A Bayesian approach for testing drought intensity trends using daily SPI data: The case of China

The standardized precipitation index (SPI), is one of the mostly used drought indicators for assessing drought duration and intensity. Drought intensity and its trends due to its effect on eco-system has attracted many researchers in fields such as meteorology, agriculture and hydrology. Here, we introduce a Bayesian trend analysis by fitting alternative stochastic-trend models based on the Normal, the Generalized Extreme Value and the Gumbel distributions. Our intention is to test the time-trend assumption in the mean and variance. We apply the above methodology using a new multi-scale daily SPI dataset in mainland China from 1961 to 2018. Based on these data series, we derive the annual total drought intensity (ATDS) for several station in east and south of China. Posterior simulations results based on the best fitting model reveal the locations and time-periods with the strongest and the weakest inter-annual trends for various aspects of ATDS's distribution. Furthermore, a model comparison experiment based on Bayes Factors compares the above stochastic-trend models with the corresponding ones with no-time trends. The results are very significant since they reveal the locations and the regions where inter-annual trends for ATDS have real grounds.