Benefits of multiple remotely sensed datasets and machine learning models to predict the Chlorophyll-a concentration in the Nakdong River, South Korea

Abstract

Abstract Machine learning models (MLMs) are increasingly used with remotely sensed data to monitor chlorophyll-a (Chl-a). MLMs require large amounts of remotely sensed data to monitor Chl-a effectively. However, weather conditions, satellite revisit cycles, and coverage constraints can impede the collection of adequate remotely sensed data. To address this, we tested whether MLMs effectively improved the predictions of Chl-a concentrations within the 16 lakes of the Nakdong River in South Korea using two remotely sensed datasets (Sentinel-2 and Landsat-8). This study evaluated four MLMs: Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), Random Forest (RF), and eXetreme Gradient Boosting (XGB) using three different input datasets: 1) two remotely sensed datasets (Sentinel-2 and Landsat-8), 2) Sentinel-2, and 3) Landsat-8. The MLP model with multiple remotely sensed datasets outperformed other MLMs affected by data imbalance. The predictive map of the spatial distribution of Chl-a generated by the MLP model highlighted areas with high and low Chl-a concentrations. In conclusion, this study emphasizes the value of multiple remotely sensed datasets in enhancing the accuracy and reliability of Chl-a estimation, mainly when using the MLP model. This also highlights the need to address data imbalances when predicting Chl-a concentrations using MLMs. These findings provide valuable insights into utilizing MLMs effectively for Chl-a monitoring.