Support Vector Machine (SVM) based Rain Area Detection from Kalpana-1 Satellite Data
Rain is one of the major components of water cycle; extreme rain events can cause destruction and misery due to flash flood and droughts. Therefore, assessing rainfall at high temporal and spatial resolution is of fundamental importance which can be achieved only by satellite remote sensing. Though there are many algorithms developed for estimation of rainfall using satellite data, they suffer from various drawbacks. One such challenge in satellite rainfall estimation is to detect rain and no-rain areas properly. To address this problem, in the present study we have used the Support Vector Machines (SVM). It is significant to note that this is the first study to report the utility of SVM in detecting rain and no-rain areas. The developed SVM based index performance has been evaluated by comparing with two most popular rain detection methods used for Indian regions i.e. Simple TIR threshold used in Global Precipitation Index (GPI) technique and Roca method used in Insat Multi Spectral Rainfall Algorithm (IMSRA). Performance of the above considered indices has been analyzed by considering various categorical statistics like Probabil ity of Detection (POD), Probability of no-rain detection (POND), Accuracy, Bias, False Alarm Ratio (FAR) and Heidke Skill Score (HSS). The obtained results clearly show that the new SVM based index performs much better than the earlier indices.