An efficient calibration method for continental-scale land surface modeling

Land surface models contain physically conceptualized parameters that require calibration for optimal model performance. However, calibration time can be prohibitive. To reduce computational time, we calibrated the VIC land surface model for a subset of the grid cells and then interpolated the parameters to the uncalibrated grid cells. In the continental United States, the "observation" to which we calibrated was the monthly runoff ratio, calculated for 1130 small basins throughout the country and interpolated to those grid cells that did not fall within a small gauged basin. The results demonstrated that this approach is sufficiently accurate and computationally efficient for large-scale applications. We examined the effect of model spatial and temporal resolutions on calibrated parameter sets to evaluate if one could calibrate at coarser resolutions and apply these parameter sets to finer resolutions, reducing computational time. Results indicated that calibrating at different temporal resolutions causes minimal changes in modeled runoff while transferring parameter sets across spatial resolutions can cause significant changes in model performance.