Application of hybrid conceptual-statistical model to estimate streamflow with consideration of groundwater variation

Conceptual models are common and efficient approaches to study short term and long term streamflow change in catchments. However, the possible role of groundwater dependent evapotranspiration on streamflow is not investigated well in literature. To fill this gap, our current study aims to investigate the importance of groundwater dependent evapotranspiration on streamflow by using several conceptual models (i.e., the ABCD type and the Budyko-type models) together with spatiotemporal groundwater level data. The performances of these models in streamflow estimation were assessed in the Harvey River Catchment of Western Australia. The results showed that monthly streamflow estimation was significantly improved, indicating the importance of groundwater level in short time scale. Yet, due to simplification and assumptions necessary for the conceptual models, these models were not able to accurately estimate peak-flow and base-flow variations in the study area. Hence, we further combine the conceptual models with a statistical module to form a hybrid conceptual-statistical framework. The result showed clear increase in the accuracy of the estimation and notable improvement in the models' performances, affirming that using only conceptual models may not be able to grasp all the aspects required for accurate streamflow estimation. The developed hybrid framework can be adapted to other catchments, particularly the areas with non-uniform or limited data.

Automated summary

Conceptual models are commonly used to study streamflow change, but their evaluation of the role of groundwater dependent evapotranspiration is lacking. To address this gap, our study used conceptual models and groundwater level data to investigate the importance of groundwater on streamflow. The results showed improved monthly streamflow estimation, but the models struggled to accurately estimate peak-flow and base-flow variations. Thus, we developed a hybrid conceptual-statistical framework which significantly improved estimation accuracy and model performance. This hybrid framework can be applied to other catchments, particularly with limited or non-uniform data.