Developing practical models for predicting chlorophyll levels in the Andong Reservoir using empirical and machine learning approaches based on integrated analysis of physical, chemical, and hydrological indicators

Abstract

Managing excessive phytoplankton biomass is a pressing issue worldwide and requires precise predictions and efficient water quality control. In this study, we employed the random forest approach to build a practical predictive model for phytoplankton biomass as determined by chlorophyll (CHL-a) levels. Using an integrative 15-year dataset, we obtained insights into CHL-a dynamics and identified important factors influencing it. Seasonal dynamics were crucial for shaping water quality, with meteorological and hydrological fluctuations playing pivotal roles. Elevated suspended solids and phosphorus levels during the summer monsoon indicated increased runoff and nutrient loading, contributing to fluctuations in nutrient ratios. CHL-a was also responsive to seasonal variations in nutrient availability, particularly in phosphorus and ammonium-nitrogen (NH4-N). These results suggest the importance of nutrient management strategies for controlling and mitigating eutrophication risk. The random forest model had practical accuracy (R2 = 0.66, RMSE = 1.85) for predicting long-term CHL-a variation and identified phosphate-phosphorus, water temperature, and NH4-N as the most important predictors. However, our findings emphasize the importance of comprehensive analyses to identify the factors driving CHL-a variation. Combining the strengths of machine learning with multifaceted insights would enhance prediction accuracy and efficiency, ultimately facilitating informed decision-making for reservoir management and environmental conservation efforts.

Keywords

• Chlorophyll-a
• Empirical analysis
• Eutrophication
• Nutrient regimes
• Random forest
• Summer monsoon