COS 6-10
Response of stream ecosystems to climate change (IV): Stream temperature modeling

Monday, August 5, 2013: 4:40 PM
101G, Minneapolis Convention Center
Ryan A. Hill, Watershed Sciences and Ecology Center, Utah State University, Logan, UT
Charles P. Hawkins, Watershed Sciences and Ecology Center, Utah State University, Logan, UT
Jiming Jin, Watershed Sciences and Ecology Center, Utah State University, Logan, UT
David Tarboton, Civil and Environmental Engineering, Utah State University, Logan, UT

Stream temperature (ST) is a primary determinant of the distribution of stream species and hence stream community composition. We therefore need ST models that can accurately predict both the current spatial variation in STs as well as how STs will respond (ΔST) to climate change (CC). We used data (yrs 1999-2008) from 569 USGS ST sites across the conterminous USA to develop a random forest model to predict mean summer ST (MSST). We used air temperature (AT), base-flow index, topography, and soil characteristics as predictors (r2 = 0.87, RMSE = 1.9 °C). We then assessed how well modeled STs predicted the distribution of stream benthic invertebrates by comparing the performance of multi-taxon niche models developed with observed and predicted MSST at 92 USGS sites. We determined how well the ST model predicted CC-related ΔST by comparing (ANCOVA) the regressions of observed and predicted ΔST (1970s to present) on observed changes in AT (ΔAT). We then used spatially downscaled climate projections to predict end-of-century ΔST at 1,197 streams from which macroinvertebrate samples were collected. 


Measured and predicted STs performed similarly for predicting biotic distributions, assessed as the SD of the observed-to-predicted ratios of taxonomic composition at the 92 USGS sites (O/P SD = 0.16 and 0.15, respectively). Plots of taxon-specific, predicted capture probabilities versus measured and predicted STs were nearly identical indicating that predicted STs represent ecologically important thermal conditions as well as measured STs. ANCOVA showed no difference between the responses of measured and predicted ΔSTs to ΔAT. Additionally, ΔSTs were positively and significantly (p < 0.05) associated with ΔAT indicating that ST predictions were precise enough to detect ST change in response to climatic variability over relatively short time periods. The ST model predicted that average MSST in USA streams will increase by 1.7 °C by the 2090s and that the upper Midwest (+2.5 °C), northern Appalachians (+2.6 °C), Cascades (+1.8 °C), and northern Rocky Mountains (+1.8 °C) will be most vulnerable to CC. Streams in the southeastern Coastal Plains and coastal Texas were predicted to be least responsive to CC (average MSST = +0.5 °C). These ST projections provide important insight regarding which streams and rivers within the conterminous USA are likely to be thermally and ecologically most vulnerable to CC.