부산대학교 도서관

The creeping characteristics of drought make it possible to mitigate drought’s effects with accurate forecasting models, which provide important information about drought (onset, duration, severity, etc.) before their occurrence. Therefore, early drought forecasting and assessment becomes the effective strategy to forewarn drought in drought disaster mitigation and water resources operation and management systems. In this study, we employed a discrete-time finite state-space hidden Markov model (HMM) as a basic framework and explored its various applicability as a forecasting tool, similarity measure, and a classifier for drought forecasting and assessment. In order to implement the different applications of HMM in drought analysis, this dissertation provided three case studies to evaluate the performances of the proposed models. The first study proposed a new probabilistic scheme to forecast droughts, i.e., the HMM-PF, which used a HMM aggregated with the future precipitation forecasts (PF) using a multi-model ensemble (MME) method provided by the Asia-Pacific Economic Cooperation Climate Center (APCC). The standardized precipitation index (SPI) was employed to analyze the drought conditions over five selected stations in South Korea. The new scheme utilized a weight-corrected post-processing based on the precipitation forecasts transformed SPI (PF-SPI) for the HMM to perform the one-month ahead probabilistic forecasting of SPI considering uncertainties. The point forecasts which were derived as the mean values of the HMM-PF forecast, as measured by forecasting skill scores, were much more accurate than those from conventional models. We also used probabilistic forecast verification criteria and found that the HMM-PF provided a probabilistic forecast with satisfactory evaluation for different drought categories. The results showed that the HMM-PF had good potential in probabilistic drought forecasting.In the second study, we proposed a HMM as a similarity measure framework for a modified analogue forecasting (MAF) approach for meteorological droughts in South Korea. The SPI is also selected as the drought index. Within the framework, the likelihood estimator was used as similarity measure to select the past SPI analogues. The MAF approach was conducted on the selected analogues to make forecasts at lead times of one and three months. The proposed model was applied to the five stations selected in the first study with precipitation data from year 1973 to 2016. Results showed significant improvement in the forecast skill of the proposed model achieved from the reference forecasts, and satisfactory performance for numerical SPI forecasting and categorical drought forecasting. The results also suggested that with lead time up to three months, the proposed model was able to provide useful information in determining the future drought categories for early drought warning. Instead of directly using the HMM, the third study employed an extension of the HMM, i.e., the Dynamic Naive Bayes Classifier (DNBC) for multi-index probabilistic drought assessment by integrating various drought indices (i.e., SPI, Streamflow Drought Index (SDI), and Normalized Vegetation Supply Water Index (NVSWI)) as indicators of different feature spaces contributing drought occurrence. Overall results showed that the proposed model is able: (1) to account for various physical forms of drought in probabilistic drought assessment, (2) to accurately detect the drought onset and termination, (3) to provide useful information on assessing the potential drought risk, and (4) to precisely capture the drought persistence in terms of drought state transition probability in drought monitoring. The major contribution of this study was the development of the HMM for various applicabilities in drought analysis. Its powerful and flexible mathematical structure is able to make statistical inferences on the stochastic drought evolution process, and has been verified as an effective tool in drought assessment and forecasting.