Authors: Halil Dertli, Daniel Hayes, Troy Zorn

Many models of stream temperature dynamics are available, but guidance on best approaches vary among literature sources. In our review of the literature, we found that one challenge is that modelers often utilize different levels of data aggregation or temperature collection intervals. More specifically, modern temperature loggers can be set to collect data at very short intervals (e.g., every 15 minutes), but some models operate on data aggregated up to daily or even weekly intervals. Thus, we sought to determine how a suite of models perform on the same data, but aggregated at different time intervals. The suite of models we evaluated ranged from relatively simple models of temperature change between two monitoring stations driven by the difference in air temperature from water temperature, to more complex models that included additional drivers such as day length, sun angle, discharge, and changes in discharge between monitoring stations. In general, all models performed better (as measured by the R^2 between observed and predicted temperature change) for data aggregated at weekly or daily time intervals as compared to hourly or sub-daily intervals. We noted, however, that the rate of increase in predictive capacity varied among models, leading to differences in the model with the highest R^2 for different time aggregations. For the hourly time interval, a relatively complex model performed best as measured by AIC and model weights. For broader time intervals, this model generally still remained as the best choice, but simpler models tended to receive higher weights than they did for hourly data. For the best model, parameter estimates varied across levels of time aggregation, indicating that a model developed at one temporal scale cannot be directly applied to another temporal scale. We recommend that investigator pay more attention to the temporal scaling of data collection and aggregation before comparing results across studies.