Spatial-temporal distributions of phytoplankton shifting, chlorophyll-a, and their influencing factors in shallow lakes using remote sensing

Understanding phytoplankton dynamics is crucial for assessing water ecosystem health. However, traditional in-situ investigations often fall short of capturing the spatial-temporal variations of phytoplankton community structure and biomass on large scales. This study introduces a novel approach that harnesses the power of machine learning models coupled with Sentinel-2 satellite data to predict phytoplankton shifting and chlorophyll-a (Chla) in three large shallow lakes in central China from 2016 to 2021. We employed an array of machine learning algorithms, including the extreme gradient boosting method (XGBoost), random forest (RF), support vector machine (SVM), and K-nearest neighbor (KNN) to retrieve Chla and phytoplankton groups clustered by hierarchical cluster analysis based on the Bray-Curtis similarity index (HCA). The influences of environmental factors, including meteorology and nutrients, on phytoplankton dynamics were then explored. Our results identified four distinct phytoplankton groups, each exhibiting unique structural characteristics according to the HCA. The RF and XGBoost, utilizing top-of-atmosphere reflectance, provided the most accurate estimations of phytoplankton shifting and Chla, respectively. The red edge-red ratio emerged as a key variable in both models, underlining the significance of red edge bands in phytoplankton monitoring. We mapped spatial-temporal patterns of phytoplankton group occurrence and average Chla concentration across the three lakes. Our findings indicated that eutrophication extended periods of cyanobacteria domination and algal blooms, particularly in bays and nearshore areas. Redundancy analysis and classification and regression tree model highlighted temperature as a significant driver of phytoplankton shifting, while nutrient levels exhibited stronger influences on Chla. Our study underscores the potential of integrating machine learning models with Sentinel-2 data to enhance the monitoring and prediction of phytoplankton dynamics in shallow lakes. This approach offers valuable insights for the early detection of algal blooms and informed management strategies, contributing to the preservation and sustainable management of aquatic ecosystems.