Speaker
Description
Land Use Land Cover (LULC) classification in semi-arid forested regions, such as the argan-dominated landscapes of the Essaouira region (Morocco), presents a significant challenge due to the high heterogeneity and spectral similarity among vegetation types. To address this, our study integrates spectral indices (MSAVI, BSI, Albedo) with textural features derived from Sentinel-2 imagery to enhance class discrimination. A multi-level segmentation workflow based on the SNIC (Simple Non-Iterative Clustering) algorithm was implemented in Google Earth Engine as part of an object-based image analysis (OBIA) approach.
Four machine learning algorithms were tested to classify segmented objects: Random Forest (RF), Extreme Gradient Boosting (XGBoost) and Support Vector Machine (SVM). Among these, the Random Forest classifier yielded the best overall performance in terms of both accuracy and class stability. The results demonstrate the effectiveness of combining OBIA, textural analysis, and machine learning to improve LULC mapping in complex ecosystems, offering a valuable tool for ecological monitoring and land management in arid and semi-arid regions.
