Intensifying extreme climatic events drive the normalization of cyanobacterial blooms in a large shallow eutrophic lake

Under anthropogenic climate change, intensifying extreme climatic events are reshaping lake cyanobacterial blooms; yet their contributions remain poorly quantified, and the causal pathways are unclear, limiting integration into attribution and early warning systems and posing risks to drinking water safety. Here, we fuse 22 years of multisource daily records, build a distributed lag nonlinear model (DLNM) with counterfactual decomposition to separate the marginal effects of extremes from the concurrent normal climate load, and develop a transferable machine learning nowcast centered on extreme exposure features. From 2003 to 2024, extreme climatic events contributed more cumulatively than the normal climate load and served as proximate triggers through predominantly lagged responses. Extreme heat events (EHEs) and extreme drought events (EDEs) generally amplified blooms, whereas extreme rainfall events (EREs) tended to suppress them, primarily by altering water temperature and internal and external nutrient loading. Centering on daily extreme exposure features, we built a nowcasting and threshold alert model that requires no contemporaneous water quality inputs. Bias corrected projections indicate monotonic increases in the frequency, persistence, and cumulative exposure of heat and compound extremes toward 2100, driving a shift from frequent to near normalized blooms; under a high emission pathway (SSP585), annual bloom duration in Lake Hongze exceeds 200 days. Even when current nitrogen and phosphorus targets are met, blooms still intensify under increasingly severe extremes. Importantly, even with aggressive phosphorus abatement, strong extreme climate forcing alone can sustain frequent blooms. We thus propose an interpretable, transferable, extreme event centric framework for bloom attribution and prediction, offering guidance for shallow eutrophic lakes worldwide.