The first consistent global gap-free daily 4 km dataset of particulate organic carbon, particulate organic nitrogen and their ratio (1998&ndash;2023)

Zhang, Yu; Liu, Huizeng; Liu, Cong; Xu, Nan; Yan, Lin; Yang, Chao; Wang, Yongquan; Wu, Guofeng; Li, Qingquan

doi:10.5194/essd-2026-453

Preprints

https://doi.org/10.5194/essd-2026-453

Preprints

29 Jun 2026

| 29 Jun 2026

Status: this preprint is currently under review for the journal ESSD.

The first consistent global gap-free daily 4 km dataset of particulate organic carbon, particulate organic nitrogen and their ratio (1998–2023)

Yu Zhang, Huizeng Liu, Cong Liu, Nan Xu, Lin Yan, Chao Yang, Yongquan Wang, Guofeng Wu, and Qingquan Li

Abstract. Particulate organic carbon (POC), particulate organic nitrogen (PON) and their ratios (POC:PON) are key to understanding particulate organic matter (POM) and marine biogeochemical cycles. However, global observations remain insufficient, with existing products suffering from spatial and temporal gaps, inconsistent methodologies, and interruptions from clouds and satellite sampling limitations. Here, we present a global gap-free daily 4 km dataset of consistent POC, PON and POC:PON from January 1998 to December 2023 (GGFD-POM), generated using a concise retrieval then reconstruction workflow. POC and PON concentrations were retrieved from ocean color data (OC-CCI v6.0) and Copernicus reanalysis physics data using newly developed Gaussian Process Regression (GPR) models. The models were trained on ~3110 matchups between in-situ observations, OC-CCI bio-optical properties, and Copernicus physical properties, achieving R² of 0.87 and 0.89, and RMSE of 1.47 mg m^-3 and 1.41 mg m^-3 for POC and PON, respectively. The Discrete Cosine Transform–Penalized Least Squares (DCT-PLS) approach was subsequently applied to reconstruct missing values in the satellite-retrieved POC and PON fields, resulting in gap-free global datasets. Based on these reconstructed products, a gap-free global POC:PON dataset was further derived. Validation using an independent in-situ dataset confirmed high accuracy of both satellite retrievals and reconstructions for POC, PON and POC:PON. Triple-collocation analysis (TCA) exhibited that the GGFD POC data outperform existing MODIS-Aqua and MULTIOBS POC datasets, reflecting the distinct strengths of the GGFD-POM product. Comparative analysis of spatiotemporal variations in POC, PON, and POC:PON further demonstrated that the gap-free dataset better captures trends and magnitudes, enhancing understanding of their roles in the global carbon and nitrogen cycles. The complete GGFD POM product dataset (1998–2023) is openly available at https://doi.org/10.11888/Ocean.tpdc.303488 (Zhang and Liu, 2026).

Received: 11 Jun 2026 – Discussion started: 29 Jun 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 12509 KB)

Supplement (1146 KB)

Download & links

Preprint (12509 KB)
Metadata XML
Supplement (1146 KB)
BibTeX
EndNote

Yu Zhang, Huizeng Liu, Cong Liu, Nan Xu, Lin Yan, Chao Yang, Yongquan Wang, Guofeng Wu, and Qingquan Li

Status: open (until 01 Oct 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on essd-2026-453', Anonymous Referee #1, 02 Aug 2026 reply
This manuscript presents a valuable and timely contribution by introducing a global gap-free daily 4 km dataset of particulate organic carbon (POC), particulate organic nitrogen (PON), and their ratio spanning more than two decades (1998–2023). The development of a consistent long-term dataset addressing spatial and temporal gaps in existing POM products is highly relevant for studies of marine biogeochemistry, carbon and nitrogen cycling, and ecosystem dynamics.
The manuscript is generally well structured, and the methodology is clearly presented. The authors provide extensive validation, which strengthens confidence in the generated dataset. The dataset itself has the potential to become a valuable resource for the scientific community.

Some minor revisions require clarification and I have a few very minor suggestions the authors may wish to consider.

Data accessibility and usability

The dataset is openly available through the provided TPDC link; however, the current download procedure requires the use of temporary FTP credentials, which are given through the data doi and an external FTP client is need to use them. I acknowledge that FTP-based access is commonly used for large scientific datasets. However, the current access instructions may be challenging for users who are unfamiliar with FTP workflows. Please provide clearer documentation regarding the download procedure, including whether registration is required, how the provided credentials should be used, and the organization of the downloaded files.
Technical corrections
Line 36: Remove “the” before POC and PON.

Line 77: Change “environment” to “environments”.

Dates are currently presented in different formats throughout the manuscript (e.g., “1st January” vs. “1 January 1998”). I suggest to use one consistent date format throughout the manuscript, including tables.

Figures in general are relatively small. Increasing the size of Figures 1, 5, 6, 8, and 10 would improve readability.

Throughout the manuscript, a space is missing between numerical values and the percentage unit (e.g., “5%” should be written as “5 %” according to journal style). Please check all occurrences (e.g., lines 170, 213, 261, 264, 265, 271, 272, 317, 326, 327, 330, 333, 345, 367, 369–371, 386, 388, 398–399, 447, 482–483).

Throughout the manuscript, a space could be placed between ± and the numerical value for readability (e.g., “±5” -> “± 5”). Please check all occurrences (e.g., lines 146, 150, 200, 369–371, 452–456).

In Table 1, chlorophyll a is written as “Chl-a”, whereas the abbreviation is written differently in the main text (Chla) (lines 278–282, 304). I suggest a consistent notation throughout the manuscript.

Table 3: The table title appears to be missing “and” between POC and PON.

Line 409: Replace “were showed” with “were shown”.

Figure 11: Some symbols/marks in panel 11d appear to be unintended artifacts and should be removed.

Reply
Citation: https://doi.org/10.5194/essd-2026-453-RC1
RC2:
'Comment on essd-2026-453', Marit van Oostende, 23 Sep 2026 reply
This study presents a daily global 4 km dataset of POC, PON and POC:PON for 1998-2023. Such a record could be useful for investigating marine POM dynamics, carbon export, nutrient cycling and long-term ocean change, particularly because comparable continuous global products are limited.
My main concern is whether the record is sufficiently consistent over time for the proposed trend applications. OC-CCI sensor composition and data coverage change throughout the record, with SeaWiFS as the only source during the early years. Figure S4 shows clear changes in coverage, while the retrieved and reconstructed global time series differ most strongly during 1998-2003. The authors need to examine whether these changes affect the product and whether the effects differ among regions and sensor periods. The reconstruction should also be tested using more realistic gap patterns and data availability from different sensor periods. The gap-simulation experiment removes pixels at random on only two dates, whereas gaps in ocean-colour data vary with location, season and sensor. The data files should identify retrieved and reconstructed values and include uncertainty information.

Comments in manuscript order:
Lines 18-21: Please explain what is meant by “consistent”. It is unclear whether this refers to the method, to consistency in space and time, or to something else. The manuscript later states that some areas remain unreconstructed when too few satellite observations are available. Please define “global gap-free” when the dataset is first introduced.

Lines 30–32: It is unclear what is meant by “better captures trends and magnitudes” or what the gap-free product is being compared with. Greater completeness does not itself demonstrate greater accuracy. The basis of this claim should be stated or the wording revised.

Lines 36-41: Please define POC, PON and POM at first use in the main text.

Lines 45-53: Because POC and PON are reported in mg m^-3, the derived POC:PON appears to be a mass ratio. Please state this explicitly, as the Redfield ratio quoted here is a molar ratio.

Lines 129-134: SeaWiFS is the only contributing sensor before April 2002 and was processed using a different atmospheric correction method. Coverage also changes as other sensors are added to the merged OC-CCI dataset. Please examine whether these changes affect GGFD across regions and time, particularly in coastal and high-latitude waters, and whether they influence the reported trends.

Lines 134-136 and Table 1: The temporal coverage of GLORYS12V1 is given as 2001-2022 in the text but as 1 January 1998-31 December 2023 in Table 1.

Lines 149-150: Please describe how the GLORYS12V1 fields were resampled to the OC-CCI grid, including the interpolation or aggregation method.

Section 2.3.1: The information provided is not sufficient to reproduce the GPR models. Please report the transformations and scaling applied to the response and predictor variables, the logarithm base, the kernel and noise model, the hyperparameter optimisation, and the software and version used. Please also explain how predictions were converted back to concentration units.

Lines 159-175: Please clarify whether feature selection and any model tuning were carried out before or after the 80/20 split, and how these steps were handled during ten-fold cross-validation. Also state whether the split and cross-validation folds were grouped by cruise, location or sampling period.

Lines 170-175; Please show how the matchups used for model development and independent validation are distributed across the study period.

Lines 173-175: The cited BCO-DMO dataset ends in July 2010, whereas the listed independent cruises took place in 2012-2014. Is this the correct dataset? If not, what source was used for the independent observations?

Section 2.3.1: Was uncertainty estimated for the individual GPR predictions? It is also unclear how users can identify predictions made under conditions that are poorly represented in the training data.

Lines 195-206: The DCT–PLS settings are not fully described. Which software was used, and how were the smoothing parameter and robust weighting specified? The treatment of the first and last 15 days is also unclear. At the ends of the record, the target day cannot be the central slice of a 31-day window. Which window and slice were used for these dates?

Lines 210-217: The gap-simulation experiment randomly removes 10% of pixels that were originally observed and is limited to two dates in 2020. It therefore does not test reconstruction across large or persistent gaps, or under lower observation densities in parts of the record. Performance should also be assessed using more realistic gap patterns and periods with different data availability.

Lines 218-225 and Table 2: he exact MODIS-Aqua and MULTIOBS products, versions and common comparison period are not given. Please also explain how the daily and weekly data were converted to monthly values and mapped to 1degree, how the common comparison domain was defined, and how many collocated months were required per grid cell.

Lines 218-225: TCA assumes that errors in the three products are independent, but this may not hold for this comparison. GGFD is derived from OC-CCI, which includes MODIS-Aqua observations for part of the record, while MODIS-Aqua is also used as a separate comparison product. MULTIOBS also uses satellite ocean-colour data. Please discuss whether the independence assumption is reasonable for these products.

Lines 235-241: Please explain how the regression slopes were converted to % yr^-1 and how uncertainty in the retrieved and reconstructed time series was accounted for. Please indicate which reported trends remain distinguishable from product uncertainty.

Lines 245-250: It is unclear to me how ΔPOC and ΔPON were estimated. Please define these uncertainties and state whether any limits or checks were applied when calculating POC:PON.

Lines 259-272 and 385-399 Figures 2, 4 and 7: It is unclear whether R^2, RMSE and MAPE were calculated using the original or log-transformed concentrations. Please give the calculation scale and metric equations. The unit of the POC:PON RMSE at line 388 should be corrected, as the ratio is dimensionless.

Figures 2, 4 and 7: The colour bar is not explained in Figure 2. In Figures 4 and 7, it is described as point density, but the numerical values are not defined. Please explain what the colour scales represent.

Table 3: Please state which correlation coefficient is reported and use consistent notation for u0 and v0 across Tables 1 and 3.

Lines 343-361: On average, retrieved pixels account for only 24.21% of the pixels available after reconstruction. It would therefore be useful for the archived files to distinguish retrieved from reconstructed values and to provide some indication of uncertainty or reconstruction quality.

Figure 5: Please state that panels (a–c) show the retrieved fields and panels (d–f) the reconstructed fields.

Figure 6: The red rectangles referred to in the text are not explained in the caption. In addition, the colour scale covers ±10%, whereas the text reports values from about −70% to +125%. Please state whether values outside the displayed range were clipped and explain the choice of colour scale.

Lines 407-430 and Figure 8: Some of the wording in this section, including “significantly lower”, “consistently superior” and “clear advantage”, seems stronger than warranted by the results shown. No statistical test is reported to support “significantly lower”, the GGFD and MULTIOBS distributions overlap, and MODIS-Aqua has lower fMSE in some regions. These statements should either be supported statistically or revised to acknowledge the overlap and regional differences.

Figure 9: The caption should identify which lines represent the retrieved and reconstructed series.

Lines 448-460; Figures 9 and S4: The difference between the retrieved and reconstructed series changes markedly around 2002-2003, particularly for POC:PON. This coincides with the transition from the SeaWiFS-only period to a multi-sensor record and with the large increase in valid observations shown in Figure S4. As Chl-a is the strongest predictor in both retrieval models, changes in OC-CCI coverage may propagate into GGFD, and gap filling does not necessarily remove such temporal inconsistencies. The authors should examine whether this pattern is related to sensor composition, observation density, upstream processing, reconstruction, or a genuine change. The existing validation supports the accuracy of individual reconstructed values under the tested conditions, but does not establish whether the temporal trends are preserved by the reconstruction or affected by changes in sensor composition and coverage in the underlying OC-CCI record. This should be assessed before the reconstructed trends are described as “more accurate”.

Lines 463-475 and Figure 10: The caption should identify the variable and data type shown in each panel. It should also explain how cells without sufficient data are displayed and distinguish them from cells where a trend was tested but was not significant. The colour scale is limited to ±20% yearr ^-, and many pixels appear to reach these limits. Please state whether values were clipped, report how frequently such large trends occur, and discuss their plausibility overall and in relation to product uncertainty, and, where available, comparable published trends.

Data and Suplement
The file I inspected contains POC, PON and POC:PON with units and fill values, but does not include a product version or processing history. Providing this metadata and making the GPR and DCT–PLS code available in a persistent repository would make the dataset easier to assess and reuse. If the code cannot be shared, further implementation details are needed to reproduce the workflow.

Figure S1: In panels (a–b), the coloured map background has no scale or explanation and the black and orange symbols are not defined in the caption. In panels (c–d), the colours and outlines of the histogram series are not identified.

Table S1: Units should be given for all concentration statistics. The maximum PON concentration is reported as 175.56 for the complete dataset but 202.26 for the training subset, which is not possible if the latter is a subset of the former. Please check these values.

Figure S2 and Table S2: Figure S2 shows nine red rectangular areas, whereas the caption refers to a single “red diamond-shaped area”. Table S2 should state the year of the 1 January and 1 July results and explain the backslashes used for the Arctic results.

Figures S4: The grey and orange series in Figure S4 are not identified.

Reply
Citation: https://doi.org/10.5194/essd-2026-453-RC2

Yu Zhang, Huizeng Liu, Cong Liu, Nan Xu, Lin Yan, Chao Yang, Yongquan Wang, Guofeng Wu, and Qingquan Li

Supplement

https://doi.org/10.5194/essd-2026-453-supplement

Data sets

Global gap-free daily 4 km dataset of particulate organic carbon, particulate organic nitrogen and their ratio (1998–2023) Yu Zhang and Huizeng Liu https://doi.org/10.11888/Ocean.tpdc.303488

Yu Zhang, Huizeng Liu, Cong Liu, Nan Xu, Lin Yan, Chao Yang, Yongquan Wang, Guofeng Wu, and Qingquan Li

Viewed

Total article views: 307 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
178	79	50	307	47	71	73

HTML: 178
PDF: 79
XML: 50
Total: 307
Supplement: 47
BibTeX: 71
EndNote: 73

Views and downloads (calculated since 29 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	70	40	4	114
Aug 2026	50	17	4	71
Sep 2026	58	22	42	122

Cumulative views and downloads (calculated since 29 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	70	40	4	114
Aug 2026	50	17	4	71
Sep 2026	58	22	42	122

Viewed (geographical distribution)

Total article views: 255 (including HTML, PDF, and XML) Thereof 255 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 24 Sep 2026

Download

Preprint (12509 KB)
Metadata XML

Short summary

This work presents GGFD-POM, the first global gap-free daily 4 km dataset covering particulate organic carbon (POC), particulate organic nitrogen (PON), and their ratio (POC:PON) from 1998 to 2023. The dataset provides continuous high-resolution data, enhancing studies of particulate organic matter (POM) dynamics in relation to the carbon cycle and biogeochemical stoichiometry.


Total:	0
HTML:	0
PDF:	0
XML:	0