Application of regional parameter estimation schemes to simulate the water balance of a large continental river

Two applications of a grid network version of the two-layer variable infiltration capacity (VIC-2L) land surface hydrology model applied at one degree resolution to the Arkansas-Red River basin are compared. In the first application, gridded parameter fields were linearly interpolated from locally estimated parameters (via a search procedure) at a set of 40 unregulated catchments distributed throughout the Arkansas-Red River basin with long-term gage records. In the second application, the grid network model was run using the same precipitation and potential evapotranspiration as in the first application, but with parameters transferred to the large area grid from the gaged catchments via regression of the locally estimated parameters on catchment soils, topographic, and climatological characteristics. In both cases, the model was run at a daily time-step, and vegetation parameters were tired (with prescribed seasonal variation but no interannual variation in the case of leaf area index). Generally, the simulations based on the regional regression transfer scheme performed significantly better than those based on the interpolated parameters; at the mouth of the Arkansas the absolute simulation error for annual flows was reduced from 7.3 to 1.3%, and for the Red, it was reduced from 12 to 1.7%. However, the basin-wide annual evapotranspiration and its seasonal cycle were relatively insensitive to the hydrological parameter estimation schemes. (C) 1997 Elsevier Science B.V.