Comparison of three land-surface schemes with the Fourier amplitude sensitivity test ( FAST )

This paper explores which are the land-surface parameters playing a key role in three surface schemes, namely the land-atmosphere interactive dynamics (LAID), the interaction soil-biosphere-atmosphere (ISBA) and the biosphere-atmosphere transfer scheme (BATS). The Fourier amplitude sensitivity test (FAST) was used for that purpose. This test estimates the relative contribution of model input parameters to the variance of surface heat fluxes. This analysis demonstrates that, for the three considered schemes, four parameters can explain most of the variance of surface heat fluxes under a broad range of environmental conditions. Soil wetness plays a predominant rôle for the heat fluxes. Roughness length is the most important parameter for the momentum flux. Leaf area index, in vegetated land, and texture, mainly in bare land, also have a significant impact on the fluxes. Roughness length is usually more important for sensible heat flux than for latent heat flux, and is mostly important under stable atmospheric conditions. Soil wetness and vegetation parameters are the dominant parameters under buoyant conditions.


Several investigations have already addressed
Collins and Avissar (1994) have first used the the importance of the different land-surface para-Fourier Amplitude Sensitivity Test (FAST) to meters in atmospheric modelling (Mintz, 1984; estimate the relative importance of land-surface Walker and Rowntree, 1977; Rowntree and parameters.The impact of microscale variability Bolton, 1978;Miyakoda and Strickler, 1981; of the most important land-surface characteristics Shukla and Mintz, 1982;Charney et al., 1977; on atmospheric turbulent heat fluxes near the Chervin, 1978;Carson and Sangster, 1981; Sud ground surface has been investigated by Li and et al., 1988;Dickinson and Henderson-Sellers, Avissar (1994).The Project for Intercomparison 1988;Sud et al., 1990;Henderson-Sellers, 1993, of Land-surface Parameterization Schemes 1996a, 1996b, 1996c).Representation of landscape (PILPS) (Henderson-Sellers et al., 1993, 1995) subgrid-scale heterogeneity has been simulated emphasizes the needs for sensitivity intercompareither by dividing each grid square of the modeled isons studies to identify which simplifications domain into multiple homogeneous regions or omissions are important to improve our (Avissar and Pielke, 1989;Koster and Suarez, understanding of the parameterizations of the 1992) or by statistical-dynamical approaches using interactions between the atmosphere and the continental surface in climate and weather forecast models.
The radiative solar and atmospheric fluxes the environmental conditions.If, on the other hand, stomatal conductance (or resistance) is para-absorbed by land-surface are mainly redistributed as latent and sensible heat.Both heat fluxes are meterized as a function of environmental conditions, the parameters controlling the stomata are the main mechanisms to turn back energy into the atmosphere from land surface.The relative those which are really relevant in computing surface heat fluxes.Because different land-surface importance of latent and sensible heat fluxes depend strongly on surface features.In bare, dry schemes use different parameterizations to represent this processes, it is interesting to investigate soils, the absorbed radiative energy is mostly used to heat the surface, turning back the energy to the the impacts of various parameters (assumed independent) in different schemes.This was the atmosphere usually as a vigorous, turbulent sensible flux.On the other hand, densely vegetated purpose of the study summarized in this paper.
This study will mainly demonstrate that soil surfaces with enough water available for evapotranspiration invest most of the radiative energy wetness, which has a crucial impact on the stomatal mechanism, mostly affects the surface in extracting subsurface water through the root system.This process of transpiration is mainly heat fluxes.controlled by leaves, opening and closing their stomata according to the environmental conditions and to the available soil wetness.2. Analysis with the Fourier amplitude sensitivity test ( FAST ) Transpiration turns energy back to the atmosphere in form of latent heat flux (Garratt, 1992, among many others).Typical sensitivity analysis of land-surface parameterizations to the variation of their parameters Between such extreme cases of dry, bare soils and densely vegetated wet terrains there are many have assumed predetermination of the range of variation of all involved parameters.This range intermediate land-surface conditions, whose behaviour is constrained by soil features, especially soil of variation must be related either to the degree of uncertainty associated with every measured wetness and texture, and by the predominant vegetation type.The parameters describing soil parameter or to its use in the parameterization scheme.Henderson-Sellers (1993) proposed the and vegetation, together with atmospheric conditions, determine the magnitude of the surface heat following sources of uncertainty for the BATS parameters, which can be considered as general fluxes and the Bowen ratio (the ratio of sensible to latent heat flux).Both ways of turning back for every land-surface scheme: (i) original global archives; (ii) association of some particular para-energy to the atmosphere, either as sensible or as latent heat, are controlled by the ability of plants meter with a particular vegetation type; (iii) use of the parameter in the land-surface parameteriz-to moderate water consumption during transpiration under unfavorable environmental condi-ation scheme; (iv) aggregation of the original data to the appropiate model resolution; and (v) the tions.As mentioned above, stomata play a key role in controlling transpiration.Assuming that it compilation procedure from different literature sources together with unclear prescriptions to is an independent parameter, Collins and Avissar (1994) showed that uncertainty in estimating this some vegetation types when measurements exist only for one type.parameter results in the most important error in estimating surface fluxes under convective Other important issues in sensitivity analysis arise when several parameters are modified simul-conditions.
However, stomatal conductance (or resistance) taneously.Do all possible combinations of parameters exist in nature?What is the frequency of is generally not considered an independent parameter.It is well accepted (Jarvis, 1976) that this appearance of every combination?Obviously, if some possible combinations appear very rarely in parameter is a function of various environmental conditions (solar radiation, air temperature, air nature, they should have a reduced weight in the analysis.In this paper, all combinations will be relative humidity, soil wetness).When stomatal conductance (or resistance) is considered an inde-assumed possible if parameters are independent.
Parameters are independent when their assigned pendent parameter, it is allowed to vary freely between within some interval, independently of value do not condition the value of the others.Of Tellus 50A (1998), 3 course, that excludes the parameters which can be Avissar (1996) to study the sensitivity of shallow convective precipitation induced by land-surface expressed as a function of others.Usually, landsurface parameters are considered independent heterogeneity to dynamical and cloud microphysical parameters.With this technique input para-only as a first approximation.For instance, one can expect that high values of soil wetness are meters are varied simultaneously through their ranges of possible values according to some given associated with dense vegetation and, consequently, with high values of leaf area index.On PDF.All input parameters are assumed to be independent.the other hand, vegetation cannot develop in dry soil.It is important to stress that the results of Each input parameter is assigned a different frequency, which determines the number of times the analysis depend greatly on the selection of the parameters as well as on their range of variation.that the whole range of the parameter is traversed.
This frequency of oscillation, different for each In current state-of-art land-surface schemes, at least 10 input parameters can be assumed to vary parameter, is analyzed in the model output to separate the response of the model to every input more or less independently.Several techniques are available for sensitivity analysis: frequency.Addition of those Fourier coefficients corresponding to a particular input parameter (1) The Monte Carlo method, which makes a random sampling of the entire input parameter frequency and its harmonics determines the total contribution of that particular input parameter to space.It becomes computationally very expensive when the dimension of the input parameters space the model output variances.The essence of this method consists of analyzing the spectrum of increases; (2) The so-called factorial stratified sampling, frequencies of model outputs generated when parameters are forced to oscillate with given linearly which is the simplest technique available.A ''high'' and a ''low'' value can be chosen for each of the independent frequencies.Finally, by scaling the relative contribution of the input parameters to input parameters, representing the range of that parameter, and the model is run for all combina-the total variance, partial variances are obtained which show the sensitivity of the model output tions of these high and low values.The number of computations grows exponentially with the parameters to the variation of the individual input parameters in their prescribed range of values.A dimension of the parameters space.If the parameters variability is distributed according to prob-description of the theory and implementation of the FAST method and approximations used in ability density functions (PDF), more than two values are needed to explore the impact of these the computer code can be found in Collins and Avissar (1994).parameters.For instance, if 10 values are needed to represent the PDF of each parameter, then the Table 1 summarizes the number of model runs required by FAST as a function of the number of total number of computations to cover all possible combinations of these parameters is 10n.Such an input parameters, which is substantially less than the number of integrations required by the Monte approach becomes prohibitive as the dimension of input parameter space grows; Carlo or the factorial stratified method (McRae et al., 1982).(3) Other stratified sampling techniques exist that are much more efficient, such as latin hyper- The FAST technique is a very powerful technique for general sensitivity analysis in mathemat-cube sampling (McKay et al., 1979).However, as the above mentioned ones, they do not provide ical models, though it has also some limitations.
$ Nonlinear algorithms connecting the input model sensitivity to individual input parameters; (4) The Fourier Amplitude Sensitivity Test parameter and output parameter spaces can distort the real sensitivity caused by some input parameter.(FAST) originally developed by Cukier et al. (1973), Cukier et al. (1975) Avissar and Mahrer (1988), the 12 587 interaction soil-biosphere-atmosphere (ISBA) 13 695 developed by Noilhan and Planton (1989) and 14 915 the biosphere-atmosphere transfer scheme (BATS) 15 1027 developed by Dickinson (1984).Their formulation and degree of complexity are somewhat different.The three models were run with steady-state atmospheric conditions.LAID was integrated until corresponding to two input parameters can interact to create harmonics not present originally, steady-state was reached, as only a few iterations are needed to solve the energy balance equations and, in this way, decrease the relative contribution to the output spectrum of the input parameters.for soil and canopy layers.The simplified version of LAID described in Collins and Avissar (1994) Whenever several ways of defining input parameters exist, those linearly related with the output was adopted here to avoid the introduction of soil moisture in all soil layers (it originally had 13 parameters will be more adequate for FAST studies. layers) as parameters and to prevent the introduction of additional parameters (as, root distribution, $ T he input parameters should be either independent or their dependency (in form of covariances porosity, percentage of sand, percentage of clay, etc.) not always independent.Furthermore, the between pairs of parameters) should be modelled.The independency of the input parameters usually importance of soil moisture al different depths depends both on the vertical distribution of roots is an a priori assumption based on the way the parameters are selected.If an input parameter is and on the integration range.However, BATS and ISBA were integrated for 24 h, as soil moisture physically related either with other input parameters or with the model forcing conditions, the was only left to evolve in the upper surface layer.
In both squemes, soil moisture in the upper layer partial variance corresponding to such input parameter will be unrealistic.
is steered by surface fluxes and by fixed soil moisture in the lower layer.The ISBA scheme $ The FAST technique provides the module of the sensitivity but not its sign.
reaches the steady-state typically before 24 h, whereas BATS can need several days.At the $ The simultaneous introduction of many parameters, some of them with low partial variance, equilibrium state, moisture flux in the upper boundary of the surface layer is compensated by usually attenuates the effect of the most sensitive parameters.As mentioned above, the nonlinear soil moisture diffusion from the lower layer.Initial soil wetness was one of the parameters considered contributions originated by interactions of pairs of parameters generate new harmonics, decreasing for the three models.Equilibrium latent and sensible heat fluxes (E and H, respectively) between the relative contribution to the output spectrum by the input parameters.
the atmosphere and the surface were computed for each atmospheric steady-state condition, and $ T he range of variation of the input parameters is usually a critical issue.Sensitivity results, were used as model outputs in the FAST algo-rithm.No precipitation of any type was consid-roughness length was not taken as in Collins and Avissar (1994), to assign the same interval to ered here.equivalent parameters in the three models.
3.1.L AID 3.2.ISBA The version of the land-atmosphere interactive dynamics (LAID) scheme used for this comparison The interaction soil-biosphere-atmosphere (ISBA) scheme, was developed by Noilhan and is comprehensively described in Collins and Avissar (1994) and Avissar and Pielke (1989).The Planton (1989) (see also Bougeault et al., 1991;Braud et al., 1993;Giordani, 1993;Noilhan et al., surface is considered to consist of two layers, a vegetation and a soil layer.Surface energy fluxes 1993; Noilhan and Lacarre ´re, 1995;Mahfouf et al., 1995) and modified by Bringfelt (1996) and by of latent and sensible heat are calculated from two energy budget equations, one for the vegetation Giard and Bazile (1997) for its operational implementation in the HIRLAM and ARPEGE sys-and one for the soil layer.No storage of heat is allowed for the canopy.Therefore, the net radiative tems, respectively.In this scheme sensible and latent fluxes are averaged according to the frac-energy absorbed by the plant canopy is released as sensible and latent heat back to the atmosphere.tional areal share of the grid square.The whole scheme makes an efficient use of the new physio-In the present study, the fourteen input parameters used with the FAST program are: rough-graphic database created for the HIRLAM system (Bringfelt et al., 1995).The land surface scheme ness length (Z o ), leaf area index (lai), soil texture (t), soil emissivity (e g ), soil albedo (a g ), vegetation treats vegetation processes, such as surface canopy resistance to water transpiration and storage and emissivity (e v ), vegetation albedo (a v ), vegetation extinction coefficient (k v ), soil surface wetness (W ), evaporation of intercepted rainfall.The land part of the scheme includes a two-layer force-restore maximum relative stomatal conductance (Cs max ), radiation factor for stomatal conductance (R st ), model for soil moisture and temperature.There is an additional equation for moisture stored in temperature factor for stomatal conductance (T st ), vapor pressure deficit factor for stomatal conduct-the canopy.
In this study, the prognostic equations for mean ance (e st ), and soil moisture potential factor for stomatal conductance (W st ).These parameters are temperature and bulk soil moisture were not solved, thus resulting in constant mean temper-summarized in Table 2, and are prescribed by continuous PDFs.Most maximum and minimum ature and bulk soil moisture equal to their initial value.As diurnal cycle is suppressed by prescribing values assigned were obtained from Collins and Avissar (1994).However, the soil thermal conduc-environmental conditions, it seems more reasonable to keep mean temperature constant.Bulk soil tivity was expressed as a function of soil texture and water content (McCumber and Pielke, 1981).moisture is kept constant to assure that some steady-state is reached when the surface soil mois-The textural classification of the U.S. Department of Agriculture (USDA) was used for the texture ture equation is integrated.Soil moisture in the surface layer (10 cm) reaches a steady value, typic-parameter (Clapp and Hornberger 1978).The discrete classification, which assigns values to the ally within 24 h of integration.
The surface resistance is expressed by a product hydraulic parameters corresponding to 11 soil types was extended to the continuous domain of a minimum value and a number of limiting factors (Jarvis, 1976;Dickinson, 1984; Jacquemin between 1 and 11, simply by interpolating linearly between values assigned to the integers.The com-and Noilhan, 1989;Thompson et al., 1981) depending on environmental conditions (radi-putation of the relative stomatal conductance introduces additional parameters (Avissar et al., ation, water stress, vapor pressure deficit and air temperature).1985; Avissar and Pielke, 1989) with their respective uncertainty range.Uncertainty was allowed in The ISBA scheme distinguishes between 2 primary and 15 secondary parameters.The 2 primary the maximum stomatal conductance, and in the environmental parameters for which the relative parameters refer to the dominant type of vegetation and soil texture.For each of the 11 soil stomatal conductance are half their maximum value.The variation range of leaf area index and texture types of the USDA classification, values are assigned to each of the 8 hydraulic secondary 15 secondary parameters.They have been introduced here to study their sensitivity in the stomatal parameters (Table 2 of Noilhan and Planton (1989) and Table A2.1 of Bringfelt (1996)).In resistance formulation.These four additional parameters can, in principle, be species dependent.Noilhan and Lacarrere (1995) the 8 hydraulic parameters are expressed as a function of percent- The maximum and minimum values assigned to the various parameters were obtained from the age of sand and percentage of clay.The 8 hydraulic parameters consequently result into just 2. The climate model described by Manzi and Planton (1994) and used by Bringfelt (1996) in the Wilson and Henderson-Sellers (1985) classification is used for assigning the vegetation secondary HIRLAM model operational implementation.The range of variation of leaf area index and roughness parameters to the 18 types of classified land use.For this FAST application, the parameters related length is the same as in the other two models.The range of variation of minimum stomatal resistance to vegetation are considered independent with the exception of the soil column depth (always equal is the same as in BATS.The 13 input parameters and their corresponding range of variation range to 1 m) and fraction of vegetation (always equal to 1). are summarized in Table 3.In this analysis, the thirteen input parameters used with the FAST program are: percentage of 3.3.BAT S clay (c), percentage of sand (s), initial soil moisture in the surface layer (W s ), initial bulk soil moisture The version of the biosphere-atmosphere transfer scheme (BATS) used for this comparison is (W d ), roughness length (Z o ), leaf area index (lai), minimum stomatal resistance (Rs min ), photosyn-comprehensively described in Dickinson (1993).
The current frozen version, BATS1e, includes thetically active radiation factor for stomatal resistance (R st ), vapor pressure deficit factor for (i) calculation of soil temperature in response to net surface heating and depending on soil heat stomatal resistance (e st ), temperature factor for stomatal resistance (T st ), soil moisture factor for capacity and thermal conductivity, (ii) calculation of soil moisture, evaporation, and surface and stomatal resistance (W st ), albedo (a) and emissivity (e).The four parameters related to stomatal resist-groundwater runoff, (iii) specification of vegetation cover in terms of fractional ground shading and ance were not originally included in the list of the Table 3. L and-surface parameters used for the ISBA scheme as input to the FAST algorithm; their maximum and minimum values determine the range of variability of every parameter as it is used in the analysis  3b in Henderson-Sellers (1993)) associated with each of these 18 classes.It uses 12 soil textural plant surface for different types of land-use, (iv) surface albedo in terms of soil moisture and classes as well as 8 color classes.There are 6 soil parameters associated with each texture class and vegetation cover, (v) plant water budget including foliage and stem water storage, intercepted precip-4 reflectances associated with each color class (see Table 3 in Dickinson (1993)).Thus, to use this itation, and transpiration as limited by stomatal resistance and soil dryness, and (vi) determination scheme a total of 26 (16+6+4) parameters must be specified for each grid point.The possibility of of foliage temperature in response to energybalance requirements and consequent fluxes of changing independently all 26 parameters could lead to some unrealistic combinations of para-heat and moisture from the foliage to the canopy air.

Parameter
meters, not always existing in nature.To avoid such inconsistencies as much as possible, the fol-The surface resistance is expressed by a product of a minimum value and a number of limiting lowing assumptions were made.
(1) The six soil parameters belonging to the factors (Jarvis, 1976) depending on environmental conditions (radiation, seasonal temperature and different textures were not allowed to vary within each of the textural classes.Thus, only one para-vapor pressure deficit).
For this FAST application, subsurface temper-meter was used to characterize a single class.This parameter ranges continuously between 1 and 12 ature, soil water in the rooting layer and soil water in the total soil column are kept constant along and it is obtained by linear interpolation to the nearest soil parameters for the non-integers values the integration.Soil water is only integrated for the surface upper layer (around 10 cm).The of textural class.The textural class, considered as a continuous parameter, has no physical signific-steady-state could not be reached after 24 h due to the slow evolving soil water, even in the thin ance but it allows to estimate the sensitivity of texture.upper layer.Longer integrations change the relative role played by soil water in the upper surface (2) The same procedure was adopted for color class.One continuous parameter, varying from 1 and rooting layers.
BATS uses 18 distinct vegetation types (Table 1 to 8, was defined using the color class table.Linear interpolations between the nearest color classes in Dickinson (1993)) when coupled to the National Center for Atmospheric Research Community were computed for non-integers values of the parameter.Texture and color parameters range Climate Model (NCAR CCM) to represent different land uses.There are 16 parameters (see between the values proposed by Wilson (1984) using the Food and Agriculture Organization Soil their corresponding variation range are summarized in Table 4. Map of the World (FAO/UNESCO, 1974).
(3) Fixed values are assigned to the following vegetation parameters.(i) Depth of the three soil layers (0.1, 1 and 10 m, respectively).They are not 4. Numerical experiments independent of initial soil water depth, which is the soil water variable used in BATS.
The number of selected input parameters was (ii) Maximum fractional vegetation cover (1).It 14 for LAID and 13 for ISBA and BATS, which is not independent of leaf area index (lai);.requires 915, 695, and 695 model runs for FAST, respectively (Table 1).).In fact the simple required by FAST, by the number of environrelation d/z 0 ~2/3 seems to be fairly representative mental conditions, and by the number of PDFs of many natural vegetated surfaces (Garratt, 1992).
computed, it turns out that the total number of (4) Two parameters not used as such in the integrations was 915×243×3=667 035 for LAID scheme were allowed to vary as well, to study and 695×243×3=527 505 for ISBA and BATS.their sensitivity as compared to the other two With the uniform distribution for all of the schemes.These parameters, which have constant input parameters, all values within the defined value assigned in BATS, are the vapor pressure range are given equal probability.deficit factor and the temperature factor of the The PDF for the normal distribution is stomatal resistance.
In summary, the following 13 input parameters (1) were used with FAST: surface soil water depth (W s ), soil water depth in the rooting zone (W r ), where m and s2 are the mean and the variance of texture (t), color (c), roughness length (Z o ), minthe distribution, respectively.The mean was set to imum stomatal resistance (Rs min ), maximum leaf be the midpoint between the maximum (x max ) and area index (lai), stem area index (sai), vapor presthe minimum (x min ) value for each parameter.The sure deficit factor for the stomatal resistance (e st ), standard deviation was defined as temperature factor for the stomatal resistance (T st ), vegetation reflectance of visible wavelengths (a vis ), s=(x max −x min )/s.
( 2 ) vegetation reflectance of infrared wavelengths (a ir ) Since the normal distribution is defined as a and factor describing the sensitivity of the stomatal continuous function in the range −2 to +2, resistance to the amount of photosynthetically the chosen mean and standard deviation satisfies active solar radiation (R st ).The maximum and the condition that 98% of the PDF is contained minimum values assigned to these parameters within the range of each parameter.were obtained either from published tables The PDF of the lognormal distribution is (Dickinson, 1993) or from the corresponding values used in the other schemes (when parameters are comparable).As the role of the reflectance turned out to be very small, no attempt was made to model the covariance between infrared and where M and s2 are the mean and variance of the normally distributed ''ln x'' variable, respectively.visible reflectances.The 13 input parameters and Table 4. L and-surface parameters used for the BAT S scheme as input to the FAST algorithm; their maximum and minimum values determine the range of variability of every parameter as it is used in the analysis

Parameter
where a and b are empirical constants.The As we are mainly interested in drawing common adopted value for these constants were 0.4 and behaviour patterns from the three considered 0.3, respectively.
schemes, the relative importance of parameters In the first set of experiments, it was assumed was considered on average for the different envirthat vegetation covers 100% of the simulated onmental conditions.The rationale for designing domain.In the second set, bare soil was assumed.our numerical experiment and looking at the Thus, in this second set, evaporation was the only results this way was to find the minimum numbers mechanism extracting water from the soil.All of parameters able to explain most of the variance parameters related to vegetation were eliminated in surface heat fluxes.and, consequently, the total number of parameters was drastically reduced.Finally, the last set of 5.1.L atent heat flux experiments, performed with BATS only, explored which environmental variables are more import-Fig.1 provides the partial variance of latent heat flux for each of the input surface parameters ant for land-surface processes.For each of the 15 used with the LAID scheme.The simulated The highest partial variance is by far contributed by the soil wetness (W ), with values well domain was assumed to be completely covered with vegetation.The points along each line repres-scattered for different environmental conditions.This is true for almost all atmospheric conditions.ent the partial variances obtained for the various sets of environmental conditions, and the dia-Leaf area index (lai) is another important parameter mainly in buoyant situations, as leaf trans-monds represent the mean partial variances.Each series of points corresponds to different distribu-piration contributes mostly to latent heat flux.
The third important parameter is the roughness tion of input parameters, namely, from left to right: lognormal, uniform and normal (see length (Z o ). High partial variance is associated with stable stratification, whereas the scheme is Section 4).
One fact is first noticeable: No large differences not sensitive to this parameter in unstable buoyant conditions.Two other plant parameters, the appear among the three distributions.This similarity was already pointed out by Collins and Avissar extinction coefficient (k v ) and the maximum relative conductance (Cs max ), have a marginal impor-(1994) and by Liu and Avissar (1996).It corroborates the robustness of the results, since the selec-tance, again in buoyant conditions.The other parameters have virtually no impact on the surface tion of the PDFs is somewhat arbitrary.heat fluxes under the environmental conditions length (Z o ) also explain most of the variance.Two other vegetation parameters, namely, the min-considered here, for the selected parameter ranges.
Results obtained by Collins and Avissar (1994) imum stomatal resistance (Rs min ) and the water vapor deficit factor for the stomatal resistance for fully vegetated terrain demonstrate that most of the variance of surface heat fluxes may be (e st ), share the rest of the variance for the latent heat flux.described by the distributions of relative stomatal conductance and surface roughness.If relative Fig. 3 represents the partial variance of latent heat fluxes for each of the input parameters used stomatal conductance is parameterized as a function of the environmental variables, as was done with the BATS scheme.The results obtained with this scheme for the simulated domain fully covered here, then it appears that soil wetness and leaf area index inherit most of the sensitivity in land-with vegetation are similar to those obtained with LAID and ISBA, except that here soil water is surface processes.
Fig. 2 shows the partial variance of latent for distributed between the surface (W s ) and the rooting layer (W r ).As in ISBA, the water vapor deficit each of the input parameters used with the ISBA scheme.Results for the simulated domain fully factor for the stomatal resistance (e st ) shows a none negligible partial variance for the latent heat covered with vegetation are basically similar to those obtained with LAID.Indeed, bulk soil wet-flux.However, what is distinctive here is the significant partial variance for texture (t).This fact ness (W d ), leaf area index (lai), and roughness was already pointed out by Wilson et al. (1987) rooting layer becomes more relevant than water in the surface layer.for the case of tundra vegetation.Texture partial variance shows differences among the three distri-Using the factorial method with the BATS scheme, Henderson-Sellers (1993) obtained results butions, reaching higher values for the lognormal distribution, which gives more weight to the consistent with ours, considering the differences of both approaches.She found that the most import-sandy textures.
The previous BATS results, obtained with a ant are vegetation roughness length, soil porosity, and the light sensitivity factor and, to a lesser root system having 80% of roots in the upper surface layer, give to moisture in the surface layer extent, soil and vegetation reflectances.It must be stressed that her results were obtained for long predominancy over water in the rooting layer.Such tendency is reversed either by distributing simulation periods, which were independent of the initial conditions and, therefore, precluded the use roots predominantly in the rooting layer, or by integrating over longer periods.Several days of of initial soil water as a parameter.Furthermore, the procedure used to assign values to the para-integration allow the slowly evolving surface soil water to reach some equilibrium determined by meter ranges, based on the upper and lower quartile values for each of the BATS variables, soil water in the rooting zone, and by atmospheric conditions.In both cases, i.e., longer integrations was somewhat arbitrary, and excluded parameters as important as the maximum leaf area index and or less roots in the upper layer, water in the the uniform distribution are more spread out, with flux for each of the input parameters used with the BATS scheme.The results obtained with this higher average values, suggesting that the greater probability of the extreme values for parameters scheme for the simulated domain fully covered with vegetation are similar to those obtained with with a uniform distribution tends to give more importance to the roughness length and less to LAID and ISBA, except that here soil water is distributed between the surface and the rooting the other vegetation parameters.
Fig. 5 shows the partial variance of sensible heat layer.flux for each of the input parameters used with the ISBA scheme.Results are basically similar to those obtained with LAID.Indeed, soil wetness, 5.3.Case of bare soil leaf area index, and roughness length also explain most of the variance.Two other vegetation para-For the bare soil case using the LAID scheme, soil moisture, followed by roughness length and meters, namely, the minimum stomatal resistance and the water vapor deficit factor for the stomatal texture, are the most important parameters for both heat fluxes.Emissivity and albedo are only resistance, share the rest of the variance for the latent heat flux.The emissivity partial variance is marginal on average.Soil albedo is far less important than in Collins and Avissar (1994), due to the significant at night for the sensible heat flux.The very simplified treatment of the canopy layer in smaller variation range selected here, to avoid any type of snow or ice albedo.ISBA, without any energy budget equation associated with it, seems to be the source of this excessive Again, soil moisture, roughness length, and texture again carry most of the variance for both partial variance.
Fig. 6 provides the partial variance sensible heat surface heat fluxes, when the ISBA scheme is used.Tellus 50A (1998), 3 Fig. 6.Same as Fig. 3, but for sensible heat flux.
Soil texture parameters show higher partial vari-and in the rooting zone layer was 50 and 500 mm, respectively.Figs.7 and 8 provide the partial ance for sensible heat flux.
Results using the BATS scheme in case of bare variance of latent and sensible heat fluxes obtained for the different cases, respectively.Each point soil show sensitivity mostly to soil water content mainly in the surface layer.Texture is the second appearing in those figures refers to each of the 15 vegetation types considered in the BATS scheme most important parameter, consistently with the vegetated case.Roughness length and color are (ice and water excluded), and the three columns for each parameter are for the three different hardly relevant for both latent and sensible heat fluxes.
PDFs.The vegetation types appear to be grouped in three categories: low vegetation (type 1, 2, 7, 9, 10, 11, 13, 16, and 17), forest (type 3, 4, 5, 6, and 5.4.Importance of environmental variables 18) and bare ground (type 8).When there is enough water in the ground, the most relevant The FAST technique was also applied for various environmental conditions ranging between the parameter for sensible heat flux is solar radiation, followed by air temperature.This is not really maximum and minimum values given in Table 5, and using the BATS model.An intermediate value surprising, given that solar radiation is the main source of energy, which is redistributed into latent was adopted for the soil color (6) and soil texture (7).The vegetation parameters were fixed (corres-and sensible heat fluxes.Temperature, as already mentioned, determines the atmospheric surface ponding to each of the vegetation types described in (Dickinson, 1993)).Soil water content was also layer stability and, consequently, the magnitude of the heat fluxes to the atmosphere.Fig. 7 shows fixed: initial water content in the superficial layer Fig. 7. Partial variances of latent heat flux obtained from the FAST analysis using the BATS scheme.Environmental variables (temperature (T), relative humidity (RH), wind (W), solar radiation (Rs) and atmospheric radiation (Ra)).are now the input parameters.Computations are carried out for each of the species contemplated in the BATS scheme.For soil color (6) and soil texture (7) an intermediate value were chosen.Initial water content in the superficial layer and in the rooting zone layer was 50 and 500 mm, respectively.that for the latent heat flux, relative humidity is included among the environmental variables.
However, we have preferred along this work to also relevant, but much less than solar radiation.In bare ground, water can be extracted only consider environmental forcing without precipitation.Otherwise, the importance of the initial soil through the surface.Thus, when the upper layer is dry, most of the energy received at the ground moisture considered as parameter would have been masked.Furthermore, the role of total pre-surface is converted into sensible heat, even if the deeper soil layers are wet.Results show that, for cipitation can change depending on its intensity along the integration period.latent heat flux, the role of solar radiation, relative humidity, temperature, and wind are about similarly important.This set of experiments was carried out using 6. Conclusions the environmental conditions appearing in Table 5.It is well acknowledged, as it has already The three land-surface processes schemes considered in this study seem to be sensitive to been stressed in the previous seccions, that soil water content is extremely important for the vari-approximately the same set of parameters.Soil wetness is the most important parameter in con-ation in the Bowen ratio.Consequently, precipitation, as main source of soil moisture, would have trolling surface heat fluxes.The amount of water in the soil plays a crucial role in the redistribution played a paramount role if it would have been of energy into the atmosphere, either as sensible Results obtained here are consistent with previous studies (i.e., Collins and Avissar, 1994;heat, or as latent heat.Roughness length controls mechanical stress and, as a result, turbulent Henderson-Sellers, 1993), which attribute to soil moisture, surface roughness, albedo and, when exchanges with the atmosphere.Its role is mostly important under stable conditions, when buoy-vegetation covers the ground, leaf area index and stomatal resistance, a key role in the land-surface ancy is small or non-existant.Vegetation parameters, mostly leaf area index (lai), constrain the fluxes.When stomatal resistance is expressed as a function of soil wetness and environmental vari-vegetation response.However, the stomatal mechanism is mainly controlled by environmental con-ables, then soil wetness becomes even more predominant.ditions, including soil moisture.Assuming optimal atmospheric conditions and soil water availability, The major production terms of turbulent kinetic energy (TKE) near the ground surface are shear the maximum transpiration is determined by lai and Rs min .The other parameters appearing in the and buoyancy (Garratt, 1992).Mechanical production by shear depends on the surface stomatal resistance (or conductance) function, which in principle are species dependent, do not momentum flux and the vertical gradient of horizontal wind speed.Production by buoyancy seem to affect very much the heat fluxes.The other land-surface parameters are relatively depends on the vertical thermal stratification, and on the sensible heat flux.If shear production is important only in some of the models, or under particular atmospheric conditions, e.g., vegetation dominant, as is the case under stable conditions, roughness length is the most important parameter albedo for sensible heat flux during day time, soil texture for BATS, emissivity for sensible heat flux for the land-surface scheme.If, on the other hand, buoyant production is dominant, as is the case during night time only in the ISBA scheme, etc.
transpiring and non-transpiring Table

Fig. 1 .
Fig. 1.Partial variances of latent heat flux obtained from the FAST analysis using the LAID scheme in a fully vegetated domain.The 3 distributions described in Section 4 (see text for details) were used to represent the variability of roughness length (Z o ), leaf area index (lai), soil texture (t), soil emissivity (e g ), soil albedo (a g ), vegetation emissivity (e v ), vegetation albedo (a v ), extinction coefficient (k v ), soil wetness (W ), maximum relative stomatal conductance (Cs max ), temperature coefficient in relative stomatal conductance (T st ), water vapor deficit coefficient in relative stomatal conductance (e st ), soil moisture coefficient in relative stomatal conductance (W st ), solar radiation coefficient in relative stomatal conductance (R st).

Table 1 .
T he minimum number of model runs expressed as partial variances of single parameters, can be very much affected by the selection of that

Table 2 .
L and-surface parameters used for the L AID scheme as input to the FAST algorithm; their maximum and minimum values determine the range of variability of every parameter as it is used in the analysis a The variation range is ±2°the model value.b The variation range is ±25% of the model value.

Table 5 .
Input values of prescribed environmental variables used for the experiments