Effects of soil wetness, plant litter, and under-canopy atmospheric stability on ground evaporation in the Community Land Model (CLM3.5)



[1] The National Center for Atmospheric Research (NCAR) Community Land Model Version 3.5 (CLM3.5) has significantly improved the simulation of hydrologic cycles compared to its earlier version (CLM3.0) owing to a series of new and modified parameterizations for canopy and soil processes. One of the key elements is the addition of a soil resistance to effectively reduce soil evaporation (Es) and improve the partitioning of evapotranspiration. This soil resistance, however, is found to be physically inconsistent under wet soil conditions and implicitly include the effects of dead leaves. A new treatment with three components are proposed here: (1) two different approaches to better reflect the soil moisture limitation to Es, the so-called α and β methods combined and a new soil resistance; (2) a new surface resistance to explicitly represent the effect of plant litter cover on water vapor transfer; and (3) an explicit consideration of the effect of under-canopy atmospheric stability on the under-canopy turbulent resistance. The effects of each modification vary locally and seasonally, and their combination leads to regional differences between CLM3.5 and our new formulations. Our new formulations tend to have higher Es over high latitudes and similar or slightly higher Es in dry regions. A larger reduction of Es by the new formulations is also found over regions with relatively wet soil and more vegetation, in better agreement with previous ET partitioning studies.

1. Introduction

[2] The NCAR Community Land Model Version 3.0 (CLM3.0) [Oleson et al., 2004] was developed as the land component of the Community Climate System Model version 3.0 (CCSM3.0), which contributed to the Fourth Assessment Report by the International Panel on Climate Change [Intergovernmental Panel on Climate Change, 2007]. The performance of this land model has been evaluated by the community [e.g., Dickinson et al., 2006; Hack et al., 2006; Lawrence et al., 2007], and main deficiencies were found in the model's hydrology. One such problem was the excess soil evaporation (Es) and canopy evaporation relative to transpiration, which led to unrealistic evapotranspiration (ET) partitioning and contributed to a drier soil than observed [Lawrence et al., 2007; Oleson et al., 2008; Stöckli et al., 2008]. For example, Stöckli et al. [2008] found that the root zone soil moisture simulated by CLM3 for a temperate forest in Indiana, USA, was lower than observation throughout the year. Too dry soil limited photosynthesis and led to a bias in the simulated vegetation cover when CLM3 was coupled to a Dynamic Global Vegetation Model [Bonan and Levis, 2006]. In order to improve the hydrological cycle of the model, several modifications on soil hydrology, plant physiology, biogeophysical and biogeochemical processes were incorporated and tested as CLM3.5 [Oleson et al., 2008].

[3] One of the key elements of those modifications is the addition of soil resistance, which was adapted from Sellers et al. [1992] to represent the resistance imposed by the molecular diffusion path in the dry part of the soil. It is an empirical formula in the following form:

equation image

where fsno is the fraction of snow covered ground and s1 is soil wetness in the topsoil layer (i.e., the ratio of volumetric soil water to its saturated value). Equation (1) enables the model to effectively reduce Es and increase moisture availbility for plants to transpire. As a result, partitioning of ET became more realistic [Oleson et al., 2008; Stöckli et al., 2008].

[4] However, we suggest that there are two issues with the soil resistance in CLM3.5. First, its value is physically inconsistent under wet soil conditions, because equation (1) imposes a resistance when the soil is wet or even saturated with water (52 s m−1 for s1 = 1 by equation (1)). To illustrate this point, the soil resistance and under-canopy aerodynamic resistance are plotted from a single grid cell offline simulation of CLM3.5 (Figure 1). The model grid is centered at the Harvard forest site in Massachusetts, USA, with 100% cover of broadleaf temperate deciduous trees. The simulated topsoil layer is nearly 80% saturated throughout the year. Figure 1 clearly shows that the soil resistance dominates over the aerodynamic resistance during the winter and early spring, making the change in atmospheric resistance (i.e., stability conditions) unimportant in calculating water vapor flux. This dominance of the soil resistance over the wet soil is physically unrealistic because it is intended for relatively dry soil only.

Figure 1.

Under-canopy aerodynamic resistance (blue dots) and soil resistance (green dots) at half-hourly time steps from a single grid offline simulation of CLM3.5. The grid box is centered at the Harvard forest in Massachusetts. The simulation was forced by atmospheric data of 2003 from Qian et al. [2006]. The y axis is cut off at 500 s m−1 for ease of display.

[5] The second issue is that the soil resistance by Sellers et al. [1992] may implicitly include the effects of dead standing grass and plant litter. Equation (1) was derived using the data from the FIFE field campaign in Kansas, USA, where standing brown leaves and plant litter were also observed [Sellers et al., 1992]. Song et al. [1997] used in situ measurements of those dead plant materials to explicitly parameterize their effect on surface fluxes in a mesoscale weather prediction model. They concluded that the influence of the dead grass and plant litter was important to improve the model performance, and plant litter should decrease the evaporation from the ground.

[6] Although the soil resistance from equation (1) can be physically unreasonable over wet soil, its reduction of Es in relatively wet soil conditions did improve ET partitioning compared to CLM3.0. Thus, we aim to reduce the excess Es by not only the soil moisture limitation but also by other physically consistent parameterizations of near-surface processes. Our approach consists of three components: (1) replacing the soil resistance in CLM3.5 with alternative soil wetness functions, either adding the so-called β factor from Lee and Pielke [1992] or a new soil resistance based on Fick's law of diffusion; (2) developing a new surface resistance to explicitly represent the effect of plant litter cover on Es; and (3) including the effect of under-canopy atmospheric stability on aerodynamic resistance between the canopy air and the ground. Details of these formulations will be provided in section 2 and the results from model simulations are presented in section 3. Finally the results are discussed and summarized in section 4.

2. New Formulations

2.1. Soil Wetness Functions

[7] In CLM3.5, evaporation from the ground surface, Eg (kg m−2 s−1), is parameterized as

equation image

where ρatm (kg m−3) is the air density, qair (kg kg−1) is the specific humidity of the canopy air over vegetated ground or that of the bottom layer of the atmospheric model over bare soil, and raw is the aerodynamic resistance (s m−1) to water vapor transfer between the ground and the air. The specific humidity at the ground surface is estimated by a parameter α and the saturation specific humidity at the ground temperature, qsat (Tg). In the CLM the ground temperature Tg (K) is equivalent to the temperature of the topsoil layer in the case of snow-free surface. α is given as [Philip, 1957; Oleson et al., 2004]

equation image

where Rwv is the gas constant for water vapor (J kg−1 K−1), g is the gravitational acceleration (m s−2), and ψ1 is the soil matric potential of the topsoil layer (mm), which is a function of soil texture, volumetric water content and matric potential at saturation [Gardner et al., 1970; Clapp and Hornberger, 1978; Oleson et al., 2004].

[8] A problem with the use of this α alone to predict the surface specific humidity has been pointed out in the literature [e.g., Kondo et al., 1990; Lee and Pielke, 1992], and it comes from the misuse of α as the relative humidity at the ground surface. The parameter α is supposed to represent the relative humidity of the air within the soil pore where water evaporates, which is not necessarily at the ground surface [Monteith, 1981; Yamanaka et al., 1998; Saravanapavan and Salvucci, 2000]. Therefore the molecular diffusion process from the soil pore to the surface within the dry part of the soil is not considered in this approach. This results in an overestimation of surface specific humidity that, in turn, leads to excess evaporation. Numerous soil resistance formulations and alternative schemes have been suggested in the literature to account for the effect of the molecular diffusion process [e.g., Deardorff, 1978; Sun, 1982; Camillo and Gurney, 1986; Kondo et al., 1990; Lee and Pielke, 1992; Sellers et al., 1992; Mihailović et al., 1993; Ye and Pielke, 1993; van de Griend and Owe, 1994; Komatsu, 2003], and the soil resistance formulation included in CLM3.5 is just one of them.

[9] Because of the two issues with the soil resistance formulation of equation (1) discussed in section 1, we remove the soil resistance term from equation (2) and instead test two different formulations. First, we include an empirical factor β from Lee and Pielke [1992],

equation image

with β as a function of volumetric water content of the topsoil layer, θ1,

equation image

where θfc is the volumetric water content at field capacity, which is defined as the water content corresponding to hydraulic conductivity of 0.1 mm d−1 [Lee and Pielke, 1992]. The β function starts to decrease gradually when θ1 becomes less than θfc by an empirically derived curve fitted to observation [Lee and Pielke, 1992]. This formulation does not contain site-specific parameters, which is favorable for implementation in global models. It also agrees well with the measurement of soil evaporation within the shallow (2 cm thick) evaporation pan filled with soil sample by Kondo et al. [1990]. Similar results near the field capacity were also reported in the chamber measurement studies by van de Griend and Owe [1994] and Aluwihare and Watanabe [2003]. Lee and Pielke [1992] did not apply α to estimate the ground specific humidity as in equation (4), but Ye and Pielke [1993] pointed out that β alone does not properly predict the negative vapor flux (dew formation) over a relatively dry surface. This problem with the Lee and Pielke [1992] method was also indicated as an overestimation of nocturnal water vapor flux observed by Mahfouf and Noilhan [1991] and Mihailović et al. [1993]. We use both α and β in order to avoid this problem and also set β as one when the specific humidity profile is favorable for dew formation (i.e., qair > αqsat(Tg)), which was also done for the soil resistance in CLM3.5. The parameter β is also taken as unity for 100% snow cover, wetland, and glacier.

[10] Even though the β approach by Lee and Pielke [1992] seems reasonable in modeling bare soil evaporation, its functional form is a result of mathematical fitting to data rather than a physical relationship between the soil moisture and evaporation flux within the dry layer. Our second approach is to test a formulation that is more directly linked to the physical process within the dry layer based on Fick's law of diffusion, although it is still empirical and simplified in various aspects.

[11] Assuming that the vapor transport within the dry layer is dictated by molecular diffusion due to the gradient of water vapor and is constant along its path, the water vapor flux through the dry layer, Ed, can be written using Fick's law of diffusion [Crank, 1975; Troeh et al., 1982],

equation image

where Do is the molecular diffusion coefficient of water vapor in the air assumed here as a constant of 2.2 × 10−5 m2 s−1, qg is the specific humidity at the ground surface (i.e., moisture roughness height) (kg kg−1), τ is the tortuosity factor used to account for the difference between the actual meandering diffusion path and the thickness of the dry layer, and L is the thickness of the dry part of the soil (m) (auxiliary material Figure S1). We use Tg (i.e., the temperature of the topsoil layer) by assuming that the dry layer develops only within the topsoil layer of CLM (1.75 cm), which seems reasonable according to previous observational studies even though it can be as thick as 5 cm [van de Griend and Owe, 1994; Yamanaka et al., 1998; Aluwihare and Watanabe, 2003]. Numerous studies have measured and formulated the effect of the soil tortuosity on gas diffusion [e.g., Penman, 1940; Currie, 1960; Millington, 1959; Troeh et al., 1982; Freijer, 1994; Jin and Jury, 1996; Moldrup et al., 1999, 2004], and they commonly used D, the reduced vapor diffusivity within the soil (m2 s−1) to replace Do/τ in equation (6),

equation image

Then the soil resistance due to molecular diffusion is defined as

equation image

The empirical expression for D by Moldrup et al. [1999] is used here because it is simple enough for use in global models and performs reasonably well compared to more complex formulations [Moldrup et al., 2004].

equation image

where θsat is the volumetric water content at saturation, b is the fitting parameter for the soil water characteristic curve and depends on soil texture [Clapp and Hornberger, 1978; Cosby et al., 1984; Oleson et al., 2004], and θr is the residual water content (with a water potential of −1.0 × 108 mm, which is the minimum (strongest) matric water potential defined in CLM). Appropriate formulations of the dry layer thickness, L, for land surface models were not found in our literature search. Thus it is derived here following the assumption that the diffusion distance would be described as a function of soil water content relative to saturation [e.g., Kondo et al., 1990],

equation image

where d1 is the thickness of the topsoil layer, e is the constant (2.718), and w is a parameter that controls the concavity of the curve. The parameter w is set as 5 for this study to have an exponential shape that has often been found in experimental studies [Mahfouf and Noilhan, 1991; Sellers et al., 1992; van de Griend and Owe, 1994; Aluwihare and Watanabe, 2003] and the dependence of L on w is plotted as a function of the soil wetness (θ1sat) in auxiliary material Figure S1. The fractional term in equation (10) varies between zero (θ1 = θsat) and unity (θ1 = 0). Assuming that the water vapor flux is the same from the soil pore to the roughness height (i.e., Ed from equation (7)) versus the water vapor flux from the roughness height to the atmosphere, the expression for the water vapor flux from the soil pore below the dry layer to the reference level in the atmosphere becomes

equation image

[12] Figure 2 compares the three approaches discussed here; the soil resistance from Sellers et al. [1992] (rsoil of equation (1)), the β from Lee and Pielke [1992] (equation (5)), and the new soil resistance (rsoil,new of equation (8)). Note that the equivalent surface resistance for the β function is

equation image

and raw is taken as 50 s m−1 for simplicity in Figure 2. It is clear that rsoil is substantially higher than the other two over wet soil. The β factor does not limit evaporation when the surface soil water content is higher than the field capacity but strongly reduces evaporation over dry soil. Lee and Pielke [1992] found that the effect of β became too strong over extremely dry soil; however, the evaporation at such dry conditions is proportionately small and the modeling error should be minimal [Lee and Pielke, 1992]. The magnitude of rsoil,new derived here is located between the other two and also within the range of other soil resistance estimations [Mahfouf and Noilhan, 1991; Sellers et al., 1992; van de Griend and Owe, 1994; Aluwihare and Watanabe, 2003].

Figure 2.

Comparison of the soil moisture limitation functions for sandy loam soil. “Sellers” refers to equation (1), “L & P” refers to equation (5) plotted as a resistance by equation (12) with raw = 50 (s m−1), and “New” refers to equation (8). The y axis is in log scale.

2.2. Litter Layer Resistance

[13] In densely vegetated areas, the ground is usually covered by plant litter such as dead leaves, twigs, fruits, and dead standing short vegetation (e.g., brown leaf grass). Plant litter is known to play major roles in both managed and natural ecosystems by physically protecting the soil surface and also as an important pool of carbon, nitrogen and other biogeochemical substances [e.g., Stigter, 1984; Sayer, 2006]. It has also been observed that plant litter can substantially reduce evaporation from the ground by covering the soil surface and by having high porosity that prevents the capillary rise of liquid water from the underneath soil [Varadan and Rao,1983; Bussière and Cellier, 1994; Putuhena and Cordery,1996; Schaap and Bouten, 1997; Findeling et al., 2003].

[14] Motivated by these observations, several models for the plant litter layer have been developed over the last couple of decades [e.g., Bristow et al., 1986; Bussière and Cellier, 1994; Schaap and Bouten, 1997; Gonzalez-Sosa et al., 1999; Ogée and Brunet, 2002; Findeling et al., 2003]. For example, Gonzalez-Sosa et al. [1999] studied the effect of litter cover by comparing their Soil-Vegetation-Atmosphere Transfer (SVAT) model with an additional plant litter layer to three years of observation on a fallow field. Their litter layer submodel was based on the detailed parameterization of Bussière and Cellier [1994] that computes movements and storage of heat and water within the litter layer. In their simulation, annual evaporation was decreased by 5–10% with the litter layer, in better agreement with the observation. The simulation of other quantities such as surface energy fluxes and soil temperatures was also improved. Schaap and Bouten [1997] developed an empirical surface resistance formulation based on in situ measurements to represent the litter layer's effect on water vapor transfer. Their study showed that the contribution of the surface (litter) resistance could be higher than aerodynamic resistance. Ogée and Brunet [2002] adapted the approach by Schaap and Bouten [1997] into their soil heat and water dynamics model. They found it was necessary for the surface resistance to depend on the litter layer moisture, not on the soil surface moisture, in order to reasonably simulate the observed heat and moisture flux over the forest floor. The same model overestimated the cumulative evaporation by 7 times compared to observation over a 3-week period when the litter layer was not considered.

[15] These studies suggest that the consideration of the litter layer in CLM3.5 would be beneficial. After a literature search for diverse litter layer modeling and our own trial and error, we choose a simple resistance-type parameterization for global implementation. Our litter resistance takes the following form:

equation image

where Llittereff is the effective litter area index (m2 m−2) which will be explained later and u* is the friction velocity (m s−1) (with u*2 representing the momentum flux above the canopy based on the Monin-Obukhov similarity theory) and used here to roughly represent the wind speed within the canopy air space [e.g., Dickinson et al., 1993]. The basic functional form of the resistance is similar to the under-canopy aerodynamic resistance (s m−1) for turbulent exchange of heat and moisture in CLM3.5,

equation image

where subscripts h and w of the resistance represent heat and water vapor, respectively, and Cs is a turbulent transfer coefficient. Equation (13) is also similar to the theoretical form of litter resistance based on molecular diffusion through the litter layer [Schaap and Bouten, 1997],

equation image

where z is the diffusion distance (m) and Q is an empirical function that reduces Do according to the water content of the medium [Freijer, 1994]. We exclude the Q function for the reason explained below and include the friction velocity considering that turbulent transport can be important within the litter layer [Bristow et al., 1986; Bussière and Cellier, 1994; Findeling et al., 2003; Novak et al., 2000]. Experimental studies on the relative contribution of the molecular diffusion and turbulent transfer are still sparse and its explicit expression is beyond the scope of this study since it may vary depending on the litter type, thickness, and other environmental conditions. The constant 0.004 in the denominator of equation (13) is chosen to yield reasonable resistance values after a series of sensitivity tests. It can be interpreted to represent a combination of Do (almost constant at the temperature range near the ground surface) and the turbulent transfer coefficient at the litter layer height. The effective litter area index, Llittereff, is the fraction of plant litter area index (Llitter m2 m−2) that is not covered by snow. It represents the strength of the physical barrier to the turbulent wind as well as an expression for the tortuous path for molecular diffusion [Troeh et al., 1982; Freijer, 1994]. Llittereff is computed as

equation image

where flittersnow is the effective snow cover of the litter layer given as

equation image

where dsnow is the depth of the snow (m) and 0.05 m is assumed to be a typical depth of the litter layer. Because of the lack of a global data set of plant litter amount and the inclusion of Llitter in the stem area index (SAI, m2 m−2) in the CLM3.5 surface data, Llitter is simply taken as one for this study, which is the minimum value for the stem plus litter area index in the global surface data set for CLM [Zeng et al., 2002; Lawrence and Chase, 2007]. Since the plant litter area is part of the stem area in CLM3.5, the water storage and evaporation from the wet portion of the plant litter is implicitly included in those from the canopy. Thus only the dry part of the plant litter needs to be considered here, and the dependency of the litter resistance on the moisture content of the litter itself, as observed by others [Schaap and Bouten, 1997], is not included. Note that taking Llitter ≡ 1 can be an underestimation for forested sites but overestimation for areas with sparse canopy such as semiarid shrubland. Partitioning the global SAI data into litter and stem area indices or linking Llitter to prognostic aboveground litter in biogeochemical models coupled with CLM is the subject of future study. The litter layer is also assumed to be isothermal with the topsoil layer thus the effect of plant litter for heat transfer or heat storage is not considered.

[16] We applied the litter resistance to ground evaporation from the vegetated tiles only (i.e., not over the bare soil tile). In those vegetated subgrid tiles, the formula for the ground evaporation with the β factor becomes

equation image

Or, with the new soil resistance, it is

equation image

For the bare soil tile, equations (4) or (11) is used. When there is no litter (Llitter is zero) or all the litter is covered by snow, equation (13) converges to zero and equations (18a) and (18b) go back to equations (4) or (11).

[17] In our CLM simulations, litter resistance ranges from 0 s m−1 with complete snow cover to over 4000 s m−1 under very weak turbulent mixing (u* ≈ 0.05 m s−1). This range is comparable to the soil resistance in CLM3.5 and to the resistance by dry litter as discussed by Schaap and Bouten [1997].

2.3. Stability Correction for Under-Canopy Turbulent Transfer Coefficient

[18] Because a thick canopy layer absorbs the incoming solar radiation before reaching the ground, the canopy air becomes warmer than the ground surface, leading to a stable thermal structure under the canopy [Niu and Yang, 2004; Miller et al., 2007]. In the case of thin canopy cover, the under-canopy air can be unstable owing to the heated ground surface [Baldocchi et al., 2000]. Miller et al. [2007] collected flux measurements in the Amazon rain forest and found that stable air under canopy suppressed the vertical mixing of heat, water vapor, and carbon dioxide. These quantities were instead transported horizontally to the open area where solar radiation can reach the ground and promote localized free convection. The eddy covariance study by Baldocchi et al. [2000] showed that the surface temperature could be higher than the air temperature by 10°–15°C. Therefore, the stability effect under canopy had to be taken into consideration in order to better predict the mass and energy fluxes in forests with relatively sparse canopy. As suggested by these studies, turbulent exchange between the surface and canopy should be modified by the stability of the under-canopy air, but it is not currently considered in CLM.

[19] The aerodynamic resistances to heat and water vapor transfer between the ground and the canopy air are already given in equation (14). They depend on friction velocity and the turbulent transfer coefficient, Cs, which is a weighted sum of the values for dense canopy and bare soil [Oleson et al., 2004; Zeng et al., 2005]

equation image

The weighting function of canopy thickness, W, is

equation image

where LAI and SAI are leaf and stem area indexes (m2 m−2), respectively. Cs,bare is a function of roughness Reynolds number [Zeng and Dickinson, 1998], but Cs,dense is currently a constant value of 0.004. This value is reasonable when one computes the drag coefficient under neutral conditions for each plant functional type (PFT) in CLM, since they range from 0.012 for the grass PFT to the minimum of 0.003 for the tree PFT. However, the use of a constant Cs,dense does not represent the stable and unstable conditions, thus it can underestimate or overestimate turbulent fluxes from the ground to the canopy air. In order to reflect this stability effect, we modified and tested the formulations for stable and unstable conditions for the above-canopy surface layer from Dickinson et al. [1993]. A stability criteria similar to the bulk Richardson number is computed for under-canopy air and used to change the value of Cs,dense (=0.004). For the stable case (i.e., the canopy air temperature Tc (K) is greater than the ground temperature Tg (K)), Cs,dense is modified as

equation image

where γ is a constant and S is a stability parameter

equation image

where h is the canopy height (m). S ranges from 10−3 to 102 depending on the temperature profile and wind speed within the canopy. We choose 10 as a threshold value of S for a very stable environment to keep Cs,dense within a reasonable range. We set γ equal to 0.5 which decreases Cs,dense to 35 ∼ 15% of the original value (0.004) for weak wind speeds under broadleaf evergreen trees, for example. We consider this change as conservative and use it throughout this study.

[20] Our sensitivity tests showed that for the unstable case the model was not sensitive to the change in Cs,dense. For example, in a sensitivity test for a broadleaf temperate forest, an increase in annual mean of Cs,dense from 0.004 to 0.0071 (increase by 78%) led to an increase of latent heat flux from the ground by only 5%. At the monthly timescale, an increase in the monthly mean of Cs,dense by 70% changed the latent heat flux from the ground by only 9% in December, and an 87% increase in Cs,dense resulted in a 4% increase in ground evaporation in July. For comparison, annual mean Cs,dense was reduced to 63% of the original value by the above stable modification and led to a 25% reduction of annual mean ground evaporation. The main reason for this insensitivity is that unstable conditions typically occur when the canopy is thin so that solar radiation reaches the ground, but at the same time the weight for Cs,dense becomes small in equation (19) owing to low LAI in equation (20). Thus the total value of Cs is weighted heavily toward Cs,bare.

[21] There are other stability criteria such as the Monin-Obukhov length scale which is used widely for surface layer flux computations including CLM [e.g., Ayra, 2001; Oleson et al., 2004]. For example, Niu and Yang [2004] incorporated a stability effect in the turbulent resistance for under-canopy sensible heat flux. They scaled the resistance by the absorption coefficient of momentum which was a function of canopy density, canopy dimension, and the Monin-Obukhov length for the under-canopy layer. Their modification improved the downward sensible heat flux in their land surface model simulation for a boreal forest. However, the Monin-Obukhov length, which is also a function of sensible heat flux, requires additional computational cost in iterations. Also as pointed out by Niu and Yang [2004], the specification of one of the parameters for their formulation, canopy dimension (lm in equation (21) from Niu and Yang [2004]), was difficult to be specified for global model use. Considering these issues, we take the above approach modified from Dickinson et al. [1993] for its simplicity and reasonable sensitivity of CLM3.5.

3. Results

3.1. Global Distributions

[22] CLM3.5 was initially spun up for 20 years from 1980 to 1999, and then eight experiments were branched from this spin-up. They were: A, CLM3.5 without the soil resistance (i.e., rsoil = 0 in equation (2)); B, CLM3.5 with the soil resistance (“control”); C, A with the β method (i.e., equation (4)); D, A with the new soil resistance (i.e., equation (11)); E, A with the stability correction (i.e., equation (21)); F, A with the litter resistance (i.e., equation (18) but without the β or rsoil,new); Ga, A with the β method, litter resistance, and stability correction (i.e., equations (4), (18a), and (21)); and Gb, A with the new soil resistance, litter resistance, and stability correction (i.e., equations (11), (18b), and (21)). Each experiment was run from 2000 to 2004, and monthly mean outputs were analyzed. All the simulations were forced by Qian et al. [2006] atmospheric data.

[23] Figure 3 compares the effects of soil resistance in the control CLM3.5 and new modifications averaged over 5 years. The soil resistance by Sellers et al. [1992] (Figure 3a), the β factor from Lee and Pielke [1992] (Figure 3b, converted to a surface resistance representation by equation (12)), and the new soil resistance (Figure 3c) share very similar spatial patterns as expected. However, the effects of the β factor and new soil resistance are more focused and greater over the dry regions compared to the soil resistance in CLM3.5. For a rough idea of the effect of the β factor, aerodynamic resistance of 250 s m−1 (global mean from the simulation) and β = 0.1 give an equivalent surface resistance of 2250 s m−1 based on equation (12). Figure 3d shows litter resistance after weighted averaging according to vegetation cover fraction; thus the values shown are lower than the actual values for vegetated subgrid tiles when the vegetation cover is less than 100% in a given grid cell. Its spatial pattern is different from that of soil resistance, because it is more effective with higher plant cover where the soil is wet enough to sustain vegetation. Litter resistance shows strong seasonal variations due to its dependence on u* and becomes largest during the growing season with thickening canopy cover (not shown). Figure 3e shows the difference of the under-canopy aerodynamic resistance (raw, equation (14)) between the control CLM3.5 and the simulation with all new modifications (experiment Gb). raw is expected to be increased by equation (21) when under-canopy air is stable, which appears as the red color in Figure 3e. The resistance values were also scaled by vegetation cover in each grid before taking the difference. The stability correction is most effective in dense boreal and tropical forest and also strongly affected by season. The maximum increase in annual mean resistance is less than 200 s m−1, but during the growing season the maximum increase can be up to 400 s m−1. There are no noticeable differences in the mean litter resistance and the under-canopy aerodynamic resistance between the experiment Ga and Gb.

Figure 3.

Five-year mean of the variables by the formulations discussed in the text: (a) soil resistance (equation (2)) from the control CLM3.5 simulation (experiment B), (b) effect of the β factor in equation (5) from the simulation with all of the modifications (experiment Ga) converted into the units of resistance (s m−1) by equation (12) with the mean aerodynamic resistance of each grid, (c) new soil resistance (equation (8)) from the experiment Gb, (d) litter resistance (equation (13)) from the experiment Gb and weight-averaged according to vegetation fraction with rlitter = 0 s m−1 over bare soil, and (e) the difference in under-canopy aerodynamic resistance (equation (14)) between the new formulations (experiment Gb, Cs,dense is given by equation (21)) and the control CLM3.5 (experiment B, Cs,dense = 0.004 in equation (19)). New soil resistance, litter resistance, and under-canopy aerodynamic resistance difference from the experiment Ga are visually identical to those plotted here from Gb.

[24] Table 1 summarizes global 5-year mean hydrological components. Overall, the improvements made by all of the new modifications together, both the experiments Ga and Gb, are similar to those achieved by the soil resistance in CLM3.5. Compared with the CLM3.5 without the soil resistance (experiment A), the control CLM3.5 (experiment B) and all of the new modifications (experiment Ga and Gb) decrease soil evaporation by about 0.2 mm d−1 and slightly increase transpiration by 0.04 ∼ 0.05 mm d−1. Changes in soil infiltration, surface and subsurface runoff are also very similar. The β method and the new soil resistance provide comparable effect, except for relatively dry regions with small vegetation cover where the new soil resistance reduces Es more to some extent. They do not seem to contribute much to the improvements at the global scale, but it rather indicates that the evaporation from the dry regions is a small component at global scale. Similarly, little change was made by the stability correction, and it suggests that Es under the thick canopy is a minor contribution to the global and yearly scale water budget. It will be shown later, however, that they are important in certain regions and timescales.

Table 1. Global 5-Year Mean Hydrologic Components Excluding Grid Cells With Glaciers, Wetlands, and Lakes
ExperimentTranspirationSoil EvaporationCanopy EvaporationTotal ET mm/dQINFLa mm/dQSURFb mm/dQDRAINc mm/d
  • a

    QINFL is infiltration to the soil.

  • b

    QSURF is surface runoff.

  • c

    QDRAIN is subsurface runoff.

A: CLM3.5 w/o Rsoil310.40520.67170.221.290.860.120.50
B: CLM3.5390.44410.47200.
C: Beta320.41510.66170.221.280.880.120.50
D: new Rsoil330.41500.63170.221.260.910.120.51
E: Stability310.40520.66170.221.280.860.120.50
F: Rlitter380.44410.48200.
Ga: All New 1390.44410.47200.
Gb: All New 2400.45400.45210.

[25] Figure 4 shows the difference between the control CLM3.5 and all of the new modifications (experiment Gb) in soil evaporation percentage (Es %) of the total ET. Months with very small (negative or positive) mean evaporation components are excluded from the computation of percentage. A simple t test was performed for the statistical confidence of the difference, and those grid cells with 95% confidence are shown as red boxes in Figure 4. The same plot with the experiment Ga is almost identical, except that Ga has slightly more grid cells with statistical significance over northern Canada, where Es % is higher than the control CLM3.5 (not shown). Global mean of the difference in Es % is −0.4% for Ga with the β formulation and −1.7% for Gb with the new soil resistance, demonstrating that the mean results from the control CLM3.5 and the combination of the new modifications are very similar. However, the difference can be more than 20% in magnitude over certain regions. In general, new modifications have lower Es % in regions with intermediate canopy thickness, higher Es % over high latitudes, and similar Es % in dry regions and places with thick canopy cover.

Figure 4.

Difference between the new formulations (experiment Gb) and the control CLM3.5 (experiment B) in soil evaporation percentage (Es%) of the total ET. Red boxes indicate the grid boxes where the difference is statistically significant at the 95% confidence level based on the t test for 5 years of monthly outputs.

[26] Sensible heat flux from the soil surface (SHs (W m−2)) is increased by the new formulations in the regions where Es (or latent heat flux from the ground (LHs (W m−2))) is decreased. SHs is decreased over those regions where the new formulations yield higher LHs. Thus the SHs difference between CLM3.5 and the new modifications exhibits almost the same pattern with that of LHs but of the opposite sign (not shown). However, the change in SHs is slightly smaller than that of LHs, especially over high latitudes. This is probably because SHs itself is slightly smaller than LHs at the high-latitude region, and also the stability correction decreases SHs in thick boreal forest. Other variables such as ground heat flux, net radiation, and snow depth do not show significant differences between control CLM3.5 and the new formulations. Differences of up to ±0.1 in volumetric water content and ±2°C in soil temperature of the top layer are found over the regions with a significant difference in Es.

3.2. Regional Average

[27] Over high latitudes such as northern Canada, northern Russia and the southern tip of South America, our modifications yield higher Es % than the control CLM3.5 (Figure 4). Figure 5a and Figure 6a show the monthly mean LHs and its percentage (%) of total LH, respectively, averaged over northern Canada. The simulations with the new formulations have higher LHs and LHs % than the control CLM3.5 because the effects of all of the new modifications are strongly limited in this region owing to the small mean vegetation cover of less than 50% and the small mean total area index (TAI = LAI + SAI (m2 m−2)) of 0.1 ∼ 0.6 m2 m−2. The largest difference of more than 10 W m−2 in LHs exists between the control and the new formulations during the summer (Figure 5a). This is mostly the result of the strong soil resistance in CLM3.5 over wet soil. It imposes 200∼300 s m−1 on surface water vapor flux even though the mean volumetric water content is above 0.3 m3 m−3 that makes the β stay near one and the new soil resistance below 70 s m−1.

Figure 5.

Monthly latent heat flux from the ground, averaged over 5 years and over each region: (a) northern Canada (60°N ∼ 80°N, 80°W ∼ 110°W), (b) Europe (46°N ∼ 60°N, 0° ∼ 32°E), (c) Amazon (10°S ∼ 10°N, 50°W ∼ 70°W), and (d) western United States (29°N∼50°N, 105°W ∼ 122°W). The legend refers to the experiments listed in section 3.1. Only the results that visually differ from the simulation of CLM3.5 without the soil resistance (experiment A, gray dashed line) are shown.

Figure 6.

Similar to Figure 5 but for mean Es % of total ET over each region. Results are not shown for very small Es values.

[28] In Europe, the new formulations (experiment Ga and Gb) yield significantly lower LHs % than control CLM3.5 (Figures 4, 5b, and 6b). The soil resistance in the control CLM3.5 is rather low (40 ∼ 220 m s−1) and so are the corresponding resistance of the β factor and the new soil resistance (0 ∼ 25 m s−1) because of the relatively wet soil. On the contrary, the litter resistance is high ranging from 500 to 1200 m s−1 owing to the high vegetation cover and the intermediate canopy thickness with TAI from 1.2 to 2.8 m2 m−2 in this region. The stability correction is still less important with this medium canopy thickness, which allows LHs to be a significant portion of the total LH flux. The reduction of LHs by the litter resistance is well reflected in the partitioning (Figures 5b and 6b) in that LHs % is reduced from 40% in the control CLM3.5 to about 30% during the growing season. The result from the new formulations agrees better with an ET partitioning study over forest stands with similar canopy thickness by Barbour et al. [2005] in which 25% of ET came from ground evaporation in a temperate rain forest with LAI less than 3 m2 m−2. Other regions with relatively high vegetation cover with intermediate TAI, such as China, give similar results (not shown).

[29] Over a region with thick canopy cover, such as the Amazon basin where the mean vegetation cover exceeds 90% and the mean TAI ranges from 4.5 to 5 m2 m−2, not only the litter resistance but also the stability correction become important, especially during the dry seasons when there is less cloud cover and more solar radiation reaching the canopy to create stable under-canopy air space (Figure 5c). The stability correction raises the under-canopy aerodynamic resistance by up to 200 s m−1 compared to the control CLM3.5. The litter resistance is also effective and imposes the resistance of 650 to 950 s m−1. The β is higher than 0.9 and the new soil resistance remains lower than 20 s m−1 throughout the year while the soil resistance in the control CLM3.5 reaches 110 ∼ 190 s m−1. The new formulations together reduce LHs by 2 W m−2 (Figure 5c) and LHs % by ∼1.5% from the control CLM3.5 (Figure 6c). Even though these numbers are small, it is likely a favorable change according to Choudhury et al. [1998], who reported that 4% of ET comes from Es in the Amazon forest.

[30] Similar results were found in western Siberia and a single point simulation over boreal forest in Canada. The control CLM3.5 estimates 20% of total ET as Es during the growing season in these regions, and it is reduced to about 15% by new formulations, which again seems to be a favorable change according to other ET partitioning studies. Grelle et al. [1997] estimated 15% of the total ET as forest floor evaporation from the measurements in two evergreen needleleaf forests in Sweden. Unsworth et al. [2004] estimated 11 ∼ 12% of the total ET as forest floor evaporation in evergreen needleleaf forests in Washington, USA.

[31] The new formulations yield slightly higher Es compared to the control CLM3.5 in the western U.S. (Figure 5d), but LHs % of the total LH is quite similar especially from July to September (Figure 6d). A mean vegetation cover of about 60% in this region allows the litter resistance to reach about 600 s m−1 on grid average. The soil resistance in the control simulation reduces LHs more by applying 600 to 1200 s m−1 resistance over both the vegetated and bare soil subgrid patches. Contributions from the β factor, new soil resistance, and stability correction are very small because the annual maximum TAI reaches only 1.8 m2 m−2. Plus, it is only during the summer when the topsoil layer becomes drier than field capacity (θ1 < 0.15 m3 m−3).

4. Discussion and Conclusion

[32] Soil resistance in CLM3.5 significantly improved the model hydrology including the ratio of Es over ET [Oleson et al., 2008; Stöckli et al., 2008], but its formulation suffers from an unrealistic value over wet soil and the implicit inclusion of the plant litter effect. In this study, we revisit the soil water limitation on Es and also look for other mechanisms that control ground evaporation. Specifically, we tested the empirical β formulation by Lee and Pielke [1992], a new soil resistance based on the Fick's law of diffusion, an explicit litter resistance, and the effect of under-canopy atmospheric stability on under-canopy aerodynamic resistance. Including all modifications leads to improvements similar to those by the soil resistance in CLM3.5 at the global scale. Each modification contributes uniquely over different regions and seasons, which lead to spatial and seasonal differences in Es % of total ET between the control CLM3.5 and our new formulations. The new formulations tend to have higher Es over high latitudes and similar or slightly higher Es in dry regions, mostly depending on the vegetation cover. A larger reduction of Es by the new formulations are found over regions with relatively wet soil and more vegetation, leading to a better agreement of the simulated Es fraction with observations or independent modeling studies [Grelle et al., 1997; Choudhury et al., 1998; Unsworth et al., 2004; Barbour et al., 2005].

[33] The stability modification provides a minor contribution at the global scale, but its effect can be significant during the dry season within dense forests. The model sensitivity to this modification can be tuned by the free parameter γ in equation (21), but more detailed study with reliable observations of surface fluxes and the temperature profile of the under-canopy air space is necessary to have a better constraint. We set the γ value to 0.5 for this study so that model sensitivity lies in a conservative range.

[34] The β factor and the new soil resistance are designed to reduce the ground evaporation only under dry soil conditions as observed by several studies [e.g., Kondo et al., 1990; van de Griend and Owe, 1994; Aluwihare and Watanabe, 2003], and our model simulations show that the two formulations perform as intended. The β factor and the new soil resistance produce similar results and it is not conclusive in this study about which of two approaches is more beneficial in global models. As noted by Lee and Pielke [1992], their formulation is based on the experiment with the soil samples of 2 cm thickness [Kondo et al., 1990], thus one needs caution to apply the β approach in land surface models whose topsoil layer is much thicker than 2 cm. It also lacks explicit expressions for the diffusion process involved in the water vapor flux within the dry layer. The new soil resistance is more directly related to the physical process, and its effect is more conservative than the β method. However, it is still empirical in obtaining the effective molecular diffusion coefficient (D) and the thickness of the dry layer (L). In addition, it has not gone through rigorous comparisons with in situ observations. Nonetheless, the main point of this study is that any formulations for soil moisture limitation on Es would provide comparable results as long as they have a physically reasonable behavior at dry and wet ends of the soil moisture. Our analysis further suggests that the effects of these modifications are small at the global scale. This is not because they fail to correctly model the molecular diffusion process in the dry layer. Rather, the evaporation in dry regions is a small component of the total global ground evaporation. It follows that improved global ET partitioning in CLM3.5, owing to a single surface resistance from the soil moisture condition, does not capture the actual diverse processes at the surface.

[35] These results should not conclude that it is no use to employ the soil moisture limitation for Es over dry soils in land surface models. One of the benefits from them is to improve the unrealistic diurnal behavior without using any surface resistances or β factor. Figure 7 shows such an example from a randomly chosen grid cell in the Sahel region for a simulation of year 2004. It is the mean diurnal cycle of simulated LH for the relatively wet month of July with 70 mm precipitation. The large spike of Es in the morning is caused by the incidence of solar radiation and resulting increase of available energy and evaporation demand combined with the overestimation of qg from α alone (section 2.1). This spike is effectively removed by both the soil resistances and the β factor while neither the litter resistance nor stability correction is effective because of the little vegetation cover. In addition, one of our single point simulations in a semiarid region shows that the β factor and the new soil resistance both effectively reduces the ground evaporation as much as the soil resistance in control CLM3.5 at daily and monthly timescales, as indicated by Figure 2 for dry soils. The validation of those small-scale model results by in situ flux data is the subject of our future study.

Figure 7.

Monthly mean diurnal cycle of the simulated latent heat flux from the ground in July 2004 for a single grid in the Sahel region centered at (15°N, 17°E). The legend refers to the experiments listed in section 3.1. Only the results that significantly differ from the simulation of CLM3.5 without the soil resistance (experiment A, gray dashed line) are shown.

[36] The litter layer parameterization is simple to fit into the current CLM configuration and is able to simulate a different constraint on Es other than soil water limitation. Its magnitude is comparable to the soil resistance in CLM3.5 as well as to available observed values by Schaap and Bouten [1997]. Applying this simple but reasonable representation of the plant litter layer as resistance over vegetated tiles exerts the largest influence among our new parameterizations.

[37] Note that the litter resistance introduced in this study can be easily modified, owing to its simplicity, when better representation of the litter layer becomes possible. For example, an empirical formula could be used to predict litter amount for all the vegetated tiles from the surface data set, or the litter resistance formulation can be coupled to the prognostic plant litter mass simulated by the new biogeochemical submodel CN-DGVM [Levis et al., 2004; Thornton et al., 2007]. Behavior of the litter resistance would be changed when Llitter becomes a spatially and/or temporally changing variable, but our basic functional form and the model sensitivity are supported by regional improvements and achievement of global results similar to the control CLM3.5. When litter resistance is linked to the CN-DGVM model, it will be able to give plant litter both biophysical and biogeochemical roles in the CLM-CN-DGVM framework. Coincidently, the effect of organic soil on the soil thermal and hydraulic properties is being simulated in the climate model context [Nicolsky et al., 2007; Lawrence and Slater, 2008]. Since the plant litter is also the source for organic soil, the importance of the litter layer in the model will be further increased. Our study additionally suggests that the litter layer should be explicitly considered in the climate/earth system model, not only for its biogeochemical but also biophysical aspect.


[38] We thank NCAR CISL for providing computing resources and technical support, and we thank Michael Brunke for helpful editing of the manuscript. Two anonymous reviewers and Christopher Castro are appreciated for their constructive comments that helped improve our manuscript. This work was supported by the National Aeronautics and Space Administration grant NNG06GA24G and the National Science Foundation grant ATM-0634762.