Interaction between topographic and process parameters due to the spatial resolution of DEMs in distributed rainfall-runoff modeling

Selecting an appropriate digital elevation model (DEM) resolution is an essential part of distributed rainfall-runoff modeling since the resolution affects parameter values and, in turn, leads to predictive uncertainty. Moreover, the DEM resolution directly determines the computational workload required for model simulation. This study conducted several experiments to clarify the interaction between topographic and process parameters due to the spatial resolution of DEMs in distributed rainfall-runoff modeling. First, five different spatial resolutions (from 50 m to 1 km) were used to analyze the effects of DEM resolution on the topographic and process parameters of a distributed rainfall-runoff model [kinematic wave method for subsurface and surface runoff (KWMSS)]. Second, parameter compatibility was tested with regard to the sensitivity of model performance to optimal parameter values for each DEM, by applying the best-performing parameter combinations for each resolution to the models based on differing resolutions. Finally, the sensitivity of model performances to artificially generated parameters (deviating ±10% from optimal parameter sets) was analyzed to determine whether fine spatial discretization yielded equally good model performance measures or indistinguishable hydrographs (i.e., equifinality). The results indicate that differing topographic parameters due to distinct DEM sizes require differing process parameters to produce identically good runoff simulations. In addition, the parameter compatibility assessment suggests that increased spatial complexity due to fine DEM resolution results in decreased identifiability in process parameters. Consequently, nonoptimal parameter values can yield acceptable model performance measures when modeling is based on DEM resolutions of 250 m and smaller. The results of the sensitivity analysis also indicate that fine spatial discretization can be a dominant factor causing equifinality as well as overparameterization in distributed rainfall-runoff modeling. These findings may provide a new perspective on the equifinality problem, which many consider to be caused by huge model parameter requirements when operating distributed models.