You are here:
VALUING AQUATIC ECOSYSTEM HEALTH AT A NATIONAL SCALE: MODELING ECOLOGICAL INDICATORS ACROSS SPACE AND TIME
Citation:
Hill, Ryan A., Chris Moore, J. Doyle, S. Leibowitz, P. Ringold, B. Rashleigh, AND E. Fergus. VALUING AQUATIC ECOSYSTEM HEALTH AT A NATIONAL SCALE: MODELING ECOLOGICAL INDICATORS ACROSS SPACE AND TIME. Social Cost of Water Pollution Workshop, N/A, Virtual, April 22 - 24, 2021.
Impact/Purpose:
Estimates of the value of aquatic ecosystems requires collaboration between social scientists (in identifying the valued biophysical units and the value people place on incremental changes in those units) and natural scientists (in describing the current and future states of the aquatic ecosystems). This work describes that collaboration.
Description:
Estimating economic benefits of Clean Water Act regulations requires quantifying the resultant changes in water quality and monetizing those changes. US EPA typically values water quality changes via stated preference (SP) using a water quality index (WQI) that reflects the suitability of lakes, rivers, and streams for a range of recreational uses. While the WQI has the desired qualities of being applicable over large geographic areas and a variety of waterbody types, it derives its value strictly from human use and thus omits, or perhaps conflates, a potentially important source of value in ecosystem health. To capture this portion of total economic value, we are pursuing a dual-index approach that will complement the WQI with a second indicator of aquatic ecosystem health. Here we describe our effort to find a suitable metric and develop models that provide the spatial resolution and predictive capability necessary for our current benefit estimation paradigm. To estimate household willingness to pay (WTP) for water quality improvements, we treat each census tract as a representative household and predict the change in water quality for every waterbody within a desired radius of the census tract centroid. This approach requires current and alternative water quality estimates at a very fine spatial scale. Indicators we might use for aquatic ecosystem health, such as the index for biotic integrity (IBI), and observed to expected ratio (O/E), are calculated using data from the National Aquatic Resource Survey (NARS). The spatially-balanced sampling design used by NARS allows the US EPA to provide statistically-representative summaries of the current status of populations of waterbodies at relatively large scales, such as ecoregions. While these data can represent the status of regions with known confidence, with about 25 observations per state, the data are too sparse for our application. We propose the use of empirical, machine learning approaches (e.g., random forest) to interpolate ecological condition metrics to several million waterbodies across the conterminous US to provide finer-scale current status estimates of ecological condition. Although similar approaches have been developed and applied in the past, this study raises several methodological questions that have not been addressed at this scale, such as: How do we communicate the accuracy and precision achieved in the descriptive ecological models to support economic analyses? How will the use of regional reference sites (i.e., sites in best-available condition) in the development of ecological indicators affect interpolations at regional boundaries and are there ways to mitigate these differences? Does the choice of spatial units make a difference? We will review past approaches for interpolating ecological condition and discuss our proposed approach to model and spatially-interpolate estimates for streams and lakes. We will also present results from a preliminary analysis that provides a proof-of-concept of our approach. An extension of this work necessary for benefits analysis is predictive modeling that will estimate water quality changes as a result of candidate regulation. We refer to these possible future ecosystem states as alternative futures. We are planning work to formulate the characteristics of the appropriate predictive models. For example, will that modeling need to be temporally dynamic or can it be static, what should be the spatial units predicted, is there a need for models of varying levels of data requirements, complexity and certainty?