Drought Forecasting of Seyhan and Ceyhan Basins Using Machine Learning Methods

Abstract

A drought is a prolonged natural disaster with numerous economic, social, and environmental consequences; it occurs when the natural water supply in a given region falls below normal levels. Precautions must be taken to mitigate the negative impacts that drought can cause. Drought probabilities can be estimated by carefully analyzing variables such as precipitation, river flow, and soil moisture with the help of various indices. In the literature, many studies have been conducted to estimate drought indices over time. In this study, drought forecasts were made for the Seyhan and Ceyhan Basins in 3-, 4-, 6-, and 12-month periods with the Standardized Precipitation Index (SPI) using precipitation data between January 1989 and July 2020. The success rates of the forecasts made in the models created with the Random Forest (RF) Algorithm, Support Vector Machine (SVM), and Artificial Neural Network (ANN) machine learning methods were statistically compared. The SVM drought forecasting model was performed in 3-month forecasts in the study. Among the machine learning methods, the ANN method was more successful than the other methods in terms of performance in 4, 6, and 12-month drought forecasts.