A1744
Title: Prediction error assessment with contaminated spatial data
Authors: Rahasya Bharaniah - Kansas State University (United States) [presenting]
Juan Du - Kansas State University (United States)
Abstract: Accurate prediction of the underlying signal from spatial data is critical in applications such as epidemiology, agriculture, and environmental science. In practice, observations are often contaminated by measurement error or systematic bias, which may also exhibit spatial dependence, leading to degraded prediction performance. Prediction methods that aim to recover a less noisy, or true, spatial process using contaminated data are studied. Several kriging-based strategies are developed that incorporate auxiliary information from related sources. Performance is assessed with emphasis on uniform prediction error over the spatial domain. Uniform error bounds are derived, and finite-sample performance is examined through simulation studies under varying contamination levels. The proposed approach is applied to Kansas ozone data, where ground-level measurements are limited but satellite-derived column-level data are abundant, to illustrate potential improvements in prediction accuracy and design-based control of uniform prediction error compared to traditional kriging methods.
