Application of machine learning approach in zoning of desert geomorphological facies

Identifying and managing geomorphological facies are fundamental aspects of geomorphological investigations. Remote sensing data is a widely used method for recognizing geomorphological facies. However, classifying desert facies using multispectral images is challenging due to the spectral similarities among them. This study focuses on the Yazd-Ardakan Plain in Central Iran, which features diverse arid land geomorphological facies, including clay plains, pavements, sand dunes, rocks, and vegetation. The classification of these facies was performed using random forest and support vector machine algorithms. Due to the spectral similarity of facies in this desert region, auxiliary data such as land surface temperature, normalized difference vegetation index, brightness temperature, and albedo were incorporated to enhance classification performance. Landsat 8 images and ground truth data were utilized, with 70% of the data allocated for training and 30% for testing. Results indicated that the overall accuracy of the support vector machine and random forest algorithms was 83.89% and 83.22%, respectively, with Kappa coefficients of 0.80 and 0.79. Both algorithms performed similarly in identifying geomorphological facies using spectral bands. However, by incorporating both spectral and auxiliary data, the Kappa coefficient and overall accuracy increased to 0.92 and 94.18% for the support vector machine algorithm and to 0.91 and 93.29% for the random forest algorithm. In conclusion, applying the random forest and support vector machine algorithms with auxiliary data led to more accurate geomorphological facies zoning, overcoming challenges posed by spectral similarities. This approach can be extended to other desert environments, providing a reliable methodology for improving landform classification and supporting natural resource management efforts.