Satellite-Based Estimation of Chlorophyll-a and Total Phosphorus in Saemangeum’s Hydrodynamically Complex Waters Using Machine and Deep Learning

The coastal waters formed by the Saemangeum Dam are difficult to monitor and predict in terms of water quality because of complex mixing of seawater and freshwater caused by artificial water gate control, high optical variability, and other factors. The Saemangeum Dam area on South Korea's west coast is a representative artificial coastal water system where inflow of watershed water, seawater exchange through water gate operation, and nutrient accumulation interact nonlinearly, frequently leading to eutrophication and algal blooms.

This research developed a forecasting system for chlorophyll-a (Chl-a) and total phosphorus (T-P) levels in the Saemangeum aquatic region by integrating geostationary satellite GOCI data with artificial intelligence methods. We combined GOCI observations from 2011 to 2020 with in situ water quality measurements from 13 sites to compare machine learning and deep learning algorithms for estimating water quality.

To identify effective input variables for the optically complex Saemangeum environment, satellite reflectance was combined with meteorological information, gate-controlled water exchange, and nutrient indicators. Seven input scenarios were designed to evaluate how progressive variable integration influences prediction performance, and representative machine learning and deep learning models were compared.

 

Results showed that scenarios incorporating nutrient-related variables yielded the most robust predictions for both chlorophyll-a and total phosphorus. While deep learning models captured complex relationships under standard evaluation, spatially independent validation highlighted that machine learning approaches maintained more stable generalization under strong spatial heterogeneity and limited training data. This finding suggests that model suitability depends on data structure and validation context rather than algorithm complexity alone.

Overall, the artificial intelligence-based water quality prediction system presented in this study can effectively monitor fluctuations in chlorophyll a and T-P in embankment reservoir waters, and can be utilized as a practical tool for early warning systems and the development of water quality management policies. It is expected to contribute to strategies for responding to algal blooms and managing large-scale artificial coastal waters.

 

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT)(RS-2025-00515357).