Development of Machine Learning-based River Algal Bloom Prediction Model and Assessment of Influencing Factors: Miho and Nakdong River Basins

At the Gangjeong-Goryeong site on the Nakdong River, cyanobacterial cell counts increased approximately 45-fold within a short span, underscoring the need to address algal blooms not just as a water quality issue, but also as a disaster management priority. This study examined how algal bloom driving factors vary by river type, focusing on chlorophyll-a (Chl-a), as a representative indicator of algal biomass. The Miho River, a smaller tributary and the Nakdong River, a large mainstem river, served as study sites. Various scenarios were constructed based on the hydrological and biochemical parameters. Predictive models for Chl-a concentration were developed using machine learning techniques such as Random Forest and XGBoost. The variable importance was derived for each model, and the prediction performance was evaluated separately. Results showed water temperature as the dominant driver in the Miho River, while biochemical factors like BOD and COD were more influential at the Gangjeong-Goryeong site. These results support the development of river -specific algal bloom prediction systems and targeted disaster response strategies.