A0867
Title: Scalable spatial skew-Gaussian process models
Authors: Kapil Gupta - Indian Institute of Management, Kozhikode (India) [presenting]
Abstract: Spatial data often exhibit skewness and heavy tails that violate the Gaussian assumptions underpinning traditional geostatistical methods. While transformation-based approaches such as trans-Gaussian kriging attempt to address this, they often suffer from bias and theoretical limitations. A Bayesian spatial modeling framework is proposed using a skewed Gaussian process model. The full conditional posteriors are derived, their normalization constants are analyzed, and closed-form expressions are provided. The model enables scalable inference through tractable likelihood decomposition and efficient MCMC sampling. The proposed methodology is validated through comprehensive simulation studies, demonstrating improved predictive performance and robustness over existing approaches.
