Department of Statistics Seminar
North Carolina State University

presents

Montserrat Fuentes

National Center for Atmospheric Research and North Carolina State University

"Spatial Prediction of Climate Fields for Ecological Models"

ABSTRACT

A basic problem in the geosciences is predicting a spatial surface from irregularly spaced measurements. Geophysical data sets are often characterized by large number of observations and a spatial structure that changes depending on location. These two features thwart the application of standard methods of spatial statistics. I will present a strategy for dealing with these problems that addresses the problems of computing standard kriging estimators models for large numbers of observations by discretizing the spatial locations and taking advantage of fast multiplication algorithms for covariance functions evaluated on a grid. A motivating problem for this work is the need to construct monthly precipitation fields for approximately 100 years for the United States. These fields along with other meteorological variables will then be used as the inputs to numerical models for vegetation (e.g. the Century model). The number of recording stations in this data base exceeds 16,000 sites and has irregular spatial coverage. To evaluate the sensitivity of the ecological models to their climate drivers we create ensembles of different versions of the predicted surface.

Friday, February 5, 1999

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.