Drought Class Transition Analysis by Markov Chains and Log-Linear Models: Approach for Early Drought Warning

In the present research, probabilistic drought characteristics, i.e., steady state probabilities of drought occurrence, drought termination and expected residence times for each severity class were studied for Mazandaran province, utilizing a modified version of Standardized Precipitation Index (SPI) for 3-, 6- and 12-month time scales and Markov chains. According to results, drought termination time and expected residence time of drought classes with increasing time series steps (from 3- to 12-month) showed increasing and decreasing trends, respectively. Furthermore, log-linear models were applied for short term prediction of drought class transition corresponding to SPI 12-month. Results also indicated that the lowest frequencies occurred during direct transitions from a given drought class to two/three classes with higher (lower) severity. Furthermore, the occurrence probability of each drought class from two previously given drought classes was calculated and validated by the log-linear model. Results of validation of the predictions during 2001 confirmed appropriateness of predictions corresponding to drought class severity for a 2-month lead time from two previous months, particularly when drought was initiating or dissipating. It is concluded that log-linear predictions of drought severity class can be used as a useful tool for early warning to farmers and water managers early in autumn.