You are here:
SMALL AREA ESTIMATION OF INDICATORS OF STREAM CONDITION FOR MAIA USING HIERARCHICAL BAYES PREDICTION MODELS
Citation:
Olsen, A R., M. Handcock, D. L. Stevens, S. Carroll, J. C. Sifneos, A. T. Herlihy, AND D J. Chaloud. SMALL AREA ESTIMATION OF INDICATORS OF STREAM CONDITION FOR MAIA USING HIERARCHICAL BAYES PREDICTION MODELS. Presented at EMAP Symposium, Kansas City, MO, May 7-9, 2002.
Description:
Probability surveys of stream and river resources (hereafter referred to as streams) provide reliable estimates of stream condition when the areas for the estimates have sufficient number of sample sites. Monitoring programs are frequently asked to provide estimates for areas that are smaller than initially planned. Typically, too few sites are available in the areas to produce estimates with even reasonable precision. This is known as the small area estimation problem in survey sampling. This paper describes one small area estimation approach based on hierarchical Bayes prediction models. EMAP's Mid-Atlantic Integrated Assessment (MAIA) stream survey from 1993-1998 provides data on over 600 sites with water quality, benthic macroinvertebrate, and fish indicators. Watersheds were delineated for each site and watershed variables were calculated. Our objective is to produce estimates for approximately 100 4th-field hydrologic units (HUCs). Our first step constructs a model relating watershed variables to an indicator of stream condition. A hierarchical Bayes model is specified that has watershed predictor variables and two forms of spatial correlation for the model error terms. One term uses the proportion of stream length that is common between two watersheds and the other term uses the euclidean distance between the two sites defining the watersheds. The model is used to predict stream condition at prediction sites selected as a very large probability sample of streams, approximately 200 sites within each HUC. For each of these sites, watersheds are automatically delineated and watershed variables determined. Predictions from the hierarchical Bayes model are used in place of measured indicators in the survey design estimation process for each HUC.