Local landscape predictors of maximum stream temperature and thermal sensitivity in the Columbia River Basin, USA

Abstract

Stream temperature regimes are important determinants of the health of lotic ecosystems, and a proper understanding of the landscape factors affecting stream temperatures is needed for water managers to make informed decisions. We analyzed spatial patterns of thermal sensitivity (response of stream temperature to changes in air temperature) and maximum stream temperature for 74 stations in the Columbia River basin, to identify landscape factors affecting these two indices of stream temperature regimes. Thermal sensitivity (TS) is largely controlled by distance to the Pacific Coast, base flow index, and contributing area. Maximum stream temperature (Tmax) is mainly controlled by base flow index, percent forest land cover, and stream order. The analysis of four different spatial scales – relative contributing area (RCA) scale, RCA buffered scale, 1km upstream RCA scale, and 1km upstream buffer scale – yield different significant factors, with topographic factors such as slope becoming more important at the buffer scale analysis for TS. Geographically weighted regression (GWR), which takes into account spatial non-stationary processes, better predicts the spatial variations of TS and Tmax with higher R2 and lower residual values than ordinary least squares (OLS) estimates. With different coefficient values over space, GWR models explain approximately up to 62% of the variation in TS and Tmax. Percent forest land cover coefficients had both positive and negative values, suggesting that the relative importance of forest changes over space. Such spatially varying GWR coefficients are associated with land cover, hydroclimate, and topographic variables. OLS estimated regression residuals are positively autocorrelated over space at the RCA scale, while the GWR residuals exhibit no spatial autocorrelation at all scales. GWR models provide useful additional information on the spatial processes generating the variations of TS and Tmax, potentially serving as a useful tool for managing stream temperature across multiple scales.