A remote sensing-based spatial prediction framework using a Naive Bayes approach for cyanobacterial blooms in eutrophic lakes of China

Three typical eutrophic lakes in China: Lake Taihu, Lake Chaohu, and Lake Dianchi. Cyanobacterial harmful algal blooms (HABs) threaten lake ecosystems and drinking water safety, requiring accurate spatial forecasting. However, existing data-driven models often provide only site-specific or whole-lake average predictions, lacking spatial continuity. Therefore, we developed a spatially explicit HAB prediction model based on a Naive Bayes (NB) approach with spatial priors for short-term forecasting and scenario analysis. Using daily meteorological data and MODIS-derived index (2000–2023), the model provided 1–7 day forecasts of HAB probability and coverage. It achieved Area Under the ROC Curve (AUC) values above 0.84 (1-day) and 0.75 (7-day), with an average 1–7 day AUC ranging from 0.79 to 0.96. Scenario analyses under seven meteorological stress scenarios (high temperature, low wind, heavy precipitation, and their combinations) revealed that Lake Taihu and Lake Chaohu were sensitive to prolonged thermal stress (30-day mean temperature ≥29°C), while Lake Dianchi responded to prolonged flushing (30-day precipitation ≥192 mm). Under a compound “thermal-flushing-stagnation” scenario, average summer–autumn HAB coverage increased to 16 %–23 % in Lake Taihu and Lake Chaohu, and surged to 50 % in Lake Dianchi. This framework provides spatially explicit forecasts and quantitative risk assessment to support eutrophic lake management and sustainable water resource planning. • Integrate remote sensing and meteorological data to support a prediction framework. • A lightweight Naive Bayes (NB) model applicable to diverse lake ecosystems. • Accurately predicts cyanobacterial harmful algal blooms (HABs) for 1–7 days ahead. • Outputs HAB probability and coverage maps directly to guide mitigation. • Quantifies HAB risks under extreme climate-coupled scenarios.