Method for Modeling High-Temporal-Resolution Stream Inflows in a Long-Term ParFlow.CLM Simulation

Traditional hydrologic modeling has compartmentalized the water cycle into distinct components (e.g. rainfall-runoff, river routing, or groundwater flow models). An integrated, process-based modeling framework assesses two or more of these components simultaneously, reducing the error associated with approximated boundary conditions. One integrated model, ParFlow.CLM, offers the advantage of parallel computing, but it lacks any mechanism for incorporating time-varying streamflow as an upstream boundary condition. Here, we present a generalized method for applying transient streamflow at an upstream boundary in ParFlow.CLM. Downstream flow values are compared to predictions by traditional runoff and routing methods as implemented in HEC-HMS. Additionally, we define a model spin-up process which includes initialization of steady-state streamflow. The upstream inflow method was successfully tested on two domains - one synthetic tilted V catchment and an idealized small stream catchment in the Brazos River Basin. The stream in the idealized domain is gaged at the upstream and downstream boundaries. Both tests assumed a homogeneous subsurface so that the efficacy of the transient streamflow method could be evaluated with minimal complications by groundwater interactions. In the tilted V catchment, spin-up criteria were achieved within 6 model years. A 25 x 25 x 66 cell model grid was run at a computational efficiency of <0.01 CPU-hours per model hour in spin-up, and an average of 0.013 CPU-hours per model hour during the transient simulation.In the idealized Brazos River catchment, the best model performance had a Nash-Sutcliffe model efficiency of 0.945 (compared to measured discharge at the outlet), which indicates good performance in spite of the simplified subsurface processes. However, for both catchments, model response to initial state of land surface variables produced discontinuous and extreme runoff values early in the simulation.