Conditional Simulation of Nonstationary Spatial-Temporal Processes

• Activity Number: 643
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Joseph Guinness*+
• Companies: The University of Chicago
• Keywords: Periodogram ; Time series ; Nonstationary ; Spatial-temporal ; Simulation

Abstract

Given a set of observations of a spatial-temporal process, one may be interested in predicting or conditionally simulating the process at unobserved locations or times. For example, when a climate model requires gridded data as an input, but there exist only irregularly spaced observations, the process must be interpolated to a grid, and conditional simulations provide a method for quantifying the uncertainty in the predictions. The temperature data from the Atmospheric Radiation Measurement program for the southern Great Plains region consists of regularly spaced time series data from a set of fixed but irregularly spaced observation locations, and we develop computational methods and a modeling framework for handling data with this structure. The high frequency with which the observations are recorded in time allows us to estimate models that are nonstationary in the time domain, and the nonstationary models fit the data substantially better than their stationary counterparts do. Our methods result in accurate predictions and realistic conditional simulations of temperatures in the region at a one minute time resolution over a one month period.