Regional Assessment of COCTS HY1-C/D Chlorophyll-a and Suspended Particulate Matter Standard Products over French Coastal Waters

Abstract

Chlorophyll-a (Chla) and suspended particulate matter (SPM) are key indicators of water quality, playing critical roles in understanding marine biogeochemical processes and ecosystem health. Although satellite data from the Chinese Ocean Color and Temperature Scanner (COCTS) onboard the Haiyang-1C/D satellites is freely available, there has been limited validation of its standard Chla and SPM products. This study is a first step to address this gap by evaluating COCTS-derived Chla and SPM products against in situ measurements in French coastal waters. The matchup analysis showed robust performance for the Chla product, with a median symmetric accuracy (MSA) of 50.46% over a dynamic range of 0.13–4.31 mg·m⁻³ (n = 24, Bias = 41.11%, Slope = 0.93). In contrast, the SPM product showed significant limitations, particularly in turbid waters, despite a reasonable performance in the matchup exercise, with an MSA of 45.86% within a range of 0.18–10.52 g·m⁻³ (n = 23, Bias = −14.59%, Slope = 2.29). A comparison with another SPM model and Moderate Resolution Imaging Spectroradiometer (MODIS) products showed that the COCTS standard algorithm tends to overestimate SPM and suggests that the issue does not originate from the input radiometric data. This study provides the first regional assessment of COCTS Chla and SPM products in European coastal waters. The findings highlight the need for algorithm refinement to improve the reliability of COCTS SPM products, while the Chla product demonstrates suitability for water quality monitoring in low to moderate Chla concentrations. Future studies should focus on the validation of COCTS ocean color products in more diverse waters.

1. Introduction

Ocean color radiometry (OCR) and its derived products are considered as essential climate variables by the Global Climate Observing System (https://gcos.wmo.int/, accessed on 10 February 2025)). Satellite-derived OCR products are critical for assessing ocean ecosystem health and productivity, monitoring the global carbon cycle, and quantifying the impacts of climate variability and change [1,2,3,4]. Two of the key OCR-derived products are chlorophyll-a concentration (Chla), the primary pigment in phytoplankton and an indicator of biomass and primary productivity [5,6,7,8], and suspended particulate matter (SPM), which influences water quality, light availability, and turbidity [9,10,11,12]. Together, these parameters provide essential insights into biogeochemical processes, coastal dynamics, and ecosystem functioning of marine environments. Coastal waters represent a key region to study as they are a major site of nutrients and sediments inputs [13], serving both human needs and ecological functions [14], and are particularly sensitive to environmental changes from both anthropogenic and natural origins [15,16].

Chla and SPM distributions in coastal waters are highly variable and vary over a broad range of time and space scales. Thus, their quantification through field sampling is considered inadequate to study water quality dynamics at large spatial and temporal scales [17]. As a result, satellite remote sensing is an efficient and cost-effective method for monitoring SPM and Chla, offering high temporal and spatial resolutions along with a synoptic view [18]. China launched its third and fourth ocean color satellites, Haiyang-1C (HY-1C) and Haiyang-1D (HY-1D), in 2018 and 2020, respectively, further increasing the amount of space-borne ocean color data available [19]. They are equipped with the Chinese Ocean Color and Temperature Scanner (COCTS), a wide-swath sensor providing daily coverage that contributes to regional and global monitoring of marine environments [20,21]. The dedicated atmospheric correction algorithm consists of a layer removal scheme [22,23] and has been validated, indicating its capacity to provide reliable radiometric data [24,25].

The retrieval of Chla and SPM concentrations from ocean color radiometry is primarily based on their relationships with water-leaving reflectance (${\rho }_w$), or remote sensing reflectance ($R_{rs}$), obtained from the top of atmosphere radiometric signal after correction of the atmospheric effects [26,27]. For Chla, one type of algorithm relies on blue-to-green band ratios [28,29], assuming that the magnitude and the shape of the reflectance spectrum are primarily governed by Chla. They are effective in clear waters but face challenges in optically complex coastal waters where other constituents, such as colored dissolved organic matter (CDOM) and SPM, influence reflectance signals [30]. On the other hand, algorithms based on red-NIR (near-infrared) bands assume negligible absorption by CDOM and non-algal particles in this region of the spectrum and perform well in Chla-dominated waters [31,32]. To improve Chla retrieval in optically diverse waters, band-switching approaches and adaptive algorithms have been developed to dynamically select the most appropriate inversion models based on water optical properties [33]. Finally, statistical approaches like neural networks have also been developed [7,34,35]. For COCTS, Chla concentration is estimated with an OC4-like model, tuned for this sensor [29]. Similarly, SPM can be estimated using empirical relationships or semi-analytical models, which either retrieve SPM concentrations directly or estimate proxies like the particulate backscattering coefficient ($b_{bp}$) [36,37]. The relationship between $R_{rs}$ and SPM is less sensitive in clear waters, for which the optical properties are dominated by other in-water constituents. In turbid waters, the sensitivity of NIR bands to SPM increases, whereas visible bands tend to saturate. Empirical models are generally simpler to implement but may perform poorly outside the range of conditions represented in their calibration datasets [38]. Band-switching methods and recent advancements in machine learning have improved the robustness of SPM retrievals across large turbidity levels [7,37,39,40]. Regarding COCTS, SPM is estimated using a proxy, deduced from four indices defined by combinations of reflectances at different bands [41]. Unfortunately, the actual expressions of the COCTS Chla and SPM models, and their calibration coefficients, are not publicly available.

While extensive research has been conducted on the evaluation of SPM and Chla standard algorithms for sensors such as MODIS [42,43], VIIRS [44], or OLCI [45], there is a need for more studies evaluating COCTS standard products, especially in coastal environments. Different Chla products derived from COCTS have been created and/or validated: ref. [21] developed an atmospheric correction algorithm for high-latitude seas and derived Chla with OC4, while [46] published a residual network Chla model dedicated to COCTS. However, this is not the case for the operational COCTS products currently distributed. Independent validation is essential for quantifying the accuracy of satellite-derived Chla and SPM, particularly in coastal waters where optical complexity and environmental variability remain challenging. This study aims to address this gap by evaluating the performance of the standard COCTS-derived SPM and Chla products using in situ measurements collected in French coastal waters. Consequently, this paper is organized as follows: Section 2 describes the in situ and satellite datasets used in this study and details the validation approach, including the matchup methodology, algorithms, and statistical metrics employed. Section 3 presents the validation results for both SPM and Chla products, while Section 4 provides a discussion on the findings and their implications for the performance of COCTS products in coastal waters and in general.

2. Materials and Methods

2.1. In Situ Data

Data from the French coastal monitoring network “Service d’Observation en Milieu Littoral” (SOMLIT) were used in the matchup analysis (https://www.somlit.fr/, accessed on 3 May 2024). This long-lasting in situ dataset, providing continuous bi-monthly data since 1997, is acquired using a standardized protocol. The dataset currently consists of twenty-two stations distributed along the French coastline, where sixteen parameters are measured or sampled [47]. The SPM concentration is determined by filtering a sufficient volume of water through pre-weighted GF/F filters (approximately 0.7 µm porosity) to deposit 0.5 to 1 mg of material on the filter, rinsing them with Milli-Q water to remove salts and drying the filters in a temperature-controlled oven at 50–70 °C for at least 6 h. Post-drying, filters are weighed again, and the SPM concentration is computed. The Chla concentration is measured by filtering from 0.1 to 1 L of seawater with GF/F filters (approximately 0.7 µm porosity), extracting pigments with 90% acetone, and measuring fluorescence before and after acidification. In this study, only surface measurements were selected, and the SPM data were quality controlled (retaining only data with quality flags 2, 6, and 7, corresponding to unique measurements, duplicate samples, and reliable measurements from an alternative protocol). After applying these criteria, 2265 and 2229 potential SPM and Chla matchups remained, respectively, with measurements being collected between September 2018 and July 2023. Figure 1 shows as red dots the locations of those in situ measurements in the French coastal waters. The Chla values vary from 0.04 mg·m⁻³ to 50.8 mg·m⁻³, with a mean value of 1.54 mg·m⁻³ (Figure 1b), while SPM values range from 0.01 g·m⁻³ to 1970.14 g·m⁻³, with a mean value of 47.28 g·m⁻³ (Figure 1c).

2.2. Satellite Data

2.2.1. COCTS

The Chinese Ocean Color and Temperature Scanner (COCTS) sensor has been in orbit aboard the HY-1C/D satellites since September 2018 for HY-1C and June 2020 for HY-1D, respectively. COCTS was firstly onboard the experimental ocean color satellite Haiyang-1A, launched in 2002, and has achieved significant improvements in its number of bands, swath width, and radiometric performance on the succeeding Haiyang-1B/C/D satellites. The sensor observes the Earth with a swath width of 3000 km and a spatial resolution of 1.1 km. COCTS comprises six visible bands and two near-infrared bands, i.e., 412 nm, 443 nm, 490 nm, 520 nm, 565 nm, 670 nm, 750 nm, and 865 nm [23]. The remote sensing data of COCTS on HY-1C/1D are available for registered users from the National Satellite Ocean Application Service (NSOAS) website (https://osdds.nsoas.org.cn/ (accessed on 10 May 2024)). Each file contains a five-minute granule of satellite data. L2A files contain $R_{rs}$ data and L2B files the derived biogeochemical products. A total of 1799 L2B files were downloaded for HY-1C, and 762 were downloaded for HY-1D. Notably, some L2A files required to derive the L2B files were missing on the NSOAS servers, reducing the potential number of matchups. Regarding the COCTS standard algorithms, Chla concentration is estimated with an OC4-like model [29], and SPM is derived using a proxy, deduced from four indices defined by combinations of reflectances at different bands [41]. However, the exact model formulations and their calibration coefficients are not publicly available. Several quality flags are included in the L2A and L2B files, such as atmospheric correction failure, sun glint, radiance saturation, high sensor view zenith angle, shallow water, or turbid water.

2.2.2. MODIS

The Moderate Resolution Imaging Spectroradiometer (MODIS) AQUA level 2 image used in this study, for a comparison with a COCTS image, was obtained from the NASA Ocean Color Archive (https://oceancolor.gsfc.nasa.gov, accessed on 7 February 2025). The MODIS AQUA sensor has been operational aboard NASA’s Earth Observing System (EOS/PM1) satellite since May 2002. The available spectral bands for this product are 412, 443, 469, 488, 531, 547, 555, 645, 667, and 678 nm.

2.3. Matchup Methodology

The matchup extraction was performed following the method of previously published works [45,48,49,50]. We considered 3 × 3 pixel windows, centered on the in situ sampling point, and matchups were selected according to the following criteria: (1) at least 6 valid pixels among the 9 pixels; (2) to ensure spatial homogeneity, the coefficient of variation (i.e., the ratio of the standard deviation over the mean value) of $R_{rs}$(565) within the window must be less than 20%; (3) the time difference between in situ and satellite measurements has to be lower than 3 h; and (4) the sun zenith angle has to be lower than 70°. Although this time window may be relatively large for SOMLIT stations located in dynamic coastal waters, we chose it to maximize the number of matchups [45,50]. Finally, for the matchups, SPM and Chla were extracted for the valid pixels in the 3 × 3 window, and the median value was calculated for comparison. A second stricter protocol was also applied by excluding matchups with negative $R_{rs}$ spectra in the visible spectrum.

2.4. Statistical Metrics

The performance of the SPM and Chla products was assessed using metrics calculated in the log-space, as this approach offers a more accurate evaluation due to the log-normal distribution of these parameters in natural environments [51]. Metrics based on the median were prioritized, as it is less affected by outliers [52,53]. The following metrics were analyzed, where $X^{est}$ represents the estimated value and $X^{obs}$ the in situ value of either SPM or Chla:

• Symmetric signed percentage bias (SSPB), representing a zero-centered percent bias that maintains symmetry between over- and under-estimation [53,54,55]. Negative values indicate under-estimation and positive over-estimation

$$SSPB=100\times sign\left(MdLQ\right)\times ({10}^{\left|MdLQ\right|}-1)$$

(1)

$$\mathrm{where}MdLQ=median\left(lo{g}_{10}(X^{est}/X^{obs})\right)$$

(2)

• Mean symmetric accuracy (MSA), representing a symmetric percentage error and penalizing over- and under-estimation in the same way [53,54,55]. This metric was designed to address the drawbacks of MAPE (mean absolute percentage error). Values closer to zero indicate better performance

$$MSA=100\times ({10}^{median\left(\right|lo{g}_{10}(X^{est}/X^{obs})\left|\right)}-1)$$

(3)

• Root mean square logarithmic error (“RMSLE”), quantifying the deviation of estimated values from the measured ones, incorporating a quadratic penalty for larger errors. Values closer to zero indicate better performance

$$RMSLE=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^n{(lo{g}_{10}\left(X_i^{\mathrm{est}}\right)-lo{g}_{10}\left(X_i^{\mathrm{obs}}\right))}^2}{n}}}$$

(4)

• The Slope from a type II linear regression on the log-transformed data accounts for errors in both observed and estimated variables.

$$lo{g}_{10}\left(X^{est}\right)=slope\times lo{g}_{10}\left(X^{obs}\right)+intercept$$

(5)

2.5. SPM Model Tested

In this study, the standard COCTS SPM product was compared with the semi-analytical approach proposed by Han et al. [39] (hereafter referred to as Han16). This model was chosen because it performs well over a wide range of SPM concentrations [45] and was also tuned for the band 670, making it applicable to COCTS. It combines two equations through weights, one tuned for low concentrations of SPM and one tuned for high concentrations. The development dataset spans 4 orders of magnitude, ranging from 0.15 to 2626 g·m⁻³. Both equations are combined when 0.03 < $R_{rs}\left(670\right)$ < 0.04 sr⁻¹:

$$SP{M}_{low}={\displaystyle \frac{391.161\times \pi \times R_{rs}\left(670\right)}{1-\frac{\pi \times R_{rs}\left(670\right)}{0.5}}}\mathrm{if}{R}_{rs}\left(670\right)<0.03s{r}^{-1}$$

(6)

$$SP{M}_{high}={\displaystyle \frac{1336.584\times \pi \times R_{rs}\left(670\right)}{1-\frac{\pi \times R_{rs}\left(670\right)}{0.3864}}}\mathrm{if}{R}_{rs}\left(670\right)>0.04s{r}^{-1}$$

(7)

$$SPM={\displaystyle \frac{W_{low}\times SP{M}_{low}+W_{high}\times SP{M}_{high}}{W_{low}+W_{high}}}$$

(8)

$$\begin{array}{cc}\hfill \mathrm{where}{W}_{\mathrm{low}}& ={log}_{10}\left(0.04\right)-{log}_{10}\left(R_{\mathrm{rs}}\left(670\right)\right),\hfill \\ \hfill W_{\mathrm{high}}& ={log}_{10}\left(R_{\mathrm{rs}}\left(670\right)\right)-{log}_{10}\left(0.03\right)\hfill \end{array}$$

(9)

3. Results

3.1. Matchup Exercise

The scatterplots in Figure 2 show the performance of COCTS standard algorithms in estimating Chla and SPM concentrations from $R_{rs}$ radiometric data. Out of the 90 potential matchups between L2B COCTS images and SOMLIT data, only 75 of the corresponding L2A files ($R_{rs}$ data) were available on the NSOAS server, reducing the number of matchups. We found 50 matchups for Chla, corresponding to in situ concentrations varying between 0.09 and 8.12 mg·m⁻³ (Figure 2a). The scatterplot shows several outliers, mostly located above the unity line, leading to a Bias (SSPB) of 53.77%, an Error (MSA) of 90.15%, and an RMSLE of 0.44. Remotely estimated and in situ values are overall well aligned, with a Slope of 1.03. On the other hand, Figure 2b shows a more dispersed distribution of 48 matchup points for SPM, with in situ concentrations varying between 0.18 g·m⁻³ and 10.52 g·m⁻³. The standard algorithm significantly overestimated SPM, leading to an Error of 120.94%, an RMSLE of 1.38, and a Slope of 3.26. The model showed a saturation value of 5011.87 g·m⁻³, which was reached in three matchups without raising any of the existing flags in the L2 product. While we observe clear differences between the performances of Chla and SPM standard models in this figure, the matchups with those variables correspond to measurements conducted at the same time and location. Thus, both models used the same radiometric data as input, and it suggests that the issues related with the SPM product might not be due to the $R_{rs}$ data. The other overestimated SPM values also presented no flags.

3.2. Chla and SPM Maps Assessment

Figure 3 shows an example of a HY-1C/COCTS image on 27 September 2018, taken above Northwestern Europe. We selected this image as it featured small cloud cover and a great diversity of SPM concentrations and gradients. Both Chla and SPM fields exhibited low concentrations offshore in the Bay of Biscay and the Atlantic Ocean and higher values in the English Channel, the North Sea, and river estuaries. Chla values were reasonable across the image, with concentrations of about 0.3 mg·m⁻³ in the Atlantic Ocean and 2.5 mg·m⁻³ in the English Channel. Chla reached about 30 mg·m⁻³ in the Thames River plume, where such concentrations are expected [56]. On the other hand, the SPM product exhibited extreme values in coastal and estuarine waters. As an example, concentrations of 5011.87 g·m⁻³ were found in the plumes of rivers like Seine or Thames, as well as along the coasts of Belgium and the Netherlands. However, SPM can only reach about 500 g·m⁻³ in the Thames estuary [57] and 200 g·m⁻³ for the Seine River [58]. Although some of these pixels were flagged as “turbid waters”, not all were. The same pixels showed realistic Chla values, suggesting that the issue likely originates from the SPM model. The SPM gradients near those pixels were very strong, increasing from values of 3 g·m⁻³ to 5011.87 g·m⁻³ within less than 6 km. Moreover, pixels with SPM values of 0.22 g·m⁻³ were found between land and saturated pixels, as shown in Figure 3b in the Thames estuary, and also flagged as “turbid waters” and “spare”. They corresponded to saturated Chla values at 64.0 mg·m⁻³. Finally, we also observed a blue band on the left of the SPM image (Figure 3b), along the 10°W meridian, corresponding to a constant SPM value of 0.22 g·m⁻³, which was not visible in the Chla image. However, they were flagged as pixels with a satellite viewing angle greater than 60°, and matchups with those pixels should not be considered. The above-mentioned issues were observed in many other COCTS images.

4. Discussion

4.1. A Stricter Matchup Exercise

To investigate the causes of SPM overestimation, we removed the matchups presenting negative $R_{rs}$ values in the visible part of the COCTS spectra (Figure 4). The number of matchups decreased to 24 for Chla and 23 for SPM, but the overall quality improved. They cover relatively clear waters: Chla varied from 0.13 to 4.31 mg·m⁻³ and SPM from 0.18 to 10.52 g·m⁻³, with respective median values of 0.75 mg·m⁻³ and 0.8 g·m⁻³. This filtering step eliminated strong outliers for Chla, now retrieved with a Bias of 41.11%, an Error of 50.46%, an RMSLE of 0.39, and a Slope of 0.93 (Figure 4a). Concerning SPM, this filtering significantly altered the matchup results, removing most of the outliers. The retrieval accuracy increased, with a new Bias value of −14% and an Error of 45.86%. However, one outlier with a value of 5011.87 g·m⁻³ was retained, inducing a high Slope value of 2.29 and an RMSLE of 0.84. Even though the associated $R_{rs}$ data did not present negative values in this case, the matchup is located in a part of the image where $R_{rs}$ values suddenly decrease from one or two orders of magnitude for all bands, compared with surrounding pixels. Excluding this outlier yielded improved statistics; the statistical metrics changed to −22.7% for the Bias, 45.5% for the Error, 0.29 for the RMSLE, and 0.96 for the Slope. One of the limitations of this study lies in the relatively small number of matchups, which may limit the representativeness of our matchup dataset. However, they provide valuable regional information on the accuracy of SPM and Chla COCTS products over specific ranges of concentrations. Moreover, the accuracy and Bias metrics used here are based on the median, which mitigates the impacts of potential outliers within the matchups [51,53].

4.2. Inter-Comparison with Another SPM Model

We applied the Han16 model to estimate SPM from the $R_{rs}$ corresponding to the 23 matchups with SOMLIT to compare with the standard COCTS product. Figure 5 shows that the outlier present for the COCTS product is not present in the scatterplot for Han16, further suggesting that the issue originates from the COCTS SPM model for this outlier and not the $R_{rs}$ product. For those matchups, Han16 retrieved SPM with a negligible Bias, an Error of 181.39%, an RMSLE of 0.46, and a Slope of 1.53. Its contrasted performance might result from the fact that the model is based on $R_{rs}$(670), whose signal is weak in relatively clear waters. Ref. [45] showed that Han16 tends to output high estimation errors when the SPM concentrations are about 1 g·m⁻³ or lower. When considering only the matchups with in situ SPM values below 1 g·m⁻³, the Bias and Error for the COCTS product were −9.79% and 31.39%, and the ones for Han16 are 0.09% and 215.96%. However, when considering matchups with SPM values greater than 1 g·m⁻³, the Bias and Error of the COCTS product are −91.1% and 91.1%, and the ones for Han16 are 0.54% and 73.82%.

To extend the comparison, we applied Han16 to estimate the SPM for the COCTS scene shown in Figure 3. Figure 6 presents a subset centered on the Thames estuary, showing the COCTS SPM (Figure 6a), Han16-derived SPM from COCTS (Figure 6b), and a concentration profile along a transect from the Thames estuary to the Belgian coast. A MODIS/AQUA image of the same day was selected as a comparison, as it is a widely used and well-validated sensor [17,49,59]. SPM derived from MODIS (Figure 6c) was also estimated with Han16, with coefficients computed for this sensor [39]. A comparison of the three maps and transect data revealed that the COCTS product systematically presented higher values than the SPM estimated with Han16 from COCTS or MODIS. As described previously, COCTS SPM values exceeded 1000 g·m⁻³ in most of the Thames plume waters and along the Belgium coastline, reaching up to 5011.87 g·m⁻³. In the same part of the section, Han16-derived SPM from both COCTS and MODIS varied between 3 and 20 g·m⁻³, showing great differences between the three products. The COCTS Han16 and MODIS Han16 transects also align well (Figure 6e), with minor differences likely attributable to the fact that the COCTS image was taken at 10h30 and the MODIS one at 12h40. This is consistent with the highly dynamic nature of coastal waters, where strong currents, tides, and fronts can be observed, and where SPM can thus present a high temporal variability [60]. Finally, the values of $R_{rs}\left(670\right)$ for COCTS and $R_{rs}\left(667\right)$ for MODIS along this same transect were in good agreement, suggesting that the COCTS SPM overestimation originates from the SPM algorithm used and not from errors in the atmospheric correction procedure.

5. Conclusions

In this study, we evaluated the quality of the standard COCTS Chla and SPM products over French coastal waters for the first time. While the Chla product showed robustness, the SPM product must be interpreted with caution. The 23 and 24 high-confidence matchups for SPM and Chla, respectively, only covered moderately turbid waters, with concentrations varying from 0.13 to 4.31 mg·m⁻³ for Chla and from 0.18 to 10.52 g·m⁻³ for SPM and respective median values of 0.75 mg·m⁻³ and 0.80 g·m⁻³. Over this dynamic range, the errors associated with Chla and SPM retrievals reached 50.46% and 45.86%, respectively. However, several issues related to the COCTS SPM product were pointed out: the matchup exercise and a comparison with the SPM model from Han et al. [39] showed that the COCTS SPM model seems to have difficulties over a wide range of concentrations and tends to saturate. This saturation leads to unrealistically strong SPM gradients in the images and limits the potential use of this product in river estuaries or turbid areas. This study represents an initial regional assessment of COCTS-derived standard products. One limitation of this study is the relatively small number of matchups, all confined to a single coastal region, which may constrain the overall representativeness of the dataset. Nevertheless, these matchups offer valuable regional insights into the accuracy of the COCTS-derived SPM and Chla products, particularly within those specific concentration ranges. These findings have important ecological implications. The evaluation results of the COCTS Chla product suggest it can serve as a useful tool for monitoring phytoplankton dynamics in oligotrophic to mesotrophic coastal waters. Such environments are particularly sensitive to nutrient loading and climatic variability, making robust spaceborne Chla estimations valuable for monitoring ecological status and for the detection of algal blooms and water quality changes [61]. On the other hand, the poor performance of the COCTS SPM product in similar conditions limits its ecological utility, particularly for applications involving turbidity monitoring, sediment transport, light availability assessments, and implications for aquaculture [11,62]. SPM plays a critical role in modulating light penetration, influencing benthic habitat suitability, and transporting nutrients and contaminants [9,63]. While the COCTS Chla standard product shows promise for estimating biological productivity, additional algorithm improvements are needed to fully support an integrated ecological monitoring in optically complex coastal waters. Finally, more evaluation exercises are necessary to validate both COCTS biogeochemical and $R_{rs}$ products over a greater diversity of optical water types. Future work should focus on expanding the number of matchups in different regions of the world, but also in different concentration ranges, to validate these products under specific conditions, such as in turbid or eutrophic waters.