Observed and Simulated Channel Network Infiltration Losses in the Semi-Arid Walnut Gulch Experimental Watershed, Arizona

Channel bed infiltration losses have a strong impact on the hydrological response of semi-arid catchments, reducing runoff downstream. Numerous studies have shown the impact of channel infiltration using in situ observations. However, quantifying its magnitude at the catchment scale is challenging, as nested watershed observations at the larger-scale are rare. As a result, computer models are often used instead, although their performance to accurately simulate these losses is unknown.

This study tries to improve our understanding of channel infiltration losses and their variability, by analyzing observed runoff events within the semi-arid ~149 km²USDA-ARS Walnut Gulch Experimental Watershed (WGEW) in southeast Arizona. Over 50 years of distributed runoff and precipitation observations are available within the WGEW. For five different channel sections, events were selected where the upstream and downstream runoff response was not affected by additional precipitation and lateral surface runoff. For these events, considerable infiltration losses are observed where no runoff occurs at the downstream locations for events with smaller upstream event accumulations. Although the infiltration response varies between events and channel sections, the loss per unit length [m³m^-1] for all sections jointly follows a power law as a function of upstream event accumulation, with largest losses occurring for the smallest accumulations. These results show that for >90% of the runoff events occurring upstream within the WGEW, all surface water infiltrates into the channel bed with no water reaching the outlet.

A second aspect of this work is to assess whether the KINEROS2 and the WRF-Hydro National Water Model (NWM) are able to properly represent the observed channel infiltration. Currently, both models are used by different operational authorities in the southwest US for flood forecasting. For KINEROS2, channel infiltration depends on antecedent conditions, soil suction, total amount of event infiltration and surface water level. For WRF-Hydro NWM, infiltration varies with surface water level only. Both models were calibrated as part of this study. Results show that KINEROS2 does not produce any runoff for a considerable number of events. However, for those events for which runoff is simulated, the model performs well. KINEROS2 slightly overestimates observed infiltration amounts, but reduces the minimal upstream accumulations that result in downstream runoff by 50%. WRF-Hydro NWM simulates most runoff events, although total accumulations are considerably underestimated. Also, almost all upstream events result in downstream runoff. Therefore, WRF-Hydro NWM underestimates infiltration amounts, which is expected due to the relatively simple relationship used to simulate infiltration as well as the applied approach to calibrate the model.