Health & Environmental Research Online (HERO)


Print Feedback Export to File
502951 
Journal Article 
Drought forecasting using artificial neural networks and time series of drought indices 
Morid, S; Smakhtin, V; Bagherzadeh, K 
2007 
Yes 
International Journal of Climatology
ISSN: 0899-8418
EISSN: 1097-0088 
27 
15 
2103-2111 
English 
Drought forecasting is a critical component of drought risk management. The paper describes an approach to drought forecasting, which makes use of Artificial Neural Network (ANN) and predicts quantitative values of drought indices - continuous functions of rainfall which measure the degree of dryness of any time period. The indices used are the Effective Drought Index (EDI) and the Standard Precipitation Index (SPI). The forecasts are attempted using different combinations of past rainfall, the above two drought indices in preceding months and climate indices like Southern Oscillation Index (SOI) and North Atlantic Oscillation (NAO) index. A number of different ANN models for both EDI and SPI with the lead times of I to 12 months have been tested at several rainfall stations in the Tehran Province of Iran. The best models in both cases have been found to include, among the others, a corresponding drought index value from the same month of the previous year. Both best models have the R-2 values of 0.66-0.79 for a lead time of 6 months, but it is also shown that the EDI forecasts are superior to those of the SPI for all lead times and at all rainfall stations. The better performance of the EDI model is illustrated by its more accurate prediction of the overall pattern of 'dry' and 'wet' conditions. The structure of the model inputs (previous rain and drought indices) does not vary with the lead time, which makes the models very convenient for the operational purposes. The final forecasting models can be utilized by drought early warning systems, which are emerging in Iran at present. Copyright (c) 2007 Royal Meteorological Society. 
drought forecasting; drought indices; artificial neural networks; Iran; rainfall; precipitation; model; temperatures; oscillation