Sutikno, Dedy Dwi Prastyo : Spatial Extreme Value Using Bayesian Hierarchical Model For Precipitation Return Levels Prediction
Dedy Dwi Prastyo S.Si., M.Si.
3RD INTERNATIONAL CONFERENCE ON RESEARCH, IMPLEMENTATION AND EDUCATION OF MATHEMATICS AND SCIENCE
Extreme value theory; Peaks Over Threshold; Bayesian hierarchical model; spatial; return level
Extreme precipitation is a rare natural phenomenon categorized as one of extreme climates indicator. It leads to natural disasters such as floods and landslides. Prediction of precipitation return level, i.e. quantile which has probability 1/m of being exceeded in m period, become quite important to reduce the negative impact caused by this extreme event. The link between observation recorded on a particular time frame and quantities of longer time scales such as return level is provided by Extreme Value Theory (EVT) commonly used to learn the behaviors of extreme events. Given the observations are recorded from several locations, the extreme events at different locations are driven by geographical and climatologically factors. Unfortunately, the data of these factors are not always available. In this study, the spatial Bayesian hierarchical model (BHM) was employed to update the information in the likelihood that is not fully described by those unobservable covariates. The proposed method was applied to predict the extreme precipitation return level in Lamongan district, East Java, Indonesia. The Peak Over Threshold (POT) scheme was used to obtain extreme observations. The prior distribution was employed to update the likelihood of Generalized Pareto Distribution as an asymptotic distribution
of exceeding resulted from POT procedure. The empirical results showed that the return level got higher for longer periods.