Chlorophyll-a Prediction Based on Machine Learning and Satellite Data in the South Sea of Korea

Abstract

Chlorophyll-a (Chl-a) is a critical indicator of phytoplankton biomass, offering key insights into changes in marine ecosystems, including algal blooms and nutrient cycling. However, predicting Chl-a concentrations using numerical models remains challenging due to the intricate interplay of climatic and geographical factors. Moreover, conventional studies relying heavily on localized marine measurement instruments face limitations in both scope and generalizability. To address these challenges, this study integrates high-resolution satellite imagery with advanced machine learning (ML) techniques to develop a robust and scalable model for predicting Chl-a concentrations in the South Sea of Korea. Multiple linear regression, random forest, and extreme gradient boosting models were evaluated, along with an averaging ensemble that combines their predictions. A feature ablation study was also conducted to assess the contribution of individual input variables. The results showed that the Ensemble model achieved the highest performance, with R2 values of 0.821 for 1-day predictions, 0.750 for 2-day predictions, and 0.713 for 3-day predictions. These findings demonstrate the potential of integrating high-resolution (250 m) satellite data with ML to achieve accurate and wide-scale predictions of Chl-a concentrations. These advancements are expected to provide foundational data for applications in marine environmental monitoring, sustainable aquaculture management, and harmful algal bloom prediction.

Keywords

• Chlorophyll-a
• ensemble
• feature ablation study
• machine learning (ML)
• satellite data
• South Sea of Korea