Assessing Land Use and Land Cover changes in Al Hoceima province, Morocco (2014–2024): a comparative analysis using machine learning algorithms

Abstract

The present work employed remote sensing data (Landsat 8 OLI images), two machine learning algorithms (Random Forest (RF) and Support Vector Machine (SVM)) and a parametric algorithm (Maximum Likelihood – MLH) within the ArcGIS environment to assess the spatiotemporal Land Use and Land Cover changes in Al Hoceima province (north-eastern Morocco) from 2014 to 2024. Based on the classification generated by the MLH algorithm, there were increases in the forest, urban area and water/river classes of 8.2%, 1.2% and 0.2%, respectively. Conversely, the vegetation and bare land classes decreased by 1% and 8%, respectively. The RF indicated that the forest and water/river classes remained stable from 2014 to 2024, vegetation decreased by 2.3%, while urban area and bare land increased by 0.6% and 1% respectively. The SVM classification revealed an increase of 1% and 0.8% in the forest and urban area, respectively. However, water/ river, vegetation and bare land decreased by 0.2%, 1% and 0.2%, respectively. Overall, there were two common trends for the three classifiers: an increase in urban area (of between 0.6% and 1.2 %) and a decrease in vegetation (of between –1% for both MLH and SVM and –2.3% for RF). In terms of accuracy evaluation, the MLH exhibits remarkable overall accuracy of 91% and Kappa coefficient (K = 0.89) followed by the SVM with 88% (K = 0.84). To validate the outcomes of the three algorithms classifier, an estimation of the Normalized Difference Vegetation Index (NDVI) changes was conducted. The results of NDVI changes support the outcomes of the RF classifier, although it gave the lowest accuracy, with 81% (K = 0.77).