Monitoring and classification of karst rocky desertification with Landsat 8 OLI images using spectral indices, multi-endmember spectral mixture analysis and support vector machine

Abstract

Karst Rocky Desertification (KRD) is the reduction of vegetative productivity of this land with the release of bedrock as a result of the full or partial transportation of the fertile soil through natural processes and human activities in karst landscapes. The purpose of this study is to reveal the effectiveness of Remote Sensing methods in monitoring, mapping and evaluating KRD. Landsat 8 OLI images were used to carry out these procedures. In monitoring this process, Karst Bare Rock Index (KBRI), Normalized Difference Rock Index (NDRI), Carbonate Rock Index 2 (CRI2), Normalized Difference Build-Up Index (NDBI), Normalized Difference Vegetation Index (NDVI), Dimidiate Pixel Model (DPM), Multi Endmember Spectral Mixture Analysis (MESMA) and Support Vector Machine (SVM) were used from the spectral indices. In order to determine KRD with spectral indexes, a strong linear relationship was tested between some indices such as DPM (R2=0,79), KBRI (R2=0,66), and NDBI (R2=0,64) and field measurements. In order to evaluate the results obtained, KRD was divided into 4 basic classes such as none, mild, moderate, and severe. According to these classification levels, it was determined that the SVM method had the highest accuracy (Kappa=0.88). According to the classification results, which have the highest accuracy in the study area, the rate of areas undergoing severe karst desertification is 40%, moderate desertification process is 17%, mild desertification is 14% and non-desertification is 29%. In the study, it was concluded that the KRD strengthens as one goes from south to north and from west to east in the research area. This study points out KRD is one of the effective ecosystem problems in the Mediterranean region, Türkiye.