Drought sensitivity mapping using two one-class support vector machine algorithms

This paper investigates the use of standardised precipitation index (SPI) and the enhanced vegetation index (EVI) as indicators of soil moisture. On the other hand, we attempted to produce a drought sensitivity map (DSM) for vegetation cover using two one-class support vector machine (OC-SVM) algorithms. In order to achieve promising results a combination of both 30 years statistical data (1978 to 2008) of synoptic stations and 10 years MODIS imagery archive (2001 to 2010) are used within the boundary of Kermanshah province, Iran. The synoptic data and MODIS imagery were used for extraction of SPI and EVI, respectively. The objective is, therefore, to explore meaningful changes of vegetation in response to drought anomalies, in the first step, and further extraction of reliable spatio-temporal patterns of drought sensitivity using efficient classification technique and spatial criteria, in the next step. To this end, four main criteria including elevation, slope, aspect and geomorphic classes are considered for DSM using two OC-SVM algorithms. Results of the analysis showed distinct spatio-temporal patterns of drought impacts on vegetation cover. The receiver operating characteristics (ROC) curves for the proposed DSM was used along with the simple overlay technique for accuracy assessment phase and the area under curve (AUC = 0.80) value was calculated.