J., Castle, R. D., Chen, L., Chierici, M., Currie, K., Evans,
W., Featherstone, C., Feely, R. A., Fransson, A., Goyet, C., Greenwood, N.,
Gregor, L., Hankin, S., Hardman-Mountford, N. J., Harlay, J., Hauck, J.,
Hoppema, M., Humphreys, M. P., Hunt, C. W., Huss, B., Ibánhez, J.
S. P., Johannessen, T., Keeling, R., Kitidis, V., Körtzinger, A.,
Kozyr, A., Krasakopoulou, E., Kuwata, A., Landschützer, P., Lauvset,
S. K., Lefèvre, N., Lo Monaco, C., Manke, A., Mathis, J. T.,
Merlivat, L., Millero, F. J., Monteiro, P. M. S., Munro, D. R., Murata, A.,
Newberger, T., Omar, A. M., Ono, T., Paterson, K., Pearce, D., Pierrot, D.,
Robbins, L. L., Saito, S., Salisbury, J., Schlitzer, R., Schneider, B.,
Schweitzer, R., Sieger, R., Skjelvan, I., Sullivan, K. F., Sutherland, S. C.,
Sutton, A. J., Tadokoro, K., Telszewski, M., Tuma, M., Van Heuven, S. M.,
Vandemark, D., Ward, B., Watson, A. J., and Xu, S.: A multi-decade record of
high-quality <inline-formula><mml:math id="M745" display="inline"><mml:mi>f</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M746" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> data in version 3 of the Surface Ocean CO<inline-formula><mml:math id="M747" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Atlas (SOCAT),
Earth Syst. Sci. Data, 8, 383–413, <ext-link xlink:href="https://doi.org/10.5194/essd-8-383-2016" ext-link-type="DOI">10.5194/essd-8-383-2016</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx3"><?xmltex \def\ref@label{{Bates and Peters(2007)}}?><label>Bates and Peters(2007)</label><?label Bates2007?><mixed-citation>Bates, N. R. and Peters, A. J.: The contribution of atmospheric acid
deposition to ocean acidification in the subtropical North Atlantic Ocean,
Mar. Chem., 107, 547–558, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2007.08.002" ext-link-type="DOI">10.1016/j.marchem.2007.08.002</ext-link>, 2007.</mixed-citation></ref>
      <ref id="bib1.bibx4"><?xmltex \def\ref@label{{Bates et~al.(2014)}}?><label>Bates et al.(2014)</label><?label Bates2014a?><mixed-citation>Bates, N. R., Astor, Y. M., Church, M. J., Currie, K., Dore, J. E.,
González-Dávila, M., Lorenzoni, L., Muller-Karger, F., Olafsson,
J., and Santana-Casiano, J. M.: A time-series view of changing surface ocean
chemistry due to ocean uptake of anthropogenic CO<inline-formula><mml:math id="M748" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and ocean acidification,
Oceanography, 27, 126–141, <ext-link xlink:href="https://doi.org/10.5670/oceanog.2014.16" ext-link-type="DOI">10.5670/oceanog.2014.16</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx5"><?xmltex \def\ref@label{{Bednar{\v{s}}ek et~al.(2019)}}?><label>Bednaršek et al.(2019)</label><?label Bednarsek2019?><mixed-citation>Bednaršek, N., Feely, R. A., Howes, E. L., Hunt, B. P. V., Kessouri, F.,
León, P., Lischka, S., Maas, A. E., McLaughlin, K., Nezlin, N. P.,
Sutula, M., and Weisberg, S. B.: Systematic Review and Meta-Analysis Toward
Synthesis of Thresholds of Ocean Acidification Impacts on Calcifying
Pteropods and Interactions With Warming, Front. Mar. Sci., 6, 227 pp.,
<ext-link xlink:href="https://doi.org/10.3389/fmars.2019.00227" ext-link-type="DOI">10.3389/fmars.2019.00227</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx6"><?xmltex \def\ref@label{{Bittig et~al.(2018)}}?><label>Bittig et al.(2018)</label><?label Bittig2018a?><mixed-citation>Bittig, H. C., Steinhoff, T., Claustre, H., Fiedler, B., Williams, N. L.,
Sauzède, R., Körtzinger, A., and Gattuso, J.-P.: An alternative
to static climatologies: Robust estimation of open ocean CO<inline-formula><mml:math id="M749" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> variables and
nutrient concentrations from T, S, and O<inline-formula><mml:math id="M750" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> data using Bayesian neural
networks, Front. Mar. Sci., 5, 1–29,
<ext-link xlink:href="https://doi.org/10.3389/fmars.2018.00328" ext-link-type="DOI">10.3389/fmars.2018.00328</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx7"><?xmltex \def\ref@label{{Bockmon and Dickson(2015)}}?><label>Bockmon and Dickson(2015)</label><?label Bockmon2015?><mixed-citation>Bockmon, E. E. and Dickson, A. G.: An inter-laboratory comparison assessing
the quality of seawater carbon dioxide measurements, Mar. Chem., 171,
36–43, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2015.02.002" ext-link-type="DOI">10.1016/j.marchem.2015.02.002</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx8"><?xmltex \def\ref@label{{Bopp et~al.(2013)}}?><label>Bopp et al.(2013)</label><?label Bopp2013?><mixed-citation>Bopp, L., Resplandy, L., Orr, J. C., Doney, S. C., Dunne, J. P., Gehlen, M.,
Halloran, P. R., Heinze, C., Ilyina, T., Séférian, R., Tjiputra,
J., and Vichi, M.: Multiple stressors of ocean ecosystems in the 21st
century: Projections with CMIP5 models, Biogeosciences, 10, 6225–6245,
<ext-link xlink:href="https://doi.org/10.5194/bg-10-6225-2013" ext-link-type="DOI">10.5194/bg-10-6225-2013</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx9"><?xmltex \def\ref@label{{Boutin et~al.(2018)}}?><label>Boutin et al.(2018)</label><?label Boutin2018?><mixed-citation>Boutin, J., Vergely, J. L., Marchand, S., D'Amico, F., Hasson, A.,
Kolodziejczyk, N., Reul, N., Reverdin, G., and Vialard, J.: New SMOS Sea
Surface Salinity with reduced systematic errors and improved variability,
Remote Sens. Environ., 214, 115–134,
<ext-link xlink:href="https://doi.org/10.1016/j.rse.2018.05.022" ext-link-type="DOI">10.1016/j.rse.2018.05.022</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx10"><?xmltex \def\ref@label{{Boyer et~al.(2013)}}?><label>Boyer et al.(2013)</label><?label Boyer2013?><mixed-citation>Boyer, T. P., Antonov, J. I., Baranova, O. K., Garcia, H. E., Johnson, D. R.,
Mishonov, A. V., O'Brien, T. D., Seidov, D., Smolyar, I. I., Zweng, M. M.,
Paver, C. R., Locarnini, R. A., Reagan, J. R., Forgy, C. C., Grodsky, A., and
Levitus, S.: World ocean database 2013, Tech. Rep., National Oceanographic
Data Center (U.S.), Ocean Climate Laboratory, 208 pp., <ext-link xlink:href="https://doi.org/10.7289/V5NZ85MT" ext-link-type="DOI">10.7289/V5NZ85MT</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx11"><?xmltex \def\ref@label{{Breiman(2001)}}?><label>Breiman(2001)</label><?label Breiman2001?><mixed-citation>Breiman, L.: Random forests, Mach. Learn., 45, 5–32,
<ext-link xlink:href="https://doi.org/10.1023/A:1010933404324" ext-link-type="DOI">10.1023/A:1010933404324</ext-link>, 2001.</mixed-citation></ref>
      <ref id="bib1.bibx12"><?xmltex \def\ref@label{{Broecker and Peng(1974)}}?><label>Broecker and Peng(1974)</label><?label Broecker1974?><mixed-citation>Broecker, W. S. and Peng, T.-H.: Gas exchange rates between air and
sea, Tellus, 26, 21–35, <ext-link xlink:href="https://doi.org/10.3402/tellusa.v26i1-2.9733" ext-link-type="DOI">10.3402/tellusa.v26i1-2.9733</ext-link>, 1974.</mixed-citation></ref>
      <?pagebreak page804?><ref id="bib1.bibx13"><?xmltex \def\ref@label{{Broull{\'{o}}n et~al.(2018)}}?><label>Broullón et al.(2018)</label><?label Broullon2018?><mixed-citation>Broullón, D., Pérez, F. F., Velo, A., Hoppema, M., Olsen, A., Takahashi, T., Key, R. M., Tanhua, T., González-Dávila, M., Jeansson, E., Kozyr, A., and van Heuven, S. M. A. C.: A global monthly climatology of total alkalinity: a neural network approach, Earth Syst. Sci. Data, 11, 1109–1127, <ext-link xlink:href="https://doi.org/10.5194/essd-11-1109-2019" ext-link-type="DOI">10.5194/essd-11-1109-2019</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx14"><?xmltex \def\ref@label{{Bushinsky et~al.(2019)}}?><label>Bushinsky et al.(2019)</label><?label Bushinsky2019?><mixed-citation>Bushinsky, S. M., Landschützer, P., Rödenbeck, C., Gray, A. R.,
Baker, D., Mazloff, M. R., Resplandy, L., Johnson, K. S., and Sarmiento,
J. L.: Reassessing Southern Ocean air-sea CO<inline-formula><mml:math id="M751" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux estimates with the
addition of biogeochemical float observations, Global Biogeochem. Cy.,
33, 1370–1388, <ext-link xlink:href="https://doi.org/10.1029/2019GB006176" ext-link-type="DOI">10.1029/2019GB006176</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx15"><?xmltex \def\ref@label{{Carter et~al.(2016)}}?><label>Carter et al.(2016)</label><?label Carter2016?><mixed-citation>Carter, B. R., Williams, N. L., Gray, A. R., and Feely, R. A.: Locally
interpolated alkalinity regression for global alkalinity estimation,
Limnol. Oceanogr.-Method., 14, 268–277, <ext-link xlink:href="https://doi.org/10.1002/lom3.10087" ext-link-type="DOI">10.1002/lom3.10087</ext-link>,
2016.</mixed-citation></ref>
      <ref id="bib1.bibx16"><?xmltex \def\ref@label{{Carter et~al.(2018)}}?><label>Carter et al.(2018)</label><?label Carter2018?><mixed-citation>Carter, B. R., Feely, R. A., Williams, N. L., Dickson, A. G., Fong, M. B., and
Takeshita, Y.: Updated methods for global locally interpolated estimation of
alkalinity, pH, and nitrate, Limnol. Oceanogr.-Method., 16,
119–131, <ext-link xlink:href="https://doi.org/10.1002/lom3.10232" ext-link-type="DOI">10.1002/lom3.10232</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx17"><?xmltex \def\ref@label{{Carton et~al.(2018)}}?><label>Carton et al.(2018)</label><?label Carton2018?><mixed-citation>Carton, J. A., Chepurin, G. A., and Chen, L.: SODA3: A new ocean climate
reanalysis, J. Clim., 31, 6967–6983,
<ext-link xlink:href="https://doi.org/10.1175/JCLI-D-17-0149.1" ext-link-type="DOI">10.1175/JCLI-D-17-0149.1</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx18"><?xmltex \def\ref@label{{Cheng et~al.(2020)}}?><label>Cheng et al.(2020)</label><?label Cheng2020?><mixed-citation>Cheng, L., Trenberth, K. E., Gruber, N., Abraham, J. P., Fasullo, J. T., Li,
G., Mann, M. E., Zhao, X., and Zhu, J.: Improved Estimates of Changes in
Upper Ocean Salinity and the Hydrological Cycle, J. Clim., 33,
10357–10381, <ext-link xlink:href="https://doi.org/10.1175/JCLI-D-20-0366.1" ext-link-type="DOI">10.1175/JCLI-D-20-0366.1</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx19"><?xmltex \def\ref@label{{Claustre et~al.(2020)}}?><label>Claustre et al.(2020)</label><?label Claustre2020?><mixed-citation>Claustre, H., Johnson, K. S., and Takeshita, Y.: Observing the Global Ocean
with Biogeochemical-Argo, Ann. Rev. Mar. Sci., 12, 23–48,
<ext-link xlink:href="https://doi.org/10.1146/annurev-marine-010419-010956" ext-link-type="DOI">10.1146/annurev-marine-010419-010956</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx20"><?xmltex \def\ref@label{{Cooley and Doney(2009)}}?><label>Cooley and Doney(2009)</label><?label Cooley2009?><mixed-citation>Cooley, S. R. and Doney, S. C.: Anticipating ocean acidification's economic
consequences for commercial fisheries, Environ. Res. Lett., 4, 8 pp.,
<ext-link xlink:href="https://doi.org/10.1088/1748-9326/4/2/024007" ext-link-type="DOI">10.1088/1748-9326/4/2/024007</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx21"><?xmltex \def\ref@label{{Hersbach(2018)}}?><label>Hersbach(2018)</label><?label C3S2017?><mixed-citation>Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horányi, A., Muñoz Sabater, J., Nicolas, J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., and Thépaut, J.-N.: ERA5 hourly data on single levels from 1979 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS), <ext-link xlink:href="https://doi.org/10.24381/cds.adbb2d47" ext-link-type="DOI">10.24381/cds.adbb2d47</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx22"><?xmltex \def\ref@label{{Denvil-Sommer et~al.(2019)}}?><label>Denvil-Sommer et al.(2019)</label><?label DenvilSommer2019?><mixed-citation>Denvil-Sommer, A., Gehlen, M., Vrac, M., and Mejia, C.: LSCE-FFNN-v1: a
two-step neural network model for the reconstruction of surface ocean <inline-formula><mml:math id="M752" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M753" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
over the global ocean, Geosci. Model Dev., 12, 2091–2105,
<ext-link xlink:href="https://doi.org/10.5194/gmd-12-2091-2019" ext-link-type="DOI">10.5194/gmd-12-2091-2019</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx23"><?xmltex \def\ref@label{{Dickson et~al.(1990)}}?><label>Dickson et al.(1990)</label><?label Dickson1990?><mixed-citation>Dickson, A. G., Wesolowski, D. J., Palmer, D. A., and Mesmer, R. E.:
Dissociation constant of bisulfate ion in aqueous sodium chloride solutions
to 250 <inline-formula><mml:math id="M754" display="inline"><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:math></inline-formula>C, J. Phys. Chem., 94, 7978–7985,
<ext-link xlink:href="https://doi.org/10.1021/j100383a042" ext-link-type="DOI">10.1021/j100383a042</ext-link>, 1990.</mixed-citation></ref>
      <ref id="bib1.bibx24"><?xmltex \def\ref@label{{Dickson et~al.(2007)}}?><label>Dickson et al.(2007)</label><?label Dickson2007a?><mixed-citation>Dickson, A. G., Sabine, C. L., and Christian, J. R. (Eds.): Guide to best
practices for ocean CO<inline-formula><mml:math id="M755" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> measurements, North Pacific Marine Science
Organization, Sidney, British Columbia, 3rd Edn., available at:
<uri>https://www.oceanbestpractices.net/handle/11329/249</uri> (last access: 22 May 2019), 2007.</mixed-citation></ref>
      <ref id="bib1.bibx25"><?xmltex \def\ref@label{{Dlugokencky et~al.(2019)}}?><label>Dlugokencky et al.(2019)</label><?label Dlugokencky2019a?><mixed-citation>Dlugokencky, E., Thoning, K., Lang, P., and Tans, P.: NOAA Greenhouse Gas
Reference from Atmospheric Carbon Dioxide Dry Air Mole Fractions from the
NOAA ESRL Carbon Cycle Cooperative Global Air Sampling Network, available at:
<uri>ftp://aftp.cmdl.noaa.gov/data/trace_gases/co2/flask/surface/</uri> (last accessed: 14 September 2020),
2019.</mixed-citation></ref>
      <ref id="bib1.bibx26"><?xmltex \def\ref@label{{Doney et~al.(2009)Doney, Fabry, Feely, and Kleypas}}?><label>Doney et al.(2009)Doney, Fabry, Feely, and Kleypas</label><?label Doney2009?><mixed-citation>Doney, S. C., Fabry, V. J., Feely, R. A., and Kleypas, J. A.: Ocean
Acidification: The Other CO<inline-formula><mml:math id="M756" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Problem, Ann. Rev. Mar. Sci., 1,
169–192, <ext-link xlink:href="https://doi.org/10.1146/annurev.marine.010908.163834" ext-link-type="DOI">10.1146/annurev.marine.010908.163834</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx27"><?xmltex \def\ref@label{{Doney et~al.(2020)}}?><label>Doney et al.(2020)</label><?label Doney2020?><mixed-citation>Doney, S. C., Busch, D. S., Cooley, S. R., and Kroeker, K. J.: The Impacts of
Ocean Acidification on Marine Ecosystems and Reliant Human Communities,
Ann. Rev. Environ. Resour., 45, 1–30,
<ext-link xlink:href="https://doi.org/10.1146/annurev-environ-012320-083019" ext-link-type="DOI">10.1146/annurev-environ-012320-083019</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx28"><?xmltex \def\ref@label{{Dore et~al.(2009)}}?><label>Dore et al.(2009)</label><?label Dore2009?><mixed-citation>Dore, J. E., Lukas, R., Sadler, D. W., Church, M. J., and Karl, D. M.:
Physical and biogeochemical modulation of ocean acidification in the central
North Pacific, P. Natl. Acad. Sci. USA, 106,
12235–12240, <ext-link xlink:href="https://doi.org/10.1073/pnas.0906044106" ext-link-type="DOI">10.1073/pnas.0906044106</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx29"><?xmltex \def\ref@label{{Fabry et~al.(2008)}}?><label>Fabry et al.(2008)</label><?label Fabry2008?><mixed-citation>Fabry, V. J., Seibel, B. A., Feely, R. A., and Orr, J. C.: Impacts of ocean
acidification on marine fauna and ecosystem processes, ICES J.
Mar. Sci., 65, 414–432, <ext-link xlink:href="https://doi.org/10.1093/icesjms/fsn048" ext-link-type="DOI">10.1093/icesjms/fsn048</ext-link>, 2008.</mixed-citation></ref>
      <ref id="bib1.bibx30"><?xmltex \def\ref@label{{Fabry et~al.(2009)}}?><label>Fabry et al.(2009)</label><?label Fabry2009?><mixed-citation>Fabry, V. J., McClintock, J. B., Mathis, J. T., and Grebmeier, J. M.: Ocean
acidification at high latitudes: The Bellwether, Oceanography, 22, 160–171,
<ext-link xlink:href="https://doi.org/10.5670/oceanog.2009.105" ext-link-type="DOI">10.5670/oceanog.2009.105</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx31"><?xmltex \def\ref@label{{Fassbender et~al.(2018)}}?><label>Fassbender et al.(2018)</label><?label Fassbender2018a?><mixed-citation>Fassbender, A. J., Rodgers, K. B., Palevsky, H. I., and Sabine, C. L.:
Seasonal Asymmetry in the Evolution of Surface Ocean <inline-formula><mml:math id="M757" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M758" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and pH
Thermodynamic Drivers and the Influence on Sea-Air CO<inline-formula><mml:math id="M759" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Flux, Global
Biogeochem. Cy., 32, 1476–1497, <ext-link xlink:href="https://doi.org/10.1029/2017GB005855" ext-link-type="DOI">10.1029/2017GB005855</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx32"><?xmltex \def\ref@label{{Fay and McKinley(2014)}}?><label>Fay and McKinley(2014)</label><?label Fay2014?><mixed-citation>Fay, A. R. and McKinley, G. A.: Global open-ocean biomes: Mean and temporal
variability, Earth Syst. Sc. Data, 6, 273–284,
<ext-link xlink:href="https://doi.org/10.5194/essd-6-273-2014" ext-link-type="DOI">10.5194/essd-6-273-2014</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx33"><?xmltex \def\ref@label{{Feely et~al.(2009)}}?><label>Feely et al.(2009)</label><?label Feely2009?><mixed-citation>Feely, R. A., Doney, S., and Cooley, S.: Ocean Acidification: Present
Conditions and Future Changes in a High-CO<inline-formula><mml:math id="M760" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> World, Oceanography, 22, 36–47,
<ext-link xlink:href="https://doi.org/10.5670/oceanog.2009.95" ext-link-type="DOI">10.5670/oceanog.2009.95</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx34"><?xmltex \def\ref@label{{Franco et~al.(2018)}}?><label>Franco et al.(2018)</label><?label Franco2018a?><mixed-citation>Franco, A. C., Gruber, N., Frölicher, T. L., and Kropuenske Artman, L.:
Contrasting Impact of Future CO<inline-formula><mml:math id="M761" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Emission Scenarios on the Extent of CaCO<inline-formula><mml:math id="M762" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula>
Mineral Undersaturation in the Humboldt Current System, J.
Geophys. Res.-Ocean., 123, 2018–2036, <ext-link xlink:href="https://doi.org/10.1002/2018JC013857" ext-link-type="DOI">10.1002/2018JC013857</ext-link>,
2018.</mixed-citation></ref>
      <?pagebreak page805?><ref id="bib1.bibx35"><?xmltex \def\ref@label{{Friedlingstein et~al.(2019)}}?><label>Friedlingstein et al.(2019)</label><?label Friedlingstein2019?><mixed-citation>Friedlingstein, P., Jones, M. W., O'Sullivan, M., Andrew, R. M., Hauck, J.,
Peters, G. P., Peters, W., Pongratz, J., Sitch, S., Le Quéré,
C., DBakker, O. C., Canadell1, J. G., Ciais1, P., Jackson, R. B., Anthoni1,
P., Barbero, L., Bastos, A., Bastrikov, V., Becker, M., Bopp, L., Buitenhuis,
E., Chandra, N., Chevallier, F., Chini, L. P., Currie, K. I., Feely, R. A.,
Gehlen, M., Gilfillan, D., Gkritzalis, T., Goll, D. S., Gruber, N.,
Gutekunst, S., Harris, I., Haverd, V., Houghton, R. A., Hurtt, G., Ilyina,
T., Jain, A. K., Joetzjer, E., Kaplan, J. O., Kato, E., Goldewijk, K. K.,
Korsbakken, J. I., Landschützer, P., Lauvset, S. K., Lefèvre, N.,
Lenton, A., Lienert, S., Lombardozzi, D., Marland, G., McGuire, P. C.,
Melton, J. R., Metzl, N., Munro, D. R., Nabel, J. E., Nakaoka, S. I., Neill,
C., Omar, A. M., Ono, T., Peregon, A., Pierrot, D., Poulter, B., Rehder, G.,
Resplandy, L., Robertson, E., Rödenbeck, C., Séférian, R.,
Schwinger, J., Smith, N., Tans, P. P., Tian, H., Tilbrook, B., Tubiello,
F. N., Van Der Werf, G. R., Wiltshire, A. J., and Zaehle, S.: Global
carbon budget 2019, Earth Syst. Sci. Data, 11, 1783–1838,
<ext-link xlink:href="https://doi.org/10.5194/essd-11-1783-2019" ext-link-type="DOI">10.5194/essd-11-1783-2019</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx36"><?xmltex \def\ref@label{{Gallego et~al.(2018)}}?><label>Gallego et al.(2018)</label><?label Gallego2018?><mixed-citation>Gallego, M. A., Timmermann, A., Friedrich, T., and Zeebe, R. E.: Drivers of
future seasonal cycle changes in oceanic <inline-formula><mml:math id="M763" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M764" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, Biogeosciences, 15,
5315–5327, <ext-link xlink:href="https://doi.org/10.5194/bg-15-5315-2018" ext-link-type="DOI">10.5194/bg-15-5315-2018</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx37"><?xmltex \def\ref@label{{Gattuso et~al.(1998)}}?><label>Gattuso et al.(1998)</label><?label Gattuso1998a?><mixed-citation>Gattuso, J.-P., Frankignoulle, M., Bourge, I., Romaine, S., and Buddemeier,
R. W.: Effect of calcium carbonate saturation of seawater on coral
calcification, Glob. Planet. Change, 18, 37–46,
<ext-link xlink:href="https://doi.org/10.1016/S0921-8181(98)00035-6" ext-link-type="DOI">10.1016/S0921-8181(98)00035-6</ext-link>, 1998.</mixed-citation></ref>
      <ref id="bib1.bibx38"><?xmltex \def\ref@label{{Goddijn‐Murphy et~al.(2015)}}?><label>Goddijn‐Murphy et al.(2015)</label><?label GoddijnMurphy2015?><mixed-citation>Goddijn‐Murphy, L., Woolf, D. K., Land, P. E., Shutler, J. D., and Donlon,
C. J.: The OceanFlux Greenhouse Gases methodology for deriving a sea surface
climatology of CO<inline-formula><mml:math id="M765" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> fugacity in support of air-sea gas flux studies, Ocean
Sci., 11, 519–541, <ext-link xlink:href="https://doi.org/10.5194/os-11-519-2015" ext-link-type="DOI">10.5194/os-11-519-2015</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx39"><?xmltex \def\ref@label{{Good et~al.(2013)}}?><label>Good et al.(2013)</label><?label Good2013?><mixed-citation>Good, S. A., Martin, M. J., and Rayner, N. A.: EN4: Quality controlled ocean
temperature and salinity profiles and monthly objective analyses with
uncertainty estimates, J. Geophys. Res.-Ocean., 118,
6704–6716, <ext-link xlink:href="https://doi.org/10.1002/2013JC009067" ext-link-type="DOI">10.1002/2013JC009067</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx40"><?xmltex \def\ref@label{{Good et~al.(2019)}}?><label>Good et al.(2019)</label><?label Good2019?><mixed-citation>Good, S. A., Embury, O., Bulgin, C., and Mittaz, J.: ESA Sea Surface
Temperature Climate Change Initiative (SST_cci): Level 4 Analysis Climate
Data Record, version 2.1, <ext-link xlink:href="https://doi.org/10.5285/aced40d7cb964f23a0fd3e85772f2d48" ext-link-type="DOI">10.5285/aced40d7cb964f23a0fd3e85772f2d48</ext-link>,
2019.</mixed-citation></ref>
      <ref id="bib1.bibx41"><?xmltex \def\ref@label{{Good et~al.(2020)}}?><label>Good et al.(2020)</label><?label Good2020?><mixed-citation>Good, S. A., Fiedler, E., Mao, C., Martin, M. J., Maycock, A., Reid, R.,
Roberts-Jones, J., Searle, T., Waters, J., While, J., and Worsfold, M.: The
current configuration of the OSTIA system for operational production of
foundation sea surface temperature and ice concentration analyses, Remote
Sens., 12, 1–20, <ext-link xlink:href="https://doi.org/10.3390/rs12040720" ext-link-type="DOI">10.3390/rs12040720</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx42"><?xmltex \def\ref@label{{Gray et~al.(2018)}}?><label>Gray et al.(2018)</label><?label Gray2018?><mixed-citation>Gray, A. R., Johnson, K. S., Bushinsky, S. M., Riser, S. C., Russell, J. L.,
Talley, L. D., Wanninkhof, R. H., Williams, N. L., and Sarmiento, J. L.:
Autonomous Biogeochemical Floats Detect Significant Carbon Dioxide
Outgassing in the High-Latitude Southern Ocean, Geophys. Res.
Lett., 45, 9049–9057, <ext-link xlink:href="https://doi.org/10.1029/2018GL078013" ext-link-type="DOI">10.1029/2018GL078013</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx43"><?xmltex \def\ref@label{{Gregor(2021)}}?><label>Gregor(2021)</label><?label Gregor2021?><mixed-citation>Gregor, L.: luke-gregor/OceanSODA-ETHZ: code, Zenodo, <ext-link xlink:href="https://doi.org/10.5281/zenodo.4455354" ext-link-type="DOI">10.5281/zenodo.4455354</ext-link>, 2021.</mixed-citation></ref>
      <ref id="bib1.bibx44"><?xmltex \def\ref@label{{Gregor and Gruber(2020)}}?><label>Gregor and Gruber(2020)</label><?label Gregor2020a?><mixed-citation>Gregor, L. and Gruber, N.: OceanSODA-ETHZ: A global gridded data set of the
surface ocean carbonate system for seasonal to decadal studies of ocean
acidification (NCEI Accession 0220059), Tech. Rep., NOAA National Centers
for Environmental Information, <ext-link xlink:href="https://doi.org/10.25921/m5wx-ja34" ext-link-type="DOI">10.25921/m5wx-ja34</ext-link>,
2020.</mixed-citation></ref>
      <ref id="bib1.bibx45"><?xmltex \def\ref@label{{Gregor et~al.(2018)}}?><label>Gregor et al.(2018)</label><?label Gregor2018?><mixed-citation>Gregor, L., Kok, S., and Monteiro, P. M. S.: Interannual drivers of the
seasonal cycle of CO<inline-formula><mml:math id="M766" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> in the Southern Ocean, Biogeosciences, 15, 2361–2378,
<ext-link xlink:href="https://doi.org/10.5194/bg-15-2361-2018" ext-link-type="DOI">10.5194/bg-15-2361-2018</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx46"><?xmltex \def\ref@label{{Gregor et~al.(2019)}}?><label>Gregor et al.(2019)</label><?label Gregor2019?><mixed-citation>Gregor, L., Lebehot, A. D., Kok, S., and Scheel Monteiro, P. M.: A
comparative assessment of the uncertainties of global surface ocean CO<inline-formula><mml:math id="M767" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
estimates using a machine-learning ensemble (CSIR-ML6 version 2019a) – have
we hit the wall?, Geosci. Model Dev., 12, 5113–5136,
<ext-link xlink:href="https://doi.org/10.5194/gmd-12-5113-2019" ext-link-type="DOI">10.5194/gmd-12-5113-2019</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx47"><?xmltex \def\ref@label{{Gruber and Sarmiento(1997)}}?><label>Gruber and Sarmiento(1997)</label><?label Gruber1997?><mixed-citation>Gruber, N. and Sarmiento, J. L.: Global patterns of marine nitrogen fixation
and denitrification, Global Biogeochem. Cy., 11, 235–266,
<ext-link xlink:href="https://doi.org/10.1029/97GB00077" ext-link-type="DOI">10.1029/97GB00077</ext-link>, 1997.</mixed-citation></ref>
      <ref id="bib1.bibx48"><?xmltex \def\ref@label{{Gruber et~al.(1996)}}?><label>Gruber et al.(1996)</label><?label Gruber1996?><mixed-citation>Gruber, N., Sarmiento, J. L., and Stocker, T. F.: An improved method for
detecting anthropogenic CO<inline-formula><mml:math id="M768" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> in the oceans, Global Biogeochem. Cy., 10,
809–837, <ext-link xlink:href="https://doi.org/10.1029/96GB01608" ext-link-type="DOI">10.1029/96GB01608</ext-link>, 1996.</mixed-citation></ref>
      <ref id="bib1.bibx49"><?xmltex \def\ref@label{{Gruber et~al.(2012)}}?><label>Gruber et al.(2012)</label><?label Gruber2012?><mixed-citation>Gruber, N., Hauri, C., Lachkar, Z., Loher, D., Frölicher, T. L., and
Plattner, G.-K.: Rapid Progression of Ocean Acidification in the California
Current System, Science, 337, 220–223, <ext-link xlink:href="https://doi.org/10.1126/science.1216773" ext-link-type="DOI">10.1126/science.1216773</ext-link>, 2012.</mixed-citation></ref>
      <ref id="bib1.bibx50"><?xmltex \def\ref@label{{Gruber et~al.(2019)}}?><label>Gruber et al.(2019)</label><?label Gruber2019?><mixed-citation>Gruber, N., Clement, D., Carter, B. R., Feely, R. A., van Heuven, S., Hoppema,
M., Ishii, M., Key, R. M., Kozyr, A., Lauvset, S. K., Lo Monaco, C.,
Mathis, J. T., Murata, A., Olsen, A., Perez, F. F., Sabine, C. L., Tanhua,
T., and Wanninkhof, R. H.: The oceanic sink for anthropogenic CO<inline-formula><mml:math id="M769" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> from 1994
to 2007, Science, 363, 1193–1199, <ext-link xlink:href="https://doi.org/10.1126/science.aau5153" ext-link-type="DOI">10.1126/science.aau5153</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx51"><?xmltex \def\ref@label{{Hauri et~al.(2016)}}?><label>Hauri et al.(2016)</label><?label Hauri2016?><mixed-citation>Hauri, C., Friedrich, T., and Timmermann, A.: Abrupt onset and prolongation of
aragonite undersaturation events in the Southern Ocean, Nat. Clim.
Change, 6, 172–176, <ext-link xlink:href="https://doi.org/10.1038/nclimate2844" ext-link-type="DOI">10.1038/nclimate2844</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx52"><?xmltex \def\ref@label{{Holte et~al.(2017)}}?><label>Holte et al.(2017)</label><?label Holte2017?><mixed-citation>Holte, J., Talley, L. D., Gilson, J., and Roemmich, D.: An Argo mixed layer
climatology and database, Geophys. Res. Lett., 44, 5618–5626,
<ext-link xlink:href="https://doi.org/10.1002/2017GL073426" ext-link-type="DOI">10.1002/2017GL073426</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx53"><?xmltex \def\ref@label{{Humphreys et~al.(2020)}}?><label>Humphreys et al.(2020)</label><?label Humphreys2020?><mixed-citation>Humphreys, M. P., Gregor, L., Pierrot, D., van Heuven, S., Lewis, E. R., and
Wallace, D. W. R.: PyCO2SYS: marine carbonate system calculations in
Python, Zenodo, <ext-link xlink:href="https://doi.org/10.5281/zenodo.3967359" ext-link-type="DOI">10.5281/zenodo.3967359</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx54"><?xmltex \def\ref@label{{Iida et~al.(2015)}}?><label>Iida et al.(2015)</label><?label Iida2015a?><mixed-citation>Iida, Y., Kojima, A., Takatani, Y., Nakano, T., Sugimoto, H., Midorikawa, T.,
and Ishii, M.: Trends in <inline-formula><mml:math id="M770" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M771" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and sea–air CO<inline-formula><mml:math id="M772" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux over the global open
oceans for the last two decades, J. Oceanogr., 71, 637–661,
<ext-link xlink:href="https://doi.org/10.1007/s10872-015-0306-4" ext-link-type="DOI">10.1007/s10872-015-0306-4</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx55"><?xmltex \def\ref@label{{Jiang et~al.(2019)}}?><label>Jiang et al.(2019)</label><?label Jiang2019?><mixed-citation>Jiang, L.-q., Carter, B. R., Feely, R. A., Lauvset, S. K., and Olsen, A.:
Surface ocean pH and buffer capacity: past, present and future, Sci.
Rep., 9, 18624, <ext-link xlink:href="https://doi.org/10.1038/s41598-019-55039-4" ext-link-type="DOI">10.1038/s41598-019-55039-4</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx56"><?xmltex \def\ref@label{{Johnson et~al.(2016)}}?><label>Johnson et al.(2016)</label><?label Johnson2016?><mixed-citation>Johnson, K. S., Jannasch, H. W., Coletti, L. J., Elrod, V. A., Martz, T. R.,
Takeshita, Y., Carlson, R. J., and Connery, J. G.: Deep-Sea DuraFET: A
Pressure Tolerant pH Sensor Designed for Global Sensor Networks, Anal.
Chem., 88, 3249–3256, <ext-link xlink:href="https://doi.org/10.1021/acs.analchem.5b04653" ext-link-type="DOI">10.1021/acs.analchem.5b04653</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx57"><?xmltex \def\ref@label{{Johnson et~al.(2017)}}?><label>Johnson et al.(2017)</label><?label Johnson2017?><mixed-citation>Johnson, K. S., Plant, J. N., Coletti, L. J., Jannasch, H. W., Sakamoto, C. M.,
Riser, S. C., Swift, D. D., Williams, N. L., Boss, E., Haëntjens, N.,
Talley, L. D., and Sarmiento, J. L.: Biogeochemical sensor performance in
the SOCCOM profiling float array, J. Geophys. Res.-Ocean., 122, 6416–6436,
<ext-link xlink:href="https://doi.org/10.1002/2017JC012838" ext-link-type="DOI">10.1002/2017JC012838</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx58"><?xmltex \def\ref@label{{Keppler and Landsch{\"{u}}tzer(2019)}}?><label>Keppler and Landschützer(2019)</label><?label Keppler2019?><mixed-citation>Keppler, L. and Landschützer, P.: Regional Wind Variability Modulates
the Southern Ocean Carbon Sink, Sci. Rep., 9, 7384,
<ext-link xlink:href="https://doi.org/10.1038/s41598-019-43826-y" ext-link-type="DOI">10.1038/s41598-019-43826-y</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx59"><?xmltex \def\ref@label{{Keppler et~al.(2020)}}?><label>Keppler et al.(2020)</label><?label Keppler2020?><mixed-citation>Keppler, L., Landschützer, P., Gruber, N., Lauvset, S. K., and Stemmler,
I.: Seasonal carbon dynamics in the global ocean based on a neural network
mapping of observations, Global Biogeochem. Cy., 34, e2020GB006571, <ext-link xlink:href="https://doi.org/10.1029/2020GB006571" ext-link-type="DOI">10.1029/2020GB006571</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx60"><?xmltex \def\ref@label{{Kwiatkowski et~al.(2020)}}?><label>Kwiatkowski et al.(2020)</label><?label Kwiatkowski2020?><mixed-citation>Kwiatkowski, L., Torres, O., Bopp, L., Aumont, O., Chamberlain, M., Christian, J. R., Dunne, J. P., Gehlen, M., Ilyina, T., John, J. G., Lenton, A., Li, H., Lovenduski, N. S., Orr, J. C., Palmieri, J., Santana-Falcón, Y., Schwinger, J., Séférian, R., Stock, C. A., Tagliabue, A., Takano, Y., Tjiputra, J., Toyama, K., Tsujino, H., Watanabe, M., Yamamoto, A., Yool, A., and Ziehn, T.: Twenty-first century ocean warming, acidification, deoxygenation<?pagebreak page806?>, and upper-ocean nutrient and primary production decline from CMIP6 model projections, Biogeosciences, 17, 3439–3470, <ext-link xlink:href="https://doi.org/10.5194/bg-17-3439-2020" ext-link-type="DOI">10.5194/bg-17-3439-2020</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx61"><?xmltex \def\ref@label{{Land et~al.(2019)}}?><label>Land et al.(2019)</label><?label Land2019?><mixed-citation>Land, P. E., Findlay, H. S., Shutler, J. D., Ashton, I. G., Holding, T.,
Grouazel, A., Girard-Ardhuin, F., Reul, N., Piolle, J. F., Chapron, B.,
Quilfen, Y., Bellerby, R. G., Bhadury, P., Salisbury, J., Vandemark, D., and
Sabia, R.: Optimum satellite remote sensing of the marine carbonate system
using empirical algorithms in the global ocean, the Greater Caribbean, the
Amazon Plume and the Bay of Bengal, Remote Sens. Environ., 235,
111469, <ext-link xlink:href="https://doi.org/10.1016/j.rse.2019.111469" ext-link-type="DOI">10.1016/j.rse.2019.111469</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx62"><?xmltex \def\ref@label{{Landsch{\"{u}}tzer et~al.(2013)}}?><label>Landschützer et al.(2013)</label><?label Landschutzer2013?><mixed-citation>Landschützer, P., Gruber, N., Bakker, D. C., Schuster, U., Nakaoka, S.,
Payne, M. R., Sasse, T. P., and Zeng, J.: A neural network-based estimate of
the seasonal to inter-annual variability of the Atlantic Ocean carbon sink,
Biogeosciences, 10, 7793–7815, <ext-link xlink:href="https://doi.org/10.5194/bg-10-7793-2013" ext-link-type="DOI">10.5194/bg-10-7793-2013</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx63"><?xmltex \def\ref@label{{Landsch{\"{u}}tzer et~al.(2014)}}?><label>Landschützer et al.(2014)</label><?label Landschutzer2014a?><mixed-citation>Landschützer, P., Gruber, N., Bakker, D. C., and Schuster, U.: Recent variability of the global ocean carbon sink, Global Biogeochem. Cy., 28, 927–949, <ext-link xlink:href="https://doi.org/10.1002/2014GB004853" ext-link-type="DOI">10.1002/2014GB004853</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx64"><?xmltex \def\ref@label{{Landsch{\"{u}}tzer et~al.(2015)}}?><label>Landschützer et al.(2015)</label><?label Landschutzer2015?><mixed-citation>Landschützer, P., Gruber, N., Haumann, F. A., Rödenbeck, C.,
Bakker, D. C., Van Heuven, S. M., Hoppema, M., Metzl, N., Sweeney, C.,
Takahashi, T. T., Tilbrook, B., and Wanninkhof, R. H.: The reinvigoration of
the Southern Ocean carbon sink, Science, 349, 1221–1224,
<ext-link xlink:href="https://doi.org/10.1126/science.aab2620" ext-link-type="DOI">10.1126/science.aab2620</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx65"><?xmltex \def\ref@label{{Landsch{\"{u}}tzer et~al.(2016)}}?><label>Landschützer et al.(2016)</label><?label Landschutzer2016?><mixed-citation>Landschützer, P., Gruber, N., and Bakker, D. C.: Decadal variations and
trends of the global ocean carbon sink, Global Biogeochem. Cy., 30,
1396–1417, <ext-link xlink:href="https://doi.org/10.1002/2015GB005359" ext-link-type="DOI">10.1002/2015GB005359</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx66"><?xmltex \def\ref@label{{Landsch{\"{u}}tzer et~al.(2020)}}?><label>Landschützer et al.(2020)</label><?label Landschutzer2020?><mixed-citation>Landschützer, P., Laruelle, G. G., Roobaert, A., and Regnier, P.: A uniform <inline-formula><mml:math id="M773" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M774" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> climatology combining open and coastal oceans, Earth Syst. Sci. Data, 12, 2537–2553, <ext-link xlink:href="https://doi.org/10.5194/essd-12-2537-2020" ext-link-type="DOI">10.5194/essd-12-2537-2020</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx67"><?xmltex \def\ref@label{{Laruelle et~al.(2013)}}?><label>Laruelle et al.(2013)</label><?label Laruelle2013?><mixed-citation>Laruelle, G. G., Dürr, H. H., Lauerwald, R., Hartmann, J., Slomp, C. P.,
Goossens, N., and Regnier, P. A.: Global multi-scale segmentation of
continental and coastal waters from the watersheds to the continental
margins, Hydrol. Earth Syst. Sci., 17, 2029–2051,
<ext-link xlink:href="https://doi.org/10.5194/hess-17-2029-2013" ext-link-type="DOI">10.5194/hess-17-2029-2013</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx68"><?xmltex \def\ref@label{{Laruelle et~al.(2017)}}?><label>Laruelle et al.(2017)</label><?label Laruelle2017?><mixed-citation>Laruelle, G. G., Landschützer, P., Gruber, N., Ti, J. L., Delille, B.,
Regnier, P., Tison, J.-L., Delille, B., and Regnier, P.: Global
high-resolution monthly <inline-formula><mml:math id="M775" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M776" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> climatology for the coastal ocean derived from
neural network interpolation, Biogeosciences, 14, 4545–4561,
<ext-link xlink:href="https://doi.org/10.5194/bg-14-4545-2017" ext-link-type="DOI">10.5194/bg-14-4545-2017</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx69"><?xmltex \def\ref@label{{Lauvset et~al.(2015)}}?><label>Lauvset et al.(2015)</label><?label Lauvset2015?><mixed-citation>Lauvset, S. K., Gruber, N., Landschützer, P., Olsen, A., and Tjiputra,
J.: Trends and drivers in global surface ocean pH over the past 3 decades,
Biogeosciences, 12, 1285–1298, <ext-link xlink:href="https://doi.org/10.5194/bg-12-1285-2015" ext-link-type="DOI">10.5194/bg-12-1285-2015</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx70"><?xmltex \def\ref@label{{Lauvset et~al.(2016)}}?><label>Lauvset et al.(2016)</label><?label Lauvset2016?><mixed-citation>Lauvset, S. K., Key, R. M., Olsen, A., van Heuven, S., Velo, A., Lin, X.,
Schirnick, C., Kozyr, A., Tanhua, T., Hoppema, M., Jutterström, S.,
Steinfeldt, R., Jeansson, E., Ishii, M., Perez, F. F., Suzuki, T., and
Watelet, S.: A new global interior ocean mapped climatology: the <inline-formula><mml:math id="M777" display="inline"><mml:mrow><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup><mml:mo>×</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:msup><mml:mi/><mml:mo>∘</mml:mo></mml:msup></mml:mrow></mml:math></inline-formula>
GLODAP version 2, Earth Syst. Sci. Data, 8, 325–340,
<ext-link xlink:href="https://doi.org/10.5194/essd-8-325-2016" ext-link-type="DOI">10.5194/essd-8-325-2016</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx71"><?xmltex \def\ref@label{{Lee et~al.(2000)}}?><label>Lee et al.(2000)</label><?label Lee2000?><mixed-citation>Lee, K., Wanninkhof, R. H., Feely, R. A., Millero, F. J., and Peng, T. H.:
Global relationships of total inorganic carbon with temperature and nitrate
in surface seawater, Global Biogeochem. Cy., 14, 979–994,
<ext-link xlink:href="https://doi.org/10.1029/1998GB001087" ext-link-type="DOI">10.1029/1998GB001087</ext-link>, 2000.</mixed-citation></ref>
      <ref id="bib1.bibx72"><?xmltex \def\ref@label{{Lee et~al.(2006)}}?><label>Lee et al.(2006)</label><?label Lee2006?><mixed-citation>Lee, K., Tong, L. T., Millero, F. J., Sabine, C. L., Dickson, A. G., Goyet, C.,
Park, G.-H., Wanninkhof, R. H., Feely, R. A., and Key, R. M.: Global
relationships of total alkalinity with salinity and temperature in surface
waters of the world's oceans, Geophys. Res. Lett., 33, L19605,
<ext-link xlink:href="https://doi.org/10.1029/2006GL027207" ext-link-type="DOI">10.1029/2006GL027207</ext-link>, 2006.</mixed-citation></ref>
      <ref id="bib1.bibx73"><?xmltex \def\ref@label{{Lef{\`{e}}vre and Taylor(2002)}}?><label>Lefèvre and Taylor(2002)</label><?label Lefevre2002?><mixed-citation>Lefèvre, N. and Taylor, A.: Estimating <inline-formula><mml:math id="M778" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M779" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> from sea surface
temperatures in the Atlantic gyres, Deep-Sea Res. Pt. I, 49, 539–554, <ext-link xlink:href="https://doi.org/10.1016/S0967-0637(01)00064-4" ext-link-type="DOI">10.1016/S0967-0637(01)00064-4</ext-link>, 2002.</mixed-citation></ref>
      <ref id="bib1.bibx74"><?xmltex \def\ref@label{{Lewis et~al.(1998)}}?><label>Lewis et al.(1998)</label><?label Lewis1998?><mixed-citation>Lewis, E., Wallace, D., and Allison, L. J.: Program developed for CO<inline-formula><mml:math id="M780" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> system
calculations,  CDIAC, <ext-link xlink:href="https://doi.org/10.15485/1464255" ext-link-type="DOI">10.15485/1464255</ext-link>, 1998.</mixed-citation></ref>
      <ref id="bib1.bibx75"><?xmltex \def\ref@label{{Lovenduski et~al.(2007)}}?><label>Lovenduski et al.(2007)</label><?label Lovenduski2007a?><mixed-citation>Lovenduski, N. S., Gruber, N., Doney, S. C., and Lima, I. D.: Enhanced CO<inline-formula><mml:math id="M781" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
outgassing in the Southern Ocean from a positive phase of the Southern
Annular Mode, Global Biogeochem. Cy., 21, 1–14,
<ext-link xlink:href="https://doi.org/10.1029/2006GB002900" ext-link-type="DOI">10.1029/2006GB002900</ext-link>, 2007.</mixed-citation></ref>
      <ref id="bib1.bibx76"><?xmltex \def\ref@label{{Maritorena et~al.(2010)}}?><label>Maritorena et al.(2010)</label><?label Maritorena2010?><mixed-citation>Maritorena, S., Fanton D'andon, O. H., Mangin, A., Siegel, D. A., D'Andon, O.
H. F., Mangin, A., and Siegel, D. A.: Merged satellite ocean color data
products using a bio-optical model: Characteristics, benefits and issues,
Remote Sens. Environ., 114, 1791–1804,
<ext-link xlink:href="https://doi.org/10.1016/j.rse.2010.04.002" ext-link-type="DOI">10.1016/j.rse.2010.04.002</ext-link>, 2010.</mixed-citation></ref>
      <ref id="bib1.bibx77"><?xmltex \def\ref@label{{Marshall and Speer(2012)}}?><label>Marshall and Speer(2012)</label><?label Marshall2012?><mixed-citation>Marshall, J. C. and Speer, K.: Closure of the meridional overturning
circulation through Southern Ocean upwelling, Nat. Geosci., 5,
171–180, <ext-link xlink:href="https://doi.org/10.1038/ngeo1391" ext-link-type="DOI">10.1038/ngeo1391</ext-link>, 2012.</mixed-citation></ref>
      <ref id="bib1.bibx78"><?xmltex \def\ref@label{{Matsumoto and Gruber(2005)}}?><label>Matsumoto and Gruber(2005)</label><?label Matsumoto2005?><mixed-citation>Matsumoto, K. and Gruber, N.: How accurate is the estimation of anthropogenic
carbon in the ocean? An evaluation of the <inline-formula><mml:math id="M782" display="inline"><mml:mi mathvariant="normal">Δ</mml:mi></mml:math></inline-formula>C<inline-formula><mml:math id="M783" display="inline"><mml:msup><mml:mi/><mml:mo>*</mml:mo></mml:msup></mml:math></inline-formula> method, Global
Biogeochem. Cy., 19, 1–17, <ext-link xlink:href="https://doi.org/10.1029/2004GB002397" ext-link-type="DOI">10.1029/2004GB002397</ext-link>, 2005.</mixed-citation></ref>
      <ref id="bib1.bibx79"><?xmltex \def\ref@label{{Mehrbach et~al.(1973)}}?><label>Mehrbach et al.(1973)</label><?label Mehrbach1973?><mixed-citation>Mehrbach, C., Culberson, C. H., Hawley, J. E., and Pytkowicx, R. M.:
Measurement of the Apparent Dissociation Constants of Carbonic Acid in
Seawater At Atmospheric Pressure, Limnol. Oceanogr., 18, 897–907,
<ext-link xlink:href="https://doi.org/10.4319/lo.1973.18.6.0897" ext-link-type="DOI">10.4319/lo.1973.18.6.0897</ext-link>, 1973.</mixed-citation></ref>
      <ref id="bib1.bibx80"><?xmltex \def\ref@label{{Millero et~al.(1998)}}?><label>Millero et al.(1998)</label><?label Millero1998?><mixed-citation>Millero, F. J., Lee, K., and Roche, M.: Distribution of alkalinity in the
surface waters of the major oceans, Mar. Chem., 60, 111–130,
<ext-link xlink:href="https://doi.org/10.1016/S0304-4203(97)00084-4" ext-link-type="DOI">10.1016/S0304-4203(97)00084-4</ext-link>, 1998.</mixed-citation></ref>
      <ref id="bib1.bibx81"><?xmltex \def\ref@label{{Monteiro et~al.(2015)}}?><label>Monteiro et al.(2015)</label><?label Monteiro2015?><mixed-citation>Monteiro, P. M. S., Gregor, L., Lévy, M., Maenner, S., Sabine, C. L., and
Swart, S.: Intraseasonal variability linked to sampling alias in air-sea CO<inline-formula><mml:math id="M784" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>
fluxes in the Southern Ocean, Geophys. Res. Lett., 42, 8507–8514,
<ext-link xlink:href="https://doi.org/10.1002/2015GL066009" ext-link-type="DOI">10.1002/2015GL066009</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx82"><?xmltex \def\ref@label{{Negrete-Garc{\'{i}}a et~al.(2019)}}?><label>Negrete-García et al.(2019)</label><?label Negrete-Garcia2019?><mixed-citation>Negrete-García, G., Lovenduski, N. S., Hauri, C., Krumhardt, K. M., and
Lauvset, S. K.: Sudden emergence of a shallow aragonite saturation horizon
in the Southern Ocean, Nat. Clim. Change, 9, 313–317,
<ext-link xlink:href="https://doi.org/10.1038/s41558-019-0418-8" ext-link-type="DOI">10.1038/s41558-019-0418-8</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx83"><?xmltex \def\ref@label{{Olafsson et~al.(2010)}}?><label>Olafsson et al.(2010)</label><?label Olafsson2010a?><mixed-citation>Olafsson, J., Olafsdottir, S. R., Benoit-Cattin, A., and Takahashi, T. T.: The
irminger sea and the Iceland sea time series measurements of sea water carbon
and nutrient chemistry 1983–2008, Earth Syst. Sci. Data, 2, 99–104,
<ext-link xlink:href="https://doi.org/10.5194/essd-2-99-2010" ext-link-type="DOI">10.5194/essd-2-99-2010</ext-link>, 2010.</mixed-citation></ref>
      <ref id="bib1.bibx84"><?xmltex \def\ref@label{{Olsen et~al.(2016)}}?><label>Olsen et al.(2016)</label><?label Olsen2016?><mixed-citation>Olsen, A., Key, R. M., van Heuven, S., Lauvset, S. K., Velo, A., Lin, X.,
Schirnick, C., Kozyr, A., Tanhua, T., Hoppema, M., Jutterström, S.,
Steinfeldt, R., Jeansson, E., Ishii, M., Pérez, F. F., and Suzuki, T.:
The Global Ocean Data Analysis Project version 2 (GLODAPv2) – an
internally consistent data product for the world ocean, Earth Syst. Sci.
Data, 8, 297–323, <ext-link xlink:href="https://doi.org/10.5194/essd-8-297-2016" ext-link-type="DOI">10.5194/essd-8-297-2016</ext-link>, 2016.</mixed-citation></ref>
      <ref id="bib1.bibx85"><?xmltex \def\ref@label{{Olsen et~al.(2019)}}?><label>Olsen et al.(2019)</label><?label Olsen2019?><mixed-citation>Olsen, A., Lange, N., Key, R. M., Tanhua, T., Álvarez, M., Becker, S.,
Bittig, H. C<?pagebreak page807?>., Carter, B. R., Cotrim Da Cunha, L., Feely, R. A., Van
Heuven, S., Hoppema, M., Ishii, M., Jeansson, E., Jones, S. D.,
Jutterström, S., Karlsen, M. K., Kozyr, A., Lauvset, S. K., Lo
Monaco, C., Murata, A., Pérez, F. F., Pfeil, B., Schirnick, C.,
Steinfeldt, R., Suzuki, T., Telszewski, M., Tilbrook, B., Velo, A., and
Wanninkhof, R.: GLODAPv2.2019 – An update of GLODAPv2, Earth Syst. Sci.
Data, 11, 1437–1461, <ext-link xlink:href="https://doi.org/10.5194/essd-11-1437-2019" ext-link-type="DOI">10.5194/essd-11-1437-2019</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx86"><?xmltex \def\ref@label{{Orr et~al.(2005)}}?><label>Orr et al.(2005)</label><?label Orr2005?><mixed-citation>Orr, J. C., Fabry, V. J., Aumont, O., Bopp, L., Doney, S. C., Feely, R. A.,
Gnanadesikan, A., Gruber, N., Ishida, A., Joos, F., Key, R. M., Lindsay, K.,
Maier-Reimer, E., Matear, R. J., Monfray, P., Mouchet, A., Najjar, R. G.,
Plattner, G.-K. K., Rodgers, K. B., Sabine, C. L., Sarmiento, J. L.,
Schlitzer, R., Slater, R. D., Totterdell, I. J., Weirig, M.-F. F., Yamanaka,
Y., Yool, A., Doney, S. C., Gnanadesikan, A., Gruber, N., Ishida, A., Joos,
F., Key, R. M., Lindsay, K., Maier-Reimer, E., Matear, R. J., Monfray, P.,
Mouchet, A., Najjar, R. G., Plattner, G.-K. K., Rodgers, K. B., Sabine,
C. L., Sarmiento, J. L., Schlitzer, R., Slater, R. D., Totterdell, I. J.,
Weirig, M.-F. F., Yamanaka, Y., Yool, A., Feely, R. A., Gnanadesikan, A.,
Gruber, N., Ishida, A., Joos, F., Key, R. M., Lindsay, K., Maier-Reimer, E.,
Matear, R. J., Monfray, P., Mouchet, A., Najjar, R. G., Plattner, G.-K. K.,
Rodgers, K. B., Sabine, C. L., Sarmiento, J. L., Schlitzer, R., Slater,
R. D., Totterdell, I. J., Weirig, M.-F. F., Yamanaka, Y., and Yool, A.:
Anthropogenic ocean acidification over the twenty-first century and its
impact on calcifying organisms, Nature, 437, 681–686,
<ext-link xlink:href="https://doi.org/10.1038/nature04095" ext-link-type="DOI">10.1038/nature04095</ext-link>, 2005.</mixed-citation></ref>
      <ref id="bib1.bibx87"><?xmltex \def\ref@label{{Orr et~al.(2015)}}?><label>Orr et al.(2015)</label><?label Orr2015a?><mixed-citation>Orr, J. C., Epitalon, J. M., and Gattuso, J.-P.: Comparison of ten packages
that compute ocean carbonate chemistry, Biogeosciences, 12, 1483–1510,
<ext-link xlink:href="https://doi.org/10.5194/bg-12-1483-2015" ext-link-type="DOI">10.5194/bg-12-1483-2015</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx88"><?xmltex \def\ref@label{{Orr et~al.(2018)}}?><label>Orr et al.(2018)</label><?label Orr2018?><mixed-citation>Orr, J. C., Epitalon, J.-M., Dickson, A. G., and Gattuso, J.-P.: Routine
uncertainty propagation for the marine carbon dioxide system, Mar.
Chem., 2018, 84–107, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2018.10.006" ext-link-type="DOI">10.1016/j.marchem.2018.10.006</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx89"><?xmltex \def\ref@label{{Pedregosa et~al.(2011)}}?><label>Pedregosa et al.(2011)</label><?label Pedregosa2011?><mixed-citation>
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, C., Thirion, B., Grisel,
O., Blondel, M., Prettenhoffer, P., Weiss, R., Dubourg, V., Vanderplas, J.,
Passos, A., and Cournapeau, D.: Scikit-learn: Machine learning in Python,
J. Mach. Learn. Res., 12, 2825–2830, 2011.</mixed-citation></ref>
      <ref id="bib1.bibx90"><?xmltex \def\ref@label{{Pierrot et~al.(2009)}}?><label>Pierrot et al.(2009)</label><?label Pierrot2009a?><mixed-citation>Pierrot, D., Neill, C., Sullivan, K. F., Castle, R. D., Wanninkhof, R. H.,
Lüger, H., Johannessen, T., Olsen, A., Feely, R. A., and Cosca, C. E.:
Recommendations for autonomous underway <inline-formula><mml:math id="M785" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M786" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> measuring systems and
data-reduction routines, Deep-Sea Res. Pt. II, 56, 512–522, <ext-link xlink:href="https://doi.org/10.1016/j.dsr2.2008.12.005" ext-link-type="DOI">10.1016/j.dsr2.2008.12.005</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx91"><?xmltex \def\ref@label{{P{\"{o}}rtner and Farrell(2008)}}?><label>Pörtner and Farrell(2008)</label><?label Portner2008?><mixed-citation>Pörtner, H.-O. and Farrell, A. P.: Physiology and Climate Change,
Science, 322, 690–692, <ext-link xlink:href="https://doi.org/10.1126/science.1163156" ext-link-type="DOI">10.1126/science.1163156</ext-link>, 2008.</mixed-citation></ref>
      <ref id="bib1.bibx92"><?xmltex \def\ref@label{{Raimondi et~al.(2019)}}?><label>Raimondi et al.(2019)</label><?label Raimondi2019?><mixed-citation>Raimondi, L., Matthews, J. B. R., Atamanchuk, D., Azetsu-Scott, K., and
Wallace, D. W.: The internal consistency of the marine carbon dioxide system
for high latitude shipboard and in situ monitoring, Mar. Chem., 213,
49–70, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2019.03.001" ext-link-type="DOI">10.1016/j.marchem.2019.03.001</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx93"><?xmltex \def\ref@label{{Resplandy et~al.(2009)}}?><label>Resplandy et al.(2009)</label><?label Resplandy2009a?><mixed-citation>Resplandy, L., Lévy, M., D'Ovidio, F., and Merlivat, L.: Impact of
submesoscale variability in estimating the air-sea CO<inline-formula><mml:math id="M787" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> exchange: Results from
a model study of the POMME experiment, Global Biogeochem. Cy., 23,
1–19, <ext-link xlink:href="https://doi.org/10.1029/2008GB003239" ext-link-type="DOI">10.1029/2008GB003239</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx94"><?xmltex \def\ref@label{{R{\"{o}}denbeck et~al.(2014)}}?><label>Rödenbeck et al.(2014)</label><?label Rodenbeck2014?><mixed-citation>Rödenbeck, C., Bakker, D. C., Metzl, N., Olsen, A., Sabine, C. L.,
Cassar, N., Reum, F., Keeling, R. F., and Heimann, M.: Interannual sea–air
CO<inline-formula><mml:math id="M788" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux variability from an observation-driven ocean mixed-layer scheme,
Biogeosciences, 11, 4599–4613, <ext-link xlink:href="https://doi.org/10.5194/bg-11-4599-2014" ext-link-type="DOI">10.5194/bg-11-4599-2014</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx95"><?xmltex \def\ref@label{{R{\"{o}}denbeck et~al.(2015)}}?><label>Rödenbeck et al.(2015)</label><?label Rodenbeck2015?><mixed-citation>Rödenbeck, C., Bakker, D. C., Gruber, N., Iida, Y., Jacobson, A. R.,
Jones, S. D., Landschützer, P., Metzl, N., Nakaoka, S., Olsen, A.,
Park, G.-H. H., Peylin, P., Rodgers, K. B., Sasse, T. P., Schuster, U.,
Shutler, J. D., Valsala, V., Wanninkhof, R. H., and Zeng, J.: Data-based
estimates of the ocean carbon sink variability – first results of the
Surface Ocean pCO2 Mapping intercomparison (SOCOM), Biogeosciences, 12,
7251–7278, <ext-link xlink:href="https://doi.org/10.5194/bg-12-7251-2015" ext-link-type="DOI">10.5194/bg-12-7251-2015</ext-link>, 2015.</mixed-citation></ref>
      <ref id="bib1.bibx96"><?xmltex \def\ref@label{{Sabine et~al.(2004)}}?><label>Sabine et al.(2004)</label><?label Sabine2004a?><mixed-citation>Sabine, C. L., Feely, R. A., Gruber, N., Key, R. M., Lee, K., Bullister, J. L.,
Wanninkhof, R. H., Wong, C. S., Wallace, D. W. R., Tilbrook, B., Millero,
F. J., Peng, T.-H., Kozyr, A., Ono1, T., and Rios, A. F.: The Oceanic Sink
for Anthropogenic CO<inline-formula><mml:math id="M789" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, Science, 305, 367–371,
<ext-link xlink:href="https://doi.org/10.1126/science.1097403" ext-link-type="DOI">10.1126/science.1097403</ext-link>, 2004.</mixed-citation></ref>
      <ref id="bib1.bibx97"><?xmltex \def\ref@label{{Sarmiento and Gruber(2006)}}?><label>Sarmiento and Gruber(2006)</label><?label Sarmiento2006?><mixed-citation>
Sarmiento, J. L. and Gruber, N.: Ocean Biogeochemical Dynamics, Princeton
University Press, Princeton, NJ, 526 pp., 2006.</mixed-citation></ref>
      <ref id="bib1.bibx98"><?xmltex \def\ref@label{{Sasse et~al.(2013)}}?><label>Sasse et al.(2013)</label><?label Sasse2013b?><mixed-citation>Sasse, T. P., McNeil, B. I., and Abramowitz, G.: A novel method for diagnosing
seasonal to inter-annual surface ocean carbon dynamics from bottle data using
neural networks, Biogeosciences, 10, 4319–4340,
<ext-link xlink:href="https://doi.org/10.5194/bg-10-4319-2013" ext-link-type="DOI">10.5194/bg-10-4319-2013</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx99"><?xmltex \def\ref@label{{Siegel et~al.(2002)}}?><label>Siegel et al.(2002)</label><?label Siegel2002?><mixed-citation>Siegel, D. A., Doney, S. C., and Yoder, J. A.: The North Atlantic spring
phytoplankton bloom and Sverdrup's critical depth hypothesis, Science, 296,
730–733, <ext-link xlink:href="https://doi.org/10.1126/science.1069174" ext-link-type="DOI">10.1126/science.1069174</ext-link>, 2002.</mixed-citation></ref>
      <ref id="bib1.bibx100"><?xmltex \def\ref@label{{Steinacher et~al.(2009)}}?><label>Steinacher et al.(2009)</label><?label Steinacher2009?><mixed-citation>Steinacher, M., Joos, F., Frölicher, T. L., Plattner, G.-K., and Doney, S. C.: Imminent ocean acidification in the Arctic projected with the NCAR global coupled carbon cycle-climate model, Biogeosciences, 6, 515–533, <ext-link xlink:href="https://doi.org/10.5194/bg-6-515-2009" ext-link-type="DOI">10.5194/bg-6-515-2009</ext-link>, 2009.</mixed-citation></ref>
      <ref id="bib1.bibx101"><?xmltex \def\ref@label{{Tagliabue et~al.(2017)}}?><label>Tagliabue et al.(2017)</label><?label Tagliabue2017?><mixed-citation>Tagliabue, A., Bowie, A. R., Boyd, P. W., Buck, K. N., Johnson, K. S., Saito,
M. A., Philip, W., Buck, K. N., Johnson, K. S., Saito, M. A., Boyd, P. W.,
Buck, K. N., Johnson, K. S., and Saito, M. A.: The integral role of iron in
ocean biogeochemistry, Nature, 543, 51–59, <ext-link xlink:href="https://doi.org/10.1038/nature21058" ext-link-type="DOI">10.1038/nature21058</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx102"><?xmltex \def\ref@label{{Takahashi et al.(2014)}}?><label>Takahashi et al.(2014)</label><?label Takahashi2013?><mixed-citation>Takahashi, T., Sutherland, S. C., Chipman, D. W., Goddard, J. G., Ho, C., Newberger, T., Sweeney, C., and Munro, D. R.: Climatological distributions of pH, <inline-formula><mml:math id="M790" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M791" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, total CO<inline-formula><mml:math id="M792" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, alkalinity, and CaCO<inline-formula><mml:math id="M793" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula> saturation in the global surface ocean, and temporal changes at selected locations,
Mar. Chem., 164, 95–125, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2014.06.004" ext-link-type="DOI">10.1016/j.marchem.2014.06.004</ext-link>,
2014.</mixed-citation></ref>
      <ref id="bib1.bibx103"><?xmltex \def\ref@label{{Takahashi et~al.(2020)}}?><label>Takahashi et al.(2020)</label><?label Takahashi2019?><mixed-citation>Takahashi, T., Sutherland, S., and Kozyr, A.:  Global Ocean Surface Water Partial Pressure of CO<inline-formula><mml:math id="M794" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> Database: Measurements Performed During 1957–2019 (LDEO Database Version 2019) (NCEI Accession 0160492), Version 9.9, NOAA National Centers for Environmental Information, Dataset, <ext-link xlink:href="https://doi.org/10.3334/CDIAC/OTG.NDP088(V2015)" ext-link-type="DOI">10.3334/CDIAC/OTG.NDP088(V2015)</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx104"><?xmltex \def\ref@label{{Takahashi et~al.(1993)}}?><label>Takahashi et al.(1993)</label><?label Takahashi1993?><mixed-citation>Takahashi, T. T., Olafsson, J., Goddard, J. G., Chipman, D. W., and Sutherland,
S. C.: Seasonal variation of CO<inline-formula><mml:math id="M795" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> and nutrients in the high-latitude surface
oceans: A comparative study, Global Biogeochem. Cy., 7, 843–878,
<ext-link xlink:href="https://doi.org/10.1029/93GB02263" ext-link-type="DOI">10.1029/93GB02263</ext-link>, 1993.</mixed-citation></ref>
      <?pagebreak page808?><ref id="bib1.bibx105"><?xmltex \def\ref@label{{Takahashi et~al.(2002)}}?><label>Takahashi et al.(2002)</label><?label Takahashi2002a?><mixed-citation>Takahashi, T. T., Sutherland, S. C., Sweeney, C., Poisson, A., Metzl, N.,
Tilbrook, B., Bates, N. R., Wanninkhof, R. H., Feely, R. A., Sabine, C. L.,
Olafsson, J., and Nojiri, Y.: Global sea-air CO<inline-formula><mml:math id="M796" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux based on
climatological surface ocean pCO2, and seasonal biological and temperature
effects, Deep-Sea Res. Pt. II, 49,
1601–1622, <ext-link xlink:href="https://doi.org/10.1016/S0967-0645(02)00003-6" ext-link-type="DOI">10.1016/S0967-0645(02)00003-6</ext-link>, 2002.</mixed-citation></ref>
      <ref id="bib1.bibx106"><?xmltex \def\ref@label{{Takahashi et~al.(2014)}}?><label>Takahashi et al.(2014)</label><?label Takahashi2014?><mixed-citation>Takahashi, T. T., Sutherland, S. C., Chipman, D. W., Goddard, J. G., and Ho,
C.: Climatological distributions of pH, <inline-formula><mml:math id="M797" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M798" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, total CO<inline-formula><mml:math id="M799" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, alkalinity, and
CaCO<inline-formula><mml:math id="M800" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">3</mml:mn></mml:msub></mml:math></inline-formula> saturation in the global surface ocean, and temporal changes at
selected locations, Mar. Chem., 164, 95–125,
<ext-link xlink:href="https://doi.org/10.1016/j.marchem.2014.06.004" ext-link-type="DOI">10.1016/j.marchem.2014.06.004</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx107"><?xmltex \def\ref@label{{Tilbrook et~al.(2019)}}?><label>Tilbrook et al.(2019)</label><?label Tilbrook2019?><mixed-citation>Tilbrook, B., Jewett, E. B., DeGrandpre, M. D., Hernandez-Ayon, J. M., Feely,
R. A., Gledhill, D. K., Hansson, L., Isensee, K., Kurz, M. L., Newton, J. A.,
Siedlecki, S. A., Chai, F., Dupont, S., Graco, M., Calvo, E., Greeley, D.,
Kapsenberg, L., Lebrec, M., Pelejero, C., Schoo, K. L., and Telszewski, M.:
An enhanced ocean acidification observing network: From people to technology
to data synthesis and information exchange, Front. Mar. Sci., 6,
1–21, <ext-link xlink:href="https://doi.org/10.3389/fmars.2019.00337" ext-link-type="DOI">10.3389/fmars.2019.00337</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx108"><?xmltex \def\ref@label{{Turk et~al.(2017)}}?><label>Turk et al.(2017)</label><?label Turk2017?><mixed-citation>Turk, D., Dowd, M., Lauvset, S. K., Koelling, J., Alonso-Pérez, F., and
Pérez, F. F.: Can empirical algorithms successfully estimate aragonite
saturation state in the subpolar North Atlantic?, Front. Mar.
Sci., 4, 385 pp., <ext-link xlink:href="https://doi.org/10.3389/fmars.2017.00385" ext-link-type="DOI">10.3389/fmars.2017.00385</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx109"><?xmltex \def\ref@label{{Uppstr{\"{o}}m(1974)}}?><label>Uppström(1974)</label><?label Uppstrom1974?><mixed-citation>Uppström, L. R.: The boron/chlorinity ratio of deep-sea water from the
Pacific Ocean, Deep-Sea Res. Oceanogr. Abstracts, 21, 161–162,
<ext-link xlink:href="https://doi.org/10.1016/0011-7471(74)90074-6" ext-link-type="DOI">10.1016/0011-7471(74)90074-6</ext-link>, 1974.
</mixed-citation></ref><?xmltex \hack{\newpage}?>
      <ref id="bib1.bibx110"><?xmltex \def\ref@label{{Watson et~al.(2000)}}?><label>Watson et al.(2000)</label><?label Watson2000?><mixed-citation>Watson, A. J., Bakker, D. C., Ridgwell, A. J., Boyd, P. W., and Law, C. S.:
Effect of iron supply on Southern Ocean CO<inline-formula><mml:math id="M801" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> uptake and implications for
glacial atmospheric CO<inline-formula><mml:math id="M802" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula>, Nature, 407, 730–733, <ext-link xlink:href="https://doi.org/10.1038/35037561" ext-link-type="DOI">10.1038/35037561</ext-link>,
2000.</mixed-citation></ref>
      <ref id="bib1.bibx111"><?xmltex \def\ref@label{{Watson et~al.(2020)}}?><label>Watson et al.(2020)</label><?label Watson2020?><mixed-citation>Watson, A. J., Schuster, U., Shutler, J. D., Holding, T., Ashton, I. G.,
Landschützer, P., Woolf, D. K., and Goddijn-Murphy, L.: Revised
estimates of ocean-atmosphere CO<inline-formula><mml:math id="M803" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> flux are consistent with ocean carbon
inventory, Nat. Commun., 11, 1–6,
<ext-link xlink:href="https://doi.org/10.1038/s41467-020-18203-3" ext-link-type="DOI">10.1038/s41467-020-18203-3</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx112"><?xmltex \def\ref@label{{Williams et~al.(2017)}}?><label>Williams et al.(2017)</label><?label Williams2017?><mixed-citation>Williams, N. L., Juranek, L. W., Feely, R. A., Johnson, K. S., Sarmiento,
J. L., Talley, L. D., Dickson, A. G., Gray, A. R., Wanninkhof, R. H.,
Russell, J. L., Riser, S. C., and Takeshita, Y.: Calculating surface ocean
<inline-formula><mml:math id="M804" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M805" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> from biogeochemical Argo floats equipped with pH: An uncertainty
analysis, Global Biogeochem. Cy., 31, 591–604,
<ext-link xlink:href="https://doi.org/10.1002/2016GB005541" ext-link-type="DOI">10.1002/2016GB005541</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx113"><?xmltex \def\ref@label{{Wolf-Gladrow et~al.(2007)}}?><label>Wolf-Gladrow et al.(2007)</label><?label Wolf-Gladrow2007?><mixed-citation>Wolf-Gladrow, D. A., Zeebe, R. E., Klaas, C., Körtzinger, A., and
Dickson, A. G.: Total alkalinity: The explicit conservative expression and
its application to biogeochemical processes, Mar. Chem., 106,
287–300, <ext-link xlink:href="https://doi.org/10.1016/j.marchem.2007.01.006" ext-link-type="DOI">10.1016/j.marchem.2007.01.006</ext-link>, 2007.</mixed-citation></ref>
      <ref id="bib1.bibx114"><?xmltex \def\ref@label{{Zeng et~al.(2014)}}?><label>Zeng et al.(2014)</label><?label Zeng2014?><mixed-citation>Zeng, J., Nojiri, Y., Landschützer, P., Telszewski, M., and Nakaoka,
S.-I.: A global surface ocean <inline-formula><mml:math id="M806" display="inline"><mml:mi>f</mml:mi></mml:math></inline-formula>CO<inline-formula><mml:math id="M807" display="inline"><mml:msub><mml:mi/><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:math></inline-formula> climatology based on a feed-forward
neural network, J. Atmos. Ocean. Technol., 31,
1838–1849, <ext-link xlink:href="https://doi.org/10.1175/JTECH-D-13-00137.1" ext-link-type="DOI">10.1175/JTECH-D-13-00137.1</ext-link>, 2014.</mixed-citation></ref>

  </ref-list></back>
    <!--<article-title-html>OceanSODA-ETHZ: a global gridded data set of the surface ocean carbonate system for seasonal to decadal studies of ocean acidification</article-title-html>
<abstract-html><p>Ocean acidification has profoundly altered the ocean's carbonate chemistry since preindustrial times, with potentially serious consequences for marine life. Yet, no long-term, global observation-based data set exists that allows us to study changes in ocean acidification for all carbonate system parameters over the last few decades. Here, we fill this gap and present a methodologically consistent global data set of all relevant surface ocean parameters, i.e., dissolved inorganic carbon (DIC), total alkalinity (TA), partial pressure of CO<sub>2</sub> (<i>p</i>CO<sub>2</sub>), pH, and the saturation state with respect to mineral CaCO<sub>3</sub> (Ω) at a monthly resolution over the period 1985 through 2018 at a spatial resolution of 1° × 1°. This data set, named OceanSODA-ETHZ, was created by extrapolating in time and space the surface ocean observations of <i>p</i>CO<sub>2</sub> (from the Surface Ocean CO<sub>2</sub> Atlas, SOCAT) and total alkalinity (TA; from the Global Ocean Data Analysis Project, GLODAP) using the newly developed Geospatial Random Cluster Ensemble Regression (GRaCER) method (code available at <a href="https://doi.org/10.5281/zenodo.4455354" target="_blank">https://doi.org/10.5281/zenodo.4455354</a>, Gregor, 2021). This method is based on a two-step (cluster-regression) approach but extends it by considering an ensemble of such cluster regressions, leading to improved robustness. Surface ocean DIC, pH, and Ω were then computed from the globally mapped <i>p</i>CO<sub>2</sub> and TA using the thermodynamic equations of the carbonate system. For the open ocean, the cluster-regression method estimates <i>p</i>CO<sub>2</sub> and TA with global near-zero biases and root mean squared errors of 12&thinsp;µatm and 13&thinsp;µmol&thinsp;kg<sup>−1</sup>, respectively. Taking into account also the measurement and representation errors, the total uncertainty increases to 14&thinsp;µatm and 21&thinsp;µmol&thinsp;kg<sup>−1</sup>, respectively. We assess the fidelity of the computed parameters by comparing them to direct observations from GLODAP, finding surface ocean pH and DIC global biases of near zero, as well as root mean squared errors of 0.023 and 16&thinsp;µmol&thinsp;kg<sup>−1</sup>, respectively. These uncertainties are very comparable to those expected by propagating the total uncertainty from <i>p</i>CO<sub>2</sub> and TA through the thermodynamic computations, indicating a robust and conservative assessment of the uncertainties. We illustrate the potential of this new data set by analyzing the climatological mean seasonal cycles of the different parameters of the surface ocean carbonate system, highlighting their commonalities and differences. Further, this data set provides a novel constraint on the global- and basin-scale trends in ocean acidification for all parameters. Concretely, we find for the period 1990 through 2018 global mean trends of 8.6&thinsp;±&thinsp;0.1&thinsp;µmol&thinsp;kg<sup>−1</sup> per decade for DIC, −0.016&thinsp;±&thinsp;0.000 per decade for pH, 16.5&thinsp;±&thinsp;0.1&thinsp;µatm  per decade for <i>p</i>CO<sub>2</sub>, and −0.07&thinsp;±&thinsp;0.00 per decade for Ω. The OceanSODA-ETHZ data can be downloaded from <a href="https://doi.org/10.25921/m5wx-ja34" target="_blank">https://doi.org/10.25921/m5wx-ja34</a> (Gregor and Gruber, 2020).</p></abstract-html>
<ref-html id="bib1.bib1"><label>Bakker et al.(2008)</label><mixed-citation>
Bakker, D. C. E., Hoppema, M., Schröder, M., Geibert, W., and de Baar, H. J. W.: A rapid transition from ice covered CO<sub>2</sub>–rich waters to a biologically mediated CO<sub>2</sub> sink in the eastern Weddell Gyre, Biogeosciences, 5, 1373–1386, <a href="https://doi.org/10.5194/bg-5-1373-2008" target="_blank">https://doi.org/10.5194/bg-5-1373-2008</a>, 2008.
</mixed-citation></ref-html>
<ref-html id="bib1.bib2"><label>Bakker et al.(2016)</label><mixed-citation>
Bakker, D. C., Pfeil, B., Landa, C. S., Metzl, N., O'Brien, K. M., Olsen, A.,
Smith, K., Cosca, C., Harasawa, S., Jones, S. D., Nakaoka, S.-I., Nojiri, Y.,
Schuster, U., Steinhoff, T., Sweeney, C., Takahashi, T. T., Tilbrook, B.,
Wada, C., Wanninkhof, R. H., Alin, S. R., Balestrini, C. F., Barbero, L.,
Bates, N. R., Bianchi, A. A., Bonou, F., Boutin, J., Bozec, Y., Burger,
E. F., Cai, W.-J., Castle, R. D., Chen, L., Chierici, M., Currie, K., Evans,
W., Featherstone, C., Feely, R. A., Fransson, A., Goyet, C., Greenwood, N.,
Gregor, L., Hankin, S., Hardman-Mountford, N. J., Harlay, J., Hauck, J.,
Hoppema, M., Humphreys, M. P., Hunt, C. W., Huss, B., Ibánhez, J.
S. P., Johannessen, T., Keeling, R., Kitidis, V., Körtzinger, A.,
Kozyr, A., Krasakopoulou, E., Kuwata, A., Landschützer, P., Lauvset,
S. K., Lefèvre, N., Lo Monaco, C., Manke, A., Mathis, J. T.,
Merlivat, L., Millero, F. J., Monteiro, P. M. S., Munro, D. R., Murata, A.,
Newberger, T., Omar, A. M., Ono, T., Paterson, K., Pearce, D., Pierrot, D.,
Robbins, L. L., Saito, S., Salisbury, J., Schlitzer, R., Schneider, B.,
Schweitzer, R., Sieger, R., Skjelvan, I., Sullivan, K. F., Sutherland, S. C.,
Sutton, A. J., Tadokoro, K., Telszewski, M., Tuma, M., Van Heuven, S. M.,
Vandemark, D., Ward, B., Watson, A. J., and Xu, S.: A multi-decade record of
high-quality <i>f</i>CO<sub>2</sub> data in version 3 of the Surface Ocean CO<sub>2</sub> Atlas (SOCAT),
Earth Syst. Sci. Data, 8, 383–413, <a href="https://doi.org/10.5194/essd-8-383-2016" target="_blank">https://doi.org/10.5194/essd-8-383-2016</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib3"><label>Bates and Peters(2007)</label><mixed-citation>
Bates, N. R. and Peters, A. J.: The contribution of atmospheric acid
deposition to ocean acidification in the subtropical North Atlantic Ocean,
Mar. Chem., 107, 547–558, <a href="https://doi.org/10.1016/j.marchem.2007.08.002" target="_blank">https://doi.org/10.1016/j.marchem.2007.08.002</a>, 2007.
</mixed-citation></ref-html>
<ref-html id="bib1.bib4"><label>Bates et al.(2014)</label><mixed-citation>
Bates, N. R., Astor, Y. M., Church, M. J., Currie, K., Dore, J. E.,
González-Dávila, M., Lorenzoni, L., Muller-Karger, F., Olafsson,
J., and Santana-Casiano, J. M.: A time-series view of changing surface ocean
chemistry due to ocean uptake of anthropogenic CO<sub>2</sub> and ocean acidification,
Oceanography, 27, 126–141, <a href="https://doi.org/10.5670/oceanog.2014.16" target="_blank">https://doi.org/10.5670/oceanog.2014.16</a>, 2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib5"><label>Bednaršek et al.(2019)</label><mixed-citation>
Bednaršek, N., Feely, R. A., Howes, E. L., Hunt, B. P. V., Kessouri, F.,
León, P., Lischka, S., Maas, A. E., McLaughlin, K., Nezlin, N. P.,
Sutula, M., and Weisberg, S. B.: Systematic Review and Meta-Analysis Toward
Synthesis of Thresholds of Ocean Acidification Impacts on Calcifying
Pteropods and Interactions With Warming, Front. Mar. Sci., 6, 227 pp.,
<a href="https://doi.org/10.3389/fmars.2019.00227" target="_blank">https://doi.org/10.3389/fmars.2019.00227</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib6"><label>Bittig et al.(2018)</label><mixed-citation>
Bittig, H. C., Steinhoff, T., Claustre, H., Fiedler, B., Williams, N. L.,
Sauzède, R., Körtzinger, A., and Gattuso, J.-P.: An alternative
to static climatologies: Robust estimation of open ocean CO<sub>2</sub> variables and
nutrient concentrations from T, S, and O<sub>2</sub> data using Bayesian neural
networks, Front. Mar. Sci., 5, 1–29,
<a href="https://doi.org/10.3389/fmars.2018.00328" target="_blank">https://doi.org/10.3389/fmars.2018.00328</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib7"><label>Bockmon and Dickson(2015)</label><mixed-citation>
Bockmon, E. E. and Dickson, A. G.: An inter-laboratory comparison assessing
the quality of seawater carbon dioxide measurements, Mar. Chem., 171,
36–43, <a href="https://doi.org/10.1016/j.marchem.2015.02.002" target="_blank">https://doi.org/10.1016/j.marchem.2015.02.002</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib8"><label>Bopp et al.(2013)</label><mixed-citation>
Bopp, L., Resplandy, L., Orr, J. C., Doney, S. C., Dunne, J. P., Gehlen, M.,
Halloran, P. R., Heinze, C., Ilyina, T., Séférian, R., Tjiputra,
J., and Vichi, M.: Multiple stressors of ocean ecosystems in the 21st
century: Projections with CMIP5 models, Biogeosciences, 10, 6225–6245,
<a href="https://doi.org/10.5194/bg-10-6225-2013" target="_blank">https://doi.org/10.5194/bg-10-6225-2013</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib9"><label>Boutin et al.(2018)</label><mixed-citation>
Boutin, J., Vergely, J. L., Marchand, S., D'Amico, F., Hasson, A.,
Kolodziejczyk, N., Reul, N., Reverdin, G., and Vialard, J.: New SMOS Sea
Surface Salinity with reduced systematic errors and improved variability,
Remote Sens. Environ., 214, 115–134,
<a href="https://doi.org/10.1016/j.rse.2018.05.022" target="_blank">https://doi.org/10.1016/j.rse.2018.05.022</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib10"><label>Boyer et al.(2013)</label><mixed-citation>
Boyer, T. P., Antonov, J. I., Baranova, O. K., Garcia, H. E., Johnson, D. R.,
Mishonov, A. V., O'Brien, T. D., Seidov, D., Smolyar, I. I., Zweng, M. M.,
Paver, C. R., Locarnini, R. A., Reagan, J. R., Forgy, C. C., Grodsky, A., and
Levitus, S.: World ocean database 2013, Tech. Rep., National Oceanographic
Data Center (U.S.), Ocean Climate Laboratory, 208 pp., <a href="https://doi.org/10.7289/V5NZ85MT" target="_blank">https://doi.org/10.7289/V5NZ85MT</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib11"><label>Breiman(2001)</label><mixed-citation>
Breiman, L.: Random forests, Mach. Learn., 45, 5–32,
<a href="https://doi.org/10.1023/A:1010933404324" target="_blank">https://doi.org/10.1023/A:1010933404324</a>, 2001.
</mixed-citation></ref-html>
<ref-html id="bib1.bib12"><label>Broecker and Peng(1974)</label><mixed-citation>
Broecker, W. S. and Peng, T.-H.: Gas exchange rates between air and
sea, Tellus, 26, 21–35, <a href="https://doi.org/10.3402/tellusa.v26i1-2.9733" target="_blank">https://doi.org/10.3402/tellusa.v26i1-2.9733</a>, 1974.
</mixed-citation></ref-html>
<ref-html id="bib1.bib13"><label>Broullón et al.(2018)</label><mixed-citation>
Broullón, D., Pérez, F. F., Velo, A., Hoppema, M., Olsen, A., Takahashi, T., Key, R. M., Tanhua, T., González-Dávila, M., Jeansson, E., Kozyr, A., and van Heuven, S. M. A. C.: A global monthly climatology of total alkalinity: a neural network approach, Earth Syst. Sci. Data, 11, 1109–1127, <a href="https://doi.org/10.5194/essd-11-1109-2019" target="_blank">https://doi.org/10.5194/essd-11-1109-2019</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib14"><label>Bushinsky et al.(2019)</label><mixed-citation>
Bushinsky, S. M., Landschützer, P., Rödenbeck, C., Gray, A. R.,
Baker, D., Mazloff, M. R., Resplandy, L., Johnson, K. S., and Sarmiento,
J. L.: Reassessing Southern Ocean air-sea CO<sub>2</sub> flux estimates with the
addition of biogeochemical float observations, Global Biogeochem. Cy.,
33, 1370–1388, <a href="https://doi.org/10.1029/2019GB006176" target="_blank">https://doi.org/10.1029/2019GB006176</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib15"><label>Carter et al.(2016)</label><mixed-citation>
Carter, B. R., Williams, N. L., Gray, A. R., and Feely, R. A.: Locally
interpolated alkalinity regression for global alkalinity estimation,
Limnol. Oceanogr.-Method., 14, 268–277, <a href="https://doi.org/10.1002/lom3.10087" target="_blank">https://doi.org/10.1002/lom3.10087</a>,
2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib16"><label>Carter et al.(2018)</label><mixed-citation>
Carter, B. R., Feely, R. A., Williams, N. L., Dickson, A. G., Fong, M. B., and
Takeshita, Y.: Updated methods for global locally interpolated estimation of
alkalinity, pH, and nitrate, Limnol. Oceanogr.-Method., 16,
119–131, <a href="https://doi.org/10.1002/lom3.10232" target="_blank">https://doi.org/10.1002/lom3.10232</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib17"><label>Carton et al.(2018)</label><mixed-citation>
Carton, J. A., Chepurin, G. A., and Chen, L.: SODA3: A new ocean climate
reanalysis, J. Clim., 31, 6967–6983,
<a href="https://doi.org/10.1175/JCLI-D-17-0149.1" target="_blank">https://doi.org/10.1175/JCLI-D-17-0149.1</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib18"><label>Cheng et al.(2020)</label><mixed-citation>
Cheng, L., Trenberth, K. E., Gruber, N., Abraham, J. P., Fasullo, J. T., Li,
G., Mann, M. E., Zhao, X., and Zhu, J.: Improved Estimates of Changes in
Upper Ocean Salinity and the Hydrological Cycle, J. Clim., 33,
10357–10381, <a href="https://doi.org/10.1175/JCLI-D-20-0366.1" target="_blank">https://doi.org/10.1175/JCLI-D-20-0366.1</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib19"><label>Claustre et al.(2020)</label><mixed-citation>
Claustre, H., Johnson, K. S., and Takeshita, Y.: Observing the Global Ocean
with Biogeochemical-Argo, Ann. Rev. Mar. Sci., 12, 23–48,
<a href="https://doi.org/10.1146/annurev-marine-010419-010956" target="_blank">https://doi.org/10.1146/annurev-marine-010419-010956</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib20"><label>Cooley and Doney(2009)</label><mixed-citation>
Cooley, S. R. and Doney, S. C.: Anticipating ocean acidification's economic
consequences for commercial fisheries, Environ. Res. Lett., 4, 8 pp.,
<a href="https://doi.org/10.1088/1748-9326/4/2/024007" target="_blank">https://doi.org/10.1088/1748-9326/4/2/024007</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib21"><label>Hersbach(2018)</label><mixed-citation>
Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horányi, A., Muñoz Sabater, J., Nicolas, J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., and Thépaut, J.-N.: ERA5 hourly data on single levels from 1979 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS), <a href="https://doi.org/10.24381/cds.adbb2d47" target="_blank">https://doi.org/10.24381/cds.adbb2d47</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib22"><label>Denvil-Sommer et al.(2019)</label><mixed-citation>
Denvil-Sommer, A., Gehlen, M., Vrac, M., and Mejia, C.: LSCE-FFNN-v1: a
two-step neural network model for the reconstruction of surface ocean <i>p</i>CO<sub>2</sub>
over the global ocean, Geosci. Model Dev., 12, 2091–2105,
<a href="https://doi.org/10.5194/gmd-12-2091-2019" target="_blank">https://doi.org/10.5194/gmd-12-2091-2019</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib23"><label>Dickson et al.(1990)</label><mixed-citation>
Dickson, A. G., Wesolowski, D. J., Palmer, D. A., and Mesmer, R. E.:
Dissociation constant of bisulfate ion in aqueous sodium chloride solutions
to 250&thinsp;°C, J. Phys. Chem., 94, 7978–7985,
<a href="https://doi.org/10.1021/j100383a042" target="_blank">https://doi.org/10.1021/j100383a042</a>, 1990.
</mixed-citation></ref-html>
<ref-html id="bib1.bib24"><label>Dickson et al.(2007)</label><mixed-citation>
Dickson, A. G., Sabine, C. L., and Christian, J. R. (Eds.): Guide to best
practices for ocean CO<sub>2</sub> measurements, North Pacific Marine Science
Organization, Sidney, British Columbia, 3rd Edn., available at:
<a href="https://www.oceanbestpractices.net/handle/11329/249" target="_blank"/> (last access: 22 May 2019), 2007.
</mixed-citation></ref-html>
<ref-html id="bib1.bib25"><label>Dlugokencky et al.(2019)</label><mixed-citation>
Dlugokencky, E., Thoning, K., Lang, P., and Tans, P.: NOAA Greenhouse Gas
Reference from Atmospheric Carbon Dioxide Dry Air Mole Fractions from the
NOAA ESRL Carbon Cycle Cooperative Global Air Sampling Network, available at:
<a href="ftp://aftp.cmdl.noaa.gov/data/trace_gases/co2/flask/surface/" target="_blank"/> (last accessed: 14 September 2020),
2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib26"><label>Doney et al.(2009)Doney, Fabry, Feely, and Kleypas</label><mixed-citation>
Doney, S. C., Fabry, V. J., Feely, R. A., and Kleypas, J. A.: Ocean
Acidification: The Other CO<sub>2</sub> Problem, Ann. Rev. Mar. Sci., 1,
169–192, <a href="https://doi.org/10.1146/annurev.marine.010908.163834" target="_blank">https://doi.org/10.1146/annurev.marine.010908.163834</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib27"><label>Doney et al.(2020)</label><mixed-citation>
Doney, S. C., Busch, D. S., Cooley, S. R., and Kroeker, K. J.: The Impacts of
Ocean Acidification on Marine Ecosystems and Reliant Human Communities,
Ann. Rev. Environ. Resour., 45, 1–30,
<a href="https://doi.org/10.1146/annurev-environ-012320-083019" target="_blank">https://doi.org/10.1146/annurev-environ-012320-083019</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib28"><label>Dore et al.(2009)</label><mixed-citation>
Dore, J. E., Lukas, R., Sadler, D. W., Church, M. J., and Karl, D. M.:
Physical and biogeochemical modulation of ocean acidification in the central
North Pacific, P. Natl. Acad. Sci. USA, 106,
12235–12240, <a href="https://doi.org/10.1073/pnas.0906044106" target="_blank">https://doi.org/10.1073/pnas.0906044106</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib29"><label>Fabry et al.(2008)</label><mixed-citation>
Fabry, V. J., Seibel, B. A., Feely, R. A., and Orr, J. C.: Impacts of ocean
acidification on marine fauna and ecosystem processes, ICES J.
Mar. Sci., 65, 414–432, <a href="https://doi.org/10.1093/icesjms/fsn048" target="_blank">https://doi.org/10.1093/icesjms/fsn048</a>, 2008.
</mixed-citation></ref-html>
<ref-html id="bib1.bib30"><label>Fabry et al.(2009)</label><mixed-citation>
Fabry, V. J., McClintock, J. B., Mathis, J. T., and Grebmeier, J. M.: Ocean
acidification at high latitudes: The Bellwether, Oceanography, 22, 160–171,
<a href="https://doi.org/10.5670/oceanog.2009.105" target="_blank">https://doi.org/10.5670/oceanog.2009.105</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib31"><label>Fassbender et al.(2018)</label><mixed-citation>
Fassbender, A. J., Rodgers, K. B., Palevsky, H. I., and Sabine, C. L.:
Seasonal Asymmetry in the Evolution of Surface Ocean <i>p</i>CO<sub>2</sub> and pH
Thermodynamic Drivers and the Influence on Sea-Air CO<sub>2</sub> Flux, Global
Biogeochem. Cy., 32, 1476–1497, <a href="https://doi.org/10.1029/2017GB005855" target="_blank">https://doi.org/10.1029/2017GB005855</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib32"><label>Fay and McKinley(2014)</label><mixed-citation>
Fay, A. R. and McKinley, G. A.: Global open-ocean biomes: Mean and temporal
variability, Earth Syst. Sc. Data, 6, 273–284,
<a href="https://doi.org/10.5194/essd-6-273-2014" target="_blank">https://doi.org/10.5194/essd-6-273-2014</a>, 2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib33"><label>Feely et al.(2009)</label><mixed-citation>
Feely, R. A., Doney, S., and Cooley, S.: Ocean Acidification: Present
Conditions and Future Changes in a High-CO<sub>2</sub> World, Oceanography, 22, 36–47,
<a href="https://doi.org/10.5670/oceanog.2009.95" target="_blank">https://doi.org/10.5670/oceanog.2009.95</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib34"><label>Franco et al.(2018)</label><mixed-citation>
Franco, A. C., Gruber, N., Frölicher, T. L., and Kropuenske Artman, L.:
Contrasting Impact of Future CO<sub>2</sub> Emission Scenarios on the Extent of CaCO<sub>3</sub>
Mineral Undersaturation in the Humboldt Current System, J.
Geophys. Res.-Ocean., 123, 2018–2036, <a href="https://doi.org/10.1002/2018JC013857" target="_blank">https://doi.org/10.1002/2018JC013857</a>,
2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib35"><label>Friedlingstein et al.(2019)</label><mixed-citation>
Friedlingstein, P., Jones, M. W., O'Sullivan, M., Andrew, R. M., Hauck, J.,
Peters, G. P., Peters, W., Pongratz, J., Sitch, S., Le Quéré,
C., DBakker, O. C., Canadell1, J. G., Ciais1, P., Jackson, R. B., Anthoni1,
P., Barbero, L., Bastos, A., Bastrikov, V., Becker, M., Bopp, L., Buitenhuis,
E., Chandra, N., Chevallier, F., Chini, L. P., Currie, K. I., Feely, R. A.,
Gehlen, M., Gilfillan, D., Gkritzalis, T., Goll, D. S., Gruber, N.,
Gutekunst, S., Harris, I., Haverd, V., Houghton, R. A., Hurtt, G., Ilyina,
T., Jain, A. K., Joetzjer, E., Kaplan, J. O., Kato, E., Goldewijk, K. K.,
Korsbakken, J. I., Landschützer, P., Lauvset, S. K., Lefèvre, N.,
Lenton, A., Lienert, S., Lombardozzi, D., Marland, G., McGuire, P. C.,
Melton, J. R., Metzl, N., Munro, D. R., Nabel, J. E., Nakaoka, S. I., Neill,
C., Omar, A. M., Ono, T., Peregon, A., Pierrot, D., Poulter, B., Rehder, G.,
Resplandy, L., Robertson, E., Rödenbeck, C., Séférian, R.,
Schwinger, J., Smith, N., Tans, P. P., Tian, H., Tilbrook, B., Tubiello,
F. N., Van Der Werf, G. R., Wiltshire, A. J., and Zaehle, S.: Global
carbon budget 2019, Earth Syst. Sci. Data, 11, 1783–1838,
<a href="https://doi.org/10.5194/essd-11-1783-2019" target="_blank">https://doi.org/10.5194/essd-11-1783-2019</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib36"><label>Gallego et al.(2018)</label><mixed-citation>
Gallego, M. A., Timmermann, A., Friedrich, T., and Zeebe, R. E.: Drivers of
future seasonal cycle changes in oceanic <i>p</i>CO<sub>2</sub>, Biogeosciences, 15,
5315–5327, <a href="https://doi.org/10.5194/bg-15-5315-2018" target="_blank">https://doi.org/10.5194/bg-15-5315-2018</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib37"><label>Gattuso et al.(1998)</label><mixed-citation>
Gattuso, J.-P., Frankignoulle, M., Bourge, I., Romaine, S., and Buddemeier,
R. W.: Effect of calcium carbonate saturation of seawater on coral
calcification, Glob. Planet. Change, 18, 37–46,
<a href="https://doi.org/10.1016/S0921-8181(98)00035-6" target="_blank">https://doi.org/10.1016/S0921-8181(98)00035-6</a>, 1998.
</mixed-citation></ref-html>
<ref-html id="bib1.bib38"><label>Goddijn‐Murphy et al.(2015)</label><mixed-citation>
Goddijn‐Murphy, L., Woolf, D. K., Land, P. E., Shutler, J. D., and Donlon,
C. J.: The OceanFlux Greenhouse Gases methodology for deriving a sea surface
climatology of CO<sub>2</sub> fugacity in support of air-sea gas flux studies, Ocean
Sci., 11, 519–541, <a href="https://doi.org/10.5194/os-11-519-2015" target="_blank">https://doi.org/10.5194/os-11-519-2015</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib39"><label>Good et al.(2013)</label><mixed-citation>
Good, S. A., Martin, M. J., and Rayner, N. A.: EN4: Quality controlled ocean
temperature and salinity profiles and monthly objective analyses with
uncertainty estimates, J. Geophys. Res.-Ocean., 118,
6704–6716, <a href="https://doi.org/10.1002/2013JC009067" target="_blank">https://doi.org/10.1002/2013JC009067</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib40"><label>Good et al.(2019)</label><mixed-citation>
Good, S. A., Embury, O., Bulgin, C., and Mittaz, J.: ESA Sea Surface
Temperature Climate Change Initiative (SST_cci): Level 4 Analysis Climate
Data Record, version 2.1, <a href="https://doi.org/10.5285/aced40d7cb964f23a0fd3e85772f2d48" target="_blank">https://doi.org/10.5285/aced40d7cb964f23a0fd3e85772f2d48</a>,
2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib41"><label>Good et al.(2020)</label><mixed-citation>
Good, S. A., Fiedler, E., Mao, C., Martin, M. J., Maycock, A., Reid, R.,
Roberts-Jones, J., Searle, T., Waters, J., While, J., and Worsfold, M.: The
current configuration of the OSTIA system for operational production of
foundation sea surface temperature and ice concentration analyses, Remote
Sens., 12, 1–20, <a href="https://doi.org/10.3390/rs12040720" target="_blank">https://doi.org/10.3390/rs12040720</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib42"><label>Gray et al.(2018)</label><mixed-citation>
Gray, A. R., Johnson, K. S., Bushinsky, S. M., Riser, S. C., Russell, J. L.,
Talley, L. D., Wanninkhof, R. H., Williams, N. L., and Sarmiento, J. L.:
Autonomous Biogeochemical Floats Detect Significant Carbon Dioxide
Outgassing in the High-Latitude Southern Ocean, Geophys. Res.
Lett., 45, 9049–9057, <a href="https://doi.org/10.1029/2018GL078013" target="_blank">https://doi.org/10.1029/2018GL078013</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib43"><label>Gregor(2021)</label><mixed-citation>
Gregor, L.: luke-gregor/OceanSODA-ETHZ: code, Zenodo, <a href="https://doi.org/10.5281/zenodo.4455354" target="_blank">https://doi.org/10.5281/zenodo.4455354</a>, 2021.
</mixed-citation></ref-html>
<ref-html id="bib1.bib44"><label>Gregor and Gruber(2020)</label><mixed-citation>
Gregor, L. and Gruber, N.: OceanSODA-ETHZ: A global gridded data set of the
surface ocean carbonate system for seasonal to decadal studies of ocean
acidification (NCEI Accession 0220059), Tech. Rep., NOAA National Centers
for Environmental Information, <a href="https://doi.org/10.25921/m5wx-ja34" target="_blank">https://doi.org/10.25921/m5wx-ja34</a>,
2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib45"><label>Gregor et al.(2018)</label><mixed-citation>
Gregor, L., Kok, S., and Monteiro, P. M. S.: Interannual drivers of the
seasonal cycle of CO<sub>2</sub> in the Southern Ocean, Biogeosciences, 15, 2361–2378,
<a href="https://doi.org/10.5194/bg-15-2361-2018" target="_blank">https://doi.org/10.5194/bg-15-2361-2018</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib46"><label>Gregor et al.(2019)</label><mixed-citation>
Gregor, L., Lebehot, A. D., Kok, S., and Scheel Monteiro, P. M.: A
comparative assessment of the uncertainties of global surface ocean CO<sub>2</sub>
estimates using a machine-learning ensemble (CSIR-ML6 version 2019a) – have
we hit the wall?, Geosci. Model Dev., 12, 5113–5136,
<a href="https://doi.org/10.5194/gmd-12-5113-2019" target="_blank">https://doi.org/10.5194/gmd-12-5113-2019</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib47"><label>Gruber and Sarmiento(1997)</label><mixed-citation>
Gruber, N. and Sarmiento, J. L.: Global patterns of marine nitrogen fixation
and denitrification, Global Biogeochem. Cy., 11, 235–266,
<a href="https://doi.org/10.1029/97GB00077" target="_blank">https://doi.org/10.1029/97GB00077</a>, 1997.
</mixed-citation></ref-html>
<ref-html id="bib1.bib48"><label>Gruber et al.(1996)</label><mixed-citation>
Gruber, N., Sarmiento, J. L., and Stocker, T. F.: An improved method for
detecting anthropogenic CO<sub>2</sub> in the oceans, Global Biogeochem. Cy., 10,
809–837, <a href="https://doi.org/10.1029/96GB01608" target="_blank">https://doi.org/10.1029/96GB01608</a>, 1996.
</mixed-citation></ref-html>
<ref-html id="bib1.bib49"><label>Gruber et al.(2012)</label><mixed-citation>
Gruber, N., Hauri, C., Lachkar, Z., Loher, D., Frölicher, T. L., and
Plattner, G.-K.: Rapid Progression of Ocean Acidification in the California
Current System, Science, 337, 220–223, <a href="https://doi.org/10.1126/science.1216773" target="_blank">https://doi.org/10.1126/science.1216773</a>, 2012.
</mixed-citation></ref-html>
<ref-html id="bib1.bib50"><label>Gruber et al.(2019)</label><mixed-citation>
Gruber, N., Clement, D., Carter, B. R., Feely, R. A., van Heuven, S., Hoppema,
M., Ishii, M., Key, R. M., Kozyr, A., Lauvset, S. K., Lo Monaco, C.,
Mathis, J. T., Murata, A., Olsen, A., Perez, F. F., Sabine, C. L., Tanhua,
T., and Wanninkhof, R. H.: The oceanic sink for anthropogenic CO<sub>2</sub> from 1994
to 2007, Science, 363, 1193–1199, <a href="https://doi.org/10.1126/science.aau5153" target="_blank">https://doi.org/10.1126/science.aau5153</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib51"><label>Hauri et al.(2016)</label><mixed-citation>
Hauri, C., Friedrich, T., and Timmermann, A.: Abrupt onset and prolongation of
aragonite undersaturation events in the Southern Ocean, Nat. Clim.
Change, 6, 172–176, <a href="https://doi.org/10.1038/nclimate2844" target="_blank">https://doi.org/10.1038/nclimate2844</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib52"><label>Holte et al.(2017)</label><mixed-citation>
Holte, J., Talley, L. D., Gilson, J., and Roemmich, D.: An Argo mixed layer
climatology and database, Geophys. Res. Lett., 44, 5618–5626,
<a href="https://doi.org/10.1002/2017GL073426" target="_blank">https://doi.org/10.1002/2017GL073426</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib53"><label>Humphreys et al.(2020)</label><mixed-citation>
Humphreys, M. P., Gregor, L., Pierrot, D., van Heuven, S., Lewis, E. R., and
Wallace, D. W. R.: PyCO2SYS: marine carbonate system calculations in
Python, Zenodo, <a href="https://doi.org/10.5281/zenodo.3967359" target="_blank">https://doi.org/10.5281/zenodo.3967359</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib54"><label>Iida et al.(2015)</label><mixed-citation>
Iida, Y., Kojima, A., Takatani, Y., Nakano, T., Sugimoto, H., Midorikawa, T.,
and Ishii, M.: Trends in <i>p</i>CO<sub>2</sub> and sea–air CO<sub>2</sub> flux over the global open
oceans for the last two decades, J. Oceanogr., 71, 637–661,
<a href="https://doi.org/10.1007/s10872-015-0306-4" target="_blank">https://doi.org/10.1007/s10872-015-0306-4</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib55"><label>Jiang et al.(2019)</label><mixed-citation>
Jiang, L.-q., Carter, B. R., Feely, R. A., Lauvset, S. K., and Olsen, A.:
Surface ocean pH and buffer capacity: past, present and future, Sci.
Rep., 9, 18624, <a href="https://doi.org/10.1038/s41598-019-55039-4" target="_blank">https://doi.org/10.1038/s41598-019-55039-4</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib56"><label>Johnson et al.(2016)</label><mixed-citation>
Johnson, K. S., Jannasch, H. W., Coletti, L. J., Elrod, V. A., Martz, T. R.,
Takeshita, Y., Carlson, R. J., and Connery, J. G.: Deep-Sea DuraFET: A
Pressure Tolerant pH Sensor Designed for Global Sensor Networks, Anal.
Chem., 88, 3249–3256, <a href="https://doi.org/10.1021/acs.analchem.5b04653" target="_blank">https://doi.org/10.1021/acs.analchem.5b04653</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib57"><label>Johnson et al.(2017)</label><mixed-citation>
Johnson, K. S., Plant, J. N., Coletti, L. J., Jannasch, H. W., Sakamoto, C. M.,
Riser, S. C., Swift, D. D., Williams, N. L., Boss, E., Haëntjens, N.,
Talley, L. D., and Sarmiento, J. L.: Biogeochemical sensor performance in
the SOCCOM profiling float array, J. Geophys. Res.-Ocean., 122, 6416–6436,
<a href="https://doi.org/10.1002/2017JC012838" target="_blank">https://doi.org/10.1002/2017JC012838</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib58"><label>Keppler and Landschützer(2019)</label><mixed-citation>
Keppler, L. and Landschützer, P.: Regional Wind Variability Modulates
the Southern Ocean Carbon Sink, Sci. Rep., 9, 7384,
<a href="https://doi.org/10.1038/s41598-019-43826-y" target="_blank">https://doi.org/10.1038/s41598-019-43826-y</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib59"><label>Keppler et al.(2020)</label><mixed-citation>
Keppler, L., Landschützer, P., Gruber, N., Lauvset, S. K., and Stemmler,
I.: Seasonal carbon dynamics in the global ocean based on a neural network
mapping of observations, Global Biogeochem. Cy., 34, e2020GB006571, <a href="https://doi.org/10.1029/2020GB006571" target="_blank">https://doi.org/10.1029/2020GB006571</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib60"><label>Kwiatkowski et al.(2020)</label><mixed-citation>
Kwiatkowski, L., Torres, O., Bopp, L., Aumont, O., Chamberlain, M., Christian, J. R., Dunne, J. P., Gehlen, M., Ilyina, T., John, J. G., Lenton, A., Li, H., Lovenduski, N. S., Orr, J. C., Palmieri, J., Santana-Falcón, Y., Schwinger, J., Séférian, R., Stock, C. A., Tagliabue, A., Takano, Y., Tjiputra, J., Toyama, K., Tsujino, H., Watanabe, M., Yamamoto, A., Yool, A., and Ziehn, T.: Twenty-first century ocean warming, acidification, deoxygenation, and upper-ocean nutrient and primary production decline from CMIP6 model projections, Biogeosciences, 17, 3439–3470, <a href="https://doi.org/10.5194/bg-17-3439-2020" target="_blank">https://doi.org/10.5194/bg-17-3439-2020</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib61"><label>Land et al.(2019)</label><mixed-citation>
Land, P. E., Findlay, H. S., Shutler, J. D., Ashton, I. G., Holding, T.,
Grouazel, A., Girard-Ardhuin, F., Reul, N., Piolle, J. F., Chapron, B.,
Quilfen, Y., Bellerby, R. G., Bhadury, P., Salisbury, J., Vandemark, D., and
Sabia, R.: Optimum satellite remote sensing of the marine carbonate system
using empirical algorithms in the global ocean, the Greater Caribbean, the
Amazon Plume and the Bay of Bengal, Remote Sens. Environ., 235,
111469, <a href="https://doi.org/10.1016/j.rse.2019.111469" target="_blank">https://doi.org/10.1016/j.rse.2019.111469</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib62"><label>Landschützer et al.(2013)</label><mixed-citation>
Landschützer, P., Gruber, N., Bakker, D. C., Schuster, U., Nakaoka, S.,
Payne, M. R., Sasse, T. P., and Zeng, J.: A neural network-based estimate of
the seasonal to inter-annual variability of the Atlantic Ocean carbon sink,
Biogeosciences, 10, 7793–7815, <a href="https://doi.org/10.5194/bg-10-7793-2013" target="_blank">https://doi.org/10.5194/bg-10-7793-2013</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib63"><label>Landschützer et al.(2014)</label><mixed-citation>
Landschützer, P., Gruber, N., Bakker, D. C., and Schuster, U.: Recent variability of the global ocean carbon sink, Global Biogeochem. Cy., 28, 927–949, <a href="https://doi.org/10.1002/2014GB004853" target="_blank">https://doi.org/10.1002/2014GB004853</a>, 2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib64"><label>Landschützer et al.(2015)</label><mixed-citation>
Landschützer, P., Gruber, N., Haumann, F. A., Rödenbeck, C.,
Bakker, D. C., Van Heuven, S. M., Hoppema, M., Metzl, N., Sweeney, C.,
Takahashi, T. T., Tilbrook, B., and Wanninkhof, R. H.: The reinvigoration of
the Southern Ocean carbon sink, Science, 349, 1221–1224,
<a href="https://doi.org/10.1126/science.aab2620" target="_blank">https://doi.org/10.1126/science.aab2620</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib65"><label>Landschützer et al.(2016)</label><mixed-citation>
Landschützer, P., Gruber, N., and Bakker, D. C.: Decadal variations and
trends of the global ocean carbon sink, Global Biogeochem. Cy., 30,
1396–1417, <a href="https://doi.org/10.1002/2015GB005359" target="_blank">https://doi.org/10.1002/2015GB005359</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib66"><label>Landschützer et al.(2020)</label><mixed-citation>
Landschützer, P., Laruelle, G. G., Roobaert, A., and Regnier, P.: A uniform <i>p</i>CO<sub>2</sub> climatology combining open and coastal oceans, Earth Syst. Sci. Data, 12, 2537–2553, <a href="https://doi.org/10.5194/essd-12-2537-2020" target="_blank">https://doi.org/10.5194/essd-12-2537-2020</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib67"><label>Laruelle et al.(2013)</label><mixed-citation>
Laruelle, G. G., Dürr, H. H., Lauerwald, R., Hartmann, J., Slomp, C. P.,
Goossens, N., and Regnier, P. A.: Global multi-scale segmentation of
continental and coastal waters from the watersheds to the continental
margins, Hydrol. Earth Syst. Sci., 17, 2029–2051,
<a href="https://doi.org/10.5194/hess-17-2029-2013" target="_blank">https://doi.org/10.5194/hess-17-2029-2013</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib68"><label>Laruelle et al.(2017)</label><mixed-citation>
Laruelle, G. G., Landschützer, P., Gruber, N., Ti, J. L., Delille, B.,
Regnier, P., Tison, J.-L., Delille, B., and Regnier, P.: Global
high-resolution monthly <i>p</i>CO<sub>2</sub> climatology for the coastal ocean derived from
neural network interpolation, Biogeosciences, 14, 4545–4561,
<a href="https://doi.org/10.5194/bg-14-4545-2017" target="_blank">https://doi.org/10.5194/bg-14-4545-2017</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib69"><label>Lauvset et al.(2015)</label><mixed-citation>
Lauvset, S. K., Gruber, N., Landschützer, P., Olsen, A., and Tjiputra,
J.: Trends and drivers in global surface ocean pH over the past 3 decades,
Biogeosciences, 12, 1285–1298, <a href="https://doi.org/10.5194/bg-12-1285-2015" target="_blank">https://doi.org/10.5194/bg-12-1285-2015</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib70"><label>Lauvset et al.(2016)</label><mixed-citation>
Lauvset, S. K., Key, R. M., Olsen, A., van Heuven, S., Velo, A., Lin, X.,
Schirnick, C., Kozyr, A., Tanhua, T., Hoppema, M., Jutterström, S.,
Steinfeldt, R., Jeansson, E., Ishii, M., Perez, F. F., Suzuki, T., and
Watelet, S.: A new global interior ocean mapped climatology: the 1° × 1°
GLODAP version 2, Earth Syst. Sci. Data, 8, 325–340,
<a href="https://doi.org/10.5194/essd-8-325-2016" target="_blank">https://doi.org/10.5194/essd-8-325-2016</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib71"><label>Lee et al.(2000)</label><mixed-citation>
Lee, K., Wanninkhof, R. H., Feely, R. A., Millero, F. J., and Peng, T. H.:
Global relationships of total inorganic carbon with temperature and nitrate
in surface seawater, Global Biogeochem. Cy., 14, 979–994,
<a href="https://doi.org/10.1029/1998GB001087" target="_blank">https://doi.org/10.1029/1998GB001087</a>, 2000.
</mixed-citation></ref-html>
<ref-html id="bib1.bib72"><label>Lee et al.(2006)</label><mixed-citation>
Lee, K., Tong, L. T., Millero, F. J., Sabine, C. L., Dickson, A. G., Goyet, C.,
Park, G.-H., Wanninkhof, R. H., Feely, R. A., and Key, R. M.: Global
relationships of total alkalinity with salinity and temperature in surface
waters of the world's oceans, Geophys. Res. Lett., 33, L19605,
<a href="https://doi.org/10.1029/2006GL027207" target="_blank">https://doi.org/10.1029/2006GL027207</a>, 2006.
</mixed-citation></ref-html>
<ref-html id="bib1.bib73"><label>Lefèvre and Taylor(2002)</label><mixed-citation>
Lefèvre, N. and Taylor, A.: Estimating <i>p</i>CO<sub>2</sub> from sea surface
temperatures in the Atlantic gyres, Deep-Sea Res. Pt. I, 49, 539–554, <a href="https://doi.org/10.1016/S0967-0637(01)00064-4" target="_blank">https://doi.org/10.1016/S0967-0637(01)00064-4</a>, 2002.
</mixed-citation></ref-html>
<ref-html id="bib1.bib74"><label>Lewis et al.(1998)</label><mixed-citation>
Lewis, E., Wallace, D., and Allison, L. J.: Program developed for CO<sub>2</sub> system
calculations,  CDIAC, <a href="https://doi.org/10.15485/1464255" target="_blank">https://doi.org/10.15485/1464255</a>, 1998.
</mixed-citation></ref-html>
<ref-html id="bib1.bib75"><label>Lovenduski et al.(2007)</label><mixed-citation>
Lovenduski, N. S., Gruber, N., Doney, S. C., and Lima, I. D.: Enhanced CO<sub>2</sub>
outgassing in the Southern Ocean from a positive phase of the Southern
Annular Mode, Global Biogeochem. Cy., 21, 1–14,
<a href="https://doi.org/10.1029/2006GB002900" target="_blank">https://doi.org/10.1029/2006GB002900</a>, 2007.
</mixed-citation></ref-html>
<ref-html id="bib1.bib76"><label>Maritorena et al.(2010)</label><mixed-citation>
Maritorena, S., Fanton D'andon, O. H., Mangin, A., Siegel, D. A., D'Andon, O.
H. F., Mangin, A., and Siegel, D. A.: Merged satellite ocean color data
products using a bio-optical model: Characteristics, benefits and issues,
Remote Sens. Environ., 114, 1791–1804,
<a href="https://doi.org/10.1016/j.rse.2010.04.002" target="_blank">https://doi.org/10.1016/j.rse.2010.04.002</a>, 2010.
</mixed-citation></ref-html>
<ref-html id="bib1.bib77"><label>Marshall and Speer(2012)</label><mixed-citation>
Marshall, J. C. and Speer, K.: Closure of the meridional overturning
circulation through Southern Ocean upwelling, Nat. Geosci., 5,
171–180, <a href="https://doi.org/10.1038/ngeo1391" target="_blank">https://doi.org/10.1038/ngeo1391</a>, 2012.
</mixed-citation></ref-html>
<ref-html id="bib1.bib78"><label>Matsumoto and Gruber(2005)</label><mixed-citation>
Matsumoto, K. and Gruber, N.: How accurate is the estimation of anthropogenic
carbon in the ocean? An evaluation of the ΔC* method, Global
Biogeochem. Cy., 19, 1–17, <a href="https://doi.org/10.1029/2004GB002397" target="_blank">https://doi.org/10.1029/2004GB002397</a>, 2005.
</mixed-citation></ref-html>
<ref-html id="bib1.bib79"><label>Mehrbach et al.(1973)</label><mixed-citation>
Mehrbach, C., Culberson, C. H., Hawley, J. E., and Pytkowicx, R. M.:
Measurement of the Apparent Dissociation Constants of Carbonic Acid in
Seawater At Atmospheric Pressure, Limnol. Oceanogr., 18, 897–907,
<a href="https://doi.org/10.4319/lo.1973.18.6.0897" target="_blank">https://doi.org/10.4319/lo.1973.18.6.0897</a>, 1973.
</mixed-citation></ref-html>
<ref-html id="bib1.bib80"><label>Millero et al.(1998)</label><mixed-citation>
Millero, F. J., Lee, K., and Roche, M.: Distribution of alkalinity in the
surface waters of the major oceans, Mar. Chem., 60, 111–130,
<a href="https://doi.org/10.1016/S0304-4203(97)00084-4" target="_blank">https://doi.org/10.1016/S0304-4203(97)00084-4</a>, 1998.
</mixed-citation></ref-html>
<ref-html id="bib1.bib81"><label>Monteiro et al.(2015)</label><mixed-citation>
Monteiro, P. M. S., Gregor, L., Lévy, M., Maenner, S., Sabine, C. L., and
Swart, S.: Intraseasonal variability linked to sampling alias in air-sea CO<sub>2</sub>
fluxes in the Southern Ocean, Geophys. Res. Lett., 42, 8507–8514,
<a href="https://doi.org/10.1002/2015GL066009" target="_blank">https://doi.org/10.1002/2015GL066009</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib82"><label>Negrete-García et al.(2019)</label><mixed-citation>
Negrete-García, G., Lovenduski, N. S., Hauri, C., Krumhardt, K. M., and
Lauvset, S. K.: Sudden emergence of a shallow aragonite saturation horizon
in the Southern Ocean, Nat. Clim. Change, 9, 313–317,
<a href="https://doi.org/10.1038/s41558-019-0418-8" target="_blank">https://doi.org/10.1038/s41558-019-0418-8</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib83"><label>Olafsson et al.(2010)</label><mixed-citation>
Olafsson, J., Olafsdottir, S. R., Benoit-Cattin, A., and Takahashi, T. T.: The
irminger sea and the Iceland sea time series measurements of sea water carbon
and nutrient chemistry 1983–2008, Earth Syst. Sci. Data, 2, 99–104,
<a href="https://doi.org/10.5194/essd-2-99-2010" target="_blank">https://doi.org/10.5194/essd-2-99-2010</a>, 2010.
</mixed-citation></ref-html>
<ref-html id="bib1.bib84"><label>Olsen et al.(2016)</label><mixed-citation>
Olsen, A., Key, R. M., van Heuven, S., Lauvset, S. K., Velo, A., Lin, X.,
Schirnick, C., Kozyr, A., Tanhua, T., Hoppema, M., Jutterström, S.,
Steinfeldt, R., Jeansson, E., Ishii, M., Pérez, F. F., and Suzuki, T.:
The Global Ocean Data Analysis Project version 2 (GLODAPv2) – an
internally consistent data product for the world ocean, Earth Syst. Sci.
Data, 8, 297–323, <a href="https://doi.org/10.5194/essd-8-297-2016" target="_blank">https://doi.org/10.5194/essd-8-297-2016</a>, 2016.
</mixed-citation></ref-html>
<ref-html id="bib1.bib85"><label>Olsen et al.(2019)</label><mixed-citation>
Olsen, A., Lange, N., Key, R. M., Tanhua, T., Álvarez, M., Becker, S.,
Bittig, H. C., Carter, B. R., Cotrim Da Cunha, L., Feely, R. A., Van
Heuven, S., Hoppema, M., Ishii, M., Jeansson, E., Jones, S. D.,
Jutterström, S., Karlsen, M. K., Kozyr, A., Lauvset, S. K., Lo
Monaco, C., Murata, A., Pérez, F. F., Pfeil, B., Schirnick, C.,
Steinfeldt, R., Suzuki, T., Telszewski, M., Tilbrook, B., Velo, A., and
Wanninkhof, R.: GLODAPv2.2019 – An update of GLODAPv2, Earth Syst. Sci.
Data, 11, 1437–1461, <a href="https://doi.org/10.5194/essd-11-1437-2019" target="_blank">https://doi.org/10.5194/essd-11-1437-2019</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib86"><label>Orr et al.(2005)</label><mixed-citation>
Orr, J. C., Fabry, V. J., Aumont, O., Bopp, L., Doney, S. C., Feely, R. A.,
Gnanadesikan, A., Gruber, N., Ishida, A., Joos, F., Key, R. M., Lindsay, K.,
Maier-Reimer, E., Matear, R. J., Monfray, P., Mouchet, A., Najjar, R. G.,
Plattner, G.-K. K., Rodgers, K. B., Sabine, C. L., Sarmiento, J. L.,
Schlitzer, R., Slater, R. D., Totterdell, I. J., Weirig, M.-F. F., Yamanaka,
Y., Yool, A., Doney, S. C., Gnanadesikan, A., Gruber, N., Ishida, A., Joos,
F., Key, R. M., Lindsay, K., Maier-Reimer, E., Matear, R. J., Monfray, P.,
Mouchet, A., Najjar, R. G., Plattner, G.-K. K., Rodgers, K. B., Sabine,
C. L., Sarmiento, J. L., Schlitzer, R., Slater, R. D., Totterdell, I. J.,
Weirig, M.-F. F., Yamanaka, Y., Yool, A., Feely, R. A., Gnanadesikan, A.,
Gruber, N., Ishida, A., Joos, F., Key, R. M., Lindsay, K., Maier-Reimer, E.,
Matear, R. J., Monfray, P., Mouchet, A., Najjar, R. G., Plattner, G.-K. K.,
Rodgers, K. B., Sabine, C. L., Sarmiento, J. L., Schlitzer, R., Slater,
R. D., Totterdell, I. J., Weirig, M.-F. F., Yamanaka, Y., and Yool, A.:
Anthropogenic ocean acidification over the twenty-first century and its
impact on calcifying organisms, Nature, 437, 681–686,
<a href="https://doi.org/10.1038/nature04095" target="_blank">https://doi.org/10.1038/nature04095</a>, 2005.
</mixed-citation></ref-html>
<ref-html id="bib1.bib87"><label>Orr et al.(2015)</label><mixed-citation>
Orr, J. C., Epitalon, J. M., and Gattuso, J.-P.: Comparison of ten packages
that compute ocean carbonate chemistry, Biogeosciences, 12, 1483–1510,
<a href="https://doi.org/10.5194/bg-12-1483-2015" target="_blank">https://doi.org/10.5194/bg-12-1483-2015</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib88"><label>Orr et al.(2018)</label><mixed-citation>
Orr, J. C., Epitalon, J.-M., Dickson, A. G., and Gattuso, J.-P.: Routine
uncertainty propagation for the marine carbon dioxide system, Mar.
Chem., 2018, 84–107, <a href="https://doi.org/10.1016/j.marchem.2018.10.006" target="_blank">https://doi.org/10.1016/j.marchem.2018.10.006</a>, 2018.
</mixed-citation></ref-html>
<ref-html id="bib1.bib89"><label>Pedregosa et al.(2011)</label><mixed-citation>
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, C., Thirion, B., Grisel,
O., Blondel, M., Prettenhoffer, P., Weiss, R., Dubourg, V., Vanderplas, J.,
Passos, A., and Cournapeau, D.: Scikit-learn: Machine learning in Python,
J. Mach. Learn. Res., 12, 2825–2830, 2011.
</mixed-citation></ref-html>
<ref-html id="bib1.bib90"><label>Pierrot et al.(2009)</label><mixed-citation>
Pierrot, D., Neill, C., Sullivan, K. F., Castle, R. D., Wanninkhof, R. H.,
Lüger, H., Johannessen, T., Olsen, A., Feely, R. A., and Cosca, C. E.:
Recommendations for autonomous underway <i>p</i>CO<sub>2</sub> measuring systems and
data-reduction routines, Deep-Sea Res. Pt. II, 56, 512–522, <a href="https://doi.org/10.1016/j.dsr2.2008.12.005" target="_blank">https://doi.org/10.1016/j.dsr2.2008.12.005</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib91"><label>Pörtner and Farrell(2008)</label><mixed-citation>
Pörtner, H.-O. and Farrell, A. P.: Physiology and Climate Change,
Science, 322, 690–692, <a href="https://doi.org/10.1126/science.1163156" target="_blank">https://doi.org/10.1126/science.1163156</a>, 2008.
</mixed-citation></ref-html>
<ref-html id="bib1.bib92"><label>Raimondi et al.(2019)</label><mixed-citation>
Raimondi, L., Matthews, J. B. R., Atamanchuk, D., Azetsu-Scott, K., and
Wallace, D. W.: The internal consistency of the marine carbon dioxide system
for high latitude shipboard and in situ monitoring, Mar. Chem., 213,
49–70, <a href="https://doi.org/10.1016/j.marchem.2019.03.001" target="_blank">https://doi.org/10.1016/j.marchem.2019.03.001</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib93"><label>Resplandy et al.(2009)</label><mixed-citation>
Resplandy, L., Lévy, M., D'Ovidio, F., and Merlivat, L.: Impact of
submesoscale variability in estimating the air-sea CO<sub>2</sub> exchange: Results from
a model study of the POMME experiment, Global Biogeochem. Cy., 23,
1–19, <a href="https://doi.org/10.1029/2008GB003239" target="_blank">https://doi.org/10.1029/2008GB003239</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib94"><label>Rödenbeck et al.(2014)</label><mixed-citation>
Rödenbeck, C., Bakker, D. C., Metzl, N., Olsen, A., Sabine, C. L.,
Cassar, N., Reum, F., Keeling, R. F., and Heimann, M.: Interannual sea–air
CO<sub>2</sub> flux variability from an observation-driven ocean mixed-layer scheme,
Biogeosciences, 11, 4599–4613, <a href="https://doi.org/10.5194/bg-11-4599-2014" target="_blank">https://doi.org/10.5194/bg-11-4599-2014</a>, 2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib95"><label>Rödenbeck et al.(2015)</label><mixed-citation>
Rödenbeck, C., Bakker, D. C., Gruber, N., Iida, Y., Jacobson, A. R.,
Jones, S. D., Landschützer, P., Metzl, N., Nakaoka, S., Olsen, A.,
Park, G.-H. H., Peylin, P., Rodgers, K. B., Sasse, T. P., Schuster, U.,
Shutler, J. D., Valsala, V., Wanninkhof, R. H., and Zeng, J.: Data-based
estimates of the ocean carbon sink variability – first results of the
Surface Ocean pCO2 Mapping intercomparison (SOCOM), Biogeosciences, 12,
7251–7278, <a href="https://doi.org/10.5194/bg-12-7251-2015" target="_blank">https://doi.org/10.5194/bg-12-7251-2015</a>, 2015.
</mixed-citation></ref-html>
<ref-html id="bib1.bib96"><label>Sabine et al.(2004)</label><mixed-citation>
Sabine, C. L., Feely, R. A., Gruber, N., Key, R. M., Lee, K., Bullister, J. L.,
Wanninkhof, R. H., Wong, C. S., Wallace, D. W. R., Tilbrook, B., Millero,
F. J., Peng, T.-H., Kozyr, A., Ono1, T., and Rios, A. F.: The Oceanic Sink
for Anthropogenic CO<sub>2</sub>, Science, 305, 367–371,
<a href="https://doi.org/10.1126/science.1097403" target="_blank">https://doi.org/10.1126/science.1097403</a>, 2004.
</mixed-citation></ref-html>
<ref-html id="bib1.bib97"><label>Sarmiento and Gruber(2006)</label><mixed-citation>
Sarmiento, J. L. and Gruber, N.: Ocean Biogeochemical Dynamics, Princeton
University Press, Princeton, NJ, 526 pp., 2006.
</mixed-citation></ref-html>
<ref-html id="bib1.bib98"><label>Sasse et al.(2013)</label><mixed-citation>
Sasse, T. P., McNeil, B. I., and Abramowitz, G.: A novel method for diagnosing
seasonal to inter-annual surface ocean carbon dynamics from bottle data using
neural networks, Biogeosciences, 10, 4319–4340,
<a href="https://doi.org/10.5194/bg-10-4319-2013" target="_blank">https://doi.org/10.5194/bg-10-4319-2013</a>, 2013.
</mixed-citation></ref-html>
<ref-html id="bib1.bib99"><label>Siegel et al.(2002)</label><mixed-citation>
Siegel, D. A., Doney, S. C., and Yoder, J. A.: The North Atlantic spring
phytoplankton bloom and Sverdrup's critical depth hypothesis, Science, 296,
730–733, <a href="https://doi.org/10.1126/science.1069174" target="_blank">https://doi.org/10.1126/science.1069174</a>, 2002.
</mixed-citation></ref-html>
<ref-html id="bib1.bib100"><label>Steinacher et al.(2009)</label><mixed-citation>
Steinacher, M., Joos, F., Frölicher, T. L., Plattner, G.-K., and Doney, S. C.: Imminent ocean acidification in the Arctic projected with the NCAR global coupled carbon cycle-climate model, Biogeosciences, 6, 515–533, <a href="https://doi.org/10.5194/bg-6-515-2009" target="_blank">https://doi.org/10.5194/bg-6-515-2009</a>, 2009.
</mixed-citation></ref-html>
<ref-html id="bib1.bib101"><label>Tagliabue et al.(2017)</label><mixed-citation>
Tagliabue, A., Bowie, A. R., Boyd, P. W., Buck, K. N., Johnson, K. S., Saito,
M. A., Philip, W., Buck, K. N., Johnson, K. S., Saito, M. A., Boyd, P. W.,
Buck, K. N., Johnson, K. S., and Saito, M. A.: The integral role of iron in
ocean biogeochemistry, Nature, 543, 51–59, <a href="https://doi.org/10.1038/nature21058" target="_blank">https://doi.org/10.1038/nature21058</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib102"><label>Takahashi et al.(2014)</label><mixed-citation>
Takahashi, T., Sutherland, S. C., Chipman, D. W., Goddard, J. G., Ho, C., Newberger, T., Sweeney, C., and Munro, D. R.: Climatological distributions of pH, <i>p</i>CO<sub>2</sub>, total CO<sub>2</sub>, alkalinity, and CaCO<sub>3</sub> saturation in the global surface ocean, and temporal changes at selected locations,
Mar. Chem., 164, 95–125, <a href="https://doi.org/10.1016/j.marchem.2014.06.004" target="_blank">https://doi.org/10.1016/j.marchem.2014.06.004</a>,
2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib103"><label>Takahashi et al.(2020)</label><mixed-citation>
Takahashi, T., Sutherland, S., and Kozyr, A.:  Global Ocean Surface Water Partial Pressure of CO<sub>2</sub> Database: Measurements Performed During 1957–2019 (LDEO Database Version 2019) (NCEI Accession 0160492), Version 9.9, NOAA National Centers for Environmental Information, Dataset, <a href="https://doi.org/10.3334/CDIAC/OTG.NDP088(V2015)" target="_blank">https://doi.org/10.3334/CDIAC/OTG.NDP088(V2015)</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib104"><label>Takahashi et al.(1993)</label><mixed-citation>
Takahashi, T. T., Olafsson, J., Goddard, J. G., Chipman, D. W., and Sutherland,
S. C.: Seasonal variation of CO<sub>2</sub> and nutrients in the high-latitude surface
oceans: A comparative study, Global Biogeochem. Cy., 7, 843–878,
<a href="https://doi.org/10.1029/93GB02263" target="_blank">https://doi.org/10.1029/93GB02263</a>, 1993.
</mixed-citation></ref-html>
<ref-html id="bib1.bib105"><label>Takahashi et al.(2002)</label><mixed-citation>
Takahashi, T. T., Sutherland, S. C., Sweeney, C., Poisson, A., Metzl, N.,
Tilbrook, B., Bates, N. R., Wanninkhof, R. H., Feely, R. A., Sabine, C. L.,
Olafsson, J., and Nojiri, Y.: Global sea-air CO<sub>2</sub> flux based on
climatological surface ocean pCO2, and seasonal biological and temperature
effects, Deep-Sea Res. Pt. II, 49,
1601–1622, <a href="https://doi.org/10.1016/S0967-0645(02)00003-6" target="_blank">https://doi.org/10.1016/S0967-0645(02)00003-6</a>, 2002.
</mixed-citation></ref-html>
<ref-html id="bib1.bib106"><label>Takahashi et al.(2014)</label><mixed-citation>
Takahashi, T. T., Sutherland, S. C., Chipman, D. W., Goddard, J. G., and Ho,
C.: Climatological distributions of pH, <i>p</i>CO<sub>2</sub>, total CO<sub>2</sub>, alkalinity, and
CaCO<sub>3</sub> saturation in the global surface ocean, and temporal changes at
selected locations, Mar. Chem., 164, 95–125,
<a href="https://doi.org/10.1016/j.marchem.2014.06.004" target="_blank">https://doi.org/10.1016/j.marchem.2014.06.004</a>, 2014.
</mixed-citation></ref-html>
<ref-html id="bib1.bib107"><label>Tilbrook et al.(2019)</label><mixed-citation>
Tilbrook, B., Jewett, E. B., DeGrandpre, M. D., Hernandez-Ayon, J. M., Feely,
R. A., Gledhill, D. K., Hansson, L., Isensee, K., Kurz, M. L., Newton, J. A.,
Siedlecki, S. A., Chai, F., Dupont, S., Graco, M., Calvo, E., Greeley, D.,
Kapsenberg, L., Lebrec, M., Pelejero, C., Schoo, K. L., and Telszewski, M.:
An enhanced ocean acidification observing network: From people to technology
to data synthesis and information exchange, Front. Mar. Sci., 6,
1–21, <a href="https://doi.org/10.3389/fmars.2019.00337" target="_blank">https://doi.org/10.3389/fmars.2019.00337</a>, 2019.
</mixed-citation></ref-html>
<ref-html id="bib1.bib108"><label>Turk et al.(2017)</label><mixed-citation>
Turk, D., Dowd, M., Lauvset, S. K., Koelling, J., Alonso-Pérez, F., and
Pérez, F. F.: Can empirical algorithms successfully estimate aragonite
saturation state in the subpolar North Atlantic?, Front. Mar.
Sci., 4, 385 pp., <a href="https://doi.org/10.3389/fmars.2017.00385" target="_blank">https://doi.org/10.3389/fmars.2017.00385</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib109"><label>Uppström(1974)</label><mixed-citation>
Uppström, L. R.: The boron/chlorinity ratio of deep-sea water from the
Pacific Ocean, Deep-Sea Res. Oceanogr. Abstracts, 21, 161–162,
<a href="https://doi.org/10.1016/0011-7471(74)90074-6" target="_blank">https://doi.org/10.1016/0011-7471(74)90074-6</a>, 1974.

</mixed-citation></ref-html>
<ref-html id="bib1.bib110"><label>Watson et al.(2000)</label><mixed-citation>
Watson, A. J., Bakker, D. C., Ridgwell, A. J., Boyd, P. W., and Law, C. S.:
Effect of iron supply on Southern Ocean CO<sub>2</sub> uptake and implications for
glacial atmospheric CO<sub>2</sub>, Nature, 407, 730–733, <a href="https://doi.org/10.1038/35037561" target="_blank">https://doi.org/10.1038/35037561</a>,
2000.
</mixed-citation></ref-html>
<ref-html id="bib1.bib111"><label>Watson et al.(2020)</label><mixed-citation>
Watson, A. J., Schuster, U., Shutler, J. D., Holding, T., Ashton, I. G.,
Landschützer, P., Woolf, D. K., and Goddijn-Murphy, L.: Revised
estimates of ocean-atmosphere CO<sub>2</sub> flux are consistent with ocean carbon
inventory, Nat. Commun., 11, 1–6,
<a href="https://doi.org/10.1038/s41467-020-18203-3" target="_blank">https://doi.org/10.1038/s41467-020-18203-3</a>, 2020.
</mixed-citation></ref-html>
<ref-html id="bib1.bib112"><label>Williams et al.(2017)</label><mixed-citation>
Williams, N. L., Juranek, L. W., Feely, R. A., Johnson, K. S., Sarmiento,
J. L., Talley, L. D., Dickson, A. G., Gray, A. R., Wanninkhof, R. H.,
Russell, J. L., Riser, S. C., and Takeshita, Y.: Calculating surface ocean
<i>p</i>CO<sub>2</sub> from biogeochemical Argo floats equipped with pH: An uncertainty
analysis, Global Biogeochem. Cy., 31, 591–604,
<a href="https://doi.org/10.1002/2016GB005541" target="_blank">https://doi.org/10.1002/2016GB005541</a>, 2017.
</mixed-citation></ref-html>
<ref-html id="bib1.bib113"><label>Wolf-Gladrow et al.(2007)</label><mixed-citation>
Wolf-Gladrow, D. A., Zeebe, R. E., Klaas, C., Körtzinger, A., and
Dickson, A. G.: Total alkalinity: The explicit conservative expression and
its application to biogeochemical processes, Mar. Chem., 106,
287–300, <a href="https://doi.org/10.1016/j.marchem.2007.01.006" target="_blank">https://doi.org/10.1016/j.marchem.2007.01.006</a>, 2007.
</mixed-citation></ref-html>
<ref-html id="bib1.bib114"><label>Zeng et al.(2014)</label><mixed-citation>
Zeng, J., Nojiri, Y., Landschützer, P., Telszewski, M., and Nakaoka,
S.-I.: A global surface ocean <i>f</i>CO<sub>2</sub> climatology based on a feed-forward
neural network, J. Atmos. Ocean. Technol., 31,
1838–1849, <a href="https://doi.org/10.1175/JTECH-D-13-00137.1" target="_blank">https://doi.org/10.1175/JTECH-D-13-00137.1</a>, 2014.
</mixed-citation></ref-html>--></article>
