Summary:
Drought is a complex natural phenomenon resulting from prolonged periods of below-average precipitation. Its gradual development over large areas makes it challenging to determine its onset, duration, and overall impact accurately. In this study, an artificial neural network (ANN) method-based model was developed to predict the standard precipitation index (SPI), an index commonly used to determine drought severity. SPI12 which reflects a meteorological drought indicator to monitor precipitation anomalies over 12-month accumulation periods were estimated based on artificial neural network (ANN) method using monthly precipitation data recorded in the 1980-2015 period for Kayseri. The North Atlantic Oscillation Index (NAOI), Mediterranean Oscillation Index (MOI), and Arctic Oscillation Index (AOI), which represent large-scale global cycles, were used as input variables in the models. A multilayer perceptron-type ANN model with a single hidden layer was chosen. The model training used 70% of the data and a scaled conjugate gradient backpropagation algorithm. The remaining 30% of the data were used for model testing and control. The activation functions of the ANN model and the number of neurons in the hidden layer were determined using the trial-and-error method. The performances of the models were evaluated using the mean Nash-Sutcliffe coefficient of efficiency (NSE), root square error (RMSE), and coefficient of determination (R2) of agreement between the estimated and observed SPI12 values. This study demonstrated that drought conditions can be successfully predicted at 3, 6, and 12 months in advance using indices that reflect large-scale global climate anomalies.