Short-term drought analysis

Drought is defined as one of the most important natural phenomena effecting almost every part of life in terms of physical, economical, developmental, technological and agricultural aspects. Therefore, many definitions have been given in the literature, classifications and indices have been proposed in order to take necessary measures in water resources management and planning. In this study, prospective drought prediction was made with rainfall data of Aksaray and Antalya provinces and a hibrit model is developed to forecast future values. In the first stage, the standardized precipitation index (SPI) values for different time scales were calculated with the monthly total rainfall data between the years 1970-2020 for Aksaray and Antalya provinces. In the second stage, models were created with Support Vector Machines Regression Function (Linear Kernel Function, Gaussian Kernel Function, Radial Based Kernel Function and Polynomial Kernel Function) for SPI3 and SPI12 values, which are formed with 3 and 12-month precipitation totals, and future predictions were made. A new hyrid model was constructed by combining Support Vector Machines and Wavelet Decomposition method to increase the predictive power of the analyzes. Ten different wavelet families (Db4, Db10, Db45, Bior3.1, Bior6.8, Coif1, Coif5, Dmey, SYM3, Haar) were used in the contructed hybrid model. Performances of models were evaluated by comparing the models with the real drought index values and it was determined that the values obtained from the hybrid model constructed by Dmey, Db45 and Bior6.8 wavelet methods were closest to the actual observation values.