Dynamic Spatial Panel Autoregressive Models with Alternative Spatial Weight and Application to Wind Energy Potentials
No Thumbnail Available
Date
2024-07-31
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
The choice of appropriate spatial weight is an important step in defining the spatial dependence structure. So far, researchers have mainly used sparse weights that were constructed based on Euclidean distances between points in a plane. In this dissertation, a dense spatial weight has been defined on the separation distances between locations, where the distances are partitioned into distinct neighborhood sectors according to the quantile values. Weights of the respective inverse quantile of separation distances are assigned to each neighborhood sector. The use of such dense spatial weight induces spatial auto-correlations
in the spatial panel data of neighboring locations as well as in the errors. Various relevant spatial panel autoregressive models have been compared alternative spatial weights using simulated and empirical dataset. The modeling of spatiotemporal process has been challenging as it has to deal with methods of joint modeling that incorporate spatial dependence and evolution over time. Also, the implementation of a joint likelihood perspective of spatiotemporal process is challenging due to invalidity of some underlying assumptions including spacetime separability of covariance function. As a solution to such problems, simultaneous spatial panel model is used after the panel component is imposed on the spatial weight using Kronecker product to perform hierarchical Bayesian inferences via MCMC Gibbs methods. The hierarchical Bayesian estimates are applied to a real dataset of wind power and to assess the effects of climate covariate and topographic elements on wind power potentials in Ethiopia. Comparison of the models reveals that the combined dynamic spatial autoregressive panel (SAC) model with random effects specification has the best predictive performance for the alternative spatial weight as compared to the lag or error spatial panel autoregressive models for the proposed spatial weight. Finally, the mean prediction of wind speed has been performed at unobserved sites using the proposed methods and its respective wind power has been calculated using Weibull function at specific values of climate and geographic covariates
Description
Keywords
Spatial Weight, Climate Covariates, Topographic Elements, Hierarchical Bayesian Inference, MCMC Method, Dynamic Spatial Panel Autoregressive Model, Wind Speed And Wind Power Predictions