Spring drought forecasting in mainland Portugal based on large-scale climatic indices

Abstract

The success of a strategy of mitigation of the effects of the droughts requires the implementation of an effective monitoring and forecasting system, able to identify drought events and follow their spatiotemporal evolution. This article demonstrates the capability of the artificial neural networks in predicting the spring standardized precipitation index, SPI, for Portugal. The validation of the models used the hindcasting, which is a technique by which a given model is tested through its application to historical data followed by the comparison of the results thus achieved with the data. The SPI index was calculated at the timescale of six months and the climate indices used as external predictors in the hindcasting were the North Atlantic Oscillation and temperatures of the sea surface. The study showed the added value of the inclusion of previous predictors in the model. Maps of the probabilities of the drought occurrences which may be very important for integrated planning and management of water resources were also developed.

Keywords

• Redes neuronais artificiais
• hindcasting
• SPI
• NAO
• SST
• SPI