National Aquatic Resource Surveys: Use of Geospatial Data in Their Design and Spatial Prediction at Nonmonitored Locations

• Activity Number: 521
• Type: Topic Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Anthony R. Olsen* and Thomas M. Kincaid and Marc H. Weber and Ryan A. Hill and Scott G. Leibowitz
• Companies: EPA Western Ecology Division and EPA Western Ecology Division and EPA Western Ecology Division and ORISE U.S. EPA Western Ecology Division and EPA Western Ecology Division
• Keywords: Aquatic Resources ; Spatial Survey Design ; Spatial prediction ; NARS ; random forest model

Abstract

The National Aquatic Resource Surveys (NARS) are surveys conducted by the US Environmental Protection Agency working in collaboration with states, tribal nations and other federal agencies. The surveys are conducted for lakes, streams, coastal waters, and wetlands. Fundamental to the survey design and analysis is the dependence on geospatial data from the National Hydrography Dataset (NHD) as a basis for the sample frame for three of the four surveys. The R package "spsurvey" is used to select a spatially-representative sample from this geospatial sample frame based on a generalized random tessellation stratified design (GRTS). Geospatial information is used in the local neighborhood variance estimator that was developed for GRTS designs, which has been shown to provide better coverage probability than the Horvitz-Thompson estimator. NARS studies provide nationally consistent and scientifically-defensible assessments of the nation's waters. We also are developing an approach to predict stream condition at non-monitored locations. Preliminary random forest models have been developed to predict stream condition for all ~5.4 million km of streams in the 48 contiguous states.