A0964
Title: Integrating multi-source geospatial information using Bayesian maximum entropy
Authors: Kinspride Duah - Utah State University (United States)
Yan Sun - Utah State University (United States) [presenting]
Brennan Bean - Utah State University (United States)
Abstract: Environmental data are often imprecise due to various limitations and uncertainties. As a result, they often consist of a combination of both precise and imprecise observations, referred to as hard and soft data, respectively. Often, in practice, soft data are characterized as intervals in a simple form to properly preserve the underlying imprecision. Bayesian maximum entropy (BME) is a generalized spatial interpolation method that processes both hard and soft data simultaneously to effectively account for both spatial uncertainty and measurement imprecision. A rigorous evaluation is presented to compare the performances of BME and kriging through both simulation and a case study of reliability-targeted design ground snow load (RTDSL) prediction in Utah. The dataset contains a mixture of hard and soft-interval observations, and kriging uses the soft-interval data by extracting the midpoints in addition to the hard data. The cross-validated results show that BME outperforms kriging on multiple error metrics. These results highlight the superior prediction accuracy of BME, particularly in the presence of soft data and/or non-Gaussian hard data.
