A comparative methodological approach for argan forest classification using Landsat imagery

Abstract

In the last decades, natural and anthropogenic pressures have caused observable changes in the argan landscape despite its significance in Morocco. Remote sensing data can be used to monitor these changes over time and provide information on vegetation health and land cover changes. This study assesses the performance of supervised methods (support vector machine, maximum likelihood, and minimum distance) and unsupervised classification method (Isodata) for mapping the argan forest in the Smimou area of Essaouira province using remote sensing data from Landsat-5 and Landsat-8 (1985 and 2019). Additionally, the impact of the resampling method and the digital elevation model (DEM) integration on the classification results have been examined. The ground truth data were collected and randomly divided into two categories: 234 samples to calibrate the classification algorithms and 340 samples for validation. Maximum likelihood supervised classification achieved an overall accuracy (OA) of 89.62% (kappa = 0.84) and 87.58% (kappa = 0.81) in 1985 and 2019, respectively. Using resampling techniques on normalized difference vegetation index (NDVI) products, aiming for a 10 m resolution, the NDVI results yielded an OA of 91.60% in 1985 and 88.85% in 2019. Further integration of DEM (30-m resolution) with NDVI, which was resampled to a 10 m resolution, achieved an OA of 92.27% and 92.37% for 1985 and 2019, respectively.