Development of drought probability forecasting models based on ENSO indices - A case study in Southern Taiwan

Although Taiwan receives much precipitation, it is an area prone to suffer droughts due to uneven temporal distribution of precipitation and small storage capacity of reservoirs. Therefore, how to make drought early warning and protection against possible losses of life and property has been an essential issue in Taiwan. This study collected the information of Pacific large scale circulation patterns, i.e., ENSO indices, and the historical rainfall records from 7 raingauges in southern Taiwan for calculating the values of three-month standardized precipitation index (SPI3) during 1950 to 2009. Based on the theory of Markov chains, this study calculated the frequencies in the intervals among different variables (e.g., ENSO indices and SPI3) for building state transition probability matrixes to establish drought probability forecasting models. Different combinations of variables make different forecasting models. Therefore, for reducing the forecasting uncertainty of a single-model, the approach of multi-model ensemble (MME) which ensembles all single-model results was used to forecast the SPI3 probability distribution 1-month ahead by using the ENSO indices at the present time as predictors. The results show that the proposed models can well perform in providing short-term early warning of droughts in southern Taiwan and can be used to assess the long-term dry or wet conditions during a given period. For example, the ratio of the occurrence probabilities of SPI3 < 0 to SPI3 > 0 in southern Taiwan is 6:4 during 2001 to 2009, which reveals that the weather pattern was prone to dry condition during the period.