Prediction of commonly used drought indices using support vector regression powered by chaotic approach

An effective water resources management requires accurate predictions of possible risks. Drought is one of the most devastating phenomena that has a certain risk of occurrence. Understanding the variability of the drought indices is of great importance in determining the spatiotemporal behavior of the drought phenomenon. Moreover, determination of the variability and short-term prediction of the drought indices enables us to take necessary steps in hydrological and agricultural issues. In this study, drought indices have been predicted via Support Vector Regression, SVR. This method originated from a linear regression method in a high dimensional feature space. SVR necessitates a special input matrix. In this study, this matrix has been constructed on the basis of Chaotic Approach, CA. Commonly used drought indices are used in the prediction stage. These indices consist of monthly Palmer Drought Severity Index, PDSI, Palmer Hydrological Drought Index, PHDI, Palmer Z-Index, ZNDX, Modified Palmer Drought Severity Index, PMDI, and Standard Precipitation Index, SPI. One-step ahead prediction has been realized for a 36-month period. Most results show that predictions of the drought indices using SVR are quite promising.