An Integrated Machine Learning and Remote Sensing Method for Predicting Cyanobacterial Blooms: A Case Study in China’s lakes along a large-scale water diversion project

Cyanobacterial blooms in lakes are a complex and challenging environmental issue worldwide. However, many existing studies on cyanobacterial bloom prediction were constrained by limited data availability, which poses significant challenges to the development of reliable predictive models. To address these limitations, in our previous study, an integrated model combining machine learning and remote sensing was developed, considering physical, chemical, climatic, and hydrologic factors. In this study, a future scenario framework was developed considering water transfer, climate change, and pollution discharge conditions. Then, the integrated model was used to predict blooms trends in Lakes Hongze and Luoma, two typical inter-basin water transfer lakes in China. The study found that an increase in the water transfer scale in the future will promote blooms growth in April and May. Climate change will increase blooms by 5.33% in Lake Hongze and 3.20% in Lake Luoma, while sewage treatment can counteract the negative effects of climate change on blooms. In addition, temperature and solar radiation were demonstrated as appropriate early-warning factors for blooms. For temperature, the thresholds for Lakes Hongze and Luoma were 16.1 °C and 15.7 °C, respectively. For solar radiation, the thresholds for two lakes were 126 W/m² and 111 W/m², respectively. This study proposed a method to improve the prediction performance of machine learning models. In addition, satellite remote sensing data were incorporated into the construction of prediction models, which provide an effective approach for managing and forecasting blooms in lake regions lacking adequate monitoring data.