Evaluation of Non-Homogeneous Hidden Markov Chain Model in Improving Monthly Precipitation Forecasts

Monthly to seasonal precipitation forecasts are important in water resources management. Hidden Markov Chain Model (HMM) has wide application in precipitation simulation due to its simplicity and advancement in computing techniques. One disadvantage of HMM is that it lacks the flexibility to model the effects of external factors on transition probability of hidden states and/or emission distributions. In this study, a recently developed Non-Homogeneous Hidden Markov Chain Model (NHMM) that explicitly incorporates observed covariates, e.g., climate index, is examined. It owns the flexibility of modeling the predictors' influence on transition dynamics of hidden states as well as the emission distribution. Nonparametric Bayesian approach is used to draw samples from the posterior distribution of model parameters. The proposed approach is tested for three rainfall stations in service area of a regional water supply agency, Tampa Bay Water, in southeastern United States. NHMM is firstly examined and compared to HMM in simulating all historical data. It is then evaluated in retrospective mode by providing 3-month ahead precipitation forecasts. To illustrate its application, improved precipitation forecasts are used as input to a real-time decision support tool (DST) for water supply system.