You are here:
Assessing the biological condition of streams and rivers in the contiguous United States with joint species distribution models
Citation:
Kopp, D., J. Stoddard, Ryan A Hill, J. Doyle, Phil Kaufmann, A. Herily, AND Steve Paulsen. Assessing the biological condition of streams and rivers in the contiguous United States with joint species distribution models. Ecological Society of America Conference, Portland, OR, August 06 - 11, 2023.
Impact/Purpose:
A central objective of biological assessments is to determine whether an assemblage of organisms is comparable in terms of diversity and composition to that of natural habitats in the region. Typically, the assemblage expected to occur under natural conditions is derived from a suite of reference sites that are minimally disturbed by human activities. Unfortunately, because of the ubiquity of human alterations, reference sites meeting minimally disturbed criteria can be difficult to find. In this presentation we propose an alternative approach to establishing biological benchmarks using joint species distribution models and data collected by the National Rivers and Streams Assessment program. We found that this approach, based on taxon-environment relationships, could elucidate new patterns in the biological condition of rivers and streams in the contiguous United States. Importantly, this study supports the development of new biological assessment techniques by establishing biological benchmarks without relying on reference sites.
Description:
Using reference sites to establish appropriate biological benchmarks to assess biological condition is challenging because of the ubiquity of anthropogenic impacts. We circumvent the need for reference sites using joint species distribution models to predict the benthic macroinvertebrate assemblage expected (E) to occur if nutrients, salinity, physical habitat, and climate were minimally influenced by human activities. We then compare the observed (O) assemblage to E and assess biological condition for each site. We used random forest models to relate natural and anthropogenic factors to observed nutrient and salinity concentrations and then predicted minimally disturbed values for each stie by setting the anthropogenic effects to zero. We obtained minimally disturbed values for physical habitat and climate from published sources. Depending on the ecoregion, random forest models explained 22 to 79% of the variation in the chemical gradients and, in general, hindcasted values for all gradients were lower than present-day observations. The species distribution models fitted using present-day conditions accurately explained occurrences (AUC > 0.70) and assemblage level characteristics. We then substituted minimally disturbed values for each gradient as predictor variables in the fitted models and generated a distribution of plausible observed-to-expected ratios (O:E) by comparing O to 1250 random samples from the posterior predictive distribution of E. We then measured biological condition as the probability that the observed assemblage is different from the assemblage expected to occur if the site was in its minimally disturbed state. We found that the assemblages observed at 57% of sites across the contiguous United States had a probability > 0.5 of being different from the expected assemblage. This presentation demonstrates an alternative method for establishing biological benchmarks without relying on reference sites and we discuss its advantages, disadvantages, and complexities. The views expressed in this presentation are those of the authors and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.