Climate-driven dynamics of surface water temperature and its coupled responses to lake heatwaves and cyanobacterial blooms in Lake Taihu.

Climate change has profoundly reshaped lake thermal regimes and ecological processes, intensifying the risks of cyanobacterial blooms. Using Lake Taihu as a representative shallow eutrophic lake, this study integrates multi-source observations and remote sensing data (2002-2024) with machine learning (XGBoost), deep learning (LG-FusionNet), causal inference, and hydrodynamic modeling to investigate the coupled responses among lake surface water temperature (LSWT), lake heatwaves (LHWs), and cyanobacterial blooms under climate change. The LSWT reconstructed by XGBoost showed strong agreement with observations (R2 = 0.91, NSE = 0.95, RMSE= 1.53 °C), exhibiting a significant warming trend (+0.25 °C/10a). The hybrid LG-FusionNet model achieved high predictive accuracy (R2 = 0.97, RMSE = 1.26 °C) and provided thermal boundary inputs for future simulations. Since 2018, heatwave frequency, duration, and persistence have intensified and tend toward year-round occurrence. Convergent cross mapping (CCM) and partial least squares structural equation modeling (PLS-SEM) analysis indicated that meteorological factors drive bloom dynamics mainly through LSWT and LHWs (β = 0.905), with nutrients as secondary mediators. Under a multi-scenario multi-model ensemble (MSME; SSP126/245/585), coupled EFDC simulations project that by 2025-2050, LSWT will rise by 1.31 °C, with 10∼11 heatwave events per year and earlier, prolonged bloom peaks (June-October) showing Chl-a increases >2 μg/L. These results revealed a positive feedback of "climate warming → intensified heatwaves → enhanced stratification and deoxygenation → cyanobacterial dominance," providing scientific insights for lake management and bloom risk mitigation under climate change.