Multi-scale analysis of thermal forcing of heatwaves on cyanobacterial blooms in Lake Taihu: mechanisms, spatial heterogeneity, and future projections.

Rising water temperatures and intensifying lake heatwaves (LHWs) are increasingly recognized as important drivers of cyanobacterial harmful algal blooms (CyanoHABs). However, the sensitivity of bloom dynamics to thermal extremes across multiple temporal and spatial scales remains poorly constrained. Focusing on Lake Taihu from 2000 to 2024, we developed an integrated analytical framework combining Freshwater Lake (FLake) thermodynamic reconstruction, lag feature engineering, and machine learning to systematically quantify thermal-bloom response and project future bloom risks. On the response relationship side, multi-scale analysis revealed that LHW-driven bloom enhancement is governed by sufficient absolute temperature and cumulative heat accumulation rather than instantaneous thermal forcing, with spatial heterogeneity in bloom sensitivity co-determined by subregion morphology, hydrodynamic connectivity, and watershed nutrient loading. A Random Forest model integrating FLake-reconstructed thermal and hydrodynamic variables achieved optimal predictive performance (Area Under the Receiver Operating Characteristic Curve, AUC = 0.807, Recall = 0.804). SHapley Additive exPlanations (SHAP) analysis identified lake bottom water temperature (LBWT) as the dominant predictive contributor, operating through a distinct threshold of 10-13 °C. On the projection side, under fixed nutrient conditions, the projected increase in bloom outbreak probability by 2100 reaches approximately 22%, 37%, and 40% under SSP1-2.6, SSP3-7.0, and SSP5-8.5, respectively. Even when nutrient reduction yields maximum marginal benefit at a 30% threshold, it cannot fully offset the independent warming-driven risk component. Collectively, these findings indicate that a nutrient-only management paradigm in the context of accelerating climate warming can only achieve limited effects. Therefore, effective long-term bloom management depends not only on local actions but also on coupling greenhouse gas emission mitigation with targeted nutrient intervention.