A1358
Title: A random quantile approach for prior covariance estimation in Bayesian maximum entropy
Authors: Kinspride Duah - Utah State University (United States) [presenting]
Yan Sun - Utah State University (United States)
Maha Moussa - Utah State University (United States)
Abstract: Bayesian maximum entropy (BME) provides a flexible framework for spatial prediction by integrating precise measurements with uncertain information. However, conventional BME (CBME) often assumes homogeneous uncertainty, limiting its performance when data exhibit strong heterogeneity such as skewness and heavy tails. A quantile-based BME (QBME) approach is introduced that leverages quantile information to better represent asymmetric uncertainty in soft data. By incorporating multiple quantiles into the prior specification, QBME improves the characterization of uncertainty and enhances prediction accuracy, particularly in data-sparse regions. A comprehensive simulation study evaluates QBME against CBME across a range of controlled scenarios. Results show that QBME consistently outperforms CBME under diverse data conditions and error metrics. The proposed method is further demonstrated using snow water equivalent (SWE) data from Montana, United States, which include both precise and interval-valued observations. QBME more effectively captures large-scale spatial patterns while preserving local variability. These findings highlight QBME as a robust extension of BME for spatial prediction with asymmetric and uncertain data.
