Impact of soil map specifications for European climate simulations

Abstract

Soil physical characteristics can influence terrestrial hydrology and the energy balance and may thus affect land–atmosphere exchanges. However, only few studies have investigated the importance of soil textures for climate. In this study, we examine the impact of soil texture specification in a regional climate model. We perform climate simulations over Europe using soil maps derived from two different sources: the soil map of the world from the Food and Agricultural Organization and the European Soil Database from the European Commission Joint Research Center. These simulations highlight the importance of the specified soil texture in summer, with differences of up to 2 °C in mean 2-m temperature and 20 % in precipitation resulting from changes in the partitioning of energy at the land surface into sensible and latent heat flux. Furthermore, we perform additional simulations where individual soil parameters are perturbed in order to understand their role for summer climate. These simulations highlight the importance of the vertical profile of soil moisture for evapotranspiration. Parameters affecting the latter are hydraulic diffusivity parameters, field capacity and plant wilting point. Our study highlights the importance of soil properties for climate simulations. Given the uncertainty associated with the geographical distribution of soil texture and the resulting differences between maps from different sources, efforts to improve existing databases are needed. In addition, climate models would benefit from tackling unresolved issues in land-surface modeling related to the high spatial variability in soil parameters, both horizontally and vertically, and to limitations of the concept of soil textural class.

Appendices

Appendix 1: Parameterization of ET and vertical water transport in TERRA_ML

This section described some selected aspects of TERRA_ML which are of particular relevance for our study. A more exhaustive documentation can be found at http://www.cosmo-model.org/content/model/documentation/core/default.htm.

1.1 Evapotranspiration

The parameterization of E is similar to that of the BATS model (Dickinson 1984). Evapotranspiration includes the following components in TERRA_ML:

• Bare soil evaporation E _b

• Plant transpiration E _p

• Evaporation from interception and the snow reservoir

Interception evaporation is negligible, as well as evaporation from snow reservoir since we concentrate on summer in the analysis. We thus focus here on E _b and E _p .

Bare soil evaporation E _b is parameterized as

$$ E_b=(1-f_i) \cdot (1-f_{\rm snow}) \cdot (1-f_{\rm{plant}}) \cdot {\rm Min} [-E_{\rm pot}(T_{\rm{sfc}});F_m ] $$

(1)

where F _m is the maximum moisture flux that the soil can sustain and $f_{\rm plant}$ is the fractional vegetation area and is given as an external parameter field. F _m is parameterized as

$$ F_m = \rho_w C_k D \frac{s_t}{(z_u z_t)^{1/2}} $$

(2)

where C _k is computed as

$$ C_k = 1+1550 \frac{D_{\rm min}}{D_{\rm max}} \cdot \frac{B-3.7+5/B}{B+5} $$

(3)

with B as defined for each soil class in Table 1 and

$$ D_{\rm min} = 2.5 \cdot 10^{-10} {\rm m/s}^2 $$

(4)

$$ D_{\rm max} = B \Upphi_0 K_0 / \rho_{wm} . $$

(5)

Here $\Upphi_0 = 0.2m$ is the soil water suction at saturation and ρ _wm  = 0.8 is the fraction of saturated soil filled by water, while B and K ₀ depend on the soil class (see Table 1). D is expressed as

$$ D=1.02 D_{\rm{max}} s_u^{B+2} (s_t/s_u)^{B_f} $$

(6)

with B _f given by

$$ B_f = 5.5 -0.8 B \left [ 1+0.1(B-4) {\rm log}_{10} \frac{K_0}{K_R} \right ] $$

(7)

with K _R  = 10⁻⁵ m/s.

In Eqs. 2, 6, s _u and s _t are average values of soil water content normalized by the volume of void ($\theta_{\rm{PV}}$) for two layers. These layers approximate Dickinson’s layers (0 − 0.1m and 0 − 1m) by setting the lower boundary (n _u and n _t ) as the lowest layer for which the lower boundary does not exceed 0.1 m and 1 m, respectively.

$$ s_{u,t} = \frac{\sum_{k=1}^{n_u,t} W_k}{\theta_{\rm{PV}} \sum_{k=1}^{n_u,t} \Updelta z_k} $$

(8)

where W _k is the water content of layer k (in meters).

Plant transpiration E _p is parameterized as

$$ E_p=f_{\rm{plant}} \cdot (1-f_i) \cdot (1-f_{\rm snow}) \cdot E_{\rm pot}(T_{\rm{sfc}}) r_a (r_a+r_f)^{-1} $$

(9)

i.e. similarly to Dickinson (1984) but with additional assumptions (see online documentation for more details). Here, atmospheric resistance r _a is given by r ⁻¹ _a  = C ^d _q |v _h | = C _A and foliage resistance is given by r ⁻¹ _f  = r′ C _F  = C _V , with $C_F = f_{\rm{LAI}} r_{\rm{la}}^{-1},$ $r_{\rm{la}}^{-1} = C' u_\star ^{1/2}$ and $r'=r_{\rm la} (r_{\rm la}+r_s)^{-1}.$ ${f_{\rm LAI}}$ is the leaf area index, given as an external parameter, and the stomatal resistance r _s is defined by

$$ r_s^{-1} = r_{\rm{max}}^{-1} + (r_{\rm min}^{-1}-r_{\rm max}^{-1}) [F_{\rm rad} F_{\rm wat} F_{\rm tem} F_{\rm hum}] $$

(10)

with r_min = 150 s/m and r_min = 4,000 s/m. The functions F describe the influence of the following conditions on the stomatal resistance: radiation (F _rad), soil water content (F _wat), ambient temperature (F _tem) and ambient specific humidity (F _hum), with F = 1 for optimal conditions and F = 0 for unfavorable conditions. In particular, we note the function describing the water limitation:

$$ F_{\rm wat} = {\rm Max} \left [ 0 ; {\rm Min} \left (1 ; \frac{\theta_{\rm{root}}-\theta_{\rm{PWP}}} {\theta_{\rm TLP}-\theta_{\rm{PWP}}} \right ) \right ] $$

(11)

where θ_root is the liquid water content fraction of the soil averaged over the root depth, θ_PWP is the permanent wilting point (see Table 1) and θ_TLP is the turgor loss point of plants, parameterized following Denmead and Shaw (1962) as

$$ \theta_{\rm TLP} = \theta_{\rm PWP} +(\theta_{\rm FC}-\theta_{\rm PWP}) \cdot (0.81 + 0.121 \arctan(E_{\rm pot}(T_{\rm sfc}) - E_{\rm pot, norm})) $$

(12)

with E _{pot, norm} = 4.75 mm/day.

For these two components, potential evaporation E _pot is expressed as:

$$ E_{\rm pot}(T_{\rm{sfc}}) = \rho C_q^d |v_h| (q^v-Q^v(T_{\rm{sfc}})) $$

(13)

where T _sfc is the temperation at the surface (uppermost soil layer for both E _b and E _p ), and Q _v is the saturation specific humidity. |v _h | is the absolute wind speed the the lowest grid level above the surface and C ^d _q is the bulk-aerodynamical coefficient for turbulent moisture transfert, calculated diagnostically.

1.2 Vertical soil water transport

The vertical water transport is based on Richards equation, which in the vertical direction is usually expressed as:

$$ \frac{\partial \theta}{\partial t} = \frac{\partial}{\partial z} \left[K_w \left(\frac{\partial \psi}{\partial z} + 1 \right) \right] $$

(14)

where θ is the soil water content and ψ is the water potential. On the right side of Eq. 14, $\frac{\partial \psi}{\partial z}$ refers to capillary forces, while 1 represents gravity. However, in TERRA_ML, this equation is expressed using only θ and not ψ. To do so, hydraulic diffusivity is introduced as $D_w = K_w \frac{\partial \psi}{\partial \theta}$ and thus

$$ K_w \frac{\partial \psi}{\partial z} = K_w \frac{\partial \psi}{\partial \theta} \frac{\partial \theta}{\partial z} = D_w \frac{\partial \theta}{\partial z} $$

(15)

This leads to the equations used in TERRA_ML, where the soil water flux is expressed as

$$ F = -\rho_w \left[ -D_w \frac{\partial \theta}{\partial z} + K_w \right] $$

(16)

and the change over time in soil water content in each layer is defined as

$$ \frac{\partial \theta}{\partial t} = \frac{1}{\rho_w} \frac{\partial F}{\partial z} . $$

(17)

Here, the vertical transport due to gravity and capillary forces is represented by K _w and D _w , respectively. Note that hydraulic conductivity and hydraulic diffusivity represent the same physical characteristics of the soil, namely the ability of water to flow into it, but they express it in different units. The presence of both variables is specific to this modelling approach. In some other land-surface models (e.g. Community Land Model, see Lawrence et al. 2011), water potential is used and hydraulic diffusivity does not appear.

In addition, runoff is parameterized for any layer with θ > θ_FC and a negative divergence of the fluxes (16).

Hydraulic diffusivity D _w and hydraulic conductivity K _w depend on the water content θ as:

$$ D_w(\theta_l) = D_0 \exp \left ( D_1 \frac{(\theta_{\rm{PV}}-\theta_l)}{(\theta_{\mathrm{PV}}-\theta_{\rm{ADP}})} \right ) $$

(18)

$$ K_w(\theta_l) = K_0 \exp \left( K_1 \frac{(\theta_{\rm{PV}}-\theta_l)}{(\theta_{\rm{PV}}-\theta_{\rm{ADP}})} \right) . $$

(19)

θ is defined for each layer as $\theta_l = \frac{W_l}{\Updelta z_l}.$

Appendix 2: Conversion of the soil maps

2.1 FAO and JRC soil maps: raw data

FAO The Soil map of the World released by the FAO is available at a resolution of 5 arc minutes and in geographical projection. The raw data used is taken from the Digital Soil Map of the World cd-rom (see http://www.fao.org/icatalog/search/dett.asp?aries_id=103540). The classification considers three classes reflecting soil texture: coarse, medium, and fine. For use in TERRA_ML, data from the top layer of the soil is considered.

JRC The Soil Geographical Database of Eurasia (SGDBE, see Lambert et al. 2002) at a scale of 1:1,000,000 is a digitized European map of the soil and related attributes. It is part of the European Soil Database, a product released in 2006 by the JRC (Morvan et al. 2008; Panagos et al. 2012) and available at http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDBv2/index.htm. It contains a large number of attributes, of which two are used in this study. These two attributes reflect the properties of the top layer of the soil, thus being consistent with the data used from the FAO soil map. The soil class is derived from the attribute “dominant surface textural class of the STU” (“TEXT_SRF_DOM”). This attribute contains the classes listed in Table 4. Non-soil classes are derived from the attribute “Soil major group code of the STU from the 1990 FAO-UNESCO Soil Legend” (“FAO90-LEV1”), which contains 28 soil categories and 6 non-soil categories. Non-soil categories are listed in Table 5. More details about the attributes is given at http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDBv2/popup/sg_attr.htm.

Table 4 Categories of the attribute “TEXT_SRF_DOM” in the JRC soil map

Table 5 Non-soil categories of the attribute “FAO90-LEV1” in the JRC soil map and their conversion into TERRA_ML classes

2.2 Conversion to TERRA_ML format: resolution and classes

FAO The conversion of the raw data into TERRA_ML classes with the desired resolution is done using the PrEProcessor of time invariant parameters (PEP) of COSMO-CLM (Smiatek et al. 2008). In this code, the number of points for each textural class (coarse, medium, fine) within a grid cell is determined and a weighted mean texture is computed. The assigned class in TERRA_ML is a function of this weighted mean. Ice, rock and peat are then added where the majority of points is belonging to one of these categories. The implementation in COSMO-CLM is also described by Doms et al. (2011).

JRC For input into COSMO-CLM, the JRC data was first resampled from its original 1 km resolution in Lamberts azimuthal projection to 1 arc second resolution in geographical projection by nearest neighbor interpolation. Classes were then converted to corresponding classes in TERRA_ML. For non-soil classes, the category “glacier” was converted to the class “ice” in TERRA_ML, while all other non-soil classes were converted to “rock”, as shown in Table 5. However, non-soil classes were attributed only where there was no data available about the soil class in the attribute “TEXT_SRF_DOM” (i.e. where the value is 0). For soil classes, a conversion scheme was defined by refering to the look-up table described by Smiatek et al. (2008) and comparing it to the definition of the classes in Table 4 based on the proportion of clay, sand and silt. The resulting conversion scheme is shown in Table 6. Comparing the legend of the two classes in Table 6 gives us confidence in the chosen scheme. Note, however, that there is no ideal conversion scheme. For instance, here no class in the JRC data is converted to the TERRA_ML class “sandy loam” (coarse to medium). Conversely, the two classes “fine” and “very fine” in JRC are converted to the same class in TERRA_ML (clay, i.e. fine). These two cases illustrate the difficulty to translate a soil dataset with given soil categories into other categories, and, therefore, the associated uncertainties. Grid points where no information on both soil and non-soil categories was available (e.g. over North Africa) were filled using the original FAO soil map.

Table 6 Conversion table of the soil classes from JRC (attribute “TEXT_SRF_DOM”) to TERRA_ML

Finally, the aggregation at the model resolution (0.44° in our case) is done using the COSMO-CLM PEP program as described by Smiatek et al. (2008). The method used in this program is the majority approach, i.e. the soil class with the higher number of points within a grid cell is attributed to that grid cell.