An Integrated Remote Sensing and Machine Learning Framework for Aquatic Habitat Mapping and Ecosystem Health Assessment

Aquatic habitat degradation necessitates assessment methodologies that integrate habitat availability with ecological status. In this research, an attempt was made to formulate a unified remote sensing and machine learning approach for assessing and mapping aquatic habitats and ecosystem health. Multispectral imagery from Sentinel-2 was integrated with field data on water quality, vegetation cover, and environmental disturbance. After preprocessing, 24 satellite images were selected, and their spectral bands and indices were extracted. Habitat maps were created through Random Forest, Gradient Boosting, and Support Vector Machine classifiers with six habitat classes. Normalized ecological indices were integrated to form an ecosystem health index, which was predicted using regression models. Performance of classification and prediction was assessed using independent observations. The accuracy of Random Forest was found to be 93.8% with a macro F1 score of 0.92 and Cohen's kappa of 0.91, while Gradient Boosting and Support Vector Machine yielded accuracies of 92.1% and 90.6%, respectively. Random Forest regression yielded R² = 0.89, RMSE = 0.061, MAE = 0.047, and MAPE = 8.4%. The Remote Sensing Score positively correlated with ecosystem health (r = 0.84, P < 0.001). Open water covered 46.1% of the mapped area. “Good” or “very good” health was noted for 51.8% of the ecosystem, while “poor” or “very poor” health was noted for 20.5% of the ecosystem. The model links habitat structure with ecosystem condition and allows spatial proof of monitoring and restoration efforts that can be used. Seasonal and site comparison verification is required to determine wider applicability and uncertainty.