As explained above, the water availability for evapotranspiration
(hereafter referred to as 
) affects considerably the
redistribution of the radiative energy absorbed at the ground surface
and, as a result, the entire PBL. However, many hydrological and
physiological parameters control 
. For instance, plant
stomata control transpiration through a chain of complex biochemical
reactions, which are apparently sensitive to solar radiation, carbon
dioxide concentration, temperature, vapor pressure difference between
the leaves and their environment, and soil water potential in the root
zone [e.g., Jarvis and Mansfield, 1981;
Willmer, 1983; Martin et al., 1983;
Avissar et al., 1985; Zeiger et al., 1987;
Collatz et al., 1991; among many
others]. The soil hydraulic conductivity is one of the parameters that
control the transport of water in the root zone and to the evaporating
surface, and, therefore, obviously affects also 
.
In general, SVATS are designed to solve for the interaction of energy,
momentum, and mass between the surface and the overlying atmosphere
(and, therefore, for 
). The degree of complexity of SVATS
varies from the earlier so-called bucket scheme [ Manabe, 1969] to
schemes which include several soil and vegetation layers, and account
for horizontal heterogeneity [e.g., Avissar and Pielke, 1989].
In the bucket scheme, a near-surface layer of soil is modeled as a
bucket that can be filled by precipitation and snowmelt (if any) and
empties by evaporation and by runoff; the latter occurs only when the
bucket is full. Its other attribute is that the evaporation rate is a
linear function of the amount of water in the bucket below some critical
value. In more recent schemes, the vegetation is treated as one or more
separate layers, scaling (usually linearly) from the size of real leaves
up to the size of a grid element of the host model (i.e., 10
km
or
more in GCMs). Usually, only three land components (soil, snow, and
vegetation) are treated explicitely. The Biosphere-Atmosphere Transfer
Scheme (BATS) suggested by Dickinson et al. [1986] and the Simple
Biosphere Model (SiB) developed by Sellers et al. [1986], which includes
a two-layer vegetation module to account for two different types of
vegetation (e.g., trees and shrubs) that may be found in a single grid
element of a GCM, are two well known examples of such schemes. Carbon
fluxes are treated only in a few SVATS, although there are ecological
models that deal with carbon uptake and release [e.g., Collatz et al.,
1991].
These SVATS are based on the concept of ``big leaf'' which implies that land is homogeneously covered by one (or sometimes more) big leaf within a grid element of the numerical atmospheric model. This ``big leaf'' usually has a single ``big stoma'' which is sensitive, in the most sophisticated parameterizations, to the environmental conditions known to have an effect on the mechanism of the stomata at the leaf scale (i.e., solar radiation, temperature, humidity, carbon dioxide, and soil water potential in the root zone). This stoma controls the plant transpiration and, as discussed above, the redistribution of energy absorbed by the vegetation.
However, at the grid resolution of current GCMs, continental surfaces are very heterogeneous. This can be readily appreciated by looking at maps of soil, vegetation, topography, or land-use patterns, for instance. The combined effect of this type of heterogeneity with the heterogeneity of precipitation results in large variations of water availability for evapotranspiration and, consequently, of surface energy fluxes. In order to better address this problem, Avissar and Pielke [1989] proposed the Patchy Land-Atmosphere Interactions Dynamics (PLAID) parameterization. This scheme assumes that the landscape heterogeneity found in each grid element of the atmospheric numerical model can be represented by a mosaic of patches (or ``tiles''), with each patch consisting of a single land use (e.g., agricultural field, bare field, built-up area, forest, body of water, etc.). Then, for each land patch, a ``big leaf'' scheme is applied. With this parameterization, each land patch of the grid element is coupled independently to the atmosphere of the model, and patches affect each other only through the atmosphere. This approach has been adopted by Koster and Suarez [1992] and Ducoudré et al. [1993] in their GCMs.
In addition to the problem of heterogeneity, state-of-the-art SVATS contain a large number of empirical constants. For instance, SiB requires 49 such constants plus a leaf angle distribution function to characterize the land surface [e.g., McNaughton, 1987]. Even an extremely sophisticated micrometeorological field experiment which permits complete control of the plant environment could not provide exact values for these constants. Also, because these constants might vary throughout the growing season and for the different vegetation found in the domain represented by a single grid element of the models, it is not clear how these constants could be combined to provide grid-scale ``representative'' values that could account for the non-linearity of the involved processes.
Even though these problems are now quite well recognized, only a few studies have been conducted in recent years to address them. Henderson-Sellers [1993] used the factorial experiment technique in conjunction with BATS to identify which of the land characteristics have a predominant impact on the redistribution of energy into turbulent sensible and latent heat fluxes at the ground surface. Collins and Avissar [1994] used the Fourier Amplitude Sensitivity Test (FAST) introduced by Cukier et al. [1973], in conjunction with the Land-Atmosphere Interactions Dynamics (LAID), a SVATS developed by Avissar and Mahrer [1982, 1988] and improved by Avissar and Pielke [1991], to investigate the same problem. In general, both studies provided similar conclusions.
These studies indicated that land-surface wetness, surface roughness, albedo, and when vegetation covers the ground, leaf area index, and plant stomatal conductance are the most important characteristics which affect the redistribution of energy at the ground surface. Among them, the land-surface wetness, which expresses the relative availability of soil moisture for evaporation in the upper soil layer, was found to have a predominant role in bare land. Correspondingly, in vegetated land, the stomatal conductance which has a very similar impact on the redistribution of energy into turbulent sensible and latent heat fluxes at the ground surface was found to be the most important land-surface characteristic. In addition, Henderson-Sellers found that mean monthly temperature and the interaction between mean monthly temperature and total monthly precipitation have a predominant impact on the behavior of BATS. She emphasized that fractional cloudiness and other environmental parameters are also important.
Collins and Avissar concluded that even though it would be interesting to compare the results of their sensitivity analysis with other schemes, they expected that the different ``big leaf''schemes should produce nearly the same results. Their claim was based on the fact that most SVATS's use the same physical principles and the solution of energy budget equations for the soil layer and, when present, the vegetation layer. The major difference between these schemes is in the parameterization of the various variables (e.g., vegetation absorbtion and transmission of radiation, stomatal conductance, soil surface wetness), which were imposed in their analysis and not calculated interactively by their model.
One of the major objective of PILPS is, precisely, to investigate this issue. As explained in Henderson-Sellers et al. [1993], PILPS consists of a multistaged intercomparison process including:
Henderson-Sellers et al. [1993] provided the framework for the documentation of land-surface schemes and discussed in detail the coupling to host model(s). They suggested an intercomparison framework for PILPS and invited all SVATS developers/users to participate in the project.
Preliminary results from the off-line intercomparisons were presented at
the 1993 International Association of Meteorology and Atmospheric
Physics (IAMAP) and International Association of Hydrological Sciences
(IAHS) joint meeting in Yokohama, Japan. These results demonstrated
that differences of sensible and latent heat flux as large as
150 Wm
are produced by the various participating schemes (20 at
that time) when forced by similar atmospheric background conditions