Topic：Bayesian Area-to-point Kriging with Expert Elicitation of a Prior for the Point Support Variogram

Speaker：Gerard B.M. Heuvelink, Department of Environmental Sciences, Wageningen University, Netherlands

Brief Introduction to the report

Spatial disaggregation is required when environmental models need input at a finer spatial support than available or when models produce output at a coarser support than desired by end-users. Area-to-point kriging provides an elegant geostatistical solution. It yields ‘mass-preserving’ predictions at point support and also quantifies the associated prediction uncertainty. However, it requires that the point-support variogram is known, which cannot uniquely be derived from only block observations. This is because the effect of the nugget variance vanishes under spatial aggregation. In this work we used expert elicitation to derive informative priors of the point-support variogram parameters. Next block observations were used in a Bayesian update to compute the joint posterior distribution, using Markov chain Monte Carlo. Area-to-point kriging and stochastic simulation were then applied with random sampling from the posterior distribution, thus ensuring that uncertainty about the point-support variogram is represented in the simulations. A case study addressed the spatial disaggregation of a MODIS air temperature image at 5km×5km grid support from 12 a.m. on August 1, 2012, for the Gelderland province, the Netherlands.

Time：9:00a.m. 22, Nov. 2013

Venue：Room 2321, IGSNRR

Host：Prof. GE Yong

Hosted by State Key Laboratory for Resource and Environment Information System