Phytoplankton absorption spectra, aph(λ), contain critical information on phytoplankton biomass, pigment composition, and community structure in aquatic ecosystems. Although hyperspectral observations can resolve these spectral features, multispectral sensors remain the main source for long term and operational lake monitoring. However, most satellite based retrievals from multispectral imagery provide aph(λ) only at a limited number of discrete bands, which restricts their use in continuous spectral analysis. To address this limitation, we developed an interpretable framework that combines Gaussian spectral decomposition and machine learning to reconstruct continuous phytoplankton absorption spectra from Sentinel-3 Ocean and Land Colour Instrument (OLCI) data in optically complex lakes. The method first decomposes measured aph(λ) spectra as a sum of Gaussian components, then reduces the spectral reconstruction problem to the retrieval of three key Gaussian magnitudes from multispectral remote sensing reflectance (Rrs(λ)), and finally reconstructs continuous aph(λ) spectra across 400–700 nm. The model was calibrated and evaluated using coincident field-measured Rrs(λ) and aph(λ) data collected from 25 lakes across China, and was further assessed using OLCI match-up observations. Validation against independent field measurements showed mean bias of 0.07 m-1, Root Mean Square Error (RMSE) of 0.28 m-1, and Absolute Percentage Difference (APD) of 44.56% over 400–700 nm. Application to OLCI imagery generated a 2016–2024 continuous aph(λ) dataset for Lake Taihu, Lake Chaohu, and Lake Erhai, revealing pronounced spatial heterogeneity and higher monthly mean aph(λ) values in summer and autumn. The proposed framework extends the utility of multispectral satellite observations for characterizing phytoplankton optical properties in inland waters and provides a practical basis for long-term monitoring of lake phytoplankton dynamics.

Total phosphorus (TP), total nitrogen (TN), and their ratio (TN:TP) are fundamental indicators of trophic status and ecological balance in lakes. However, these constituents lack optical activity and are influenced by complex environmental and hydrological processes, limiting the accuracy and generalizability of their remote sensing estimation. To address this issue, environmental and morphometric variables were incorporated alongside optical information to provide stronger physical constraints, and deep learning was employed to model nonlinear relationships between multiple drivers and nutrient dynamics, enabling the simultaneous estimation of TP and TN in lakes worldwide. Specifically, global in-situ TP and TN measurements were first compiled and spatially and temporally matched with Landsat surface reflectance (SR), ERA5-Land environmental variables, and lake morphometric attributes, resulting in a large multi-source dataset (N=15,227) for model development. Considering the tabular structure and heterogeneous distributions of these variables, six deep learning models designed for tabular data (GANDALF, NODE, TabNet, AutoInt, DANets, FT-Transformer) were systematically trained and evaluated. Among them, FT-Transformer was identified as the optimal model, with the highest average R2 (0.74 for both TP and TN), highest regression slopes (0.77 for both TP and TN), and lowest errors (MAE = 1.55 for TP and 1.39 for TN; MAPE = 30.27% for TP and 23.54% for TN). This performance was comparable to tree-based baselines, while supporting multi-output learning and providing model interpretability. SHAP analysis revealed that environmental and morphometric variables contributed more than optical features, underscoring the essential role of multi-source information in nutrient retrieval. Independent validation further confirmed the model robustness and generalization on unseen data (regression slope = 0.84 for TP and 0.78 for TN). Building upon this reliability, this model was applied to multi-temporal Landsat SR images to derive spatial patterns and temporal dynamics of TP, TN and TN:TP, jointly indicating nutrient status and limitation regimes to support lake management. By integrating multi-source data with deep learning, this study provides a practical framework for the remote sensing of non-optically active constituents (NOACs), enabling accurate, interpretable, and generalizable estimations while preserving the spatial continuity and high-resolution advantages of satellite observations.

Variations in the mass ratio of nitrogen to phosphorus (TN/TP) are closely linked to algal bloom outbreaks in lakes. Traditional studies predominantly rely on laboratory or in situ observations, lacking long-term, whole-scene monitoring. This study, based on remote sensing and intelligent algorithms, optimized direct derivation, indirect derivation, and various machine learning methods to, for the first time through satellite-enabled panoramic observation, obtain the spatiotemporal distribution of TN/TP in 28 lakes of Yangtze-Huaihe region (YHR) over 21 years and analyze its effects on phytoplankton. The results show that the Extreme Gradient Boosting-based direct derivation method yielded the best validation performance, with an R² of 0.59 and RMSE of 11.34. Spatially and temporally, TN/TP predominantly ranged between 15-25, with a multi-year average of 20.49. 61% of the lakes exhibited declining trends in TN/TP ratios, with all small lakes (<100 km²) showing consistent decreases. Seasonally, TN/TP ratios demonstrated lower values in summer-autumn and higher values in winter-spring. Leveraging the spatiotemporal observational advantages of remote sensing, it was found that TN/TP at downstream outlets was 2.9% higher than upstream inflow areas in 71% of the lakes, with the most pronounced difference in autumn (6.6%). Based on reported nutrient limitation thresholds, 63% of the lake areas were co-limited by nitrogen and phosphorus (P), while 36% were P-limited, with P-limited areas being higher in winter and spring, reaching 49%. By screening continuous observation cases, we found that Floating Algae Index became less sensitive to TN/TP when TN/TP exceeded 15 in all seasons, suggesting TN/TP = 15 as a feasible threshold for eutrophication control. Remote sensing-derived TN/TP patterns reveal the need for enhanced small lake monitoring, with the proposed locally adapted P-limitation threshold offering science-based guidance for eutrophication control.

Harmful algal blooms (HABs) in eutrophic lakes threaten aquatic ecosystems and public health, but short-term chlorophyll-a (Chl-a) prediction remains challenging because satellite observations are often discontinuous and bloom dynamics are regulated by interacting environmental and ecological factors. This study developed a two-stage Extreme Gradient Boosting (XGBoost) framework for satellite-derived Chl-a reconstruction and short-term prediction in Lake Taihu. The first-stage model (XGBoost_Interp) reconstructed Chl-a on dates without valid GOCI/GOCI-II retrievals and generated a continuous daily Chl-a series for 2011–2024, achieving good performance (R2 = 0.83, RMSE = 5.72 μg/L, MAPE = 9.16%). Using the reconstructed series and environmental variables, the second-stage model (XGBoost_Pred) predicted subsequent Chl-a with comparable accuracy (R2 = 0.83, RMSE = 5.62 μg/L, MAPE = 8.97%). However, independent validation from January to March 2025 showed lower accuracy (R2 = 0.69, RMSE = 9.45 μg/L) and some overestimation, suggesting that further calibration and evaluation are still needed when valid satellite observations are unavailable for extended periods. SHapley Additive exPlanations (SHAP) identified wind speed, temperature, atmospheric pressure, day of year, and antecedent Chl-a as important predictors. After lake-specific recalibration, the framework also performed well in Lake Chaohu, indicating its practical potential for other large eutrophic lakes. Nevertheless, broader application requires further validation with nutrient-related variables, multi-sensor data, high-frequency field observations, and more lakes with different environmental conditions.

Diurnal dynamics of particulate organic carbon (POC) play a critical role in lake carbon cycling but remain poorly quantified at regional scales. Using hourly observations from the Geostationary Ocean Color Imager (GOCI) (08:16-15:16, UTC+8; 500 m spatial resolution), we examined POC variability across 52 lakes in the middle and lower reaches of the Yangtze and Huai River (MLYHR) basin in China (2011-2021). Four diurnal POC metrics, including the daily mean values of POC (POCmean), the coefficient of variation of POC (POCcv), relative difference between afternoon and morning POC (dPOC), and centroid of hourly POC (cPOC), showed substantial spatial and temporal variability. Both POCmean and POCcv peaked in summer and autumn, with maximum values in September (5.60 mg/L), and declined during winter and spring. Negative dPOC values (-6.48 +/- 15.39%) were found in 73.96% of pixels, indicating afternoon-dominated POC conditions, and cPOC was tightly clustered at 11.49 +/- 0.17 h (UTC+8). The Fuzzy C-Means clustering method identified three distinct diurnal POC patterns: Type 1 (decreasing), Type 2 (increasing), and Type 3 (midday peak). Interannual analyses suggested post-2014 trend shifts in diurnal POC metrics, with metric-dependent uncertainty. Driver analysis revealed that temperature and solar radiation primarily control diurnal POC dynamics, while wind influences POCcv by modulating the vertical movement and mixing of POC. Structural equation modeling highlighted distinct mechanisms underlying each diurnal pattern, particularly for dPOC and cPOC. These findings can provide a basis for optimizing lake carbon flux models by incorporating daytime diurnal POC dynamics, and highlight the necessity of incorporating diurnal-scale processes into environmental monitoring and policy frameworks.

Taihu Lake is one of the most severely eutrophic large lakes worldwide, characterized by a complex ecosystem where algal blooms and aquatic vegetation coexist, and it serves as a critical regional water source. Identifying the channels through which algal blooms invade vegetated areas and understanding their underlying mechanisms is essential for effective lake management. In this study, multi-source remote-sensing data were used to extract the spatial patterns of algal bloom frequency and aquatic vegetation over the past 22 years. We detected the major encroachment routes of algal blooms into vegetated zones and revealed the long-term interaction mechanisms between algae and vegetation. Algal blooms were persistently concentrated in the western and northern bays, exhibiting an overall fluctuating upward trend. Aquatic vegetation was mainly distributed in four eastern bays, remained relatively stable before 2015, declined markedly in 2016 due to dredging and shoreline modifications, and then gradually recovered. A stable negative response relationship was observed between algal blooms and aquatic vegetation, jointly regulated by seasonal climate and nutrient structure. High temperatures and intense precipitation in summer facilitated bloom expansion, whereas low temperatures in winter favored vegetation recovery. Meanwhile, lower TN to TP ratios suppressed vegetation growth and enhanced bloom encroachment during summer. Three primary channels of algal bloom encroachment into vegetated areas were identified: the northern shoreline of Gonghu Bay, the surroundings of Dongshan Island, and the southern marginal zone. Among them, the water intake in northern Gonghu Bay represents the most vulnerable zone requiring priority control. Although bloom risk is relatively low for the central and southern water intakes, maintaining the existing hydrodynamic regime and vegetation structure remains essential to prevent vegetation degradation, and enhanced monitoring is needed to identify in situ bloom occurrences within vegetated regions.

Eutrophication in shallow plain lakes is highly dynamic and often characterized by short-term trophic fluctuations that are difficult to resolve using annual or single-sensor satellite observations. The Eastern Plain Lake Zone (EPL) of China contains numerous shallow lakes and reservoirs embedded in densely populated and intensively cultivated lowland catchments, making it a representative region for high-frequency eutrophication monitoring. Here, we developed a monthly Trophic State Index (TSI) monitoring framework for lakes and reservoirs in the EPL using Harmonized Landsat–Sentinel (HLS) observations. The XGBoost model using combined spectral features achieved the best validation performance (R2 = 0.90, RMSE = 5.23), and provided the highest trophic-state classification accuracy, with an overall accuracy of 0.73 and a Kappa coefficient of 0.66. Lakes showed substantially higher trophic levels than reservoirs, with mean TSI values of 57.32 ± 6.89 and 45.00 ± 9.22, respectively. Temporally, significantly decreasing and increasing TSI trends accounted for about 15% and 5%, respectively. Natural lake TSI reflects integrated climatic, anthropogenic, and hydro-morphological influences, whereas reservoir TSI variability may be more closely associated with hydrological regulation and morphometric conditions. Our results show that HLS observations provide a robust basis for monthly eutrophication monitoring in optically complex shallow lake regions. The EPL-focused framework highlights the value of virtual satellite constellations for detecting short-term trophic deterioration and supporting region-specific lake and reservoir management.

Algal blooms are becoming more frequent and intense in lakes worldwide, but how bloom intensity and timing co-vary at the global scale is unclear. Here we analyze two decades of Moderate Resolution Imaging Spectroradiometer satellite observations for 4085 lakes ( > 20 square kilometres) to compare changes in intensity (fractional floating algal cover) and timing (start and end dates) of surface algal bloom. We find that intensity and timing often change independently: about 71% of lakes show increasing intensity, mainly associated with higher population density and agricultural pressure, whereas temperature and wind better explain shifts in bloom timing, especially in cold regions. Under a medium-emission scenario, tropical lakes show rapid intensification with modest timing shifts, while cold-region lakes exhibit regionally contrasting timing changes. This decoupling may alter lake food webs and carbon cycling, underscoring the need for region-specific management strategies under climate change.

Total nitrogen (TN) and total phosphorus (TP) serve as key indicators of aquatic eutrophication. While the monitoring stations are continuous in time, there are too few points in some lakes, and many lakes with low attention do not have monitoring stations, making it difficult to reflect the overall pattern of the lake area and multi-lake regions. This study integrated Moderate Resolution Imaging Spectroradiometer (MODIS) remote sensing data with machine learning techniques to estimate TN and TP concentrations across 26 lakes in the Yangtze-Huaihe Region (YHR) of China over a 20-year period (2003-2023). The Extreme Gradient Boosting (XGB) algorithm was identified as the top performer, achieving coefficient of determination (R2) values of 0.56 (RMSE = 1.14 mg/L) for TN and 0.58 (RMSE = 0.06 mg/L) for TP estimation. The long-term reconstruction revealed that TN levels consistently ranging from 1.2 to 2.0 mg/L, with 70% of the retrieved values exceeding the 1.5 mg/L threshold after 2015. TP concentrations showed a progressive increase, generally fluctuating between 0.10 and 0.16 mg/L levels throughout the study period. Notably, the lakes exhibited a higher trophic state index for phosphorus (mean TSITP = 74.30) was demonstrated compared to nitrogen (mean TSITN = 61.13), with spatial analysis identifying external riverine inputs as the primary phosphorus source. These results align with environmental parameters derived from other remote sensing observations, providing reliable data to support water management decisions in the YHR and offering a transferable framework for global eutrophication monitoring.

Hydrological processes drive the transport of phosphorus (P) from soil to surface water. This study is the first to use space-based observations to examine P storage in watershed lakes. Based on the vertical distribution characteristics of the total P (TP) concentration in multiple eutrophic lakes, a remote sensing estimation method for water column integrated P storage in eutrophic lakes was proposed using machine learning. The results showed that the TP profile followed a quadratic distribution that was primarily influenced by chlorophyll-a in shallow water and suspended particulate matter (SPM) in deep water. Based on this observation, a water column TP mass estimation algorithm was developed using extreme gradient boosting (XGBoost) to estimate surface TP, combined with an adjusted floating algae index (FAI) and near-infrared band. The algorithm achieved R-2 >= 0.6, with the error increasing with depth. Then, water depth and lake spatial information were added to the algorithm, the average P storage of 35 large lakes in the Jianghuai region was calculated as 5347 t, and the lake area explains 85% of the P storage. The modeled P storage in Lake Taihu and Chaohu exhibited increasing trends that were mainly driven by the water level. This study is the first to observe lake P storage from space and to help elicidate the P cycle in shallow eutrophic lakes. At present, the Yangtze River Basin exports large amounts of P, lakes reduce P loss in the basin and enrich, and there is still great potential for the recycling and utilization of P resources.

Column-integrated algal biomass has been recognized as a more logical proxy for the evaluation of lake eutrophication. Here, an algorithm with a 3-step framework is put forward for algal biomass mapping in 3 lakes of China (Lake Hongze, Lake Taihu, and Lake Chaohu). It can be summarized in step 1: inversion of surface chlorophyll a (Chl a ), step 2: inversion of diffuse attenuation coefficient of the photosynthetic active radiation [ K d (PAR)], and step 3: estimation of algal biomass with a pretrained generalized additive model. The proposed algorithm outperforms the result-oriented and process-oriented methods in terms of accuracy in 3 lakes (the root mean square error [RMSE] values for datasets of Lake Hongze, Lake Taihu, and Lake Chaohu were 5.09, 8.21, and 3.90 mg/m 2 , respectively). Validated with match-up satellite data, the algorithm generates acceptable results (RMSE = 5.69 mg/m 2 , mean absolute percentage error = 30.9%, N = 16). Another important discovery is that the extremum of algal biomass of the entire lake (B tot ) does not always coincide with that of total surface Chl a . For example, the maximum total surface Chl a was recorded in 2016, whereas the maximum B tot of Lake Hongze was observed in 2020. For Lake Taihu, 3 peaks of B tot appearing in 2017, 2019, and 2021, respectively, did not coincide with those of total surface Chl a . For Lake Chaohu, the interannual B tot followed a bimodal pattern that differed from the pattern of interannual total surface Chl a . The proposed algorithm plays an indispensable role in broadening the horizon for algal biomass inversion.

Aquatic vegetation (AV) is crucial for maintaining lake ecosystem's stability. Transformation of certain types of AV (e.g., submerged aquatic vegetation (SAV)) into other lake features (e.g., algal bloom) may cause a shift in lake's stable state. However, such transformations in Chinese lakes remain largely unknown. Using 0.23 million Landsat images from 1988 to 2023, a comprehensive investigation was conducted to reveal the long-term conversions between SAV, floating and emergent AV (FEAV), and non-AV (e.g., water or algal bloom) across 4375 Chinese lakes (with area > 1 km(2)). Results show that from 1988-1993 to 2018-2023, the total lake AV area experienced a reduction of 0.7 x 10(3) km(.)(2) Despite a countrywide net increase in FEAV (+0.9 x 10(3) km(2)), it could not offset the sharp decline in SAV (-1.6 x 10(3) km(2)). The SAV decline was mainly observed in Eastern Plain lake region. Nationwide, 3.6 x 10(3) km(2) of SAV have transitioned into non-AV, with 78.3% of this area transformation occurring in Eastern Plain. In contrast, many non-AV areas on the Tibetan Plateau saw an emergence of SAV, expanding at +17.9 km(2) per year. However, mutual conversions between SAV and FEAV were minimal. National FEAV gains were primarily from non-AV conversions (+1.3 x 10(3) km(2)), occurring mainly in Eastern Plain, while SAV-to-FEAV transitions accounted for only 23.8% of total FEAV gains. Driving forces analysis shows that lake eutrophication dominated SAV reduction, whereas FEAV variations were primarily influenced by eutrophication, temperature, and lake water area changes. These findings could provide important insights for lake restoration management in China.

Pond water surfaces (PWS) play a crucial role in the ecological sustainability and development of coastal zones. However, in these areas, different types of PWS have significant differences, and the absence of a universal pond classification system complicates the analysis of PWS characteristics. To address this bottleneck, this study introduces a refined PWS classification system for coastal zones, including landside clustering aquaculture ponds (LCAP), marine aquaculture ponds (MAS), salt pans (SP), landscape ponds (LP), and natural ponds (NP). A multifeatured fusion object-oriented (MFFO) method for PWS was established using Sentinel-2 images, based on the Google Earth Engine. Consequently, 10-m resolution PWS classification data were formed in the coastal zone of Jiangsu, China. Results showed that: (1) The total surface area of PWS was 904.01 km2, which accounted for 7.08 % of the study area, reaching 83.18 % overall accuracy. The functional types of PWS can be categorized as aquaculture, landscaping, water storage, and salt drying. (2) Regarding different PWS functional types, significant differences were demonstrated in terms of remote sensing features and geographical patterns. Remote sensing features revealed that LCAP, MAS, and SP differ greatly across various spectral bands, whereas NP varied substantially in shape characteristics, and LP exhibited distinct spatial distribution. Geographically, LCAP and SP were mostly distributed in coastal mudflats, LP were mainly situated in cities, NP were largely distributed in rural and mountainous areas, and MAS were situated on the ocean surface. Above all, the PWS classification system and data products developed in this study reveal the diverse relationships among “functional types-remote sensing features-geographical patterns” regarding PWS, implicating a crucial foundation for clarifying the ecological functional value of PWS and appropriate planning in the coastal zone.

Water salinity characterizes the physicochemical properties of natural water, serving as an essential parameter for assessing lake water quality. However, the efficiency of remote sensing inversion of water salinity is limited as salinity is a non-optically active parameter, leading to the lack of a pixel-scale lake salinity dataset. Conventional function models based on salinity tracers or single lakes have low regional applicability, while machine learning algorithms can effectively capture the nonlinear relationship between radiance and salinity, providing large-scale inversion opportunities. Our study constructed an extreme gradient boosting (XGB) salinity model, which was used to generate the Inner Mongolia lake salinity (IMSAL) dataset with Sentinel-2 remote sensing reflectance. The IMSAL dataset contains 928 raster scenes with 10-meter spatial resolution for eight lakes from 2016 to 2024. Cross-validation and independent validation with measured and published literature-recorded salinities confirmed the good consistency and reliability. This dataset provides invaluable information on spatial patterns and long-term variations in lake salinity useful to prevent lake salinization and facilitate the lake management for sustainable ecosystem development.

The complex sources of particulate organic carbon (POC) in lakes introduce substantial variability in its correlation with optically active constituents (OACs) like chlorophyll-a and suspended minerals. This variability poses a significant challenge in developing long-term POC concentration records from multi-satellite observations, particularly when bridging historical and modern sensors in optically complex waters. The Medium Resolution Imaging Spectrometer (MERIS) and Ocean and Land Colour Instrument (OLCI) perform well in retrieving the bio-optical parameters related to POC in global inland waters. In this study, we utilize data from the Geostationary Ocean Color Imager (GOCI), which has similar spectral bands, as a connection to validate the consistency of the atmospherically corrected remote sensing reflectance (R-rs(lambda)) of MERIS and OLCI using field measurements. An inter-sensor validation of R-rs(lambda) has also been conducted based on the match-up pairs of satellite data. A blended POC model suitable for both non-algal particles (NAP)-dominated (Type 1) and phytoplankton-dominated (Type 2) waters has been proposed and developed using the multi-sensor R-rs(lambda) data. POC concentrations of 110 lakes in the middle and lower reaches of the Yangtze River and Huai River (MLYHR) basin in China are retrieved based on multi-sensor R-rs(lambda) data from 2003 to 2023. The results demonstrate that: 1) the R-rs(lambda) of MERIS, GOCI, and OLCI show good consistency from the green to near-infrared (NIR) bands, 2) the proposed POC inversion model has been validated using field-measured POC with good performance (root mean square error (RMSE) of 1.49 mg/L for Type 1 and 4.52 mg/L for Type 2), and 3) significant seasonal variations in POC in the region have been observed, with high values in summer and low values in winter. Small lakes with areas of similar to 10-50 km(2) have higher mean POC values (7.63 +/- 0.69 mg/L) than those in large lakes (e.g., in lakes ranging from 100 to 500 km(2), the POC concentration is 5.91 +/- 0.81 mg/L). The multi-sensor strategy for estimating POC in optically complex waters provides a practical way to fully utilize the spectral bands and long-term observations from multiple satellites.

Lakes are widely recognized as sentinels of climate change,playing a crucial role in global hydrological and biogeochemical cycles.There is compelling evidence that climate change has led to significant increases in lake surface water temperatures over recent decades[1].One of the most pertinent consequences of increasing lake temperature is the alteration of thermal stratifica-tion[2].