Assessing land cover change detection in a developing country with machine and deep learning algorithms: a case study

This study examines the evolution of land cover in the Fez province of Morocco through a detailed analysis of Land Cover Change Detection (LCCD). Using satellite imagery from Landsat-8 and Landsat-9, we generated high-resolution land cover maps by applying a range of machine learning and deep learning techniques. The analysis spans the period from 2014 to 2022, aiming to identify the most suitable algorithm for accurate LCCD. The methods tested in this research include Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbor (KNN), Multilayer Perceptron (MLP), and Convolutional Neural Network (CNN). Upon thorough comparison, SVM emerged as the top-performing model, yielding the most reliable classification results, while MLP demonstrated comparatively limited accuracy in this context. Our findings also underscore significant land use transitions, particularly an expansion of urban and industrial zones accompanied by a decline in agricultural lands trends largely attributed to anthropogenic pressures. These patterns are intricately linked to broader environmental shifts, including those related to climate. By analyzing spatial and temporal variations in land cover, this research contributes to a better grasp of the environmental transformations occurring in Fez, offering valuable knowledge to support evidence-based planning and the promotion of sustainable land use strategies.