Drought Forecasting Using MLP Neural Networks

Abstract

For the past decades, drought has affected the natural environment of large areas of Peninsular Malaysia and drought monitoring and identification play an important role in the planning and management of natural resources and water resource systems in the country. Standardized precipitation index (SPI) has been used as a conventional tool to identify and monitor drought occurrences. However, to reduce and mitigate the adverse effects of drought impacts, effective forecasting of future droughts is necessary. In this paper, average long term monthly rainfall data for eight stations covering both the dry and wet seasons from Selangor river basin in Malaysia have been used to derive the SPI values for durations of 3 to 9 months. These drought indicators were used as time series for drought forecasting for the basin using the multi-layer artificial neural networks model. Results show that more accurate predictions are achieved using SPI of longer durations, i.e. 6 and 9 months.