Representation of subgrid-scale variability in large-scale hydrological models using hydrgeomorphic units

The performance of large-scale (104-106 km2) hydrological models that use a coarse grid resolution (i.e. >10 km2) can be improved by incorporating subgrid variability of hydrological processes and model forcing. We propose a modeling framework that represents hydrologic variability at the subgrid scale using hydromorphologic units. The approach is tested and evaluated using field data from a catchment (10 km2), considered as a single grid cell in a large-scale hydrological model. The framework consists of four steps. First, the hydromorphologic units (i.e., units representing subgrid variability) are mapped according to geomorphological and hydrological properties. These units, with an area of approximately 2 km2, have relatively homogenous and distinctive physical properties and hydrological behavior. The units are characterized with a set of lumped measurable properties; and the processes can be described by simple lumped equations. The second step is to identify the process-based lumped model for the units. The lumped model considers two key processes; infiltration and Hortonian runoff, with three parameters. The lumped model is parameterized using an artificial dataset created by a physically-based high-resolution model. This results in a lookup table containing an extensive set of lumped model parameters for a large range of characteristics regarding unit geometry and rainfall (approximately 65,000 scenarios). The next step is to implement the subgrid modeling framework to the field catchment. Lumped model parameters are estimated for each hydromorphologic unit by reading from the lookup table. Discharge is simulated for each unit and summed to retrieve the hydrologic response of the catchment. The final step is to evaluate the improvement of model performance when using our framework. This is done by comparing the model results from runs with and without hydromorphologic units as subgrid units. For the runs without subgrid units, rainstorm characteristics and physical properties are lumped over the catchment. The results show improvement in hydrological prediction when incorporating the hydromorphologic units at the subgrid level. Importantly, prediction errors in both total and peak discharge are reduced compared to those without considering subgrid units. Improvement of model performance is most remarkable when the storms exhibit a high degree of spatial variation (e.g. convective rainstorms). The model performance decreases for events with only a small amount of generated runoff; this is due to spatial processes in runoff generation not being considered in the lumped model