Application of hybrid artificial neural network algorithm for the prediction of standardized precipitation index

The application of wavelet transformation has become a popular area of interest in hydrological modeling as it enables the use of spectral and temporal information contained in input data. Drought modeling is one such area that is still far from complete, considering the stochastic nature of drought characteristics per every drought events. This study therefore aims to predict a drought index, i.e. the Standardized Precipitation Index (SPI), using artificial neural network (ANN) and a hybrid ANN with wavelet analysis (WA-ANN) using four main inputs: precipitation, potential evapotranspiration, Southern Oscillation Index, and Nino 4 index for Brisbane, Australia. For WA-ANN, the four inputs were decomposed into three detail and one approximation levels using Daubechies-4 (db4) orthogonal mother wavelet. The evaluation of prediction performance showed that WA-ANN outperformed ANN model with an increased accuracy by 49.89% based on Root Mean Squared Error values.