Mapping chronic environmental pressure in Goa using machine learning and deep learning

Coastal landscapes experience overlapping environmental pressures that require integrated spatial assessment. This study combines satellite and geospatial indicators to map chronic environmental pressure in Goa, India. Fifteen predictors represent atmospheric composition, terrain, land-surface conditions, land use, and human activity. Time-filtered products cover available observations within a nominal 2018–2023 period, alongside static and differently dated products. Of 89 expert-informed reference records, 81 supplied predictor information for three tree-based algorithms and four neural-network architectures. Their probabilities were combined into an environmental-pressure classification. Low-, moderate-, and high-classes accounted for 56.75%, 30.54%, and 12.71% of valid mapped pixels, respectively. Higher scores characterised urban and industrial corridors, whereas forested eastern areas generally recorded lower scores. Night-time lights had the largest contribution to the high-class extreme gradient boosting Shapley additive explanations. Model scores align with the expert-informed reference classes. The integrated maps provide a relative screening framework in selecting locations for further environmental observation.