Jeffrey Michael Sadler , Alison Paige Appling , Jordan S. Read, Samantha Kay Oliver , Xiaowei Jia , Jacob Zwart , Vipin Kumar

Deep learning models can accurately predict many hydrologic variables including streamflow and water temperature; however, these models have typically predicted hydrologic variables independently. This study explored the benefits of modeling two interdependent variables, daily average streamflow and daily average stream water temperature, together using multi-task deep learning. A multi-task scaling factor controlled the relative contribution of the auxiliary variable’s error to the overall loss during training. Our experiments examined the improvement in prediction accuracy of the multi-task approach using paired streamflow and temperature data from sites across the conterminous United States. Our results showed that the best performing multi-task models performed better overall than the single-task models in terms of Nash-Sutcliffe efficiency. The improvement of the multi-task models relative to the single-task models had a seasonal trend with the multi-task models making larger improvements in the high-flow seasons. The multi-task scaling factor was consequential in determining to what extent the multi-task approach was beneficial and a naïve selection of this factor led to worse-performing multi-task models for stream temperature. Our findings indicate that, when configured properly, a multi-task approach could make more accurate predictions of interdependent hydrologic variables.