Sensitivity of Air2stream Stream Temperature Prediction Accuracy to the Length of the Calibration Period

Abstract

Stream temperature is widely considered a “master variable” in aquatic ecosystems. Stream temperature models are often used to evaluate historic thermal habitat suitability where measurements are lacking and to quantify the potential effects of future climate scenarios. A major barrier to the widespread application of models is the sparse coverage and limited duration of observed stream temperature records to support model calibration. This study quantified the sensitivity of the hybrid stream temperature model Air2stream to the length of the calibration record. The study involved 23 hydrometric stations in British Columbia, Canada, including rain-dominated, snow-dominated, hybrid rain-snow, and nivo-glacial regimes within regulated and unregulated systems. Air2stream was calibrated using two approaches. The first used all possible subsets of consecutive years ranging from one to eight years in length, depending on data availability at each site, followed by validation on the data not used for calibration. The second used the years 2021 and 2022 for validation after using all possible subsets of consecutive years up to and including 2020 for calibration. Air2stream generally performed best for rain-dominated and hybrid regimes and worst for nival regimes. Even just two years of calibration data produced validation root-mean-square errors of less than 2ÅãC for all but one site, regardless of hydrologic regime or regulation status, and calibrating with three or four years provided similar performance in validation to longer calibration periods.