A1524
Title: Extending Bayesian maximum entropy using random sets theory for interval-valued spatial data
Authors: Thamali Amanda Gamlath Dissanayakalage - Utah State University (United States) [presenting]
Abstract: Bayesian Maximum Entropy (BME) is a powerful framework for spatial prediction when both exact measurements (hard data) and uncertain observations (soft data) are available. One of the main challenges in BME is defining the prior covariance structure when the soft data are provided as intervals. In many applications, these intervals are simplified into single-point values (e.g., midpoint) but doing so could overlook important uncertainty information. A random-set based method is introduced for building prior covariances in BME. Each interval observation is viewed as the realization of a one-dimensional random set, and its dynamics are described by the center and radius. As a result, covariance expressions are obtained that systematically capture interval uncertainty in a relatively simple format. An extended simulation study demonstrates that this approach extends the standard BME framework and provides a more flexible and informative method for spatial prediction under uncertainty.
