Using evapotranspiration signatures to assess evapotranspiration realism in rainfall-runoff models

Abstract. Conceptual rainfall-runoff models are typically evaluated using streamflow data. Many studies have shown the benefits of moving from a conventional single-objective function to consideration of multiple signatures of streamflow behaviour, allowing more specific diagnosis of model deficiencies. In this study, we extend this approach beyond streamflow, using actual evapotranspiration (AET) signatures to assess AET dynamics in traditionally calibrated conceptual rainfall-runoff models. We calibrated models to streamflow only and separately to streamflow jointly with AET, then evaluated AET realism using AET signatures defined across various temporal scales. This was done for five models at 14 Australian sites spanning a variety of conditions, with each site co-located with a flux tower. Our results show that incorporating AET data into calibration significantly improves aspects of AET dynamics in models, particularly monthly variability and the degree to which AET and potential evapotranspiration are synchronous or asynchronous. This improvement extends even to the independent evaluation period in split sample testing. However, other signatures were not well improved, including aspects of seasonal and event-scale timing, in addition to interannual variability. Future research could explore a wider range of calibration strategies to assess whether these deficiencies can be calibrated away or are inherent to the models. Overall, these findings suggest that commonly used conceptual rainfall-runoff models struggle with many aspects of AET dynamics, even when AET information is included in the calibration. We recommend that future model evaluations examine a wider range of measures, aiming to characterise performance against non-streamflow variables in a more holistic manner.