Sentinel-2 Random Forest Mapping of Mangrove Cover: SWIR Band Importance in the Nuxco Sub-Basin, Mexico

Mangrove monitoring is essential because of the high socioecological value of these ecosystems and the continuing decline in their extent. This study estimated mangrove forest cover in the Nuxco Sub-basin on the southern Pacific coast of Mexico and assessed the relative importance of spectral bands and indices as predictors. A May 2024 Sentinel-2 Level-2A composite was generated in Google Earth Engine (GEE) and classified with Random Forest (RF) using six spectral bands and six spectral indices, five of them mangrove-specific. The model achieved an overall accuracy of 91.2% (95% CI 85.8–94.7%), κ = 0.895, and an F1-score of 0.889 for the mangrove class. Within the predictor set adopted in this study, the SWIR bands showed higher relative importance than the mangrove-specific spectral indices (mean decrease in Gini impurity: B11 = 80.97; B12 = 80.27). A bands-only model reached 88.7% overall accuracy and a mangrove F1-score of 0.903, against 77.4% and 0.721 for an indices-only model. The estimated mangrove area was 165.6 ha. This ranking is conditional on the predictor combination evaluated; within it, raw reflectance outperformed the index transformations derived from it. The documented workflow provides a transferable approach for local mangrove monitoring in heterogeneous coastal environments.