As climatic conditions change globally, so too will stream thermal regimes, with implications for water quality and habitat suitability for aquatic life. Stream temperature measurements are sparse in many regions, motivating the development of models that are able to extrapolate to past and future climatic conditions to support decision-making for aquatic resource management. This study assesses the performance of air2stream, a hybrid, at-a-site stream temperature model that was developed to simplify the data requirements of process-based models while maintaining their predictive performance. The air2stream model requires only time series of daily mean air temperature and stream discharge as input variables, and was calibrated for 23 streams in British Columbia, Canada, using data recorded at Water Survey of Canada gauging stations for the available periods of record up to 2020. Daily mean air temperature time series were interpolated to each monitoring site from the ERA-5 gridded surface data product. Air2stream was validated with data from the years 2021 and 2022, which included an extreme summer heat wave and autumn drought conditions that fall outside the range of conditions observed during the calibration period. The validation results were compared to those of a set of linear mixed-effects models with the same predictor variables, as well as a simplified version of air2stream that only uses air temperature as an input variable. The air2stream model produced higher errors during the extreme weather conditions compared to the calibration period, though its performance under extreme conditions overall remained superior to that of the statistical models and the simplified air2stream model. The results highlight the importance of representing hydrological and thermal processes and their seasonal variation in models for predicting stream temperature under changing climatic conditions.
