Global assessment of merged multi-sensor ocean-colour chlorophyll-a products

Abstract

Chlorophyll-a concentration (chl-a) is important to assess the health and state of ocean ecosystems. With the availability of global ocean-colour chl-a estimates that now span 25 years, there has been a concerted effort to produce merged data products from different satellite sensors to assess changes in chl-a over the global ocean for long periods of time. However, to date, the performance of these merged chl-a products has not been thoroughly assessed. To perform such an assessment, we assembled a large global in situ dataset of quasi-autonomous spectrophotometrically-derived chl-a that resulted in >13,000 satellite match- ups and then filtered them to produce the highest quality data. The suite of merged ocean-colour chl-a products assessed using the in situ chl-a included two Ocean Colour - Climate Change Initiative (OC-CCI) versions (OC-CCI v5 and OC-CCI v6), two GlobColour products and the Copernicus Marine Environment Monitoring Service (CMEMS) GlobColour L3 and L4 and CMEMS-CCI products. The results confirm that spectrophotometrically-derived chl-a estimates can achieve considerably larger numbers of satellite match ups and lower root mean squared errors in validation than those obtained from discrete estimates of chl-a. Using these data, all of the satellite products (except the GlobColour L4 gap-filled one) exhibited similarly consistent results with the in situ chl-a data with a mean relative percentage difference of 30%. Residuals (differences between in situ and satellite product data) were not homogeneously distributed across chl-a ranges however, with mainly negative residuals at low and high chl-a and mainly positive residuals at intermediate chl-a. These results illustrate that absolute biases of the order of 20%–50% still affect these merged products in specific parts of the chl-a range.

1 Introduction

As the ocean warms, we need to observe and understand what is driving long-term changes in phytoplankton and the feedbacks that these changes can have on marine ecosystems and biogeochemical cycles. Earth Observation data are expected to play an increasingly important role in a growing range of climate services. Satellite ocean colour has provided unprecedented observations of changes in surface phytoplankton biomass in space and time (Behrenfeld et al., 2006). As such, ocean colour is classified as an Essential Climate Variable (ECV) for which a Climate Data Records can be created (GCOS, 2021).

Building a continuous ocean-colour climate record from sensors that have different orbits, spatial resolution and spectral wavebands, without introducing biases, artefacts or discontinuities is a major challenge (Hammond et al., 2017). For ocean-colour data to be included in the climate record, it is required to produce consistent, stable, error-characterised global products from multi-sensor data archives, that meet the Global Climate Observing System (GCOS) programme requirements (GCOS, 2021). Careful evaluation of the ocean-colour product is needed to identify and minimise effects of data processing choices and changing sensors and to maximise the true climate signal (Merchant et al., 2015).

To this end, a number of satellite ocean-colour programmes have been funded to produce continuous time series from an array of different ocean colour sensors. These include the Ocean Colour Climate Change Initiative (OC-CCI) and the Global Ocean Colour for Carbon Cycle Research record (GlobColour) both funded by the European Space Agency (ESA) and the European Copernicus Marine Environment Monitoring Service Ocean Colour (CMEMS or C3S). These products use slightly different procedures, algorithms and sensors. Chl-a is classified by GCOS as an ECV (GCOS, 2021), which is the most commonly-used metric of phytoplankton biomass, and is produced by all these satellite ocean-colour programmes operationally. Yet, comparatively few validation studies of these merged multi-mission ocean colour products have been conducted using high quality in situ data. This is because accuracy assessments of satellite products are limited by the extent of the spatial and temporal distribution of the in situ data (Melin et al., 2016). Some time series exist for a few locations, notably AERONET-OC (Zibordi et al., 2006), but the network of observations is primarily remote-sensing reflectance (R_rs) rather than chl-a per se, and the spatial distribution of the validation data points is predominantly in coastal waters. Even when large global in situ chl-a data bases are used, there is sparse coverage in open-ocean regions (Gregg and Casey, 2004). Of those that have been conducted, for OC-CCI, Brewin et al. (2015) used in situ R_rs from the NOMAD dataset as input to an array of empirical and semi-analytical models to compute chl-a and to select the most accurate algorithms to be included in the OC-CCI product. The NOMAD data has global coverage and the results illustrated that empirical band ratio algorithms such as OC and CI variants performed better than semi-analytical algorithms. For GlobColour, Garnesson et al. (2019) reported that merging single sensor data provides improved daily spatial coverage when OC5 is used rather than on merged R_rs data, which is used for the OC-CCI product. They also reported that from 2012 to 2018, the spatial coverage of the GlobColour products was improved, compared to the OC-CCI product in both the open ocean and coastal areas. The GlobColour products were validated using a global in situ dataset, though the details of this and validation statistics are lacking (GlobColour, 2020).

The use of spectrophotometric quasi-autonomous underway measurements of chl-a has shown real promise for evaluating merged multi-mission ocean colour products (Brewin et al., 2016), due to the highly accurate estimate of chl-a and the ability to sample at spatial scales comparable with satellite pixels. Furthermore, global compilations of these datasets, with the majority coming from the Atlantic Ocean, offer the potential for large numbers of match-ups (Graban et al., 2020; Tilstone et al., 2021; Jordan et al., 2025). There are growing number of papers that have used underway inline spectrophotometric systems for the validation of ocean colour (Werdell et al., 2013; Brewin et al., 2016; Tilstone et al., 2021; Ford et al., 2021). The objective of this study is to quantify and compare the performances of various merged products together, using a large consistent, globally-compiled dataset of high quality in situ chl-a data derived from carefully calibrated spectrophotometric underway measurements, especially in the sparsely sampled oligotrophic-mesotrophic ocean. For this, we assembled a global data base of these measurements, reviewed them for quality and utilised the measurements with the lowest uncertainty to validate a range of multi-sensor merged ocean colour products.

2 Data and methodology

2.1 In-situ data

2.1.1 Study areas and sampling regimes

2.1.2 Discrete measurements of chl-a using HPLC

High Performance Liquid Chromatography (HPLC) was used for discrete measurements total chl-a. The HPLC methods for the AMT campaigns are detailed in Jordan et al. (2025). Water samples were collected using two different methods during AMT campaigns: an underway flow-through system and Niskin bottle rosettes deployed on the ship’s rosette. The EXPORTS NP, NAAMES 1-4, GP505 and GP506 campaigns followed the NASA HPLC-sampling protocols (https://oceancolor.gsfc.nasa.gov/fsg/hplc/). Calibration was conducted using individual pigment standards from absorption coefficients commonly employed by various laboratories, ensuring consistency and comparability in the analysis of pigments. The processing method is further described in Hooker et al. (2005), van Heukelem and Hooker (2011) and Hooker et al. (2012).

Tara Oceans and Tara Polar expeditions made discrete measurements of chl-a that were derived from filters extracted in 100% methanol, disrupted by sonification and clarified by filtration. Extraction time lasted 2 h, and analysis by HPLC was carried out the same day. The procedure is explained in detail in Ras et al. (2008).

2.1.3 Underway quasi-autonomous spectrophotometric estimates of chl-a

In-situ estimates of chl-a were obtained from underway spectrophotometry. Using this technique, hyperspectral attenuation-absorption meters (Wetlabs ACS, spectral range 400–750 nm) were used and connected to the ship’s flow-through system, with seawater continuously pumped from the sea surface and circulated through the instrument. Measurements were made through a sequence of filtered (0.2 micron) and unfiltered time intervals and spectral particulate absorption coefficients (a_p(λ)) were then calculated following well-established procedures (Dall’Olmo et al., 2009; Slade et al., 2010; Dall’Olmo et al., 2012). Underway a_p(λ) data from the campaigns described above, conducted from 2007 to 2022, were downloaded from the NASA SeaBASS archive (Werdell and Bailey, 2002, https://seabass.gsfc.nasa.gov/). Figures 1, 2 show the locations of these in situ data classified by campaign and season respectively. Data collection took place across various seasons and regions. The majority of data were acquired during autumn in the Atlantic, summer in the Arctic, and across all seasons in the Pacific. Data for the Indian Ocean were collected during austral spring, autumn and winter, while data for the Red Sea were obtained during winter. None of the data in the compiled in situ match-up database were used in algorithm calibration or optimisation in any of the product lines.

FIGURE 1

FIGURE 2

Absorption-based chl-a estimates were determined using the line-height method (Davis et al., 1997; Boss et al., 2007) applied to the absorption peak at 676 nm using the empirical relationship given in Equation 1:where 0.014 mg m⁻³ is the nominal chl-a specific absorption coefficient at 676 nm and a_ph (676) is the corresponding phytoplankton absorption spectral coefficient. The n subscript is used to identify the line-height chl-a as a ‘nominal’ estimate, prior to calibration with discrete HPLC samples (Graban et al., 2020). Any measurement where a_p (420) < 0 or where chl_ACSn< 0 were removed. These filters removed a small fraction of remaining data (∼1% or less). The majority of the ACS data were available at 1 min sample binning, with the exceptions being EXPORTS (15 min) and GP5-05/GP5-06 (60 min). For GP5 cruise, the low resolution sample spacing and binning resulted in a low number of data points.

HPLC estimates of total chl-a (chl_HPLC) were used to calibrate (de-bias) and validate the nominal chl_ACSn concentrations. Only near-surface measurements (depth < 6 m) were used for de-biasing. The ACS vs. HPLC match-up procedure used in the de-biasing is based on previous analyses of AMT data (Brewin et al., 2016; Graban et al., 2020; Tilstone et al., 2021). For each chl_HPLC sample, chl_ACSn measurements acquired within a ±15 min window around the HPLC sampling time were identified, and the median value used in the match-up to reduce high-frequency noise in chl_ACSn often associated with bubbles. If no chl_ACSn measurements were acquired within the ±15 min window, the match-up pair was disregarded.

For each campaign, the relative residuals between chl_HPLC and their concurrent averaged chl_ACSn measurements were calculated using Equation 2, as:

De-biased total chl-a concentrations, chl_ACS, were then calculated from Equation 3, as follows:where δ is the median of the relative residuals for each campaign and quantifies the offset between chl_ACSn and chl_HPLC. The robust standard deviation of the relative residuals, σ, was also calculated for each campaign to provide a measure of the random uncertainty of the chl_ACSn estimates from Equation 4:where P₈₄ (rres) and P₁₆ (rres) are the 84^th and 16^th percentiles of the relative residual distributions. The de-biasing of the ACS chl-a assumes that there is a linear relationship between chl_HPLC and the phytoplankton absorption at 676 nm (a_ph (676)) in Equation 1. This was previously shown to be a robust approach for the AMT cruises (Brewin et al., 2016; Graban et al., 2020; Jordan et al., 2025). To justify this approach across the extended dataset in this study, we replicated the methodology in Jordan et al. (2025) and fitted a power-law relationship to each ACS dataset, given in Equation 5 of the form:where A is a proportionality constant and B is the power-law exponent (Brewin et al., 2016). A and B were derived using Type 1 linear regression on log₁₀-transformed data, using the 95% confidence interval (2σ) to define fit uncertainties, A_unc and B_unc. Figure 3 shows the distributions of chl-a (i.e., chl_ACS) for each campaign and for the entire dataset.

FIGURE 3

Table 1 summarises the in situ data statistic for the HPLC vs. ACS match-ups from each campaign. The median chl_ACS bias relative to chl_HPLC (δ) ranged from −12% to 16%, with an average value of −3.4%. The relative uncertainty (σ) of the de-biased chl_ACS ranged from 7% to 59% with an average value of 14%. All but three of the cruises (AMT23, EXPORTS NP, Tara Oceans) had B indistinguishable from 1 within the 95% confidence interval, supporting the linear de-biasing approach. It is also noted that AMT23 and EXPORTS NP had a low number of HPLC samples. In previous studies, median (δ) and standard deviation (σ) of relative percentage residuals between chl_ACS and chl_HPLC thresholds of < 10% have been used. In this paper, we used δ and σ thresholds of < 12% and < 27% respectively, which therefore excluded EXPORTS NP, Tara Oceans and Tara Polar. Figures and statistical metrics with (Supplementary Figures S1-S4) and without (Figures 4–7) the EXPORTS NP, Tara Oceans and Tara Polar, are presented.

FIGURE 4

FIGURE 5

FIGURE 6

FIGURE 7

Divergences between HPLC- and absorption-derived chl-a can be attributed to differences in instrument calibration and characteristics, and are discussed further in Brewin et al. (2016) and Graban et al. (2020). Small biases and random uncertainties (δ and σ typically below 10%) have previously been considered acceptable to use chl_ACS for validation purposes (Tilstone et al., 2021).

2.2 Merged satellite products

2.2.1 OC-CCI

The OC-CCI is part of the ESA Climate change Initiative programme which aims to produce multi-sensor global datasets from satellite data that are consistent, stable, error-characterised and fulfill ocean-colour GCOS ECV requirements. The final goal is to generate a time series of ECVs that can be used for climate research and modelling (Sathyendranath et al., 2019). The OC-CCI product merges R_rs data from SeaWiFS, MERIS, MODIS, VIIRS and OLCI. Each sensor is processed independently from level 1b using the best-performing atmospheric-correction algorithm (Muller et al., 2015a; Muller et al., 2015b). The data undergo band-shifting and bias-correction processes to minimise calibration and band differences between sensors, and to align to the reference sensor, MERIS, before merging. Merged data are classified in to 14 different optical water types (OWT), and the best-performing chl-a algorithms (Table 3) for each OWT are applied to the merged R_rs. The resulting chl-a values are averaged, weighted by water-type relative membership, to generate the chl-a product. Two OC-CCI data products v5 and v6 were analysed in this work (Table 2). The OC-CCI v5 chl-a is generated from a blended algorithm combination of OCI, OCI2, OC2, OC3, OCx and OC5 (OC-CCI, 2020). In OC-CCI v6, chl-a is estimated as a blend of OCI, OCI2, OC2 and OCx (OC-CCI, 2022). These chl-a algorithms are given in Table 3. OC-CCI v6 also includes the novel POLYMER AC, addition of OLCI 3B data, updated Idepix pixel flagging, updated MERIS 4th reprocessing data, new System Vicarious Gains for OLCI and MERIS. Both versions were downloaded from https://climate.esa.int/en/projects/ocean-colour/data/.

2.2.2 GlobColour

GlobColour is a project aimed at producing a continuous, comprehensive, validated, merged global ocean-colour dataset. The project initiated as an ESA Data User Element Project. Currently, GlobColour services continue within the framework of the Copernicus Marine Environment Monitoring Service (CMEMS) and the REGICOLOUR project. The GlobColour product merges data from SeaWiFS, MERIS, MODIS, VIIRS NPP, OLCI-A, VIIRS JPSS-1 and OLCI-B. Each sensor is acquired independently at level 2, then undergoes spatial and temporal binning schemes. Two GlobColour data products were acquired (Table 2): GlobColour AVW and GlobColour GSM, for which the functional form of the chl-a algorithms are given in Table 3. GlobColour uses different merging techniques for the production of merged products: simple averaging (AV), weighted averaging (AVW), and the GSM model (Maritorena et al., 2002; Maritorena et al., 2010). The GlobColour AVW averages the chl-a from each sensor’s specific chl-a algorithm. The GSM product is calculated from merged and weighted R_rs from the available sensors and then inverts the merged R_rs spectrum to derive chl-a (Maritorena and Siegel, 2005). The GSM is an inherent optical property model from which chl-a is derived. The AV product is currently only distributed as on single-sensor mode only sensor products, rather than a multiple sensor product, so it was not considered for this work. GlobColour data were downloaded from https://hermes.acri.fr/. From 2009, GlobColour also contributes to the Copernicus Marine Environment Monitoring Service (CMEMS). The CMEMS GlobColour products are described in Section 2.2.3.

2.2.3 CMEMS

The Copernicus Marine Environment Monitoring Service (CMEMS) is the marine component of the Copernicus Programme. It provides physical and biogeochemical ocean products from in situ, models and satellite observations on a global and regional European scale. Three CMEMS data products were acquired to be assessed (Table 2): CMEMS OC-CCI, CMEMS GlobColour L3 and CMEMS GLobcolour L4. All data were downloaded from https://data.marine.copernicus.eu/products.

The CMEMS OC-CCI/C3S product (OCEANCOLOUR GLO BGC L3 MY 009 107) provides daily global chl-a concentration, Phytoplankton Functional Type and R_rs at 412, 443, 490, 510, 560 and 665 nm. The data are directly extracted from the input OC-CCI/C3S daily composites and repackaged for compliance with the CMEMS format standards, with production overseen by Brockmann Consult. Algorithms for chl-a and R_rs are selected based on performance evaluations. To estimate chl-a, a combination of OC2, OCx, OCI, and OCI2 algorithms is used (see Table 3) depending on the water types present in each pixel. For R_rs, the POLYMER AC model is utilised for all sensors except SeaWiFS. Further information, including a description of the OC-CCI processing chain is given in OC-CCI (2014).

The CMEMS GlobColour L3 product (OCEANCOLOUR GLO BGC L3 MY 009 103) provides a chl-a multi-sensor daily product integrating chl-a data obtained from different sensors, employing a consistent methodology for each sensor. For oligotrophic waters, the product utilises the CI algorithm (Hu et al., 2012), whereas for mesotrophic and coastal waters, it relies on the OC5 algorithm. The OC5 algorithm is a modified version of the OC3 and OC4 algorithms, specifically tailored for complex waters (Gohin et al., 2002). Each sensor’s OC5 algorithm is calibrated accordingly. The blended product combines OC5 and CI-Hu data using a merging technique similar to NASA’s approach, with a smooth transition between concentrations of 0.15–0.2 mg m⁻³ to ensure seamless integration.

The CMEMS GloColour L4 product (OCEANCOLOUR GLO BGC L4 NRT 009 102) is a “cloud-free” level-4 daily interpolated product involving multiple sensors, multiple chl-a algorithms and an interpolation scheme. L4 refers to products that undergo a process of temporal averaging or interpolation to fill in missing data values. The interpolation procedure utilised encompasses techniques such as Optimal Interpolation and DINEOF. This interpolated daily product has been introduced since 2016. Significant improvements were made to the chl-a “cloud-free” product in 2018, benefiting from enhanced data quality and coverage of the daily non-interpolated upstream data.

2.3 Match-up procedure

3 Results

3.1 Independent match-up analysis

The scatter plots of in situ chl_ACS versus satellite chl-a from the seven different products for the independent matchups without the Tara and EXPORTS NP datasets are given in Figure 4, and including these datasets, in Supplementary Figure S1. The relative residual percentages between chl_ACS and chl_satellite were also calculated as [(chl_satellite - chl_ACS)/chl_ACS] × 100. The independent match-up residuals for each merged product without the Tara and EXPORTS NP are shown in Figure 5, and including these datasets, in Supplementary Figure S2. The number of independent match-ups was >13,000 for all merged products, highest for the gap-filled CMEMS GlobColour L4 (N = 61,307) and lowest for the GlobColour GSM product (N = 13,048). The highest number of match-ups for a product without gap-filling was provided by OC-CCI v6 (N = 17,925; Table 5) and these were reduced by ∼50% when the Tara and EXPORTS NP data were excluded (Table 6).

For the independent match-ups, all merged data products exhibited a similar agreement with the in situ ACS data (Tables 5, 6). The distribution of residuals that was homogeneously distributed around the zero line, with underestimates at low and high chl-a (Figure 5) and overestimates in the mid chl-a range, which was more pronounced when the Tara and EXPORTS NP data were included (Supplementary Figure S2). All products had regression slopes close to one and intercepts close to zero, with GlobColour AVW and GSM having slopes closest to one and CMEMS OC-CCI L3, OC-CCI v5 and v6 having the smallest intercept. Comparing OC-CCI v5 and v6, the differences appeared to be insignificant. Similarly, OC-CCI v6 and CMEMS-CCI produced nearly identical results. GlobColour AVW tended to underestimate chl-a concentrations at low values more than the other merged products, likely due to suboptimal performance of OC4 and OC3 within that range. To address this issue, CMEMS GlobColour L3 incorporated a colour-index algorithm specifically designed for low chl-a concentrations. The performance of the CMEMS GlobColour L4 gap-filled product was slightly worse than the CMEMS GlobColour L3 in all metrics except RPD. The GlobColour GSM algorithm exhibited an overestimation in chl-a. The residuals were high between 0.1 and 0.2 mgm⁻³ chl-a (Figure 5), especially from the Tara Oceans and EXPORTS NP datasets (Supplementary Figure S2). If these data were removed from the analysis, though N is reduced, r is even closer to one, Ψ, δ and Δ are reduced and there is a large reduction in RPD for all products (Tables 5, 6).

The residual values also varied significantly across the range in chl-a values. The products both under- and over-estimated chl-a in oligotrophic waters (chl-a < 0.1 mg m⁻³; Figure 5). However, the degree of overestimation varied, with GlobColour GSM and CMEMS OC-CCI L3 being the most significant (median RPD +21.5 and +10.4%) and CMEMS GlobColour L4 and GlobColour AVW showing the least deviation (median RPD -2.5% and +0.1%). For mesotrophic waters (chl-a concentration between 0.1 and 1 mg m⁻³), the analysis of median RPD values across the seven products revealed distinct trends. CMEMS OC-CCI L3 and OC-CCIv6 both had a median RPD of +8.8%, indicating a moderate tendency to overestimate chl-a concentrations. Similarly, OC-CCv5 showed a slight overestimation with a median RPD of +9.6%. GlobColour AVW had a median RPD of −7.1% and −4.7%, showing a modest underestimation. The most substantial underestimation in the mid chl-a range was observed for CMEMS GlobColour L4, which had a median RPD of −21.8%. The analysis for eutrophic waters (chl-a mg m⁻³) showed consistent underestimation for all products. The GlobColour GSM product exhibited the lowest bias in the range with a median RPD of −19.5%. In contrast, OC-CCv5 and CMEMS GlobColour L4 presented the largest negative median 386 values, −35.5% and −58.8% respectively, suggesting a higher level of deviation. Other products such as CMEMS OC-CCI L3, CMEMS GlobColour L3, GlobColour AVW, and OC-CCv6 showed median values around −30%, reflecting a moderate bias.

3.2 Coincident match-up analysis

The scatter plots of in situ chl_ACS versus chl_satellite for the coincident match-ups across the seven products excluding the Tara and EXPORTS NP data are presented in Figure 7 (and including these data in Supplementary Figure S3). The relative percent residuals versus in situ chl_ACS from the different merged products for the coincident match-ups with; Supplementary Figure S4) and without Tara and EXPORTS NP data are given in Supplementary Figures S4, S8, respectively. The number of match-ups was 11,116 for the coincident analysis including Tara and EXPORTS NP data (Table 7), which is below the minimum number obtained for any L3 product in the independent analysis (N = 13,048). Excluding the Tara and EXPORTS NP data, the data set was reduced by >50%, but the quality of the resulting match-ups were far higher (Tables 1, 8).

As with the independent match-up analysis, the analysis of the coincident match-ups also showed that all products performed similarly. The slope was >0.92 for all products, the values closest to one were again obtained for the GlobColour products. Similarly the intercept was low for all products and smallest for the OC-CCI products. The bias was lowest for the GlobColour AVW and CMEMS L3 and L4 products. The RMS was consistently <0.195 for all products and lowest for OC-CCI v6. The L4 gap-filled GlobColour product were not different in performance to the L3 (non-gap filled) products (Table 7), unlike for the independent match-up analysis (Table 5), suggesting that its filled values were the reason for the lower performance observed in the independent match-up analysis. All statistics, especially Ψ and RPD, but not I, improved for all products when excluding the Tara and EXPORTS NP datasets (Table 8). This final match-up dataset is principally comprised of data from the Atlantic Ocean, during the period from September to November and in oligotrophic to mesotrophic waters (Table 9).

In summary, the impact of removing the Tara and EXPORTS NP data on the performance of the products was large. When including these data, the mean RPD over all products was 49% for the independent and 45% for the coincident match-up data, where as it was 34% and 31% when these data were removed.

To assess the overall ranking of all statistical metrics, we deployed the statistical classification scheme of (Brewin et al., 2015) which is a points scoring classification that ranks the performance of the products objectively based on all of the statistical metrics computed. Using this approach, a score is assigned for each statistical metric that varies from zero to two, and then all scores were summed. The total score for each products was then normalised by the average score of all products. A score of one indicates the performance of a product is average with respect to all models, a score greater than one indicates that a product is performing better than the average and a score of less than one indicates that the product is performing worse than average. Figure 8 plots the normalised scores of the independent and coincident match-ups and shows that the OC-CCI products are >1, and over the range of statistics used, have a slightly higher score.

FIGURE 8

Brewin et al., 2015) for (A) independent and (B) coincident match-ups (excluding Tara and EXPORTS NP datasets). Whiskers are the standard error of the statistical score index.

3.3 Spatial and temporal performance distribution

The spatial and temporal dependencies of the residuals for the coincident match-ups were analysed to assess differences between the products. This analysis is useful for identifying systematic biases or variations in the performance of the products, which could be attributed to temporal shifts introduced by sensor degradation or changes in sensors and where in the ocean basins the performance of the products differ.

Both the Tara and EXPORTS NP datasets exhibited a large range in residual values (Supplementary Figure S5), which is related to the quality of the in situ data and to rule out any bias in the residuals, these data were removed from the subsequent analysis. Some patterns emerge in the relative difference between the in situ and satellite multi-mission products. The performance of all products was similar with no apparent trend in the relative differences related to a specific sensor(s) (Figure 9), except for the CMEMS GlobColour L4 gap filled product which had consistently higher residuals (Figure 9F). The GlobColour AVW, CMEMS GlobColour L3 and L4 products had higher residuals during the Sentinel-3A era especially for the AMT27 and NAAMES 3 data collected in 2017 (Figures 9C,E,F). The highest relative difference for all products were with the AMT19 and 24 data sets collected during the MERIS - MODIS-Aqua and MODIS-Aqua – Suomi-VIIRS missions.

FIGURE 9

The match-up residuals map (Figure 10) showed where spatial dependencies arise due to regional differences, which can impact the over-all accuracy of the satellite-derived products. The OC-CCI v6 and CMEMS-CCI products exhibited higher relative differences in both the western, eastern North and Tropical Atlantic, but were lower in the North and South Atlantic gyres (Figures 10A,B). By contrast, GlobColour and CMEMS GlobColour products tended to under-estimate chl-a over most of the Atlantic Ocean except the far western and eastern range of the North Atlantic of the in situ dataset (Figures 10C–F). For most products, the mean RPD values tended to be higher in the mid-latitude ranges compared to the equatorial and polar regions. For example, the mean residual value of CMEMS OC-CCI L3 was 23% in the [23.5^◦N, 66.5^◦N] latitude range and [60^◦E, 0^◦E] longitude range, while the means were lower at other latitudinal bands.

FIGURE 10

The tropical regions generally exhibited moderate RPD values, with some products, such as GlobColour GSM, showing relatively higher means compared to other latitude bands (Figure 10D; Supplementary Figure S5). Some products, such as CMEMS OC-CCI L3 and GlobColour AVW, exhibited extremely high relative differences in specific longitude ranges like [60^◦E, 0^◦E] and [60^◦E, 180^◦E], potentially indicating regional biases or data-quality issues. Residual values across different latitude and longitude ranges could be influenced by factors such as regional oceanographic conditions, data quality, processing methods, and potential biases or limitations in the data sources or algorithms used by different products.

4 Discussion

There is a pressing need to produce consistent and highly accurate long-term global chl-a datasets, so that regional to global trends in phytoplankton biomass can be assessed under the shadow of climate change. This poses a grand challenge for the satellite ocean-colour community that needs to merge a multi-decadal time series built from sensors that differ in performance and calibration. Even if high observational stability is achieved, success in pursuing this challenge can be difficult to demonstrate, unless sea-truthing for the entire time series is available, which is rarely the case (Merchant et al., 2015).

4.1 Requirement for high quality data sets for ocean colour validation

HPLC was adopted as the gold standard for the measurement of total chl-a, as well as discrete spectrophotometric chl-a (Mueller et al., 2003) due to the low uncertainty for the determination of chl-a and associated pigment derivatives. There has been a concerted effort to produce large data sets that cover as much of the global ocean as possible, both spatially and temporally (Werdell and Bailey, 2005). Within HPLC chl-a databases, the HPLC protocols used can vary widely between laboratories (particularly for extraction, storage and detection procedures), due to each lab refining their methods to detect specific phytoplankton groups within regional waters. To ensure that the differences in HPLC total chl-a between laboratories are < 15%, there have been a series of inter-comparison exercises to assess differences between laboratories and to recommend updates in the protocols used to reduce them (Claustre et al., 2004). Recently, differences reported have been around 10% for chl-a (Canuti, 2023).

There has been a drive by the Space Agencies for laboratories to supply Fiducial reference measurements for the validation of satellite products, which are in situ data traceable to metrology standards, with associated uncertainties (Banks et al., 2020). To this end, Canuti (2023) computed uncertainties associated for HPLC chl-a from data from two laboratories were within the differences between laboratories, though the reporting of uncertainties associated with HPLC data is not common place when these data are uploaded to databases. More recently the determination of chl-a by other protocols was reviewed to enable other high quality methods, such as underway spectrophotometry, to be included to facilitate the collection of data semi-autonomously to increase the number of available match-ups with satellite ocean colour (IOCCG Protocol Series, 2018). The advantages of this method is that it not only collects a high throughput of data (every minute) over the ship’s track, but there is less variance in the protocols and methods used to run the systems (Slade et al., 2010; Brewin et al., 2016). To ensure consistency and quality in the in situ ACS chl-a data, the same accepted protocol and methodology was used for the in situ ACS chl-a. Post-processing de-biasing of the data was applied to all of the datasets using HPLC chl-a. A median filter was then applied to the data to remove artefacts that can arise from bubbles. This method also facilitates computation of propagated uncertainties, which have a broad distribution across different measurement conditions (Jordan et al., 2025). There is a need to develop new validation metrics that explicitly incorporate the uncertainties associated with the in situ observations. In this way, the uncertainties can then be used to further screen the quality of the data used for validation by either using them in the weighted mean relative differences between in situ and satellite observations, or providing metrics using measurements within specific uncertainty bounds, or assessing the impact of excluding data within a range of uncertainties on the computed metrics. In the dataset we collated for this paper, the uncertainties were between 7 and 58% and enabled us to screen the quality of data. Since both EXPORTS NP, Tara-Oceans and Tara-Polar were >27% with high relative residuals, analysis was also performed without these data (Tables 5, 7). The uncertainties of the ACS chl-a data used for validation are therefore less than the expected differences between in situ and satellite ocean colour chl-a for the open-ocean (∼35%).

4.2 Differences in coverage between merged ocean-colour products

ECVs require long-term time series with high spatial and temporal coverage able to distinguish true climate signatures, whilst minimising the effects of changes in satellite sensors and nuances in data processing (GCOS, 2021). Of the products tested in this study, the CMEMS GlobColour Level 4 product returned the highest number of match-ups (N = 61,307), but had the highest I, Ψ, Δ and δ and the lowest S and r. The GlobColour GSM returned the lowest number of match-ups (N = 13,048; Table 7). The CMEMS GlobColour Level 4 product provides an interpolated daily cloud-free chl-a product which has less conservative flagging compared to OC-CCI, that resulted in a higher N. Specifically, the CMEMS GlobColour Level 4 product utilises averaging and interpolation to fill in missing data values. The methods include optimal interpolation and a data-interpolating empirical orthogonal function (DINEOF), which is why the number of returned match-ups is ∼4 times higher than the other products that do not use gap-filling routines. Yet, this increase in coverage appeared to degrade the performance of the product when we compared it to independent in situ data (Figure 5; Tables 5, 6). The other ocean-colour merged products do not employ gap-filling methods, yet OC-CCI v6 and CMEMS OC-CCI Level 3 returned the next highest number of match-ups. OC-CCI v6 utilises the POLYMER AC with an updated idepix pixel flagging, plus the addition of Sentinel-3B to improve the spatial and temporal coverage of observations. A number of previous studies have demonstrated that POLYMER increases the data coverage both at global, basin and regional scales (Muller et al., 2015a; Tilstone et al., 2021). Improvements in data coverage in regions affected by haze and cloud edge effects have been documented in the Arabian Sea (Al-Naimi et al., 2017) and Red Sea (Racault et al., 2015). The reason for this increase in coverage is that POLYMER can operate under sun glint and thin clouds (Steinmetz et al., 2011; Tan et al., 2019). In addition, during the periods when MERIS and Sentinel-3A OLCI became available, the OC-CCI data have also been shown to increase data coverage due to multiple sensor over-passes (Belo Couto et al., 2016). Not only does the POLYMER atmospheric correction with OC-CCI data increase the coverage, it also leads to accurate chl-a values (Tables 5–8). By comparison, the GlobColour AVW product utilises the atmospheric correction available for each single satellite sensor that it then uses to generate the merged product using weighted averaging (Garnesson et al., 2019). Previous work suggested that GlobColour AVW has better spatial coverage, by a factor of 2.8, compared to OC-CCI, due to the less conservative flagging strategy it employs (Garnesson et al., 2019). Consistently, for the independent match-ups, OC-CCI v6 had lower I, Ψ and Δ compared to GlobColour AVW but not for δ, and the slope of the AVW product was closer to one (Tables 5, 6). These statistics differed slightly for the coincident match-ups (Tables 7, 8). We found that OC-CCI v6 resulted in ∼35% more match-ups compared to GlobColour AVW. The GlobColour GSM product is generated using merged normalised R_rs data to which the Maritorena and Siegel (2005) algorithm is applied. The AC used in the processing chain is NASA AC. Theoretically, the coverage should be similar to GlobColour AVW, but was slightly (∼2%) lower, which meant that OC-CCI v6 had ∼37% more match-ups compared with GlobColour GSM. This has implications for trend analysis of the products both in the global ocean and at regional scales.

Current research using merged ocean colour products have illustrated contrasting trends. At a Global level, analysis of 23 years of OC-CCI data from 1998 to 2020 revealed that there is a significant positive trend in surface Chl-a, which has increased by 0.67% ± 0.37% yr⁻¹ (Yu et al., 2023). Analysis of Global CMEMS CCI data over the period from 1998 to 2017 also reported an increasing trend in chl-a similarly of 0.6% ± 0.01% year⁻¹. Analysis of these CMEMS CCI data at regional scales showed increasing trends in chl-a in the North Atlantic and Arctic Ocean, but decreasing trends around Pacific Islands. The trends reported are ultimately dependent on the accuracy of the merged chl-a product data in each region. For example, in our analysis though the CMEMS GlobColour Level 4 gap filled product has an improved coverage which therefore results in a larger number of match-ups, the tendency to under-estimate chl-a globally and especially in the mesotrophic chl-a range, could result in a decreasing trend in chl-a for these regions.

4.3 Differences in the performance of the merged ocean-colour products based on independent and coincident match-up analyses

The in situ chl-a dataset covered ranges in values typical of oligrotrophic (<0.1 mg m⁻³), mesotrophic (0.1–1 mg m⁻³) and eutrophic (>1 mg m⁻³) conditions (Tables 1, 9; Figures 1, 7). However, the in situ data were biased seasonally, mainly towards autumn in the boreal north and spring in the austral south. There were fewer data in winter and summer, with the latter having the lowest representation (Figure 2). For both independent and coincident match-ups, the highest number of data were found towards the lower end of the chl-a range, with the maximum density being at ∼0.05 mg m⁻³ (Figures 5, 7). There were two orders of magnitude fewer points at the higher end compared to the lower end of the chl-a range. The match-ups therefore represent the performance of the different merged products principally in oligotrophic to mesotrophic conditions, predominantly in the Atlantic Ocean and secondarily in the North Pacific and Arctic Oceans during boreal autumn and austral spring (Figures 1, 2). For the independent match-ups, OC-CCI v6 exhibited the lowest bias and intercept, with a correlation coefficient (r) close to 1 (Table 5). By contrast, the bias and intercept of the GlobColour AVW and GSM were higher, and although CMEMS GlobColour L4 product had the highest N, it also had the lowest slope and r, and highest bias, RMS and RPD (Table 5). For the coincident match-ups, the pattern in the statistics changed slightly. OC-CCI v6 still exhibited the lowest RMS and intercept, and r closest to 1, but the GlobColour products (except GSM), had the smallest bias and RPD, with the slope closest to 1 (Table 7). Over all statistics, the performance of OC-CCI v6 and CMEMS CCI products were marginally more accurate compared to the other products (Figure 8).

The observed differences in the validation results are due to several key differences in the processing steps and product characteristics of the OC-CCI and GlobColour GSM and AVW merged products. The first is the atmospheric correction used and how it is implemented. The second is the chl-a algorithms employed (Table 4). As highlighted in the previous section, OC-CCI uses the POLYMER atmospheric correction that improves R_rs data coverage in areas affected by sun glint and haze (Steinmetz et al., 2011). The GlobColour AVW and GSM products instead use the default atmospheric correction procedure applied by each space agency and for AVW, no additional bias correction is applied. For SeaWiFS, MODIS-Aqua and VIIRS, this is the NASA AC model (Gordon, 1997; Bailey et al., 2010; Siegel et al., 2000; Ahmad et al., 2010; Ibrahim et al., 2018). The NASA AC can respond differently under specific observation and illumination geometries, aerosol types and water types (Gordon, 2021; Li et al., 2022; Wang et al., 2022). In the open ocean, the assignment of atmospheric particles of marine origin is effective over large areas of the global ocean, but can be erroneous where there is a significant influence of atmospheric dust (e.g., Pabortsava et al., 2017), smoke from wildfires (Tang et al., 2011), volcanic ash (Whiteside et al., 2023) and anthropogenic aerosols (Doron et al., 2011) in remote areas of the ocean. For POLYMER, a polynomial is used to model the spectral shape of the reflectance from the atmosphere. In addition, the POLYMER model utilises a spectral matching of R_rs over a large range of backscattering values and chl-a concentrations (Steinmetz et al., 2011), representative of the range of chl-a values used in this study. POLYMER accurately reproduces scattering by aerosols in the atmosphere 520 under a wide range of conditions (Tan et al., 2019). A comparison between NASA AC and POLYMER applied to both Sentinel-3A and -3B data over the Atlantic Ocean from 2016 to 2019 showed that POLYMER performed better over all visible bands, though improvement in NASA AC is expected once more in situ data become available for updating the vicarious calibration (Pardo et al., 2023).

Further to the differences in atmospheric correction models and procedures, employing different chl-a algorithms will also influence the accuracy of the independent and coincident match-ups analyses. In principle, OC-CCI, GlobColour AVW and CMEMS-GlobColour utilize similar chl-a algorithms (Table 4), but the way that they are applied in the processing steps, varies greatly. There are fundamental differences in the way that chl-a is computed between OC-CCI and GlobColour. OC-CCI firstly merges R_rs and computes chl-a from the merged R_rs product. GlobColour GSM also uses merged and normalised R_rs across the multi-mission time series whereas the AVW product uses chl-a computed from each ocean colour sensor and then post-merges the chl-a product. Though both OC-CCI and Globcolour GSM use merged R_rs to compute chl-a, the algorithms that are applied are very different. OC-CCI applies one of 14 different chl-a algorithm combinations based on the OWT for a specific pixel or area and then merges the OWT to produce a coherent and consistent map of chl-a. By contrast, GlobColour GSM applies one algorithm, the GSM to the merged R_rs.

There are subtle differences between OC-CCI v5 and v6 whereby OC3 and OC5 are not utilised in v6 resulting in small differences in statistical parameters between them for both the independent (Tables 5, 6) and coincident (Tables 7, 8) match-ups with v6 performing slightly better (Figure 8). For the independent match-up analysis, the greater N for v6 creates a slightly higher scatter around the 1:1 line, that cause the slope to become closer to 1 and switch the intercept from being negative in v5, to positive in v6 (Tables 5, 6). For the coincident match-up analysis, the effect is similar, but the intercept for v6 is reduced (Tables 7, 8). When all statistical parameters are considered using the metric indices, there are no differences between v5 and v6 (Tables 5-8).

Of the GlobColour products evaluated, GlobColour AVW performed slightly better across all statistical metrics compared to the other GlobColour products (Tables 5-8). This is surprising as generally CMEMS GlobColour is considered to be an evolution beyond OC3 and OC4 to provide more accurate chl-a concentrations in both coastal, shelf and oligotrophic environments. The GlobColour GSM product exhibited a comparatively high relative percent difference compared to the other merged products (Tables 5–8). The GSM was originally developed for Case 1 waters (Maritorena et al., 2002). For case 2 waters, it solves a_ph(λ) in the presence of total suspended material (TSM) and a_CDOM(λ). This IOP parameterization may not be appropriate for many of the areas covered by the in situ data, especially the Indian Ocean, the Arctic and Antarctic as evidenced through a growing number of past and previous studies (Huang et al., 2013; Kolluru et al., 2021; Ortega-Retuerta et al., 2010; Tilstone et al., 2013; Tilstone et al., 2017; Wojtasiewicz et al., 2018), though locally adapted versions of the algorithm are documented to improve performance (Kuchinke et al., 2009; Organelli et al., 2016; Tiwari and Shanmugam, 2011).

More subtle differences between the products also arise from the derivation of the merged products and the flagging schemes that are used. The GlobColour products use a specific flagging scheme on each sensor, whereas the OC-CCI approach uses a more constrained flagging method. GlobColour AVW was slightly lower performing than OC-CCI v5 and v6. There are few independent studies that compare the OC-CCI and GlobColour merged products and from these, the salient patterns are not completely clear. For example, Pramlall et al. (2023) found that in Case 1 waters, OC-CCI was more accurate than GlobColour with higher r and lower bias and RMSE. Yu et al. (2023) compared both products against a large global in situ dataset and found that GlobColour had a slightly higher correlation (r = 0.77) compared to OC-CCI (r = 0.74). So in general, OC-CCI tends to have an advantage in Case 1 waters by prioritising consistency and minimising noise, while GlobColour’s less conservative flagging gives it better spatial coverage in cases where the same satellites and atmospheric corrections methods are used. Their relative performance can vary regionally, especially in Case 2 waters. For accurate analysis of climate trends, GCOS specify that the mean relative error in the chl-a product should be below 30% (GCOS, 2021). Based on the coincident analysis using the same data, the RPD of all seven products was close to 30% and was 30% for GlobColour AVW, 31% for OC-CCI, 33% for GlobColour GSM.

In addition to the validation of the products, other characteristics for the suitability of maintaining and extending the time series such as bias correction, computation of uncertainties, quality assurance, frequency of reprocessing and novel product features should also be considered. For a further discussion on the suitability of multi-mission ocean colour products to produce high quality ECV time-series, the reader is referred to the Supplementary Material to this manuscript and Supplementary Table S1.

4.4 Future recommendations

5 Conclusion

The primary objective of this study was to evaluate the performance of various multi-sensor merged chl-a products derived from ocean-colour sensors. To achieve this, we used an in situ dataset based on AC-S measurements, which is widely recognised as the most comprehensive and globally representative collection of co-located measurements of in situ chl-a and optical properties taken during underway sampling. We quantified the statistical performance of two OC-CCI versions (v5, v6), two GlobColour products (AVW, GSM) and three CMEMS (GlobColour L3, L4, CMEMS-CCI) products against the most comprehensive in situ chl-a dataset to date based on spectrophotometric quasi-autonomous underway measurements. The in situ dataset consisted of nine AMT campaigns, global Tara Oceans and Arctic Tara Polar campaigns, campaigns in the north Atlantic (four NAMES), North Pacific (EXPORTS NP), and sub-tropical Pacific (GP5-06, TAO2012) and provided >13,000 in situ chl-a-satellite match-up data points. Further screening of these data reduced the match-up data set to the Atlantic and Pacific Oceans, during September to November that covered oligotrophic to mesotrophic waters. Using these data, all merged products exhibited a similar performance, though there were differences in some of the statistical metrics that were quantified. The independent analysis shows the gap-filled CMEMS GlobColour L4 achieving many more match-ups but degraded performance when compared to non-gap-filled products, whereas the coincident analysis suggests the filled values drive that degradation. That nuance is valuable and could be framed more explicitly as a trade-off between coverage and fidelity. Overall all statistical metrics used, OC-CCI v6 and CMEMS CCI had a marginally improved performance compared to the other products.

References

• 1

  AhmadZ.FranzB. A.McClainC. R.KwiatkowskaE. J.WerdellJ.ShettleE. P.et al (2010). New aerosol models for the retrieval of aerosol optical thickness and normalized water-leaving radiances from the SeawiFS and MODIS sensors over coastal regions and open oceans. Appl. Opt.49, 5545–5560. 10.1364/AO.49.005545

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  Al-NaimiN.RaitsosD. E.Ben-HamadouR.SolimanY. (2017). Evaluation of satellite retrievals of chlorophyll-a in the arabian gulf. Remote Sens.9, 3. 10.3390/rs9030301

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BaileyS. W.FranzB. A.WerdellP. J. (2010). Estimation of near-infrared water-leaving reflectance for satellite ocean color data processing. Opt. Express18, 7521–7527. 10.1364/OE.18.007521

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  BanksA. C.VendtR.AlikasK.BialekA.KuuskJ.LerebourgC.et al (2020). Fiducial reference measurements for satellite ocean colour (FRM4SOC). Remote Sens.12, 1322. 10.3390/rs12081322

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  BehrenfeldM. J.O’MalleyR. T.SiegelD. A.McClainC. R.SarmientoJ. L.FeldmanG. C.653et al (2006). Climate-driven trends in contemporary ocean productivity. Nature444, 752–755. 10.1038/nature05317

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  BehrenfeldM. J.MooreR. H.HostetlerC. A.GraffJ.GaubeP.RussellL. M.et al (2019). The 650 north Atlantic aerosol and marine ecosystem study (naames): science motive and mission overview. Front. Mar. Sci.6, 122. 10.3389/fmars.2019.00122

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  Belo CoutoA.BrotasV.MelinF.GroomS.SathyendranathS. (2016). Inter-comparison of OC-CCI chlorophyll-a estimates with precursor data sets. Int. J. Remote Sens.37, 4337–4355. 10.1080/01431161.2016.1209313

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  BossE. S.CollierR.PegauW.LarsonG.FennelK. (2007). Measurements of spectral optical properties and their relation to biogeochemical variables and processes in Crater Lake, national park, oregan. Hydrobiologia574, 149–159. 10.1007/s10750-006-2609-3

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  BossE.PicheralM.LeeuwT.ChaseA.KarsentiE.GorskyG.et al (2013). The characteristics of 659 particulate absorption, scattering and attenuation coefficients in the surface ocean; contribution of the tara 660 oceans expedition. Methods Oceanogr.7, 52–62. 10.1016/j.mio.2013.11.002

  • CrossRef
  • Google Scholar

• 10

  BrewinR. J. W.SathyendranathS.MullerD.BrockmannC.DeschampsP.-Y.DevredE.et al (2015). The ocean colour climate change initiative: III. A round-robin comparison on in-water 670 bio-optical algorithms. Remote Sens. Environ.162, 271–294. 10.1016/j.rse.2013.09.016

  • CrossRef
  • Google Scholar

• 11

  BrewinR. J. W.Dall’OlmoG.PardoS.van Dongen-VogelsV.BossE. S. (2016). Underway spectrophotometry along the Atlantic meridional transect reveals high performance in satellite chlorophyll retrievals. Remote Sens. Environ.183, 82–97. 10.1016/j.rse.2016.05.005

  • CrossRef
  • Google Scholar

• 12

  CampbellJ. W. (1995). The lognormal distribution as a model for bio-optical variability in the sea. J. Geophys. Res. Oceans100, 13237–13254. 10.1029/95jc00458

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  CanutiE. (2023). Phytoplankton pigment in situ measurements uncertainty evaluation: an HPLC interlaboratory comparison with a european-scale dataset. Front. Mar. Sci.10, 1197311. 10.3389/fmars.2023.1197311

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  CazzanigaI.KwiatkowskaE. (2018). “Sentinel-3 OLCI chlorophyll index switch for low-chlorophyll waters algorithm theoretical basis document,” in EUMETSAT Report. EUM/RSP/DOC/18/1028360.

  • Google Scholar
  • View reference in article

• 15

  ClaustreH.HookerS. B.Van HeukelemL.BerthonJ.-F.BarlowR.RasJ.et al (2004). An intercomparison of HPLC phytoplankton pigment methods using in situ samples: application to remote 678 sensing and database activities. Mar. Chem.85, 41–61. 10.1016/j.marchem.2003.09.002

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  Dall’OlmoG.WestberryT. K.BehrenfeldM. J.BossE.SladeW. H. (2009). Significant 682 contribution of large particles to optical backscattering in the open ocean. Biogeosciences6, 947–967. 10.5194/bg-6-947-2009

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  Dall’OlmoG.BossE.BehrenfeldM.WestberryT. (2012). Particulate optical scattering coefficients 680 along an Atlantic meridional transect. Opt. Express20, 21532–21551. 10.1364/OE.20.021532

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  DavisR.MooreC.ZaneveldJ.NappJ. (1997). Reducing the effects of fouling on chlorophyll estimates derived from long-term deployments of optical instruments. J. Geophys. Res. Oceans102, 5851–5855. 10.1029/96jc02430

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  DoronM.BelangerS.DoxaranD.BabinM. (2011). Spectral variations in the near-infrared ocean reflectance. Remote Sens. Environ.115, 1617–1631. 10.1016/j.rse.2011.01.015

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  FordD.TilstoneG. H.ShutlerJ. D.KitidisV.LobanovaP.SchwarzJ.et al (2021). Wind speed and mesoscale features drive net autotrophy in the south Atlantic Ocean. Remote Sens. Environ.691, 112435. 10.1016/j.rse.2021.112435

  • CrossRef
  • Google Scholar

• 21

  GarnessonP.ManginA.Fanton d’AndonO.DemariaJ.BretagnonM. (2019). The CMEMS GlobColour chlorophyll a product based on satellite observation: multi-sensor merging and flagging strategies. Ocean Sci.15, 819–830. 10.5194/os-15-819-2019

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  GCOS (2021). The status of the global climate observing system 2021: the GCOS status report (GCOS-240). Geneva, Switzerland:World Meteorological Organization publication. Available online at: https://library.wmo.int/viewer/57596/?offset=2#page=3&viewer=picture&o=bookmark&n=0&q=.

  • Google Scholar
  • View reference in article

• 23

  GlobColour (2020). GlobColour product user guide. GC-UM-ACR-PUG-01, version 4.2.1. Tech. Rep. 10.48670/moi-0027

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  GohinF.DruonJ. N.LampertL. (2002). A five channel chlorophyll concentration algorithm 700 applied to SeaWiFs data processed by SeaDAS in coastal waters. Int. J. Remote Sens.23, 1639–1661. 10.1080/01431160110071879

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  GordonH. R. (1997). Atmospheric correction of ocean color imagery in the Earth observing system era. J. Geophys. Res. Atmos.102, 17081–17106. 10.1029/96JD02443

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  GordonH. R. (2021). Evolution of ocean color atmospheric correction: 1970–2005. Remote Sens.13, 5051. 10.3390/rs13245051

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  GrabanS.Dall’OlmoG.GoultS.SauzedeR. (2020). Accurate deep-learning estimation of chlorophyll-a concentration from the spectral particulate beam-attenuation coefficient. Opt. Express28, 24214–24228. 10.1364/OE.397863

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  GreggW. W.CaseyN. W. (2004). Global and regional evaluation of the SeaWiFS chlorophyll data 710 set. Remote Sens. Environ.93, 463–479. 10.1016/j.rse.2003.12.012

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  HammondM. L.BeaulieuC.SahuS. K.HensonS. A. (2017). Assessing trends and uncertainties in satellite-era ocean chlorophyll using space-time modeling. Glob. Biogeochem. Cycles31, 1103–1117. 10.1002/2016GB005600

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  HookerS. B.HeukelemL. V.ThomasC. S.ClaustreH.RasJ.BarlowR.et al (2005). The second SeaWiFS HPLC Analysis Round-Robin Experiment (SeaHARRE-2). Tech. Rep.NASA TM-2005- 212785, 1–72.

  • Google Scholar
  • View reference in article

• 31

  HookerS. B.ClementsonL.ThomasC. S.SchluterL.AllerupM.RasJ. (2012). The Fifth SeaWiFS HPLC Analysis Round-Robin Experiment (SeaHARRE-5). Rep. Greenbelt, MA NASA Goddard 716 Space Flight Center. Tech. rep.1–108.

  • Google Scholar
  • View reference in article

• 32

  HuC.LeeZ.FranzB. (2012). Chlorophyll a algorithms for oligotrophic oceans: a novel approach based on three-band reflectance difference. J. Geophys. Res. Oceans117, C01011. 10.1029/2011JC007395

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  HuC. M.FengL.LeeZ. P.FranzB. A.BaileyS. W.WerdellP. J.et al (2019). Improving satellite global chlorophyll a data products through algorithm refinement and data recovery. J. Geophys. Res. Oceans124, 1524–1543. 10.1029/2019JC014941

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  HuangJ.ChenL.ChenX.SongQ. (2013). Validation of semi-analytical inversion models for inherent optical properties from ocean color in coastal yellow sea and East China Sea. J. Oceanography725 (69), 713–725. 10.1007/s10872-013-0202-8

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  IbrahimA.FranzB.AhmadZ.HealyR.KnobelspiesseK.GaoB.-C.et al (2018). Atmospheric correction for hyperspectral ocean color retrieval with application to the hyperspectral imager for the coastal ocean (HICO). Remote Sens. Environ.204, 60–75. 10.1016/j.rse.2017.10.041

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  IOCCG Protocol Series (2018). Inherent Optical Property Measurements and Protocols: Absorption Coefficient. Editors VandermeulenR. A.ChavesJ. E.IOCCGD.IOCCG Ocean Optics and Biogeochemistry Protocols for Satellite Ocean Colour Sensor Validation (Canada: NS), 1.0.

  • Google Scholar
  • View reference in article

• 37

  JacksonT.VolpeG.SathyendranathS.GroomS. (2021). Product validation and intercomparison 730 report (Plymouth. Tech. rep., Plymouth Marine Laboratory (PML) and ESA Climate Office). Available online at: https://climate.esa.int (Accessed July 08, 2026).

  • Google Scholar
  • View reference in article

• 38

  JordanT. M.Dall’OlmoG.TilstoneG.BrewinR. J. W.NencioliF.AirsR.et al (2025). A compilation of surface inherent optical properties and phytoplankton pigment concentrations from the Atlantic meridional transect. Earth Syst. Sci. Data17, 493–516. 10.5194/essd-17-493-2025

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  KarlusichP.JoseJ.IbarbalzF. M.BowlerC. (2020). Phytoplankton in the tara ocean. Annu. Rev. Mar. Sci.12, 233–265. 10.1146/annurev-marine-010419-010706

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  KolluruS.GedamS. S.InamdarA. B. (2021). A neural network approach for deriving absorption 738 coefficients of ocean water constituents from total light absorption and particulate absorption coefficients. Comput. Geosciences147, 104678. 10.1016/j.cageo.2020.104678

  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  KuchinkeC. P.GordonH. R.HardingL. W.VossK. J. (2009). Spectral optimization for constituent retrieval in case 2 waters II: validation study in the Chesapeake Bay. Remote Sens. Environ.113, 610–621. 10.1016/j.rse.2008.11.002

  • CrossRef
  • Google Scholar

• 42

  LiQ.JiangL.ChenY.WangL.WangL. (2022). Evaluation of seven atmospheric correction algorithms for olci images over the coastal waters of qinhuangdao in bohai sea. Regional Stud. Mar. Sci.56, 102711. 10.1016/j.rsma.2022.102711

  • CrossRef
  • Google Scholar
  • View reference in article

• 43

  LiuY.RottgersR.Ramirez-PerezM.DinterT.SteinmetzF.NothigE.-M.et al (2018). Underway spectrophotometry in the fram strait (European Arctic Ocean): a highly resolved chlorophyll a data source for complementing satellite ocean color. Opt. Express26, A678–A696. 10.1364/OE.26.00A678

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 44

  MaritorenaS.SiegelD. A. (2005). Consistent merging of satellite ocean color data sets using a bio-optical model. Remote Sens. Environ.94, 429–440. 10.1016/j.rse.2004.08.014

  • CrossRef
  • Google Scholar
  • View reference in article

• 45

  MaritorenaS.SiegelD. A.PetersonA. R. (2002). Optimization of a semianalytical ocean color model for global-scale applications. Appl. Opt.41, 2705–2714. 10.1364/AO.41.002705

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  MaritorenaS.Fanton d’AndonO. H.ManginA.SiegelD. A. (2010). Merged satellite ocean color data products using a bio-optical model: characteristics, benefits and issues. Remote Sens. Environ.114, 1791–1804. 10.1016/j.rse.2010.04.002

  • CrossRef
  • Google Scholar

• 47

  MelinF.SclepG.JacksonT.SathyendranathS. (2016). Uncertainty estimates of remote sensing´ reflectance derived from comparison of ocean color satellite data sets. Remote Sens. Environ.762 (177), 107–124. 10.1016/j.rse.2016.02.014

  • CrossRef
  • Google Scholar
  • View reference in article

• 48

  MerchantC. J.de LeeuwG.WagnerW. (2015). Selecting algorithms for Earth observation of climate within the european space agency climate change initiative: introduction to a special issue. Remote Sens. Environ.162, 239–241. 10.1016/j.rse.2015.02.017

  • CrossRef
  • Google Scholar

• 49

  MorelA.AntoineD. (2011). “Pigment index retrieval in case 1 waters,” in Algorithm Technical Baseline Document 2, 9. Laboratoire de Oceangraphie, Villefranche, France. European Space Agency.

  • Google Scholar

• 50

  MuellerJ. L.FargionG. S.McClainC. R. (2003). Biogeochemical and bio-optical measurements and data analysis. Ocean Optics Protocols for Satellite Ocean Colour Sensor Validation, 768, 1–9. Washington, United States:National Aeronautical and Space Administration.

  • Google Scholar

• 51

  MullerD.KrasemannH.BrewinR. J. W.BrockmannC.DeschampsP.-Y.DoerfferR.et al (2015a). The ocean colour climate change initiative: I. An assessment of atmospheric correction algorithms based on in-situ measurements. Remote Sens. Environ.162, 242–256. 10.1016/j.rse.2013.11.026

  • CrossRef
  • Google Scholar

• 52

  MullerD.KrasemannH.BrewinR. J. W.BrockmannC.DeschampsP.-Y.DoerfferR.et al (2015b). The ocean colour climate change initiative: II. Spatial and seasonal homogeneity of atmospheric correction algorithms. Remote Sens. Environ.162, 257–270. 10.1016/j.rse.2015.01.033

  • CrossRef
  • Google Scholar

• 53

  OC-CCI (2014). System prototype description, ref: D3.1, 11/02/14, issue: 1.1. ESA/ESRIN, AO 1/6207/09/I-LG.

  • Google Scholar

• 54

  OC-CCI (2020). OC-CCI D4.2 - Product user guide for v5.0 dataset. Tech. Rep. (1), 1–50.

  • Google Scholar

• 55

  OC-CCI (2022). OC-CCI D4.2 - Product user guide for v6.0 dataset. Tech. Rep. (1), 1–50.

  • Google Scholar

• 56

  OrganelliE.ClaustreH.BricaudA.SchmechtigC.PoteauA.XingX.et al (2016). A novel near-real-time quality-control procedure for radiometric profiles measured by bio-argo floats: protocols and performances. J. Atmos. Ocean. Technol.33, 937–951. 10.1175/JTECH-D-15-0193.1

  • CrossRef
  • Google Scholar

• 57

  Ortega-RetuertaE.SiegelD. A.NelsonN. B.DuarteC. M.RecheI. (2010). Observations of chromophoric dissolved and detrital organic matter distribution using remote sensing in the Southern Ocean: validation, dynamics and regulation. J. Mar. Syst.82, 295–303. 10.1016/j.jmarsys.2010.06.004

  • CrossRef
  • Google Scholar

• 58

  O’ReillyJ. E.MaritorenaS.MitchellB. G.SiegelD. A.CarderK. L.GarverS. A.et al (1998). Ocean color chlorophyll algorithms for SeaWiFS. J. Geophys. Res. Oceans103, 24937–24953. 10.1029/98JC02160

  • CrossRef
  • Google Scholar

• 59

  O’ReillyJ. E.et al (2000). “Ocean color chlorophyll-a algorithms for SeaWiFS, OC2 and OC4,” in Seawifs Postlaunch Calibration and Validation Analyses: Part 3. Editors HookerS. B.FirestoneE. R. (Greenbelt: Greenbelt, MD: NASA Goddard Space Flight Center), 9–23.

  • Google Scholar

• 60

  PabortsavaK.LampittR. S.BensonJ.CroweC.McLachlanR.Le MoigneF. A. C.et al (2017). Carbon sequestration in the deep Atlantic enhanced by Saharan dust. Nat. Geosci.10, 189–194. 10.1038/ngeo2899

  • CrossRef
  • Google Scholar

• 61

  PardoS.TilstoneG. H.BrewinR. J. W.Dall’OlmoG.LinJ.NencioliF.et al (2023). Radiometric assessment of olci, viirs, and modis using fiducial reference measurements along the Atlantic meridional 795 transect. Remote Sens. Environ.299, 113844. 10.1016/j.rse.2023.113844

  • CrossRef
  • Google Scholar

• 62

  PauthenetE.MartinezE.GorguesT.RoussillonJ.DrumetzL.FabletR.et al (2024). Contrasted trends in chlorophyll-a satellite products. Geophys. Res. Lett.51, e2024GL108916. 10.1029/2024GL108916

  • CrossRef
  • Google Scholar

• 63

  PramlallS.JacksonJ. M.KonikM.CostaM. (2023). Merged multi-sensor ocean colour chlorophyll product evaluation for the British Columbia Coast. Remote Sens.15 (3), 687. 10.3390/801rs15030687

  • CrossRef
  • Google Scholar

• 64

  RacaultM.-F.RaitsosD. E.BerumenM. L.BrewinR. J. W.PlattT.SathyendranathS.et al (2015). Phytoplankton phenology indices in coral reef ecosystems: application to ocean-color observations in the Red Sea. Remote Sens. Environ.160, 222–234. 10.1016/j.rse.2015.01.019

  • CrossRef
  • Google Scholar

• 65

  RasJ.ClaustreH.UitzJ. (2008). Spatial variability of phytoplankton pigment distributions in the subtropical south Pacific Ocean: comparison between in situ and predicted data. Biogeosciences5, 353–369. 10.5194/bg-5-353-2008

  • CrossRef
  • Google Scholar

• 66

  SathyendranathS.BrewinR. J. W.BrockmannC.BrotasV.CaltonB.ChuprinA.et al (2019). An ocean-colour time series for use in climate studies: the experience of the ocean-colour climate change initiative (oc-cci). Sensors19 (19), 4285. 10.3390/s19194285

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67

  SiegelD. A.WangM.MaritorenaS.RobinsonW. (2000). Atmospheric correction of satellite ocean color imagery: the Black pixel assumption. Appl. Opt.39, 3582–3591. 10.1364/AO.39.003582

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68

  SiegelD. A.BuesselerK. O.BehrenfeldM. J.Benitez-NelsonC. R.BossE.BrzezinskiM. A.et al (2016). Prediction of the export and fate of global ocean net primary production: the exports science plan. Front. Mar. Sci.3, 22. 10.3389/fmars.2016.00022

  • CrossRef
  • Google Scholar

• 69

  SladeW. H.BossE.Dall’OlmoG.LangnerM. R.LoftinJ.BehrenfeldM. J.et al (2010). Underway and moored methods for improving accuracy in measurement of spectral particulate absorption and attenuation. J. Atmos. Ocean. Technol.27, 1733–1746. 10.1175/8192010JTECHO755.1

  • CrossRef
  • Google Scholar

• 70

  SteinmetzF.DeschampsP.-Y.RamonD. (2011). Atmospheric correction in presence of sun glint: application to meris. Opt. Express19, 9783–9800. 10.1364/OE.19.009783

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 71

  SunagawaS.AcinasS. G.BorkP.BowlerC.Tara Oceans Coordinators, EveillardD.et al (2020). Tara oceans: towards global ocean ecosystems biology. Nat. Rev. Microbiol.18, 428–445. 10.1038/s41579-020-0364-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 72

  TanJ.FrouinR.RamonD.SteinmetzF. (2019). On the adequacy of representing water reflectance by semi-analytical models in ocean color remote sensing. Remote Sens.11 (23), 2820. 10.3390/rs1123282

  • CrossRef
  • Google Scholar

• 73

  TangW.WeisJ.PerronM. M. G.BasartS.LiZ.SathyendranathS.et al (2011). Widespread phytoplankton blooms triggered by 2019–2020 Australian wildfires. Nature597, 370–375. 10.1038/s41586-021-03805-8

  • CrossRef
  • Google Scholar

• 74

  TilstoneG. H.LotlikerA. A.MillerP. I.AshrafP. M.KumarT. S.SureshT.et al (2013). Assessment of modis-aqua chlorophyll-a algorithms in coastal and shelf waters of the eastern Arabian Sea. Cont. Shelf Res.65, 14–26. 10.1016/j.csr.2013.06.003

  • CrossRef
  • Google Scholar

• 75

  TilstoneG.Mallor-HoyaS.GohinF.Belo CoutoA.SaC.GoelaP.et al (2017). Which ocean colour algorithm for meris in north west European waters?Remote Sens. Environ.189, 132–151. 10.1016/j.rse.2016.11.012

  • CrossRef
  • Google Scholar

• 76

  TilstoneG. H.PardoS.Dall’OlmoG.BrewinR. J. W.NencioliF.DessaillyD.et al (2021). Performance of ocean colour chlorophyll a algorithms for sentinel-3 olci, modis-aqua and suomi-viirs in open-ocean waters of the Atlantic. Remote Sens. Environ.260, 112444. 10.1016/j.rse.2021.112444

  • CrossRef
  • Google Scholar

• 77

  TiwariS. P.ShanmugamP. (2011). An optical model for the remote sensing of coloured dissolved organic matter in coastal/ocean waters. Estuar. Coast. Shelf Sci.93, 396–402. 10.1016/j.ecss.2011.05.010

  • CrossRef
  • Google Scholar

• 78

  van HeukelemL.HookerS. B. (2011). “The importance of a quality assurance plan for method validation and minimizing uncertainties in the HPLC analysis of phytoplankton pigments”, in Phytoplankton Pigments: Characterization, Chemotaxonomy and Applications in Oceanography. Editors S. Roy, C. A. Llewellyn, E. S. Egeland, G. Johnsen (eds) (Cambridge Environmental Chemistry Series.Cambridge University Press) 195–256. 10.1017/CBO9780511732263.009

  • CrossRef
  • Google Scholar

• 79

  WangJ.WangY.LeeZ.WangD.ChenS.LaiW. (2022). A revision of NASA SeaDas atmospheric correction algorithm over turbid waters with artificial neural networks estimated remote-sensing reflectance in the near-infrared. ISPRS J. Photogrammetry Remote Sens.194, 235–249. 10.1016/j.isprsjprs.2022.10.014

  • CrossRef
  • Google Scholar

• 80

  WerdellP. J.BaileyS. W. (2002). in The Seawifs bio-optical Archive and Storage System (Seabass): Current Architecture and Implementation, NASA Tech. Memo. 2002-211617. Editors FargionG. S.McClainC. R. (Greenbelt, Maryland: NASA Goddard Space Flight Center), 45.

  • Google Scholar

• 81

  WerdellP. J.BaileyS. W. (2005). An improved in-situ bio-optical data set for ocean colour algorithm 853 development and satellite data production validation. Remote Sens. Environ.98, 122–140. 10.1016/j.rse.2005.07.001

  • CrossRef
  • Google Scholar

• 82

  WerdellP. J.ProctorC. W.BossE.LeeuwT.OuhssainM. (2013). Underway sampling of marine inherent optical properties on the tara oceans expedition as a novel resource for ocean color satellite data product validation. Methods Oceanogr.7, 40–51. 10.1016/j.mio.2013.09.001

  • CrossRef
  • Google Scholar

• 83

  WhitesideA.DupouyC.SinghA.BaniP.TanJ.FrouinR. (2023). Impact of ashes from the 2022 Tonga volcanic eruption on satellite ocean color signatures. Front. Mar. Sci.9, 1028022. 10.3389/fmars.2022.1028022

  • CrossRef
  • Google Scholar

• 84

  WojtasiewiczB.Hardman-MountfordN. J.AntoineD.DufoisF.SlawinskiD.TrullT. W. (2018). Use of bio-optical profiling float data in validation of ocean colour satellite products in a remote ocean region. Remote Sens. Environ.209, 275–290. 10.1016/j.rse.2018.02.057

  • CrossRef
  • Google Scholar

• 85

  YuS.BaiY.HeX.GongF.LiT. (2023). A new merged dataset of global ocean chlorophyll-a concentration for better trend detection. Front. Mar. Sci.10, 1051619. 10.3389/fmars.2023.1051619

  • CrossRef
  • Google Scholar

• 86

  ZhaiD.BeaulieuC.KudelaR. M. (2024). Long-term trends in the distribution of ocean chlorophyll. Geophys. Res. Lett.51, e2023GL106577. 10.1029/2023GL106577

  • CrossRef
  • Google Scholar

• 87

  ZibordiG.HolbenB.HookerS.MelinF.BerthonJ.-F.SlutskerI.et al (2006). A network for standardized ocean color validation measurements. Eos Trans. AGU87, 293–297. 10.1029/2006EO300001

  • CrossRef
  • Google Scholar