A model to predict stream water temperature across the conterminous USA
Stream water temperature (ts) is a critical water quality parameter for aquatic ecosystems. However, ts records are sparse or nonexistent in many river systems. In this work, we present an empirical model to predict ts at the site scale across the USA. The model, derived using data from 171 reference sites selected from the Geospatial Attributes of Gages for Evaluating Streamflow database, describes the linear relationship between monthly mean air temperature (ta) and ts. Multiple linear regression models are used to predict the slope (m) and intercept (b) of the ta–ts linear relation as a function of climatic, hydrologic and land cover characteristics. Model performance to predict ts resulted in a mean Nash–Sutcliffe efficiency coefficient of 0.78 across all sites. Application of the model to predict ts at additional 89 nonreference sites with a higher human alteration yielded a mean Nash–Sutcliffe value of 0.45. We also analysed seasonal thermal sensitivity (m) and found strong hysteresis in the ta–ts relation. Drainage area exerts a strong control on m in all seasons, whereas the cooling effect of groundwater was only evident for the spring and fall seasons. However, groundwater contributions are negatively related to mean ts in all seasons. Finally, we found that elevation and mean basin slope are negatively related to mean ts in all seasons, indicating that steep basins tend to stay cooler because of shorter residence times to gain heat from their surroundings. This model can potentially be used to predict climate change impacts on ts across the USA.