Title
|
* Model-Assisted and Design-Based Sampling Approaches In Sampling of Natural Resources
|
Date / Time / Room
|
Sponsor
|
Type
|
08/14/2002
2:00 PM -
3:50 PM
Room: S-New York Ballroom A
|
Section on Statistical Consulting*, Section on Statistics & the Environment*, Section on Survey Research Methods*
|
Topic Contributed
|
Organizer:
|
Loveday L. Conquest, National Research Center for Statistics and the Environment/University of Washington
|
Chair:
|
Loveday L. Conquest, National Research Center for Statistics and the Environment/University of Washington
|
Discussant:
|
3:25 PM - Gretchen Moisen, U.S. Department of Agriculture
|
Floor Discussion
|
3:45 PM
|
Description
An important issue for sampling of natural resources is that of optimizing designs for population and process inference. As an illustration, a sampling strategy for lake acid neutralizing capacity (ANC) may be aimed at estimating the average ANC at a point in time or the trend through time (population), or predicting the amount of ANC at an unsampled lake or future ANC levels at sampled or unsampled lakes (process).
Model-based designs can be optimized for process estimation; e.g., a design resulting from minimizing kriging variance can provide good predictors for unsampled units. Designs that are not model-based can be optimized for design-based objectives; e.g., a spatially stratified design may admit design-unbiased estimators of average ANC over a region. These designs may be "in conflict" in the sense that it is unlikely that a single design will result in optimality for both strategies. However, model-assisted sampling techniques may provide a framework to build designs optimal to both objectives. For example, a particular model can suggest the inclusion probabilities for the sampling units. The four talks will discuss various aspects and applications of design-based and model-assisted approaches for monitoring the natural environment, including salmon in Oregon coastal streams, and monitoring for streams/rivers and forest resources.
|
|