Flood forecasting is essential to protect our communities, reducing risk and avoiding infrastructure damage. Distributed hydrological models are often used for this purpose, usually after being calibrated at gauged watershed, these results are then extrapolated to ungauged ones, which are the majority in the U.S. However, this extrapolation typically overlooks the limitations carried into the model by the loss of detail in the spatial representation of river networks. Model calibration hides this structural limitation by giving more weight to other variables rather than the detail of representation. Although we have little information about these limitations and implications, prior studies have shown that less detail of the river network can impact the performance of the forecast affecting the predictive reliability of the model when its applied to other regions. This study investigates the importance of model discretization scales (DS), which is the level of spatial detail used to divide a watershed into hillslopes and river links, can affect the watershed characteristics and simulated flow responses. I analyzed the Smoky Hills watershed (Kansas, U.S.), here I compared geomorphological features like width function, saturated hydraulic conductivity, average slope distribution, travel time to streams, and stream network density and simulated flows across six DS ranging from fine (accumulated area of 0.1 ??2 closer to reality and therefore our benchmark – BDS) and coarser (70 ??2 corresponding to USGS HUCs 12 and HYDRUS average hillslope areas). For each scale, I approached the simulation by using 110 uniform rainfall events with a constant runoff formulation of the Hillslope Link Model (HLM). Model outputs were evaluated at 400 random control points across the watershed. Our results show that coarser DS can alter important geomorphological watershed characteristics that can carry issues into the simulations. Also, simulation shows that as the DS increases, peak flows decrease while time-to-peak significantly increases, indicating slower and longer hydrographs. Finally, it was seen that an optimization of the routing parameters can help emulate the results of our DSB (benchmark) but in some cases this parametrization used unnatural values for the zone of study, weighing on the importance of the river network in the models. Our results raise concerns about using pre-imposed discretization scales and highlight potential limitations when extrapolating flood forecasts to ungauged watersheds. This study highlights the need to account for discretization effect when developing a model and transferring it across regions and scales
Simón Martinez Rendon (Sun,) studied this question.