Comparative analysis of Sentinel-2 and PlanetScope imagery for chlorophyll-a prediction using machine learning models

The application of high spatial resolution remote sensing technology enables the detailed capture of information from water bodies for water quality assessment. In this study, we compare two satellite remote sensing data on water quality assessment, focusing on chlorophyll-a (Chl-a) due to its importance in monitoring eutrophication and algal boom potential. We developed three scenarios to select key features, aiming to optimize the retrieval of Chl-a for a lake in North Carolina, USA. Utilizing five machine learning models, namely linear regression (LR), least absolute shrinkage and selection operator (LASSO), extreme gradient boosting (XGBoost), random forest (RF), and support vector regression (SVR), we constructed inversion models. The results revealed that the XGBoost model exhibited the highest prediction capacity for Chl-a concentration retrieval using Sentinel-2 (S2) data (R 2 = 0.64, RMSE = 8.58 micrograms per liter (μg/l), bias = −0.09). On the other hand, the SVR model demonstrated better predictive performance for Chl-a concentration retrieval using PlanetScope (PS) data (R 2 = 0.71, RMSE = 8.15 μg/l, bias = 0.46). Consequently, spatiotemporal maps of Chl-a concentration across the reservoir were generated using the best-performing machine learning models: XGBoost for S2 data and SVR for PS data. These maps were created to visualize the distribution and variation of Chl-a concentrations over time and space. This study contributes valuable insights into the difference between two satellite sensors with varying spatial resolution in assessing inland water quality and offers a comparative analysis of multiple inversion methods. The outcomes of our research provide guidance for enhancing inland water quality monitoring practices on small water bodies, emphasizing the importance of selecting optimal inversion models based on satellite data sources. The findings contribute to advancing our understanding of the complexities associated with remote sensing technologies and their applications in water quality assessments, ultimately facilitating improved monitoring and management strategies for small inland water bodies. • Comparative analysis of Sentinel-2 and PlanetScope for Chl-a prediction. • PlanetScope shows good performance in predicting Chl-a despite temporal lag. • Methodology is adaptable to similar lakes, offering a scalable monitoring framework. • Machine learning models capture the complex nature of inland water quality. • Feature engineering tested three scenarios to optimize model prediction accuracy.