Comparison of different algorithms for land cover mapping in sensitive habitats of Zagros using Sentinel-2 satellite image: (Case study: a part of Ilam province)

The western forests and rangelands of Iran in Zagros habitats have mainly been destroyed by various reasons in recent years. The preparation of the land cover map in these sites is the first step to protect them and to prevent further destruction. The aim of this research was to select the best algorithm for land cover mapping in a part of Ilam site using the Sentinel-2 image. After providing Sentinel-2 the supervised classification of it was performed by seven different algorithms (maximum likelihood, minimum distance from the average, mahalanobis distance, spectral angle mapper, spectral correlation mapper, support vector machine, neural network). For accuracy assessment of the land cover maps, the stratified random points were created and found in the field. In the field visit, after determining the current land cover of each point in the plot area, the real land cover of each point was compared with the defined land cover of the same point in the pixel area based on classification results and the accuracy of the algorithms was evaluated. The results showed that the support vector machine algorithm had the highest accuracy in providing the land cover map with a general accuracy of 79% and a Kappa index of 0.70. The analysis of the land cover map obtained from this algorithm showed that the dense forest area was 319.64 ha, semi-dense forest area was 361.44 ha and sparse forest area was 1832.36 ha from the total area of the study area (16085.31 ha). Also, the rangeland area was 7352.78 ha, the garden area was 62.32 ha, the agricultural area was 658.42 ha and understorey agriculture was 4504.64 ha. For optimal management of this sensitive ecosystem, land cover mapping using this algorithm in certain temporal intervals is essential to investigate the forests and rangelands change and to control the human-made land uses.