Bayesian Hierarchical Spatial-Temporal Models for Wind Prediction

spatial-temporal processes;nonstationarity;nonseparability;Bayesian inference

Chen, Li ; Dr. Gary M. Lackmann, Committee Member,Dr. John F. Monahan, Committee Member,Dr. Jerry M. Davis, Committee Co-Chair,Dr. Sujit K. Ghosh, Committee Member,Dr. David A. Dickey, Committee Member,Dr. Montserrat Fuentes, Committee Chair,Chen, Li ; Dr. Gary M. Lackmann ; Committee Member ; Dr. John F. Monahan ; Committee Member ; Dr. Jerry M. Davis ; Committee Co-Chair ; Dr. Sujit K. Ghosh ; Committee Member ; Dr. David A. Dickey ; Committee Member ; Dr. Montserrat Fuentes ; Committee Chair

University:North Carolina State University

关键词: spatial-temporal processes;    nonstationarity;    nonseparability;    Bayesian inference;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/4963/etd.pdf?sequence=2&isAllowed=y

美国|英语

来源: null

【 摘 要 】

Wind fields along coastlines are composed of many features that are spatially and temporally complex in nature, and they are well recognized as nonstationary spatial-temporal processes. The difficulties of spatial-temporalmodelingcan be overcome by using separable spatial-temporal processes. This subclass of spatial-temporal processes has severaladvantages, including rapid fitting and simple extensions of many techniques developed and used in time series analysisand classical geostatistics. However, these separable models are not always realistic. An empirical test forseparability is proposed to aid in understanding spatial-temporal structure.Also a new class of nonseparable stationary covariance models is introduced. A special case in this new nonseparable covariance class is the Matérncovariance modelwith a scale parameter. The scale parameter is used to take into account the change of units between the spatial and temporal domain. For nonstationarity, we represent a nonstationary process as a mixture of local orthogonal stationary spatial-temporal processes. Compared to the moving-cylinder spatial-temporal kriging method by Haas, this new nonstationary model is simultaneously defined everywhere. The empirical test for separability can be used to determine the covariance structure of local stationary spatial-temporal processes.We apply the methodology to thewind fields over Chesapeake Bay. The goal of this application is to evaluate the ability of MM5, a mesoscale meteorological model, to forecast the wind fields over the Bay. Simple models are specified for observed wind data andMM5 output in terms of the underlying true wind process which is nonstationary and nonseparable, and estimate them in aBayesian way.Then the numerical model evaluation consists of comparing the observed wind values with theirpredictive distributions given the MM5 output. Moreover, the improved wind fields can be simulated via the Bayesianposterior distribution of the underlying true winds.