Estimating Chlorophyll-a Concentrations in South Korean Rivers: A Machine Learning Approach Using Satellite Imagery

Algal blooms are characterized by the mass proliferation of cyanobacteria in eutrophic water bodies under environmental conditions such as elevated water temperature and nutrient concentrations, thereby destabilizing aquatic ecosystems and degrading water quality. To quantitatively assess algal bloom occurrence, chlorophyll-a concentration is widely used as a representative water quality indicator reflecting eutrophication levels and algal biomass. In South Korea, algal blooms frequently occur during summer as elevated water temperatures and sustained nutrient inputs create conditions favorable for algal growth. In particular, the Four Major River basins of South Korea serve as primary drinking water sources and play a critical role in water resource management, making variations in chlorophyll-a concentrations an important indicator for the early identification of potential water quality deterioration. However, existing water quality monitoring systems rely primarily on in situ observations, which limits their ability to capture spatiotemporal variability across extensive river reaches.

As a foundational step to address these limitations, we developed and evaluated a machine learning–based model to estimate chlorophyll-a concentrations using Sentinel-2 satellite data in combination with in situ water quality observations. Satellite spectral information and water quality variables, including water temperature, dissolved oxygen (DO), and turbidity, were used as input features, and a Random Forest (RF) algorithm was applied to develop a chlorophyll-a concentration estimation model. Based on test set validation, the RF-based model achieved an R² of 0.737, an RMSE of 13.07 (mg/m³), and an MAE of 5.83 (mg/m³), showing a reasonable level of agreement with observed chlorophyll-a concentrations.

This study confirms the applicability of combining satellite remote sensing data with in situ water quality observations for estimating chlorophyll-a concentrations. In addition, we present an analysis framework that can be extended to short-term chlorophyll-a prediction by incorporating information from previous time steps. This approach can be used to estimate and predict changes in chlorophyll-a concentrations, providing information to support future water quality management and monitoring strategy development.