The short-lived radionuclide thorium-234 (234Th, t1/2=24.1 d) is widely used to estimate the magnitude of POC that escapes the upper ocean layers (e.g., the euphotic zone) (Waples et al., 2006). 234Th is the decay product of uranium-238 (238U, t1/2=4.47×109 yr). While uranium is conservative and proportional to salinity in well-oxygenated seawater (Chen et al., 1986; Ku et al., 1977; Owens et al., 2011), thorium is not soluble in seawater and it is scavenged by particles as they form and/or sink along the water column. As a consequence, a radioactive disequilibrium between 238U and 234Th can be observed, mainly in the upper layers of the water column, which at first approximation, is proportional to the numbers of particles exported and hence can be used to estimate particle and elemental export fluxes.

A one-box scavenging model (see review by Savoye et al., 2006, and references therein) is commonly applied to calculate 234Th export rates. Steady-state (SS) or non-steady-state (NSS) conditions are assumed depending on the conditions at the sampling time and the possibility to reoccupy locations within an adequate timescale. Le Moigne et al. (2013b) reported 234Th fluxes from both types of models in their database with flux integration depths spanning from the surface down to 300 m, although the most common integration depths were between 100 and 150 m. The choice of export depth when using the 234Th technique is not trivial. Rosengard et al. (2015) provide recommendations to the various manners of choosing the export depth in order to integrate the 234Th fluxes. Once the 234Th export flux is estimated, it is multiplied by the ratio of POC to particulate 234Th activity in sinking particles to obtain the POC flux. The sinking particles from which the ratio is measured should, ideally, be collected at the depth where the export has been estimated and represent the pool of particles that are driving the export of organic carbon.

The determination of the POC/Th ratio has been historically attained by assuming that sinking carbon is driven by large particles, generally >50 µm in size (researchers also use 51, 53, or 70 µm, depending on the mesh supplier) whereas organic carbon within small particles is assumed to remain suspended and therefore not contribute to the export flux (Bishop et al., 1977; Fowler and Knauer, 1986). However, recent studies have shown that small particles can be significant players in the particle export and should not be disregarded (Alonso-González et al., 2010; Durkin et al., 2015; Le Gland et al., 2019; Puigcorbé et al., 2015; Richardson, 2019), particularly in oligotrophic regions. The most common methods to obtain the particulate fraction to measure the POC/Th ratio are (i) in situ pumps (ISPs), which can allow for sampling different particle sizes; (ii) collection bottles (CBs) such as Niskin bottles, providing bulk particles, i.e., >0.7 or 1 µm particles; (iii) sediment traps (STs); and although less common (iv) marine snow catchers. In some instances various methods have been used in combination (Cai et al., 2010; Maiti et al., 2016; Puigcorbé et al., 2015).

Different sampling devices have been shown to provide differences in POC/Th ratios, usually within a factor of 2 to 4 (Buesseler et al., 2006). The differences can be related to the collection of different particle pools and/or the enhanced presence of swimmers. STs collect sinking particles and may suffer from hydrodynamic discrimination and undersample slow-sinking particles (Gustafsson et al., 2004). CBs sample both sinking and suspended particles similar to ISPs. ISPs filter large volumes of water and have been suggested to potentially undersample some of the fast-sinking particles (Lepore et al., 2009) and sample neutrally buoyant C-rich aggregates (i.e., non-sinking but with high POC/Th ratios) (Lalande et al., 2008). Biases due to washout of large particles when using ISPs (Bishop et al., 2012) or aggregate collapse induced by their high cross-filter pressure (Gardner et al., 2003) may further enhance these differences. The presence of swimmers can also be an important bias of POC/Th ratios when not thoroughly removed, since they skew measurements towards higher values because of their high POC proportion compared to 234Th (Buesseler et al., 1994; Coale, 1990).

Despite the significant body of literature available on POC/Th ratios, more than 10 years after the review by Buesseler et al. (2006) we still cannot explain the variability of the POC/Th ratios with depth, time, particle type and size, or sinking velocity easily or at a global level. Changes with size and depth have been the most extensively examined. The relation between POC/Th ratio and particle size has been assessed before, with results suggesting that there is not a direct relationship. Previous studies have reported increasing ratios with increasing particle size (Benitez-Nelson et al., 2001; Buesseler et al., 1998; Cochran et al., 2000), which has been interpreted as an effect of the volume-to-surface area ratio of the particles, due to 234Th being surface bound whereas C would be contained within the particles (Buesseler et al., 2006). Yet, a number of studies have reported the opposite trend (i.e., decreasing ratio with increasing particle size; Bacon et al., 1996; Hung et al., 2010; Planchon et al., 2013; Puigcorbé et al., 2015) or no clear change with size (Hung and Gong, 2010; Lepore et al., 2009; Speicher et al., 2006). Depth is another factor that has been considered when assessing the variability of POC/Th, since particles are produced in the surface layer and are remineralized on their transit along the water column (Martin et al., 1987). POC/Th ratios have been found to be attenuated with depth (Jacquet et al., 2011; Planchon et al., 2015; Puigcorbé et al., 2015). This is due to (in no order or importance) decreasing autotrophic production with increasing water depth, preferential C loss compared to 234Th through remineralization processes, changes in superficial binding ligands along the water column, and/or scavenging of 234Th during particle sinking resulting in enhanced particulate 234Th activities (Buesseler et al., 2006; Rutgers van der Loeff et al., 2002), leading to significant variability in the attenuation rates. Theoretically, high sinking velocities may limit the variations in POC/Th ratios with depth, owing to shorter residence times limiting the impacts of biotic and abiotic processes. However, using specifically designed STs that segregate particles according to their in situ sinking velocities, Szlosek et al. (2009) observed no consistent trend between POC/Th ratios and sinking velocities.

The truth is that numerous processes can impact the POC/Th ratios apart from particle size or depth, such as particle composition or aggregation–disaggregation processes mediated by physical or biological activity (Buesseler and Boyd, 2009; Burd et al., 2010; Maiti et al., 2010; Szlosek et al., 2009), which adds a level of complexity to the prediction of their variability in the ocean. Yet, due to the significance of the POC/Th ratios for the accuracy of the 234Th flux method, the effort should be made to constrain the factors that will impact its variability, and a number of environmental and biogeochemical parameters can be assessed with that goal at a global scale. Among others, surface productivity, phytoplankton composition, zooplankton abundance, mixed-layer depth, dust inputs to the surface ocean, and ice cover (Buitenhuis et al., 2013; Mahowald et al., 2009; Moriarty et al., 2013; Moriarty and O'Brien, 2013) are all potential candidates to test their global patterns against POC/Th ratio variability.

Our dataset is archived in the data repository PANGAEA^® (http://www.pangaea.de), 10.1594/PANGAEA.911424 (Puigcorbé, 2019). Latitude, longitude, and sampling dates are reported. When dates of the individual stations were not reported in the original publications, we allocated the midpoint of the sampling period as the sampling date. The same was done when the specific sampling coordinates were not available (see details in the comments related to the dataset; 10.1594/PANGAEA.902103; Puigcorbé, 2019). The database consists of 9318 measurements of POC/Th ratios in the ocean. Particles were collected using in situ pumps (ISPs), water collection bottles (CBs), and sediment traps (STs). We refer to “bulk” (BU) for particles sampled using CBs and ISPs with a pore size filter of 0.2–1 µm. For this group of samples, particles >0.7 µm were collected using GFF filters and >1 µm using QMA filters. In some particular cases other types of filters, with a different pore size (e.g., 0.2, 0.45, or 0.6 µm) might have been used (see database for details). Hereafter, we use >1 µm for the bulk particles. We refer to “small particles” (SPs) for particles usually collected using ISPs on a 1–50 µm mesh size and “large particles” (LPs) for particles usually collected using ISPs on mesh size >50 µm (see details on other size ranges also used in the database). Finally, some POC/Th ratios were measured in sinking particles sampled using sediment traps (STs). Figure 1 shows the global distribution of POC/Th ratios grouped by these four categories: BU, LP, SP, and ST. The POC/Th ratios were obtained from particles collected at various depths from the surface to >5500 m (Fig. 2). All the information on locations, dates, depth, size fractions/device (BU, SP, LP, and ST), and references is included as metadata in the online database and presented in Table 1.

Our database covers POC/Th measurements sampled between 1989 and 2018, including unpublished data from our laboratories or graciously made available to us by colleagues and data available in online databases. Figure 3 shows the number of POC/Th measurements available per year. In the years 1997, 2004, 2005, 2008, 2010, 2011, and 2013, the number of POC/Th measurements was >500. This highlights dedicated carbon export programs such as the Joint Global Ocean Flux Study (JGOFS) (Buesseler et al., 1998, 1992, 1995, 2001; Murray et al., 1996, 2005), the VERTIGO (Vertical Transport in the Global Ocean) voyages in the Pacific Ocean (Buesseler et al., 2008b), and the GEOTRACES program (Mawji et al., 2015; Schlitzer et al., 2018), as well as the maintained effort of the time series stations (Kawakami et al., 2004, 2010, 2015; Kawakami and Honda, 2007). Sampling effort also varied depending on the month of the year (Fig. 3b), with late spring–summer months being the most highly sampled in both hemispheres. The Northern Hemisphere has been largely sampled in September, May, and June (49, 10, and 5 times more data than in the Southern Hemisphere, respectively), whereas the Southern Hemisphere has been more sampled in December and February (5 and 4 times more data than in the Northern Hemisphere, respectively), with no data available for the months of July and August and only five data points in September (austral winter). For the rest of the months, the Northern Hemisphere presents 1.4–1.8 times more data than the Southern Hemisphere. In the equatorial region (taken as the latitudes between -10 and 10∘ N) major sampling efforts took place in May, with no data collected in January and just eight data points available from December. The monthly distribution is, therefore, globally biased towards the warmer and more productive seasons, leaving the winter months largely undersampled, particularly in the Southern Hemisphere.

The global variability of POC/Th ratios looking at six different depths horizons (50, 100, 200, 500, 1000, and > 1000 m) and grouped by climatic zones (polar >66.5∘, subpolar 66.5–50∘, temperate 50–35∘, subtropical 35–23.5∘, and tropical 23.5∘ N–23.5∘ S) is presented in Fig. 4. A PERMANOVA analysis was conducted to examine the data, and the results indicate that all the depth horizons defined here were significantly different (p<0.05). Significant differences were also found between climatic zones, except between the temperate and subtropical zones and between the subtropical and the tropical zones, when considering all the data together. Statistical differences between zones within a certain depth range are shown in Fig. 4.

In general, we observe a reduction in POC/Th ratios with depth, previously reported by others (Buesseler et al., 2006), and likely mainly due to the remineralization of carbon along the water column. The decrease is particularly marked in the upper 200 m, where biological processes affecting the ratios are more intense, and then it smoothes below that depth horizon as the strength of these processes is more limited below the euphotic zone. It is worth noticing that some studies, particularly in coastal areas, presented extremely large POC/Th ratios (>100 µmoldpm-1, not included in Fig. 4). These high ratios are not always discussed in the publications, but the presence of live zooplankton (Buesseler et al., 2009; Savoye et al., 2008; Trull et al., 2008), especially in BU, ST, and LP fractions, when not picked out can be the cause for those high values and should be considered with caution.

Regarding the climate zones, there is significant variability, but, in general, large POC/Th ratios occur more often in productive and high-latitude regions relative to low-latitude tropical areas, particularly in the upper 200 m (Fig. 4). When looking at the different types of sampling methods, the link between latitude and magnitude of the ratio seems to be clear for ST and BU but quite variable for LP and SP (Fig. 5). High POC/Th ratios are usually associated with the presence of large phytoplankton groups, such as diatoms, which are dominant in high-latitude areas with no nutrient limitations, or where zooplankton populations are large and there is a significant input of fecal pellets, which should also have high POC/Th ratios. Low ratios, on the other hand, are commonly observed in warm oligotrophic areas where productivity is limited and the main phytoplanktonic groups are picoplankton (Buesseler et al., 2006). Exceptions do exist, but they are usually found in coastal areas where other factors could be influencing the planktonic community (e.g., seasonal upwelling, continental influence, river inputs).

The 234Th approach has been used to derive an export model at the global scale that uses sea surface temperatures and net primary productivity from satellite products (Henson et al., 2011). The parametrization for this model has large uncertainties in the cold regions (low sea surface temperature), which lead to a reduced estimate of the global biological carbon pump (∼5 GtCyr-1) compared to other satellite-derived export models (9–13 GtCyr-1; Dunne et al., 2007; Laws et al., 2011). A recent study by Puigcorbé et al. (2017a) estimated POC export fluxes in the North Atlantic using in situ data for the 234Th method and compared it to three different satellite-derived export models: Dunne et al. (2007), Henson et al. (2011), and Laws et al. (2011). The conclusion was that, overall, the geographical trends were captured by all the approaches, but the absolute values between them could reach important discrepancies. In that study, the authors advised a revision of the parametrization of the models going beyond sea surface temperatures in order to adjust to specific ocean bioregions. This database sets a strong background to develop that parametrization and contribute to similar modeling efforts to constrain the global carbon export fluxes as done by Henson et al. (2011).

This database provides the global POC/Th ratios sampled from all the oceans up until 2018. The sampling coverage is significant but it is not evenly distributed. Areas such as the China Sea, Arabian Sea, northwestern Mediterranean Sea, central Pacific, and high latitudes of the Atlantic Ocean are well represented, whereas other areas, such as the oligotrophic gyres, west Pacific, or the Southern Ocean, present important gaps. The data are not evenly distributed between seasons either, with most of the sampling taking place during spring and summer in both hemispheres, which is also when the export fluxes are expected to be larger. High seasonality in undersampled areas could potentially bias our global view of the POC/Th ratios and have an impact on the Th-derived carbon export flux estimates.

It would be beneficial for future efforts to obtain data for those undersampled areas with high seasonality to better characterize the expected variability in the ratios within those areas and to cover a larger span of seasons in order to better understand the seasonality of POC/Th and thus be able to translate it more accurately to the global POC export estimates.