You are here:
Fish Assemblage Indicators for the National Rivers and Streams Assessment: Performance of model-based vs. traditionally constructed multimetric indices
Citation:
Peck, D., K. Blocksom, AND A. Herlihy. Fish Assemblage Indicators for the National Rivers and Streams Assessment: Performance of model-based vs. traditionally constructed multimetric indices. National Water Quality Monitoring Conference, Tampa, FL, May 02 - 06, 2016.
Impact/Purpose:
The development of multimetric indices (MMIs) for use in assessing the ecological condition of rivers and streams has advanced in recent years with the use of various types of modeling approaches to factor out the influence of natural variability and improve performance. New modeling approaches may be constrained by data requirements and difficulty of applying the model construction process to new data. This necessitates a careful evaluation of the degree of improvement provided by alternative modeling approaches. We used data collected for the 2008-09 National Rivers and Streams Assessment (NRSA)to develop two variants of a fish MMI for use in assessing ecological condition within aggregated ecoregions of the conterminous USA. We will be comparing the performance of each MMI variant to determine if model-based MMIs can out-perform the simpler MMI variants enough to warrant their continued use in NRSA despite the increased complexity in construction, cost and effort associated with constructing the models, and constraints in their transferability to outside users.
Description:
The development of multimetric indices (MMIs) for use in assessing the ecological condition of rivers and streams has advanced in recent years with the use of various types of modeling approaches to factor out the influence of natural variability and improve the performance. Associated with this improved performance are several constraints related to the required data and the ease of applying the construction process to new data. We wanted to determine if modeled MMIs for fish assemblages demonstrated better performance over traditional MMIs developed from the same data. We used data collected for the 2008-09 National Rivers and Streams Assessment (NRSA) to develop two variants of a fish MMI for use in assessing ecological condition within aggregated ecoregions of the conterminous USA. We classified candidate metrics into eight categories: Nonnatives, taxonomic composition, habitat guild, migratory strategy, reproductive guild, taxa richness, tolerance to disturbance, and trophic guild. We constructed a model-based MMI by adjusting metric responses using regression models constructed with a random forest algorithm. We constructed a traditional MMI by adjusting candidate metrics only for watershed area using linear regression. For both MMI variants, we evaluated candidate metrics for range, comparison of least-disturbed sites to more highly disturbed sites (responsiveness), and repeatability (variability among sites vs. variability within sites). For each MMI variant, we used metrics that had passed these screens and calculated all possible MMIs that included at least one metric from each category. We evaluated all MMIs for responsiveness, repeatability, and pairwise correlation (maximum and mean) among component metrics. For each variant, we selected a final MMI that had high responsiveness, a reasonable value for repeatability, low mean pairwise correlation among component metrics, and a maximum pairwise correlation among component metrics that was <0.7. We present the performance characteristics of each MMI variant and discuss implications of adapting either MMI variant for use in current and future NRSA efforts. This is an abstract for a proposed presentation and does not necessarily reflect EPA policy. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.