Air temperature data source affects inference from statistical stream temperature models in mountainous terrain

• Published in: Journal of hydrology: X Vol. 22; p. 100172
• Main Authors: Isaak, Daniel J., Horan, Dona L., Wollrab, Sherry P.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.01.2024; Elsevier
• Subjects: Air temperature; Gridded climate data set; Mountains; Statistical model; Stream temperature; Thermal sensitivity; Mountains; Stream temperature; Thermal sensitivity; Air temperature; Statistical model; Gridded climate data set
• ISSN: 2589-9155; 2589-9155
• DOI: 10.1016/j.hydroa.2024.100172

•Temperature is a master variable controlling many processes in mountain streams.•Statistical stream temperature models using air temperature are often developed.•A variety of air temperature data sources are used as covariates in these models.•The performance of multiple sources was assessed at 13 mountain stream sites.•Model performance and bias were affected by air temperature data source. Instream temperatures control numerous biophysical processes and are frequently the subject of modeling efforts to understand and predict responses to watershed conditions, habitat alterations, and climate change. Air temperature (AT) is regularly used in statistical temperature models as a covariate proxy for physical processes and because it correlates strongly with spatiotemporal variability in water temperatures (Tw). Air temperature data are broadly available and sourced from sensors paired with Tw sites, remote weather stations, and gridded climate data sets—often with limited recognition of the tradeoffs these sources present and how microclimatic variation in topographically complex mountain environments could affect model inference. To address these issues, we collected daily Tw records at 13 sites throughout a mountain river network, linked the records to AT data from 11 sources available across much of North America, and fit linear regression models to assess predictive performance and the consistency of parameter estimation. Although the predictive accuracy of these models was generally high, estimates of the AT slope parameter, which is commonly interpreted as thermal sensitivity, varied substantially depending on the AT data source. These results have implications for the comparability of estimates among Tw studies and highlight the challenges that modeling stream temperatures in mountain landscapes presents. Although no AT data source is ideal, some are more advantageous than others for specific use cases and we provide general recommendations on this topic.