Author(s): Simon Martinez Rendon; Nicolas Velasquez Giron; Witold Krajewski
Linked Author(s):
Keywords: Threshold Watershed Resolution Geomorphology Flood
Abstract: Flood forecasting is essential to mitigate risks, design hydraulic structures and infrastructure, and protect communities. Usually, we obtain forecasts using distributed hydrological models calibrated at gauged watersheds and extrapolated to the ungauged ones (the majority). However, in the extrapolation process, we usually ignore potential limitations due to a poor representation of the river network due to coarse discretization scales (DS). A poor representation of the river network may have profound effects on the performance of our forecasts at different scales and on the validity of its extrapolation. Nevertheless, we have little information regarding these effects and their implications. This work explores the model DS implications when decomposing the watersheds into hillslopes and links (runoff and river elements, respectively). We analyzed the Smooky Hills watershed (Kansas, U. S. ) by comparing geomorphological features and simulated flows using six DS with hillslope average areas ranging between 0.1 km^2 (closer to reality and benchmark or BDS) and 70 km^2 (USGS HUCs 12 and HYDRUS areas). The geomorphological features include the width function, saturated hydraulic conductivity, average slope distribution, travel time to streams, and stream network density. For each DS, we ran a constant runoff formulation of the Hillslope Link Model (HLM) using uniform rainfall (110 rainfall events). We obtained simulations for each DS using realistic parameters for the runoff (70%), overland velocity (0.1 m ⋅ s−1), and channel velocity (0.3 m ⋅ s−1). Then, we searched for the overland and channel velocity parameters at each DS, providing the best performance compared to DSB. We assessed the geomorphological features and model simulations at 400 random points of control (~80 per magnitude). Our findings indicate that DS resolution affects watershed geomorphology by reducing variability in certain features and significantly increasing water travel time. From the simulations, we found peak flows decrease and time to peak increases for higher DS. After the DS parameter search, we found discrepancies among the optimal parameters for each control point and objective function. Moreover, we obtained more considerable differences when comparing coarser DS networks with the DSB. Our results raise a red flag when using pre-imposed discretization scales, highlighting potential limitations during the extrapolation of flood forecasts at ungauged watersheds.
Year: 2025