Drought forecasting using an aggregated drought index and artificial neural network
Early forecasting of future drought conditions during continuing dry periods can improve water resources management strategies. In this study, a drought forecasting approach is developed and presented using an aggregated drought index (ADI) and artificial neural network (ANN) using a monthly time step. The use of ADI forecasts the overall availability of water resources beyond the traditional forecasting of rainfall deficiency to represent future drought conditions. The paper compares two types of ANN; namely, recursive multi-step neural networks (RMSNN) and direct multi-step neural networks (DMSNN). The results show that the RMSNN approach is slightly better than the DMSNN approach for forecasts with lead time up to 3 months. The DMSNN approach gives slightly better results than the RMSNN approach when forecast lead time is over 3 months, and can give reasonable results up to 6 months ahead of forecasts.
Keywords: aggregated drought index, artificial neural network, drought, drought forecasting, water resources management, Yarra River catchment