Evaluating the performance of machine learning algorithms in monitoring temporal changes in land cover and land use in mountainous areas: The Ourika basin as a case study

Changes in land use and land cover (LULC) represent a major environmental challenge resulting from rapid population growth, necessitating accurate monitoring and assessment of their impacts. This study aims to evaluate the effectiveness of three machine learning algorithms, namely Support Vector Machines (SVM), Decision Trees (CART), and Random Forests (RF) in classifying land cover patterns and land use in the Ourika Mountain Basin for the periods 1987 and 2025 using Landsat 5 TM and 9 OLI satellite data via Google Earth Engine (GEE). The results showed a clear superiority of the random forest (RF) algorithm in terms of accuracy and consistency, as it recorded the highest values for overall accuracy (OA) and kappa coefficient (KC) for both years, with an overall accuracy of 93% and a kappa coefficient of 0.91 for 1987, and increased to 95% and 0.94, respectively, for 2025. Based on these results, the classification map produced by the RF algorithm was adopted for temporal change analysis. The change analysis revealed significant environmental shifts, represented by a notable decline in natural areas of forests and pastures by 10% of the total area of the basin (equivalent to 5831 hectares). In contrast, there has been a steady expansion in agricultural land, urban areas, and bare land. These changes highlight the increasing human pressures that are contributing to the acceleration of environmental degradation within the Ourika basin. This study provides an effective methodology for monitoring temporal changes and analyzing environmental transformations and can be a valuable tool to support natural resource management and the development of effective strategies for environmental planning and sustainable management of natural resources in similar mountainous areas