Inputs of land to the atmosphere occur in models on spatial scales of tens to hundreds of km. Yet many, if not all, of the land processes determining these inputs occur on substantially smaller spatial scales, from leaf to field or, at most, landscape spatial scales. Initial parameterizations of land surfaces in climate and mesoscale models have assumed homogeneous conditions of the land surface over a model grid square, either with some particular assumed surface type or an average over the actual surface types. Recent studies have modified land-surface parameterizations to include some aspects of heterogeneity and have studied what differences result from the inclusion of such heterogeneities. In particular, Koster and Suarez [1992] and Avissar and Pielke [1989] have proposed representing the land surface within a model in terms of a mosaic of homogeneous surfaces, each being some fraction of the total grid square. Prognostic variables, such as soil moisture and soil temperature, are then carried separately for each of these subelements. Other studies have emphasized the importance of the spatial variability of precipitation and soil moisture and have proposed statistical parameterizations to include such [e.g., Johnson et al., 1993]. Eltahir and Bras [1993a] have shown, in addition, the importance of spatial variability of rainfall interception and suggested a statistical parameterization for that.
Avissar [1992] further suggests the need for probability distributions for a wide range of surface parameters, including leaf area index, topographic roughness, and stomatal conductances. Bonan et al. [1993], using another land-surface model, consider the dependence of surface fluxes on statistical distributions of parameters. They find a considerable sensitivity to these assumptions but that the greatest sensitivity is to leaf area index. Avissar [1993] has reported on observed variability of stomatal conductances. Collins and Avissar [1994] and Li and Avissar [1994] have carried out additional sensitivity studies for statistical distributions of parameters and report greatest sensitivity to variability of stomatal conductance, roughness, and, in the latter case, leaf-area index. Seth et al. [1994] consider treating heterogeneity of precipitation and land surface elements in terms of a submesh under a given GCM grid square, and also find sensitivity to details of these distributions.
Another question [ Pielke et al., 1991] is the importance and parameterization of mesoscale circulations for vertical fluxes from the surface. Dalu and Pielke [1992] indicate with a linear perturbation approach that such fluxes should be comparable to the fluxes from small-scale motions conventionally parameterized in large-scale models. More detailed analyses and parameterizations of this issue have been developed by Zeng and Pielke [1994] and Pielke et al. [1994].
The need for appropriate data is at least as great a practical difficulty as is the conceptual formulation of land processes. Webb and Rosenzweig [1993] describe a new soils database for use with climate models. Contributions of remote sensing to provide better data are described below.
Progress is being made toward incorporation of adequately realistic land processes in NWP models. Various surface moisture stores (prognostic variables) are part of these prescriptions and are currently entirely model-generated, as derived from modeled precipitation, radiative and other surface fluxes. The patterns and intensity of the precipitation producing floods in the United States in the summer of 1993 were apparently sensitive to antecedent soil moisture [ Betts et al., 1994]. However, considerable advances are being made in approaches, in part based on remote sensing, to provide observations that can be assimilated with model values. Methods to provide surface radiative fluxes from a network of operational satellite sensors have been developed under the auspices of the WCRP GEWEX SRB (surface radiative budget) initiative. Pinker and Laszlo [1992] and Darnell et al. [1992] have developed algorithms suitable for this purpose. Use of earth radiation budget satellites (i.e., Earth Radiation Budget Experiment [ERBE]) for determining surface radiation has also been developed [e.g., Breon et al., 1994]. An ISLSCP/GEWEX project has just begun to develop a global data set for soil moisture and evaporation using surface observations of precipitation, satellite infrared surface radiative fluxes, and ECMWF assimilated surface meteorological fields as inputs and with a number of land surface schemes from various participants. This project includes an extensive validation effort using whatever observations are available.
Smith et al. [1994] use 4-1/2 months of preceding surface meteorological data and surface radiation estimated from cloud cover to initialize soil moisture in a mesoscale model. Time-varying surface temperatures can be closely related to evapotranspiration. Tarpley [1994] uses geostationary satellite data on the morning increase of surface temperatures to estimate, for clean sky conditions, evapotranspiration over related sites in Kansas. Gutman [1944] describes the land surface data sets now routinely available from National Oceanic and Atmospheric Administration (NOAA) operational satellites.
The use of remote sensing to derive surface biophysical properties as needed for climate and NWP models has made substantial advances in the context of FIFE, over Konza Prairie, Kansas, as reported in , 97, D17 and summarized by Sellers and Hall [1992]. Sellers et al. [1992] demonstrate with the FIFE data that surface canopy conductances are nearly linearly proportional to near-infrared minus visible reflectance differences and can be readily monitored by satellite. Shuttleworth [1994] reviews the overall past and future field programs being carried out as part of GEWEX. Methods are also being developed to characterize different land cover types by remote sensing globally and at the level of detail needed in climate models [e.g., Running et al., [1994].