Predicting Algae Occurrence and Identifying Influencing Factors in Two Adjacent Weirs Installed Upstream and Downstream Using Machine Learning Models

This study aimed to identify the primary drivers of spatial and temporal variations in chlorophyll-a (Chl-a) concentrations and to improve the prediction accuracy of Chl-a concentrations at two adjacent weirs-Seungchon Weir (SC Weir) and Juksan Weir (JS Weir)-in the Yeongsan River, Korea. Utilizing over a decade of high-frequency monitoring data combined with a gradient boosting (GB) regression model, we investigated the conditions leading to high Chl-a concentrations, which are indicative of algal blooms. The results indicate that JS Weir frequently experiences extreme Chl-a concentrations (> 100 μg/L) under low discharge conditions (< 20 m³/s). Seasonal analysis revealed that diatoms dominated algal blooms at JS Weir during the colder months, which contradicts the conventional expectation of more intense summer blooms given the typically low temperatures and light availability in winter. Discharge rate and water temperature exhibited inverse relationships with Chl-a concentrations, and the GB model revealed a lagged multiday discharge effect. Water temperature and total organic carbon at JS Weir were identified as the most influential predictors of Chl-a concentrations. Parameters from SC Weir also showed high importance, confirming upstream-to-downstream connectivity along the 22-km reach between SC Weir and JS Weir. Integrating upstream data enhanced predictive accuracy for downstream bloom conditions. This research provides a foundation for coordinated weir operation-optimizing sluice gate control and upstream nutrient management-to mitigate algal blooms in regulated river systems. These management strategies can improve water quality and protect ecosystem health.