Land use/cover classification in an arid desert-oasis mosaic landscape of China using remote sensed imagery: Performance assessment of four machine learning algorithms

The importance of land use and cover change (LUCC) has gradually attracted more attention due to its influence on the climate and ecosystem. Consequently, the necessity of accurate LUCC mapping has become increasingly apparent. Over the past decades, although a large number of machine learning algorithms have been developed to improve the accuracy and reliability of remote sensing image classification, especially for LUCC classification, there is a lack of studies that assess the performance of machine learning algorithms in arid desert-oasis mosaic landscapes. In this study, the main objective is to provide a reference for the extraction of LUCC information in dryland regions with oasis-desert mosaic landscapes by comparing the performances of the k-nearest neighbor (KNN), random forest (RF), support vector machine (SVM) and artificial neural network (ANN) for the LUCC classification of the Dengkou Oasis, China. Landsat-8 Operational Land Imager (OLI) image data were used with spectral indices and auxiliary variables that were derived from a digital terrain model to classify 7 different land cover categories. The highest overall accuracy was produced by the ANN (97.16%), which was closely followed by the RF (96.92%), SVM (96.20%), and finally KNN (93.98%); statistically similar accuracies were obtained for the ANN, SVM and RF. The RF algorithm performed well across several aspects, such as stability, ease of use and processing time during the parameter tuning. Overall, the random forest algorithm is a good first choice method for land-cover classification in this study area, and the elevation and some spectral indices, such as the NDVI, MSAVI2 and MNDWI, should be used as variables to improve the overall accuracy.