Application of log-linear models for analysis of the SPI drought class transitions in old weather stations of Iran during 20th century

Successful water management to cope with drought requires an understanding of the governing processes and causes. In this study, the frequency of drought class transitions in the first half of 20th century were compared with those of the second half based on the time series of standardized drought index (SPI) for four old weather stations of Iran including Tehran, Mashhad, Esfahan, and Bushehr.. A time window of 12 months was used to compute the SPI. Three-dimensional log-linear models were used to analyse the SPI drought class transition.Fitting log-linear models to the contingency tables of transition frequency from a given drought class to each class were evaluated by means of likelihood ratio of chi-square (G2). Odds ratio and respected confidence interval were estimated for both half of 20th century in the stations interesed. The results showed that log-linear models have an acceptable fit to the observed frequencies of transition at all stations at 5% level. In addition, with an exception for Tehran station, the transition frequencies of moderate and severe drought classes has in increased in the rest of the stations during the second half of 20th century compared to first half. The unconditional probability of moderate and severe droughts in some of stations have been increased up to 3 times during the second half of 20th century compared to the first half.