MONITORING FOREST DEGRADATION OVER FOUR DECADES USING REMOTE SENSING AND MACHINE LEARNING CLASSIFICATION ALGORITHMS IN BOUSKOURA, MOROCCO

Abstract

Abstract. A large portion of Morocco’s forest ecosystem is being damaged by human activity and climate change, which makes it crucial for the country to monitor forest dynamics and develop measures to counter these effects. The objective of this study was to determine how the forest cover has changed over the past four decades in Bouskoura forest, Morocco. Based on Remote Sensing (RS), Geographic Information System (GIS) and machine learning algorithms. Throughout the process, Spatial data such as Landsat 5 TM, Sentinel-2B, and spectral indices, including NDVI, NDWI, NDBI, and MSAVI2 were used to train/validate Random forest (RF), support vector machine (SVM), and K-nearest neighbor (KNN) classifiers. By comparing the performance of the three classifiers for all four periods, the RF method was the most effective with an overall accuracy of 0.99 and kappa coefficient of 0.99 for 1991, 2001 and 2011, and an overall accuracy of 0.99 and kappa coefficient of 0.98 for 2021 . Therefore, the RF was selected as a method of examining time variations. The results indicated that forests covered an area of 20,41 km2 in 1991 which has decreased to 18,96 km2 in 2021, a loss of 1,45 km2 (7.10%) in four decades. The highest forest loss was 2,69 km2 during 1991–2001, 2,12 km2 during 2001–2011, 1,40 km2 during 2011–2021. And the highest forest gain was found to be 3,75 km2 during 2011–2021, 0,61 km2 during 2001–2011, 0,43 km2 during 1991-2001. Recent declines in forest degradation attest to the benefits of initiatives to conserve the environment taken by the country.