Comparative Analysis Of Machine Learning Algorithms For Land Use And Land Cover Mapping: Case Study Of Berrechid-Settat Region, Morocco

This study analyses the spatiotemporal dynamics of Land Use and Land Cover (LULC) in the Berrechid-Settat area of Morocco throughout three reference years: 2010, 2015, and 2023. Satellite images from Landsat 7 (ETM+) and Landsat 8 OLI were processed using the Google Earth Engine (GEE) platform to facilitate quick access, preprocessing, and analysis of extensive datasets. To classify LULC changes and assess the efficacy of machine learning models, Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) were examined. These models were used to categorise five principal LULC classes: water bodies, forests, urban regions, vegetation, and barren lands. Our findings indicated that Random Forest consistently yielded the highest classification accuracy, achieving an Overall Accuracy (OA) of 91.84% and a Kappa Coefficient (KC) of 0.86 in 2023, thereby affirming its efficacy for multi-temporal land use and land cover mapping. The Decision Tree exhibited competitive performance in 2010 (87.36% OA, a KC of 0.79) but showed diminished stability in later years. The SVM displayed middling performance, particularly excelling in the classification of urban areas (about 94%) but exhibiting reduced accuracy for forest regions. This analysis emphasises the efficacy of GEE and Python libraries in analysing large satellite imagery and the proficiency of DT and RF models in land use and land cover classification. The results can guide regional planning and land management policies, fostering sustainable development.