Assessing the Significance of Spatial Allocation and Resampling Algorithm of Landcover Input in Hydrometeorological Model (WRF-Hydro/Noah-MP) Simulating Watershed Responses

During the past decade, hydrologic science community pursued a modeling system that enables continuous hydrologic simulations across large domains at high spatiotemporal resolution. This led to the convergence of hydrologic modeling and land surface modeling (LSM). In such hydrometeorological modeling, parameterization is spatially distributed and relies on land data inputs including land-cover/vegetation classification (satellite-based) and a soil texture class (with large consolidations). It is well known that land cover (LC) input and its numerical parameterization control the calculation of land-air-vegetation flux exchange in LSMs. While certain hydrologic processes (e.g. evapotranspiration, infiltration, interflow, and percolation) tend to become a dominant process that drives the streamflow simulation, their water flux calculation is directly affected by LC inputs.

LC inputs for the hydrometeorological models are generally upscaled from finer resolution dataset, and resampling algorithms easily affect the accuracy of spatial allocation and representativeness of LC classes in a modeling domain. This statement also applies to the National Water Model (NWM) that predicts streamflow for the entire continental US and uses a renowned hydrometeorological model (WRF-Hydro/Noah-MP) as a core module. Considering the key role LC input plays in water flux calculations and the converging nature of watershed processes, inaccurate spatial allocation of LC input can have a cascading effect on runoff/streamflow simulations.

We present the experimentation of how much spatial allocation of LC input affects the water flux calculations and watershed runoff simulation using WRF-Hydro/Noah-MP standalone setup. We use four different LC schemes (default NWM LC resampling, advanced "Area Preserving" resampling, random allocation, and single representative LC type assigned to each catchment), and examine the simulated catchment water balance and hydrologic signatures to indicate the watershed response behavior. Preliminary results showed that LC allocation did not necessarily bring cascading effect. Instead, influence from LC allocation seemed to diminish through model simulation steps, as simulated streamflow patterns showed lower sensitivity compared to catchment water balance.