Parameter Estimations For Locally Coupled Land Surface-Atmosphere Models,

Abstract

As land-surface modeling moves from the off-line mode to the coupled mode, it is also highly desirable to extend the off-line calibration of land-surface models to coupled applications. Using the NCAR SCCM as an example, this study proposed and implemented some effective schemes for the application of automatic parameter estimation procedures in a locally coupled environment, where other relevant issues such as parameterization tests, sensitivity analyses, and off-line calibrations were also involved. A parameterization deficiency having serious negative impacts on the performance of the NCAR SCCM was identified and rectified in this work, which led to significantly improved model performances and formed the basis for the subsequent sensitivity analysis and calibration experiments. To facilitate the calibration studies, an independent sensitivity analysis was conducted to identify some sensitive model parameters, followed by a multi-objective sensitivity analysis using the MOGSA algorithm to obtain better understanding of the model. Some off-line calibrations using the NCAR LSM were also conducted for comparison purposes. In the locally coupled environment, both land-surface and atmospheric variables/parameters were involved in the calibration processes of 14 different predesigned calibration cases. In brief, the results show that atmospheric parameters are of critical importance for the calibration of a coupled land surface-atmosphere model, and atmospheric forcing variables generally contain more useful information for calibration than land-surface fluxes/variables. In the coupled environment, step-wise calibration schemes, with land-surface and atmospheric parameters optimized successively in the off-line and coupled modes, respectively, appear to be superior to the single-step calibration schemes which optimize land-surface and atmospheric parameters simultaneously in the coupled environment, in that the former can provide better converged optimal solutions with less uncertainties. In addition, the results also show that better optimization effects can be achieved in the partially decoupled environment by replacing the model-generated precipitation and net radiation with the corresponding observations to drive the land-surface part of the model, indicating the dominant importance of precipitation and radiation in a coupled land surface-atmosphere model.