Land Surface Heterogeneity in the Cooperative Atmosphere Surface Exchange Study (CASES-97). Part II: Analysis of Spatial Heterogeneity and Its Scaling

Abstract

Multiscale atmospheric forcing data at 1-, 5-, and 10-km scales from the 1997 Cooperative Atmosphere Surface Exchange Study (CASES-97) were used to drive three uncoupled land surface models: the National Center for Atmospheric Research Land Surface Model version 1 (NCAR LSM), the Oregon State University Land Surface Model (OSU LSM), and the Japan Atomic Energy Research Institute's Atmosphere?Soil?Vegetation Model (SOLVEG). The data included high-resolution, gauge-corrected precipitation estimates from dual-polarization radar, with the experimental period covering the spring green-up process. The effects of increasing scale on modeled estimates of the domain mean flux were more pronounced when there was greater heterogeneity of the land surface in terms of the surface vegetation, although this result was model dependent. All models made use of prescribed parameters based on dominant land cover at the different scales, not effective parameters, leading to situations where the coarser-scale flux response fell outside the subset of the 1-km estimated flux range, largely a result of the disparity between the characteristics of the different land surfaces (e.g., bare soil, winter wheat, grasses, and urban). The scaling effects on the statistical distribution of fluxes were more pronounced when there was greater heterogeneity with respect to land cover. The NCAR LSM includes a ?capping? scheme to soil moisture resistance and leads to a noticeable difference in its response when compared with SOLVEG and OSU LSM, which adopt a more gradually varied surface resistance scheme to soil moisture dryness. The outcome of this study suggests the importance of incorporating actual land surface characteristics in land surface models, and tries to advance important issues regarding land surface parameterization schemes. LSMs need to endogenously simulate vegetation dynamics that respond to actual environmental conditions. Current datasets that describe the dominant land surface and that are used to parameterize many LSMs are inadequate because they treat the land surface as stationary heterogeneous, when in fact the land surface is nonstationary heterogeneous.