Assessing the Ability of supervised Classification of Landsat 8 Satellite Images in Mapping Rangelands Plant Community (Case Study: Rangelands of Southern Yazd Province)

Investigating spatial and temporal changes in the composition of plant species and communities is an essential step in assessing pasture health conditions, understanding the evolutionary processes of the local ecosystem, and developing rangeland management strategies. The aim of this study is to evaluate the ability of supervised classification of Landsat 8 satellite images in mapping the plant communities in the southern rangelands of Yazd province. To do so, we selected and sampled 90 training samples from areas that showed a homogeneous composition of plant species up to a minimum radius of 60 meters from the central point in 2015, and then; plant communities were separated based on the prevalence of the percentage of cover crown. A Landsat 8 satellite image of OLI sensor on May 29, 2015 was used after geometric, atmospheric and radiometric corrections. In the present study, the final accuracy of six supervised classification algorithms including parallel classification algorithms, the minimum distance, the Mahalanobis distance, the maximum likelihood, the neural network and support vector machine with radial kernel were examined in the separation and determination of the range of the plant community. Based on the results, the maximum likelihood algorithm and the neural network had the highest accuracy with the final accuracy of 96.4% and 84.8% and Kappa coefficient of 0.95 and 0.82, respectively. In general, the results of this study showed that in order to differentiate and classify different plant communities in the study area, the maximum likelihood algorithm has good results and combining field studies with remote sensing is a very suitable capability in segregating and classifying the plant types and communities.