Remote Sensing and U-Net-Based Prediction of Cyanobacterial Bloom Responses to Warming in Lake Taihu

In view of the limitations of existing studies, in which remote sensing extraction of algal blooms is easily affected by cloud interference, and mechanistic models are constrained by excessive parameters and inadequate representation of nonlinear relationships, resulting in limited timeliness and accuracy, this study took Taihu Lake as the study area and constructed a research framework of bloom extraction-scale matching-spatial prediction-scenario response based on Landsat imagery and gridded meteorological data, constructing the relationship between meteorological factors and algal blooms using machine learning methods. First, the Tasseled Cap transformation (TCap) and Floating Algae Index (FAI) were combined to extract the spatial distribution and area of algal blooms, while cloud interference was addressed to improve recognition stability under complex background conditions. Next, the spatial scales of bloom rasters and meteorological factors were unified to build a matched bloom-meteorological dataset. On this basis, a U-Net model driven by multiple meteorological factors was developed to predict remote-sensing-based bloom distribution/extent patterns under three warming scenarios. The results showed that: (1) the combination of TCap and FAI improved the accuracy and efficiency of bloom extraction; FAI was more sensitive but tended to overestimate bloom area, whereas TCap was more stable under cloud interference; (2) the U-Net model achieved an overall accuracy of 95% and a prediction accuracy of 88%; and (3) bloom area increased under all three warming scenarios, and the extent of expansion generally became more pronounced with increasing warming magnitude, although the response was not strictly monotonic across all cases. Based on the seasonal mean bloom-area increase relative to the baseline condition (S0), the warming response was strongest in spring, followed by summer and autumn, and weakest in winter. This study can provide a reference for cyanobacterial bloom early warning and water environment management in Lake Taihu.