Application of Hidden Markov Model to Probabilistic Assessment of Future Drought Based on RCP 8.5 in Korea

Abstract

In order to consider inherent uncertainty of drought, probabilistic approaches are needed to assess and forecast drought. However, most of the drought indices are not available for probabilistic assessment and forecasting, because of a pre-defined threshold to represent the drought severity. This study employed the Hidden Markov Model (HMM) to assess and forecast drought which can consider the inherent uncertainty embedded in daily precipitation data. Firstly, the Hidden Markov Model (HMM) based drought index (HMDI) was proposed to assess drought severity instead of using predefined threshold. Using daily precipitation data (1973-2012) of 55 stations in Korea provided by the Korea Meteorological Administration (KMA), the HMM was applied after changing the daily data to monthly and taking moving average of 3, 6, and 12 months windows. Secondly, using synthesized monthly precipitation data (2013-2100) based on Representative Concentration Pathway 8.5 (RCP 8.5) scenario which was provided by the Climate Change Information Center (CCIC), to assess future drought in Korea. When the Standardized Precipitation Index (SPI) which is one of representative indices using a pre-defined threshold, only one value can be used as a criterion to determine the drought severity. However, the HMDI can classify the drought condition considering inherent uncertainty in the data and shows the probability of each drought condition at a particular point in time even in future.