Introduction of Probabilistic Drought Prediction to Korea

Abstract

The advantage of probabilistic prediction has been verified and acknowledged for several decades so people are making use of the probabilistic prediction in lots of fields, including hydrometeorology. One of the biggest advantages is that it can take into account various events through uncertainty in the predicted value, especially for long-term predictions which have large uncertainties. In Korea, however, the drought prediction is still performed in a deterministic approach. Therefore, the purpose of this study is to apply the probabilistic drought prediction to Korea and then further propose a method to improve the prediction technique. Accordingly, this study developed an ensemble drought prediction (EDP) system focusing on the hydrological drought measured by natural streamflow in eight basins in Korea. Because of the natural characteristic of drought, it only can be measured indirectly through the hydroclimatic variables. In order to measure the hydrological drought, the streamflow was converted to standardized runoff index (SRI) which is a kind of drought index considering regional characteristics and various time scales for the hydrological drought. Then to generate EDP distribution for 1-month ahead monthly drought prediction, the streamflow simulations of an ESP (Ensemble Streamflow Prediction) were converted to SRI. The deterministic prediction was done by the expected value of EDP distribution, and the probabilistic one was derived by the probability driven from the distribution. Moreover, to improve EDP, soil moisture index (SMI) satellite data provided by APEC climate center (APCC) were used to update EDP via the Bayes' theorem. The regression between SRI and SMI was used as a likelihood function that updates the EDP distribution. Additionally, the APCC precipitation probability forecast was used to update EDP using the PDF ratio method. As a result, three main conclusions were drawn as follows.

(1) The probabilistic drought prediction was 52% better than the deterministic on average in terms of prediction skills. When predicting the short-term drought, the probabilistic approach outperformed even more.

(2) Updating EDP using soil moisture information the via Bayes' theorem makes skill to be improved by 20% on average. It can be said that the soil moisture information corrects EDP if the likelihood function is valid and accurate.

(3) Reflecting the precipitation forecast to EDP via the PDF ratio yielded 6% better performance only for the non-irrigation period. From this, it was found again that reflecting informative data can make better the drought prediction.