Retrieval and spatiotemporal variation of chlorophyll-a concentration in Baiyangdian Lake based on machine learning

Baiyangdian Lake, the largest shallow freshwater lake in northern China, plays a critical ecological role in the Beijing-Tianjin-Hebei region. However, it faces increasing eutrophication pressure due to anthropogenic activities and climate change. Chlorophyll-a (Chl-a) concentration is a key indicator of eutrophication and algal biomass. This study evaluated the performance of nine machine learning models for retrieving Chl-a concentrations in Baiyangdian Lake using Sentinel-2 Multispectral Instrument (MSI) imagery and in-situ monitoring data from September 2021 to August 2022. Among the models tested, tree-based algorithms including Decision Tree (DT), Random Forest (RF) and Gradient Boosting (GB) delivered superior overall performance for Chl-a retrieval, exhibiting stronger capacity to capture the complex nonlinear spectral responses in shallow turbid inland lakes compared with neural networks, distance-based, kernel and linear regression models. The results reveal significant seasonal variation in Chl-a, with peak concentrations occurring in autumn (September–October) due to elevated water temperature and nutrient accumulation, and lower values in summer following ecological water replenishment. Spatially, Chl-a exhibited a heterogeneous distribution, with higher concentrations in the southwestern, influenced by agricultural runoff, and lower concentrations in north and southeast, reflecting ecological management. The study highlights the effectiveness of ensemble learning models for Chl-a retrieval in optically complex shallow lakes and provides valuable insights for water quality monitoring and management strategies in Baiyangdian Lake and similar freshwater ecosystems.