Evaluation and Prediction of Meteorological Drought Conditions Using Linear Time-Series and Genetic Programming Models

Abstract

Over the years, a number of prediction methods have been proposed for the evaluation of probability of hydrological–meteorological variables or drought indices. In this study, the precipitation data recorded in four stations of northwestern Iran over the period 1960–2014 were used to develop the time-series and genetic programming (GP) models. Comparison of the observed and predicted data showed that although both models have acceptable accuracy in predicting precipitation, the time-series models had lower errors than the GP models. So, the autoregressive and periodic autoregressive moving average models were chosen as the superior models for annual and monthly series, respectively. Therefore, the Standard Precipitation Index (SPI) and Z-Score Index (ZSI) were used to assess the drought conditions. According to the results, the SPI recognised a higher percentage of historical and prediction periods as drought conditions than ZSI. The validation of indices showed that the ZSI was more capable for detecting the drought and wetness conditions. The trend analysis of SPI and ZSI showed significant decreasing trends in different stations at all-time scales, except yearly in Urmia and all-time scales in Zanjan, which statically had no significant trend. In conclusion, given the current precipitation trends, the droughts are increasing in both severity and numbers.