Many natural resource agencies routinely collect digital stream temperature data. However, a lack of coordinated sampling efforts typically results in temperature observations within a stream network that are spatially clustered, non-random, and autocorrelated. Recently, a new class of spatial statistical models that account for network topology (i.e., flow direction and volume) has been developed to address these issues (Ver Hoef et al. 2006).
To test the application of these models, we assembled a large stream temperature database (n = 780 records) spanning a 14-year period from 1993 – 2006 for the 6,900 km2 Boise River basin in central Idaho. Predictor variables that potentially affected stream temperature were quantified using automated GIS routines to obtain measurements of geomorphic features (e.g., elevation, channel slope, and valley confinement), satellite imagery to estimate solar radiation changes from wildfire, and climate stations to provide information on stream flow and air temperatures. Spatial models with four fixed-effect predictors and a mixed model spatial error structure accounted for 93% and 86% of the variation in summer mean and maximum stream temperatures, respectively, during the 14-year study period.
The spatial temperature models yield more accurate parameter estimates than traditional, non-spatial models and offer much improved predictive ability. These models are being used to map past and future thermal conditions and suitable habitat distributions for native salmonid species in this basin, but could also be used to better understand factors that affect stream temperature, determine compliance with water quality standards, or optimize temperature sampling designs. Preliminary results from this project have been presented at several scientific meetings and a peer-reviewed manuscript has been drafted for publication.