Annual detection of wetlands using optical indices and supervised

Abstract

Wetland ecosystems play a crucial role in water regulation, biodiversity preservation, and climate mitigation. However, their detection and monitoring remain challenging, especially in dynamic Mediterranean environments. This study presents a comparative evaluation of three supervised classification algorithms (Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Tree (CART)) to detect wetlands in the Tangier Tetouan Al Hoceima region using Sentinel-2 imagery. A stack of spectral indices (NDVI, NDWI, and MNDWI) was used to generate annual reference maps covering the 2020-2024 period. The results show that the spatial extent of detected wetlands varies with environmental conditions and the classification algorithm applied. RF demonstrates higher temporal stability, while SVM tends to overestimate wetland coverage. The combined use of the three indices improves overall classification accuracy. These findings suggest that a multi-model strategy can enhance the robustness of wetland detection in the face of climate change.