You are here:
Improving models of biological condition to support existence valuation of freshwater ecosystems
Citation:
Hill, Ryan A, J. Doyle, E. Fergus, S. Leibowitz, Chris Moore, P. Ringold, J. Renee Brooks, AND Phil Kaufmann. Improving models of biological condition to support existence valuation of freshwater ecosystems. Fifth Annual Workshop on Integrated Assessment Models and the Social Costs of Water Pollution, Washington D.C., DC, September 07 - 09, 2022.
Impact/Purpose:
The EPA currently uses the Water Quality Ladder (WQL) to estimate benefits associated with proposed policies under the Clean Water Act. The WQL is primarily designed to capture use-based values and may not fully capture all benefits associated with water quality improvements. Current work by EPA economists seeks to capture the existence value of freshwater ecosystems more fully and accurately in the conterminous U.S. As part of this effort, ORD ecologists have developed models to predict a measure of biological condition that will complement the WQL in an upcoming national state preference survey. However, several issues were identified with these initial models, including marked shifts in biological condition at ecoregion boundaries that are the result of how the metric is constructed. This presentation will detail efforts to improve these models, including strategies to remove harsh ecoregion boundaries and improve overall model performance. The results of this work will support efforts by NCEE economists to improve national valuation of freshwater ecosystems when conducting analysis of proposed Clean Water Act regulations and contributes to a FY22 subproduct under SSWR.1.2.2.
Description:
The U.S. Environmental Protection Agency (EPA) is developing a national stated preference study to more accurately capture the existence portion of total economic value of water quality regulation within the conterminous U.S. In preparation for the stated preference survey, we compared metrics of biological condition. This comparison used focus groups to select an index of taxonomic completeness for use in the survey because, along with other benefits, it was more easily understood by participants. The selected index uses modeling to estimate the composition of aquatic taxa expected (E) at an assessed site in the absence of significant human-related stressors. This expectation can then be compared to the composition of observed (O) taxa as a ratio (O/E) to estimate taxonomic loss due to human activities in a watershed. However, recent work has also shown that people’s willingness to pay for a given improvement in water quality generally depends on baseline conditions near them. Accounting for this dependence requires water quality information on streams and lakes from across the conterminous U.S. and at a density sufficient to estimate conditions near survey respondents; a feature that is currently missing among existing national datasets of biological condition. To impart the coverage needed for a national analysis of Clean Water Act regulations, we used several thousand samples from EPA’s National Rivers and Streams Assessment (NRSA) and National Lakes Assessment to model and spatially interpolate observed values of macroinvertebrate O/E to 1.1 million and 290,000 unsampled streams and lakes, respectively. These initial models had poor-to-moderate performances and displayed marked shifts in interpolated O/E scores at ecoregional boundaries; a consequence of the use of reference sites that are specific to each region to construct the original O/E assessments. To address these shortcomings, we returned to the original macroinvertebrate taxonomic data and developed 212 individual models to spatially predict the probability of capturing each taxon at any given stream within the conterminous U.S. Critically, these models were developed using all sites (i.e., stressed-to-reference conditions) and were national in scale (i.e., no regions used). We calculated the area under the receiver operating curve (AUC) to assess each model. AUC compares all pairwise combinations of true presences and true absences and counts the proportion of times the predicted probability of the former (phenomenon of interest) is greater than the latter. Assessments of the national models found that, on average, they were better at predicting taxa occurrences within regions than regional models we developed for comparison. Further, they outperformed the original models used by EPA to calculate E in NRSA. Comparisons with the original NRSA O/E ratio also suggested that our national models performed almost as well at predicting conditions at reference sites. Future work will expand this effort to lakes as well as explore using the national models to predict reference conditions at stressed sites. The views expressed in this abstract are those of the author(s) and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.