Comparison of different approaches for predicting SPI

Drought is one of the most important climatic phenomena which occur in all climate conditions and regions of the earth. When the drought remains for a longtime in a region, it will affect all environmental factors in that region. Drought forecasting, therefore plays an important role in designing and management of natural resources and water resources systems, assessing plant water requirement, etc. In recent decades, (ANNs) have shown great ability in modeling and forecasting nonlinear and non-stationary time series. In this study, two types of artificial neural networks, i.e. Multi Layer Perceptron and Radial Basis Function, and ARIMA time series were applied for drought forecasting. The rainfall data of Now-deh station on Khormaloo River in Golestan province (Iran) were used. Drought conditions were calculated using SPI in short time and long time periods. Among 41 years SPI data, the first 33 years data were selected for training of models and the last 8 years data were used as test data. The results showed that artificial neural networks were able to forecast the SPI and drought conditions with higher accuracy. Meanwhile ARIMA model had also significant results for forecasting.