Integrating machine learning and remote sensing for long-term monitoring of chlorophyll-a in Chilika Lagoon, India

Abstract

Chlorophyll-a (Chla) is recognized as a key indicator of water quality and ecological health in aquatic ecosystems, offering valuable insights into ecosystem dynamics and changes over time. This study aimed to to develop and validate a robust ML model for estimating Chla using Landsat data, produce a time series of Chl a maps, and analyze the spatiotemporal variability of Chla in Chilika Lagoon, Asia’s largest brackish water lagoon. Nine ML regression models, including Extreme Gradient Boost, Support Vector Regression, Random Forest, and Bagging Regression, were evaluated using Landsat imagery and field data. After extensive hyperparameter tuning, the Bagging Regression model achieved the highest estimation accuracy, with an R² of 0.8776 and a Root Mean Square Error of 0.9190 µg/L. This optimized model was subsequently applied to generate a time series of Chla maps for Chilika Lagoon from 2014 to 2023, revealing notable seasonal and spatial variability. Chla concentrations peaked during summer months and were generally higher in the lagoon’s northwestern region, gradually decreasing towards the southern area. This approach holds promise for precise Chla monitoring in diverse lagoon environments and may aid in the assessment and management of similar coastal and inland lake ecosystems worldwide.