Model-Based Sampling Design on Stream Networks

• Activity Number: 456
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Dale Zimmerman*+
• Companies: The University of Iowa

Abstract

The quality of inferences for statistical models on stream networks is affected substantially by the spatial configuration of the sites where measurements are taken. Moreover, a sampling design that yields high-quality inferences of one kind, e.g. estimation of regression parameters, may yield sub-par inferences of another kind, e.g. spatial prediction. In this talk, I propose design criteria corresponding to several kinds of inferences, and I obtain and compare optimal designs with respect to each criterion for a relatively small (6-point) design on a stream network of order 3. The effects that the assumed mean function, the spatial covariance function (flow-only versus flow-and-distance models), and the strength of spatial correlation have on optimal design are investigated.