Satellite-derived ocean color products are frequently affected by substantial data gaps due to cloud contamination, sun glint, and sensor-related limitations, posing a major challenge for long-term environmental monitoring. To address this, most existing gap-filling efforts have focused on reconstructing inverted variables including chlorophyll-a concentration (CHL) and sea surface temperature. However, this practice is limited by the inflexibility of reconstructing each variable separately. Since CHL and other bio-optical variables are derived from the remote sensing reflectance (R-rs) spectrum, reconstructing R-rs offers a more general solution. In this study, we propose a spectrally-consistent reconstruction of R-rs using the Discrete Cosine Transform with Penalized Least Squares (DCT-PLS) method, applied to MODIS-Aqua daily data over the South China Sea (SCS) for the period of 2010-2020. The reconstructed R-rs dataset shows high agreement with in-situ measurements (R-2 > 0.70 for most bands) and expands the number of valid satellite-in-situ matchups by more than four times, especially in persistently cloudy offshore areas. Compared with direct CHL reconstruction (RMSD = 0.63, MAPD = 49.1%), CHL retrieved from reconstructed Rrs shows improved accuracy (RMSD = 0.50, MAPD = 36.4%) based on validation with more than 400 in-situ samples. The reconstructed R-rs also enhances long-term trend detectability and signal-to-noise ratios in data-sparse regions. Furthermore, derived bio-optical parameters (e.g. phytoplankton absorption, detrital and gelbstoff absorption, and diffuse attenuation coefficients) exhibit consistent trends, reinforcing the reliability of the reconstructed R-rs. Our results demonstrate that reconstructing gap-free R-rs datasets not only address data loss but also support the generation of consistent downstream products, enabling reliable long-term monitoring in data-sparse marine environments.

The decreasing water transparency of the Baltic Sea over the last century has been documented by in situ measurements of the Secchi disk depth ( z SD ) as well as by satellite measurements over recent decades. While the decrease in transparency for the Baltic Sea in general has become negligible from about 2013 onwards, satellite-detected light attenuation of the violet and blue light ( k d 412) has increased, particularly from 2013 onwards. In the Gulf of Bothnia (particularly in the Bothnian Bay), eastern Gulf of Finland and eastern Gulf of Riga, k d 412 has increased significantly, indicating a dramatic increase in the concentration of colored dissolved organic matter (CDOM). These findings are partly supported by in situ CDOM data. The increase in CDOM concentration is probably due to climate change induced shifts in hydrology and greening of the land, all related to global warming. The increase in CDOM concentration and its effect on light availability may cause significant ecological changes and deterioration of the aquatic food chains.

Previous studies have consistently reported underestimation of satellite-derived chlorophyll-a concentration (CHL) in the Southern Ocean (SO). While this bias is often attributed to the region's unique bio-optical properties, the potential contribution of atmospheric correction errors has been comparatively underexplored. This study evaluated the performance of three mainstream CHL retrieval algorithms (OC2, OC3, OCI) in high-latitude waters of the SO, mainly poleward of 63 degrees S, using three types of remote-sensing reflectance (Rrs) data: in-situ Rrs, NASA-distributed standard VIIRS Rrs, and VIIRS Rrsobtained from a recently developed Cross-Satellite Atmospheric Correction (CSAC) algorithm. Our results show that when in-situ Rrswere used as inputs, the mean absolute percent difference (MAPD) values between measured and estimated CHL by OC2, OC3, and OCI were 24.4%, 23.0%, and 30.2%, respectively. Using NASA-distributed VIIRS Rrs increased the MAPD values to 41.7%, 50.4%, and 55.4%, respectively. In contrast, when CSAC-derived Rrs were used, the MAPD values reduced to 25.4%, 22.5%, and 21.4%, respectively. This is because CSAC-derived Rrs showed much stronger agreement with in-situ Rrs than NASA-distributed Rrs, with MAPD values reduced by approximately 50% across the visible bands. These results suggest that imperfect atmospheric correction plays a big role in the systematic underestimation of CHL in the SO from ocean color remote sensing. Additionally, CSAC was able to recover CHL retrievals under challenging observation conditions, such as moderate sunglint and straylight, which were previously masked in the standard VIIRS product, thereby substantially improving data coverage. Therefore, we anticipate that incorporating CSAC into the production of ocean color products will yield more accurate, better spatially covered products in this critical region.

Lidar is a key technique for 3D ocean observation, and improving its penetration depth has long been a central objective in system design and optimization. However, multiple scattering induces lateral photon redistribution, causing time-of-flight–based penetration depth estimates to be overestimated due to misinterpretation of delayed photons as deeper signals. Therefore, the actual contribution of improvements in key lidar system parameters—including laser pulse energy, receiver aperture (D), receiver field of view (FOV), and transmitted wavelength—to penetration depth enhancement should be critically reassessed. Here, a semi-analytical Monte Carlo model is developed to simulate lidar backscattering signals in typical Case-1 waters, assuming a vertically homogeneous water column and neglecting surface wave effects. Simulations are performed across a broad range of chlorophyll-a concentrations (Chl-a) and for multiple observational configurations, including spaceborne, airborne, shipborne, and underwater platforms. Photon step lengths and corresponding physical depths are simultaneously tracked to quantify effective penetration depth. The results demonstrate that system lidar parameter optimization markedly improves effective penetration depth under weak multiple-scattering conditions, whereas its impact becomes marginal when multiple scattering is strong. Notably, in coastal waters, the conventionally assumed benefits of enlarging the FOV or shifting the transmitted wavelength toward the green band are generally ineffective in enhancing effective penetration depth. Sensitivity analyses using different scattering phase functions yield consistent conclusions. This study provides a quantitative reassessment of lidar system parameter optimization under multiple-scattering conditions, refines the understanding of penetration-depth enhancement in optically complex waters, and offers theoretical guidance for the design and performance evaluation of oceanic lidar systems.

Remote sensing reflectance (R-rs) is a fundamental property in satellite ocean color remote sensing, which is critical for retrieving optical-biogeochemical properties and data-driven atmospheric correction algorithms. In this study, with three criteria applicable to similar to 91% of the global ocean, we compiled a database of the highest quality R-rs (HQ(MODISA)-R-rs) of oceanic waters based on 20+ years of ocean color measurements by the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Aqua satellite. While removing a large number of daily "standard" data products, our evaluation showed that the criteria for the highest-quality R-rs (CHQR) improved MODIS R-rs data consistency with benchmark in situ R-rs datasets, such as those from MOBY and AERONET-OC. After applying CHQR, analysis of imagery products in the South Pacific Ocean revealed that the coefficient of variation (CV) of R-rs among pixels reduced from 0.042 (standard quality control) to 0.030, along with enhanced temporal consistency, which indicates that this approach effectively filters abnormal data products. While such a dataset played a key role in the development of the cross-satellite atmospheric correction algorithm (Lee et al., 2024), we here further demonstrate that applications of HQ(MODISA)-R-rs have similar to 21.0% of oceanic areas between 50 degrees S and 50 degrees N showing reversed long-term trends of R-rs compared to the trend based on the standard R-rs product. We anticipate that this highest-quality R-rs database would not only improve our evaluation and understanding of long-term changes in various R-rs-derivative bio-optical properties of the global ocean, but also help to obtain consistent products among various satellite ocean color missions.

Remote sensing reflectance ( R-rs ) derived from satellite ocean color measurements often suffers from degraded quality and limited coverage under challenging observation conditions, such as strong sunglint (SG) and high solar zenith angle (SOZA). To address this challenge, using Visible Infrared Imaging Radiometer Suite-Suomi National Polar-orbiting Partnership (VIIRS-SNPP) measurements as an example, we developed an algorithm following the cross-satellite-atmospheric-correction concept (named CSACVIIRS) that enables the generation of high-quality (HQ) R-rs from VIIRS-SNPP measurements under such conditions. Cross-satellite atmospheric correction (CSAC) is a data-driven system that converts top-of-atmosphere (TOA) measurements from a sensor to Moderate Resolution Imaging Spectroradiometer (MODIS)-equivalent R-rs . Evaluation results show that CSACVIIRS markedly improves the consistency between VIIRS R-rs and the highest-quality MODIS-Aqua (HQMA) Rrs for the visible bands, where the mean absolute percentage difference (MAPD) is reduced by 40%-60% for measurements under strong SG and high SOZA conditions. In addition, CSACVIIRS enables the recovery of valid R-rs in areas previously masked out in the standard VIIRS products, thereby significantly (up to similar to 50%) increasing the volume of usable data. These findings further highlight that CSAC is a promising approach to achieve high quality and consistent ocean color products across different satellite missions, which is important for the generation of long-term time series of the global ocean from multiple satellites

Remote-sensing reflectance (Rrs) is a fundamental parameter in ocean-color remote sensing. A new active and direct approach for measuring Rrs using white-light lidar spanning 400 to 720 nm is proposed, and its feasibility is investigated through radiative-transfer simulations under the assumption of optically homogeneous water bodies. The results demonstrate that the fraction of multiple scattering in the lidar signal is the dominant factor controlling the measurement accuracy of Rrs. Under calm sea-surface conditions and Chl between 0.01 and 5.00 mg/m3, simulations show that the mean absolute percentage error (MAPE) of Rrs measurements stays below 10% when the receiving footprint is large enough. This requirement is readily met for an airborne platform at a detection altitude (H) of 3,000 m with a receiver field of view (FOV) greater than 20 mrad, and even for a spaceborne platform at H = 500 km with FOV = 0.5 mrad. By contrast, low-altitude platforms suffer from markedly increased errors due to insufficient collection of multiple-scattered photons. Further analysis indicates that water inherent optical properties (IOPs) influence measurement performance by modulating photon step size and the number of scattering events, resulting in necessitating larger receiving footprints in clearer waters and in blue-green bands. Among the representative scattering phase functions (SPFs) examined, SPF effects are relatively minor. A preprocessing strategy that truncates near-surface signals is further proposed to suppress sea-surface-reflection interference. These findings provide guidance for optimizing white-light lidar systems for active Rrs measurement and for supporting ocean-color satellite calibration.

Accurate retrieval of optically active components (OACs), including chlorophyll-a (Chl-a) concentration, suspended particulate matter (SPM) concentration, and the absorption coefficient of colored dissolved organic matter (CDOM) [ag(443)], from satellite observations remains challenging in aquatic remote sensing. Although deep learning has shown promise for this goal, its black-box nature produces physically inexplicable outputs and makes it vulnerable to the ill-posed inversion problem arising from overlapping spectral information among the OACs, which is further compounded by strong variability in their specific absorption and scattering properties. To address these limitations, this study proposes the Ocean Color Radiative Transfer Network (OCRT-Net), which embeds an analytical model of ocean color in the network architecture as a fully differentiable layer. This structural design constrains the internal latent space to reconstruct physically consistent inherent optical properties (IOPs) at every forward pass, guaranteeing optical closure by architecture rather than by post-hoc regularization. The model was optimized using a two-stage transfer learning strategy: it was pre-trained on HydroLight-simulated spectra (N = 12,000) to establish optical closure, followed by fine-tuning on the global GLORIA in situ dataset (N = 2049) to capture real-world optical variability. Independent validation across Sentinel-3 OLCI, Sentinel-2 MSI, and SNPP VIIRS configurations demonstrated robust retrieval performance for OACs spanning several orders of magnitude (uncertainty: 35%–55%), outperforming several established empirical and semi-analytical benchmark algorithms. Application to satellite matchups and full-scene OLCI imagery over ten optically diverse lakes in China confirmed that OCRT-Net produces physically consistent retrievals and spatially coherent OAC distributions, free of the noise-induced artifacts and physically implausible values commonly associated with purely data-driven approaches. By bridging radiative transfer theory with data-driven optimization, OCRT-Net moves beyond the black-box paradigm, offering an interpretable and sensor-transferable framework for operational water quality monitoring across inland and coastal waters.

Accurate and well-characterized regional, basin and global scale measurements of oceanic Net Primary Production (NPP) are critical for understanding the role and response of ocean ecosystems to rising atmospheric CO2 levels and global warming. Currently global NPP estimates from satellite observations, which underpin global ocean carbon cycling and environment studies, continue to suffer from substantial uncertainties due to several methodological and observational limitations. These include: (1) the reliance on satellite-derived phytoplankton biomass fields generated using algorithms that do not consistently achieve high accuracy across all regional and global scales; (2) limited measurements of phytoplankton photosynthetic quantum yields (ϕ), which are currently obtained primarily from research vessel observations; and (3) the lack of adequate methods for scaling local in-situ ϕ measurements to regional and basin-wide scales. To address these challenges, we have utilized the Absorption-based Productivity Model (AbPM), which leverages the inherent optical absorption properties of phytoplankton derived from remotely sensed reflectance, rather than relying on phytoplankton biomass as an input. Additionally, we apply a novel bio-optical classification framework, the Bio-Optical Measurement and Evaluation System (BIOMES) to scale sparse in-situ estimates of ϕ for deriving global maps of NPP. Finally, using a global collection of in-situ NPP datasets we assess the performance of AbPM and those more widely used biomass-based models.

The subtlety of oceanic color shifts, compounded by noise in satellite records, has made it uncertain whether the open ocean is undergoing a change akin to terrestrial greening. Using optical indices derived from a new, stringently screened MODIS-Aqua remote sensing reflectance dataset, we show that the ocean (60° S–60° N) has become markedly greener over the past two decades, with 73% of latitudinal bands exhibiting a shift toward greener hues and 24% showing statistically significant trends. By separating optical signals from different constituents, we attribute this greening primarily to increases in phytoplankton pigments, accompanied by smaller increases in other optically active components. We suggest the observed greening results from a complex interplay of warming-driven competitive advantages of picophytoplankton, mixed layer dynamics and atmospheric dust deposition. Importantly, we find that previously perceived chlorophyll-a concentration (Chl) decline in mid-low latitudes (40° S–40° N) are largely driven by radiance data of lower-certainty; removing these data reverses the trend. Moreover, the water corresponding to these data is not warming, further challenging the conventional causal mechanism that links warming to reduced Chl. Collectively, these results reveal that Earth’s largest biome is undergoing a subtle yet detectable greening in response to climate change.

Monitoring dissolved organic carbon (DOC) concentrations in coastal waters is critical to elucidating carbon dynamics, thereby facilitating the understanding carbon dynamics and quantifying lateral carbon fluxes to the open ocean. Machine learning combined with ocean color remote sensing provides an effective approach to estimate optically inactive substances such as DOC in complex coastal environments. In this study, we compiled a comprehensive training dataset in Google Earth Engine by paring Sentinel-2 radiometric measurements with in situ DOC measurements from coastal waters of Southeast China. Based on this dataset, a machine learning-based model, termed AutoGluon-DOC, was developed to retrieve DOC from Rayleigh-corrected top-of-atmosphere reflectance (ρrc(λ)) and auxiliary information (e.g., acquisition time). The model achieved a median absolute percentage error of ∼10 % on the validation dataset. Application of AutoGluon-DOC to Sentinel-2 images over Dongshan Bay, a mangrove-estuary-aquaculture composite system, revealed a clear spatial gradient of decreasing DOC from the bay head toward the open sea. DOC spatial variability was primarily regulated by tidal fluctuations and Chl in the bay head, while aquaculture activities dominated in the bay mouth. Satellite-derived DOC series also exhibited a distinct seasonal cycle, with higher concentrations in summer and lower in winter, mainly associated with biological production and episodically enhanced by river discharge during extreme rainfall events. These results underscore the effectiveness of AutoGluon-DOC for high-resolution monitoring of DOC dynamics in complex coastal environments, with a framework adaptable to diverse regions as training datasets expand.

This study presents a new satellite ocean color data record of Secchi depth (Z(SD)) observations from the Visible Infrared Imaging Radiometer Suite (VIIRS). As part of the NOAA enterprise satellite data processing, the decade-long Z(SD) data are derived from the visible and near-infrared reflectance measurements over oceanic, coastal, and inland waters. Based on in situ data, the model is excellent in generating low-uncertainty Z(SD) data with an absolute percentage difference (APD) of 15%-29%. The satellite and in situ matchups confirm reliable satellite retrievals with APD = 19%-26% over the Z(SD) range of 0.1-60 m. Although the product uncertainties are dependent on optical water types, assessments show that the satellite Z(SD) estimations are very reliable, especially where Z(SD) >= 1 m. This new satellite product has enabled the ability to access Level-2 daily Z(SD) imagery and information as well as Level-3 data aggregated on daily to monthly scales. Our examination indicates that the ocean transparent windows are situated at 443 and 486 nm in the vast open oceans and 551 nm for most coastal waters. They shift to the red band at 671 nm for extremely turbid environments, such as large river estuaries. From a global perspective, the Z(SD) data extend from less than half a meter in nearshore environments to >70 m in the South Pacific Gyre, while demonstrating a strong dependency on the optical water types. Short-term fluctuations over time are registered in the satellite Z(SD) daily and monthly data from almost every aquatic environment. Trend analyses reveal significant increases in water transparency over many regions, especially the open ocean. We stress the necessity of normalizing satellite Z(SD) estimations to eliminate the uncertainties induced by different solar-zenith angles. The present satellite products can be further improved by accounting for the limitations imposed by the multispectral reflectance data with hyperspectral ocean color spectra.

Upwelling in the Equatorial Pacific nurtures an expansive, westward-stretching chlorophyll-rich tongue (CRT), supporting 18% of the annual global new production. Surrounding the CRT are the oligotrophic subtropical gyres to the north and south, which are suggested to be expanding under global warming. Yet, how this productive CRT has changed, expanding or contracting, remains unknown. By applying the empirical mode decomposition (EMD) method to 20-year monthly measurements of chlorophyll-a concentration from MODIS-Aqua satellite (2002-2022), we demonstrate that the CRT exhibited a significant westward extension, at an average expanding rate of 1.87 ( ± 0.82) × 105 km2/yr. The westward extension of the CRT is attributed to strengthened equatorial upwelling and a strengthened South Equatorial Current from 2002 to 2022, driven by intensified easterly trade winds as the Pacific Decadal Oscillation predominantly remains in its negative phase during this period. Interestingly, EMD analysis on central locations of the Pacific gyres suggested simultaneous extension of the gyres and the CRT during 2002-2022, with the gyres extending poleward. Our findings imply a broader cover of productive water along the equator, while its impact on tropical climate, ecosystems, and carbon cycle deserves further investigation. Based on MODIS chlorophyll data of 2002-2022, this study reveals that the chlorophyll-rich tongue in the equatorial Pacific is extending westward, likely a result of internal variability related to the Pacific Decadal Oscillation. The findings imply a broader cover of productive water along the equator, which may greatly impact the tropical climate and ecosystems.

Based on a relatively large dataset having concurrent measurements of remote sensing reflectance (Rrs) and absorption coefficients collected in the marginal seas of China (MSC) over the past two decades (2003-2021), we evaluated two widely used semi-analytical algorithms (SAAs) for retrieving the absorption properties in the MSC from both field-measured and satellite data. The SAAs are the quasi-analytical algorithm version 6 (QAA_v6) and the generalized inherent optical properties model (GIOP), while the satellite data are from the moderate resolution imaging spectroradiometer on the aqua satellite. The water body was classified following a proposed system, also separated via the trophic level (oligotrophic, mesotrophic, and eutrophic) based on chlorophyll-a concentration, so the performance of these two SAAs was evaluated for the different water types. For the water types we evaluated, both QAA_v6 and GIOP are found to have different applicable water types in retrieving absorption properties, and the relatively reliable retrieval results of absorption properties are mainly in the water types 3-13, and at the wavebands shorter than 500 nm. In addition, the retrieval performance of these two SAAs for adg (the sum of colored dissolved organic matter and non-pigmented particulate matter) and aph (phytoplankton pigment) in the eutrophic waters in the MSC still needs to be improved. Before 500 nm, MODIS-Aqua data can provide reliable anw (non-water absorption), adg, and aph with QAA_v6; and can provide reliable anw and aph with GIOP. This study provides what we believe to be a new and more detailed perspective for evaluating the retrieval of absorption properties using these two SAAs in the MSC, and our results suggest that water types should be considered in improving the estimation of intermediate variables in these SAAs.

Carbon dioxide (CO2) is the most important greenhouse gas in the atmosphere, playing a crucial role in the greenhouse effect and climate change. Lidar, with its high spatiotemporal resolution and high-precision detection capabilities, has become an essential tool for remote sensing of CO2. However, precise temperature information is required for CO(2)retrieval. Studies showed that for both differential absorption lidar (DIAL) and spectroscopic lidar, a CO(2)concentration measurement error of 2.0-3.5 ppm would result from each 1 K temperature deviation. Therefore, using nonreal-time and non in situ temperature data can lead to significant CO(2)retrieval errors. In this study, a column-averaged CO(2 )spectroscopy lidar is proposed, which enables simultaneous measurements of CO(2 )concentration, temperature, and semi-heavy water (HDO, isotopic water vapor). First, a model combining five Lorentzian functions with a binomial background was proposed through spectral decomposition. Second, through theoretical analysis, the fitting parameters were reduced from 18 to 5. Finally, theoretical analysis shows that the model achieves system biases of less than 0.1 ppm for CO2, 0.1 K for temperature, and 0.06 ppm for HDO. Considering Poisson noise, the error distributions of CO2, temperature, and HDO under different optical distances and signal-to-noise ratios (SNRs) were studied. This technology will advance the development of CO(2 )flux remote sensing and is expected to play a crucial role in ecosystem research, atmospheric environmental monitoring, and greenhouse gas emission reduction policies.

Bathymetric lidar, with its deep penetration, continuous day-and-night operation, and high accuracy, is an important tool for remotely sensing bottom depths. However, the strong forward scattering of the laser beam during transmission in water introduces substantial multiple scattering components into the lidar signal reflected by sea bottom, leading to a peak shift and signal broadening. The peak shift leads to an overestimation of the bottom depth, while the signal broadening makes peak extraction more challenging. To quantitatively study this impact, a semianalytic Monte Carlo (MC) simulation is applied to model seabed reflected signals. By statistically analyzing the peak position bias (termed as Bias) and full-width at half-maximum (termed as FWHM) of the seabed lidar reflected signals across four platforms-spaceborne, airborne, shipborne, and underwater-empirical models are established to relate Bias and FWHM to scattering efficient (b), bottom depth ( $z_{\mathbf {m}}$ ), and lidar receiver footprint ( $r_{\mathbf {s}}$ ). Here, $r_{\mathbf {s}}$ represents the radius of the footprint of the lidar receiver on the water surface. Furthermore, the effects of different scattering phase functions and the absorption coefficient are analyzed. This study shows that the Bias and FWHM are influenced by $b, z_{\mathbf {m}}$ , and $r_{\mathbf {s}}$ . For lidar systems with an $r_{\mathbf {s}}$ of dozens of meters, measuring deeper depths in water with higher b can result in a bottom depth overestimation of nearly 4% and an FWHM broadening exceeding 28 ns solely due to multiple scattering effects. This article provides a theoretical basis for correcting and evaluating bathymetric lidar data, thereby improving the accuracy and applicability of bathymetric lidar results.

Solar radiation in the ultraviolet (UV) bands plays an important role in marine biogeochemical processes, and at the same time, measurements of a satellite sensor in the UV help the data processing of ocean color satellites. However, historically, satellite ocean color missions lack UV measurements; only in recent years have there been satellite sensors, such as PACE OCI, to provide a direct measurement of radiance in the near-blue UV (nbUV) domain. To address the limitation of earlier measurements, a deep-learning-based system (termed UVISRdl) has been previously introduced to estimate remote-sensing reflectance (Rrs) of the nbUV bands at 360, 380, and 400 nm from Rrs(visible). In this study, as PACE OCI offers global-ocean hyperspectral Rrs products from UV to visible bands, we leveraged this opportunity to comprehensively evaluate the performance of this UVISRdl system and compare the Rrs(nbUV) among VIIRS, OCI, and SGLI. It is found that the Rrs(nbUV) values from VIIRS and OCI exhibit high consistency, with mean absolute unbiased relative difference (MAURD) ranging from ~0.23-0.30 at 360 nm, ~0.21-0.22 at 380 nm, and ~0.17-0.20 at 400 nm, while the SGLI shows lower consistency compared to the former two (MAURD = ∼0.47 at 380 nm). More importantly, the consistency assessment metrics in Rrs(nbUV) between VIIRS and OCI are nearly the same, regardless of whether the OCI Rrs(nbUV) were derived from UVISRdl or measured directly. These findings demonstrate UVISRdl's potential for extending global-scale UV reflectance back into periods lacking direct UV observations, enabling the generation of long-term remote-sensing products, and deepening our understanding of the interactions between UV radiation and biogeochemical processes in the global ocean.

Phytoplankton functional types (PFTs) found in natural aquatic environments play different roles in the biogeochemical cycles of different elements. However, commonly used methods for identifying PFTs have inherent limitations. In this study, based on a large dataset (1747 samples) collected from 2004 to 2019 in the South China Sea and adjacent Taiwan Strait, which had concurrent measurements of the spectral absorption coefficient of phytoplankton and chlorophyll a concentration of nine PFTs (PFTs Chla ), along with depth and time information, a reliable support vector regression (SVR) model was developed to retrieve these nine PFTs Chla in the water column. These PFTs included diatoms, dinoflagellates, haptophytes_8, haptophytes_6, chlorophytes, cryptophytes, Prochlorococcus , Synechococcus , and prasinophytes. The independent validation results indicated that the SVR model outperformed the traditional PFTs Chla retrieval algorithms, with an average mean bias of −14.2%, an average mean absolute unbiased relative difference of 60.3%, and an average coefficient of determination of 0.56. The predicted PFTs Chla values and their error distributions in the water column were subsequently analyzed. Finally, the SVR model was found to be applicable to most PFTs Chla retrieval in the East China Sea.