On the separate effects of soil and land cover on Mediterranean ecohydrology: Two contrasting case studies in Sardinia, Italy


Corresponding author: N. Montaldo, Dipartimento di Ingegneria Civile Ambientale e Architettura, Università di Cagliari, Via Marengo 3, I-09123 Cagliari, Italy. (nmontaldo@unica.it)


[1] Natural mixed ecosystems (grass and woody vegetation) and managed grasslands are the dominant contrasting ecosystems of semiarid regions. These two types of land covers are known to differ in their responses to water stress, as the trees demonstrate both greater stress resistance and greater ability to tap into deeper water sources. In this study, the contrasting influences of vegetation differences (grassland versus mixed ecosystems) and soil differences (deeper alluvial valley soils versus shallow upland soils) on evapotranspiration (ET) and CO2 exchange dynamics have been examined. Data from two representative case study sites within the Flumendosa river basin on Sardinia were obtained. At both sites, land-surface and CO2 fluxes were estimated by eddy covariance instruments on micrometeorological towers. Fluxes at the two ecosystems were compared, and the effect of the vegetation cover was examined with the help of an ecohydrologic model to control for the different soil influences. The results show that the water and carbon fluxes in these ecosystems are more controlled by soil differences during the late spring, when the deeper soil depth leads to a doubling of the available moisture and an increase of 48% in the mixed natural vegetation transpiration. The system then switches to vegetation control in the summer as the presence or absence of drought-tolerant trees is the dominant imprint on continued transpiration and photosynthesis. In fact, total grassland ET in the summer is only 20% as large as the mixed vegetation ET in the summer.

1. Introduction

[2] Water limitation has a strong imprint on the soil, atmosphere, and vegetation interactions in Mediterranean regions [Baldocchi et al., 2004; Montaldo et al., 2008]. Mediterranean ecosystems are commonly heterogeneous savanna-like ecosystems, characterized by contrasting plant functional types (PFTs, e.g., grass, shrubs, and trees) competing with each other for water use [Naveh, 1967; Joffre and Rambal, 1993; Scholes and Archer, 1997; Baldocchi et al., 2004; Detto et al., 2006; Montaldo et al., 2008]. Two main subtypes of ecosystems can be distinguished here [Scholes and Archer, 1997]: grasslands and mixed grass-woodlands. Even though there is no universal relationship between vegetation and soil patterns [Ehrenfeld et al., 1997; Singh et al., 1998; Laurance et al., 1999] in water-limited conditions, we could expect to see some relationship between soil water storage capacity and vegetation type and density [Knoop and Walker, 1985; Buckland et al., 1997; Fernandez-Illescas et al., 2001; Bedford and Small, 2008] with direct consequences on the observed vegetation-type distribution. Bedford and Small [2008] demonstrated that in a semiarid ecosystems in central New Mexico, the influence of soil properties on vegetation may vary with landscape position and form, highlighting an increase of woody vegetation dimension and canopy density when moving from uplands of a hillslope (with thin coarse textured soils) to alluvial fans (with deep soils of finer texture).

[3] Additionally, the properties and patterns of soil and vegetation have been influenced over history by human activities [e.g., Vinelli, 1926; Dregne, 1986; Verheye and de la Rosa, 2005]. For example, deforestation activities have often been most intensive along the plains and alluvial river valleys, where deep soils with greater water retention capability are well suited for agriculture, whereas more natural woody vegetation (trees and shrubs) persists in the steep hillslopes and mountain areas, where soil thickness is low and less attractive for agriculture. In water-limited locations, the agriculture in the lowland areas is often pasture for grazing.

[4] The contrasting water relations of the two main PFTs (grass and woody vegetation) in these savanna-like water-limited ecosystems impacts significantly the land surface fluxes (e.g., evapotranspiration (ET)) and carbon balances [Joffre and Rambal, 1993; Scholes and Archer, 1997; Baldocchi et al., 2004; Williams and Albertson, 2004; Kurc and Small, 2007] and is therefore of importance to water resources and ecosystem management. To investigate the relationship between vegetation cover types and ET and carbon fluxes, a growing number of studies have compared measured land surface fluxes over grasslands with those over mixed grass-woodland ecosystems [Baldocchi et al., 2004; Williams and Albertson, 2004; Scott et al., 2006; Kurc and Small, 2007; Giambelluca et al., 2009; Teuling et al., 2010; Wolf et al., 2011]. However, these studies have typically neglecting the effect of soil depth and properties in the flux comparisons.

[5] Relatively to water-limited ecosystems, in particular, Kurc and Small [2007] compared micrometeorological observations in shrubland (Larrea tridentate) and grassland field sites (distance between the sites of 5 km) in New Mexico. They observed similar ET at the two sites, where in this case both were characterized by deep soils (more than 1 m). However, ecosystem carbon assimilation was notably higher at the grassland, where vegetation cover percentage (50%) was higher than in the shrubland (30%). However, the sites of Kurc and Small [2007] are in much drier climate conditions (mean annual precipitation of 230 mm) than those typical of the Mediterranean climate. Additionally they studied C4 grass species, which have greater water use efficiencies (WUEs) than the C3 grasses that are common in Mediterranean regions.

[6] Williams and Albertson [2004] compared a grass site and a mixed grass-woodland site (vegetation cover consisting of 50% herbaceous plant and 20% trees, Acacia erioloba), in Botswana, Africa (mean annual precipitation of 400 mm). There was a distance between the sites of less than 2 km, and the soil depth of >1 m in both sites. However, in contrast to Kurc and Small [2007], they observed higher carbon assimilation at the mixed site and similar ET at the two sites. Their measurements spanned a 1-month dry-down period that followed a significant rain event. It is also worth noting that these ecosystems are in semiarid regions, and plants, e.g., Acacia erioloba, are typical of South Africa and Botswana.

[7] In water-limited Mediterranean climates (mean annual precipitation of 559 mm, in California), Baldocchi et al. [2004] compared two sites (with a distance of less than 2 km), a grassland and an oak (Quercus douglasii, typical of California) savanna woodland (tree cover percentage of 40%), with similar soil depths (about 1 m). They observed slightly greater rates of ET from the grassland during spring (due to high leaf area index), and by late spring and summer, this switched to greater rates of ET from the woodland. Baldocchi et al. [2004] did not address CO2 exchange and the resulting plant WUEs.

[8] The effect of the soil depth and properties on transpiration of competing vegetation types is instead much less investigated. Tromp-van Meerveld and McDonnell [2006] investigated a forest transect along a hillslope of the Panola Mountain Research Watershed (Georgia), demonstrating an effect of hillslope position on transpiration, with water limitation apparent only on the shallow soils of the upper hillslope. However, this interesting effort did not investigate the competition between PFTs and CO2 exchange quantification.

[9] Hence, this paper addresses the need to (1) clarify the role of the vegetation type (i.e., grassland versus mixed grass-woodland) on ET and CO2 exchange dynamics and (2) understand the effect of the soil on these dynamics, because Mediterranean basins are usually characterized by contrasting soil thickness, with deep soil in alluvial valleys and thin soils on the steep hillslopes. The work is readily generalized beyond the region, as this geomorphologic influence addressed here is characteristic of many other semiarid ecosystems.

[10] The island of Sardinia is strongly representative of the Mediterranean region. It has a relatively low fraction of urbanized area and also of irrigated agriculture (around 7%) mainly in major plains where agricultural activities are concentrated. Moreover, historic deforestation has been concentrated along river valleys and mainly directed for pasture purposes [Vinelli, 1926], leaving a contrast of grasslands in the valleys and natural heterogeneous (mixed grass-woodland) ecosystems on the steeper slopes. We consider two representative case study sites within the Flumendosa river basin on Sardinia: one is a mixed grass-woodland on a shallow soil (15–40 cm), and the other is a grassland on a somewhat deeper soil (>1.5 m). These soils are more shallow than most general savannah, as is typical of Mediterranean basins. These shallow soils are characterized by more rapid hydrologic response to atmospheric forcing [Montaldo et al., 2008] than that of deep soils. During May 2005 to September 2005, two micrometeorological towers with eddy covariance instrumentation [e.g., Brutsaert, 1982; Garratt, 1992] were used to monitor land surface fluxes of energy, water, and CO2 [Detto et al., 2006; Montaldo et al., 2008] on the two sites. We focus on the growing season, because it is the most critical for understanding the relationships between vegetation and water and CO2 exchanges. These relationships are of great importance to water resources planning in Sardinia in light of the marked recent decrease of water availability [Montaldo et al., 2005, 2008].

[11] First, we compare observed land surface fluxes at the two ecosystems characterized by differences of both vegetation cover and soil depth, and then we investigate the isolated effect of the vegetation by separating out the soil effects with the aid of a model. In support of the comparison, we use the coupled land surface model (LSM) and vegetation dynamic model (VDM) of Montaldo et al. [2008]. This model is extended here to include a subroutine for CO2 exchange rate estimation, a key element of land-atmosphere interactions. The use of the coupled ecohydrologic model (LSM+VDM) allows for investigation of the relative controls of the soil and vegetation on the CO2 and water balance dynamics.

2. Case Studies

[12] The two sites are located in the municipal territories of neighboring villages, Nurri and Orroli, Italy, both in the Flumendosa river basin on the island of Sardinia. The distance between the sites is approximately 4 km. The climate is a typical Mediterranean—maritime. Mediterranean climates are characterized by their cool, wet winters and warm, dry summers. These climates are found throughout the Mediterranean region, parts of California, South America, and several other parts of the globe. The mean historical (1922–2007) annual precipitation is 643 mm (meteorological station in the village of Nurri), and the mean historical monthly precipitations varies from 93 mm in December to 11 mm in July. Historical air temperature has a mean annual value of 14.6°C; mean monthly values range between 23.7°C in July and 7.1°C in January. Micrometeorological fluxes monitored at the Nurri and Orroli sites are compared for the May 2005 to September 2005 period, when both towers were fully operational.

2.1. Orroli Site

[13] The Orroli site is a Mediterranean natural patchy mixture of woody vegetation and grass on a shallow (15–40 cm thick) silt loam soil (19% sand, 76% silt, 5% clay, bulk density of 1.38 g cm−3, and porosity of 53%). The dominant trees are wild olive (Olea sylvestris) of height approximately 3.5–4.5 m and occasional cork oaks (Quercus suber) of height approximately 6–7 m. Shorter stature plants include shrubs (Asparagus acutifolius and Rubus ulmifolius), creepers of the wild olive trees (Crataegus azarolus and Smilax aspera), and C3 herbaceous (grass) species (Asphodelus microcarpus, Ferula comunis, Bellium bellidioides, Scolymus hispanicum, Sonchus arvensis, Vicia sativa, Euphorbia characias, Daucus carota, and Bellis perennis; monocotyledons: Avena fatua and Hordeum murinum) that are present in live form only during wet seasons and reach heights of approximately 0.5 m. Details about the vegetation characteristics, directly measured in the site, are in Montaldo et al. [2008]. The root zone depth is coincident with the soil depth for these thin soils (on average 0.25 m), as noted by exploratory excavation.

[14] A 10 m micrometeorological tower was instrumented and put into service in May 2003 to measure latent heat (LE), sensible heat fluxes (H), and CO2 exchange (Fc) through standard eddy covariance methods [e.g., Brutsaert, 1982; Garratt, 1992], net radiation (Rn), surface temperature of the different PFTs, photosynthetically active radiation (PAR), soil heat flux (G), precipitation, and soil moisture (θ) within the root-zone at half-hour time step. Leaf area index (LAI) was also measured indirectly and periodically through a ceptometer (Accupar model PAR-80; Decagon Devices Inc., WA, USA) [e.g., Kovacs et al., 2009]. The details are given in Detto et al. [2006] and Montaldo et al. [2008].

[15] The distribution of the fraction of vegetation cover of the site is estimated from multispectral high spatial resolution Quickbird satellite images [Detto et al., 2006]. The combined use of the flux footprint model of Detto et al. [2006] and the high-resolution satellite images supports the interpretation of the eddy covariance surface fluxes in the context of the mean fraction of woody vegetation cover in the relevant flux footprint of the micrometeorological observations. The fraction of woody vegetation in the flux footprint varies mainly in the range of 0.1–0.22 with the peak of the distribution close to 0.15 [Detto et al., 2006; Montaldo et al., 2008]. Instead, the fraction of woody vegetation cover of the radiometer flux footprint is 0.38. The details are given in Detto et al. [2006] and Montaldo et al. [2008].

2.2. Nurri Site

[16] The grass site in Nurri is at 39°41′11.51″ N, 9°12′57″ E. The soil is mainly silty clay loam (17.7% sand, 52.2% silt, and 30.3% clay), and the soil depth is more than 1.5 m. The root zone depth of 0.6 m was noted by exploratory excavations.The site is located over an alluvial plain valley of the Mulargia river, a sub-basin of the Flumendosa. Grasses are C3 fodder plants, mainly Hordeum vulgare spontaneum, and their maximum height was 70 cm during the spring season.

[17] Soil moisture was monitored periodically in 2003 and 2004 using both gravimetric method and time domain reflectometer (TDR) Tektronix 1502C (Figure 1) in the upper 30 cm of the soil.

Figure 1.

Comparison of observations and model values of (a) ET and (b) soil moisture in the Nurri (grass-deep) site. (a) “obsEB” and “obsEC” are the observed ET estimated as residual term of the energy balance method and the observed ET using the eddy covariance method, respectively. (b) “obsTDR” and “obsgrav” are the observed soil moisture using TDR and the observed soil moisture using the gravimetric method, respectively. In Figure 1, the shaded gray area around the model calibrated values indicates the part of the ensemble of the θ predictions generated with the Monte Carlo simulation framework between the 5th and 95th percentiles. The thick black line in the upper right corner of the panels indicates the comparison period.

[18] At this site, a 7 m tower was instrumented to measure land-atmosphere fluxes. The tower was fully instrumented and operational from May to September 2005 with a Licor-7500 CO2/H2O infrared gas analyzer and a RM Young 81000 3D sonic anemometer at 6 m above the ground, a CNR-1 (Kipp&Zonen) integral radiometer at 7 m above the ground, a Campbell Scientific HMP45 C temperature and relative humidity probe, two soil heat flux probes HFT3 (REBS), two Campbell Scientific TCAV soil temperature probes, and an Environmental Monitoring ARG100 tipping bucket rain gauge; a CR23X (Campbell Scientific Inc., Logan, Utah) datalogger recorded average values of each variably every half hour.

[19] In addition, for this site, LAI was also monitored periodically through a ceptometer (Accupar model PAR-80; Decagon Devices Inc.).

[20] We underline the point that micrometeorological fluxes monitored at the Nurri and Orroli sites can be fully compared only for the May 2005 to September 2005 period, when both towers were operational. Unfortunately, soil moisture was not monitored at the Nurri site during this common measurement period. However, as soil moisture was monitored in the previous period, when basic meteorological (nonflux) data were also available, an ecohydrologic model was calibrated and validated for the 2003–2004 period and then used to reconstruct θ time series for the May 2005 to September 2005 period analysis.

3. Ecohydrologic Model

[21] The use of an ecohydrologic model is necessary for two reasons: (1) for understanding the impact of the differences in soil depth and type between the two sites on the CO2 and water balance dynamics, and (2) for reconstructing the θ records at the Nurri site for the May 2005 to September 2005 flux comparison period (Figure 1).

[22] The model is an upgraded version of that in Montaldo et al. [2008] with an added subroutine for estimating Fc. The model couples a LSM and a VDM. The LSM is based on the force-restore method [Noilhan and Planton, 1989; Albertson and Kiely, 2001; Montaldo and Albertson, 2001], and the VDM computes changes in biomass over time from the difference between the rates of biomass production (photosynthesis) and loss, as occurring through respiration and senescence [e.g., Larcher, 1995; Cayrol et al., 2000; Nouvellon et al., 2000], separately for woody vegetation and grass components. The essence of this coupling is that the VDM provides LAI evolution through time for each PFT, which is then used by the LSM for computations of the energy partitioning between soil and vegetation.

[23] The underlying model is described in Montaldo et al. [2005, 2008]. Model parameters are defined in Table 1; model meteorological inputs are given in Table 2; and main model equations are described in Tables 3 and 4. In Montaldo et al. [2008], the coupled model was applied successfully to the Orroli site.

Table 1. Model Parameters (VDM-LSM Model) for the Orroli (Mix-Shallow) and Nurri (Grass-Deep) Sites
ParameterDescriptionOrroli SiteNurri Site
GrassWoody Vegetation
LSM-VDM Parameters
rs,min (s m−1)Minimum stomatal resistance100280100
Tmin (°K)Minimum temperature272.15272.15270.15
Topt (°K)Optimal temperature295.15292.15300.15
Tmax (°K)Maximum temperature313.15318.15313.5
θwp (−)Wilting point0.0800.08
θlim (−)Limiting soil moisture for vegetation0.200.150.20
ω (HPa−1)Parameter of the relationship between canopy resistance and vapor pressure deficit0.010.010.01
Only VDM Parameters
cg (m2 gDM−1)Specific leaf areas of the green biomass in growing season0.010.0050.01
cd (m2 gDM−1)Specific leaf areas of the dead biomass0.010.005 
ke (−)PAR extinction coefficient0.50.5 
ξa (−)Parameter controlling allocation to leaves0.60.55 
ξs (−)Parameter controlling allocation to stem0.10.1 
ξr (−)Parameter controlling allocation to roots0.40.35 
Ω (−)Allocation parameter0.80.8 
ma (d−1)Maintenance respiration coefficients for aboveground biomass0.0320.00090.026
ga (−)Growth respiration coefficients for aboveground biomass0.320.450.16
mr (d−1)Maintenance respiration coefficients for root biomass0.0070.0020.007
gr (−)Growth respiration coefficients for root biomass0.10.10.1
Q10 (−)Temperature coefficient in the respiration process2.522.5
da (d−1)Death rate of aboveground biomass0.0230.00450.023
dr (d−1)Death rate of root biomass0.0050.0050.005
ka (d−1)Rate of standing biomass pushed down0.230.350.23
QN (−)Soil respiration coefficient related to temperature1.21.2
R10 (mmol CO2 m−2 s−1)Reference respiration at 10°C12.3
Only LSM Parameters
zom,v (m)Vegetation momentum roughness length0.050.50.05
zov,v (m)Vegetation water vapor roughness lengthzom/7.4zom/2.5zom/7.4
zom,bs (m)Bare soil momentum roughness length0.0150.015
zov,bs (m)Bare soil water vapor roughness lengthzom/10zom/10
θs (−)Saturated soil moisture0.530.55
b (−)Slope of the retention curve88.5
ks (m s−1)Saturated hydraulic conductivity5 × 10−61 × 10−6
|Ψs| (m)Air entry suction head0.790.356
drz (m)Root zone depth0.250.6
Table 2. Model Meteorological Inputs
Input DescriptionSymbolUnit
PrecipitationΠm s−1
Air temperatureTa°C
Wind velocityum s−1
Incoming shortwave radiationRswinW m−2
Air relative humidityRH%
Photosynthetically active radiationPARmmol s−1 m−2
Table 3. Main Equations of the Land Surface Model of Montaldo et al. [2008]
Soil water balance inline image
 where θrz is the soil moisture of the root zone, drz is the root zone depth, Ibs is the infiltration rate on bare soil, Iwv and Igr are the throughfall rates infiltrating into the soil covered by woody vegetation and grass, respectively, qD is the rate of drainage out of the bottom of the root zone, Ebs is the rate of bare soil evaporation, Ewv and Egr are the rates of transpiration of woody vegetation and grass, respectively, fv,WV is the fraction of green woody vegetation area per unit of ground area, fv,gr is the fraction of green grass vegetation area per unit of ground area, and fbs (= 1−fvt,WVfvt,gr) is the fraction of bare soil, where fvt,wv and fvt,gr are the total woody vegetation and grass vegetation area, repectively (including dead or not transpiring vegetation). EWV and Egr are estimated distinctly using the Penman-Monteith equation [e.g., Brutsaert, 1982, p. 224].
Sensible heat flux inline image
 with CH the heat transfer coefficient, and Ts and Ta the surface and air temperatures, respectively.
Net radiation inline image
 with Rlwin estimated by equations 6.10 of Brutsaert [1982], α albedo, ε emissitivity, and σ the Stefan-Boltzmann constant.
Soil heat fluxG = RnH − LE
Canopy resistance inline image
  inline image
  inline image
Surface temperature inline image
 with T2 the mean Ts value over 1 day τ, and CT the soil thermal coefficient.
Table 4. Equations of the Vegetation Dynamic Model of Montaldo et al. [2008]
Ecophysiological termEquations
Biomass components inline image for both woody vegetation and grass.
  inline image for both woody vegetation and grass.
  inline image for both woody vegetation and grass.
  inline image for woody vegetation only.
 where Bg, Bs, Br, and Bd are the green leaves, stem, living root, and standing dead biomass compartments, respectively; Pg is the gross photosynthesis; aa, as, and ar are allocation coefficients to leaves, stem, and root compartments (aa+as+ar = 1); Rg, Rs, and Rr are the respiration rates from leaves, stem, and root biomass, respectively; Sg, Ss, and Sr are the senescence rates of leaves, stem, and root biomass, respectively; and La is the litter fall.
Photosynthesis inline image
  inline image
AllocationFor woody vegetation,
  inline image
  inline image
 For grass,
  inline image
Respiration inline image, with Tm = mean daily temperature.
Senescence inline image
Litter fall inline image
Leaf area index inline image, where LAI and LAId are the green and dead leaf area index, respectively.

[24] Paralleling the approach for ET estimation [Montaldo et al., 2008], a three-component approach is implemented for estimating the total net CO2 flux [e.g., Wang et al., 2007]:

display math(1)

where Fc,WV and Fc,gr are carbon exchange of woody vegetation and grass, respectively, and Rbs is the soil respiration. Carbon exchange rates for of each PFT (e.g., Fc,wv and Fc,gr) are computed as the difference between Pg and Rg (see Table 4). Soil respiration is estimated as a function of the temperature through the following equation [Novick et al., 2004; Ruehr and Buckmann, 2009]:

display math(2)

where R10 is the reference respiration rate at 10°C, and QN is the soil respiration sensitivity to temperature. Note that any impact of soil moisture on the respiration rate is neglected here.

4. Comparisons of Micrometeorological Observations in Nurri and Orroli Sites

[25] First, the micrometeorological observations at the two sites are compared. We subdivide the May 2005 to September 2005 comparison period into two analysis periods: IP (May 16 to June 19; Figure 2), which coincides with the end of the 2005 spring, characterized by green vegetation cover at the Nurri site and by water-stressed grass at the Orroli, and the subsequent IIP (Figure 2) when grass was dead at both sites due to the dry conditions typical of the Sardinian summer, yet the trees of the Orroli site remained green and active. Hereafter for easier reading, we refer to the Nurri site as “grass-deep” and the Orroli site as “mix-shallow” to highlight the combination of cover type and soil depth as needed to support the interpretation of the results.

Figure 2.

Comparison of (a) net radiation and (b) albedo observations at the two sites. The IP and IIP periods are also indicated. (c) Relative frequencies of albedo values at the two sites are given.

4.1. Energy Balance

[26] The comparison of energy fluxes is made at the daily time scale, including daylight hours only [Crago and Brutsaert, 1996], and in units of mm d−1 of water for clear consideration of the impact on the soil water balance. Net radiation fluxes were higher in the mix-shallow site, especially during IIP (Figure 2). Note that incoming solar radiation levels (not shown for brevity) were very similar (correlation coefficient of 0.98) due to the proximity of the two sites; however, there was a significant difference in albedo values (estimated as reflected divided by incoming shortwave radiations) due to the differences in land cover (Figure 2b). In the grass-deep site, the highest albedo values (around 0.27) are estimated during the warmest days of the summer (day of year (DOY) 200–230), and when green vegetation was totally absent, the soil surface was very dry (Figure 1) and slightly cracked due to the clay component of the soil, and albedo reached values typical of desert dry soils [Brutsaert, 1982]. In contrast, at the mix-shallow site, the wild olive (Olea sylvestris) remained active in the dry conditions [Detto et al., 2006] and significantly impacted the land cover properties and, therefore, the albedo of this site, producing near constant values of albedo for the whole period. The notable difference between the albedo of the two sites is depicted in Figure 2c, where the corresponding relative frequencies are reported. Albedo of the mix-shallow site was significantly lower than the albedo of the grass-deep site and was characterized by a unimodal distribution with peak around 0.13, whereas in the grass-deep site, albedo showed a wider range and a trimodal distribution with peaks for 0.17, 0.2, and 0.25 due to the higher variability of land use during the observed period. The albedo behavior effect was to decrease Rn during the inactive period at the grass-deep site to approximately 70% of the Rn measured at the mix-shallow site (Figure 2a). Net radiation comparison results are consistent with the results of the Baldocchi et al. [2004] field experiments (the radiometer flux footprint of woody vegetation cover fractions in the oak savanna field of Baldocchi et al. [2004] and in the mix-shallow site are similar ≈40%).

[27] The energy balance was reasonably well closed for both sites (Figure 3). For the mix-shallow site (Figure 3a), the linear regression between LE + H and RnG showed a slope of 0.85 and an intercept of 0.45 mm d−1, the correlation coefficient was 0.94, and the energy balance ratio (equal to the ratio between total LE + H and total RnG of the period considered; Wilson et al. [2002]) was 0.93. For the grass-deep site (Figure 3b), the linear regression between LE + H and RnG showed a slope of 0.82 and an intercept of 1.13 mm d−1, the correlation coefficient was 0.88, and the energy balance ratio was 1.10. These statistics are in accordance with literature values [e.g., Wilson et al., 2002; Baldocchi et al., 2004], confirming the reliability and accuracy of the micrometeorological measurements for both sites.

Figure 3.

Energy balance closure of observed flux at the (a) Orroli (mix-shallow) and (b) Nurri (grass-deep) sites. For the mix shallow site: linear regression slope of 0.85 and intercept of 0.45 mm d−1, correlation coefficient of 0.94; for the grass-deep site: linear regression slope of 0.82 and intercept of 1.13 mm d−1, correlation coefficient of 0.88.

[28] The comparison of latent and sensible heat fluxes observed in Nurri site with those of Orroli site is provided in Figure 4. Interestingly, during IP, H was higher in the mix-shallow site than in the grass-deep site due to the higher surface temperatures (more than 4°C during the warmest hours of the day) observed in the mix-shallow site (correlation coefficient of 0.15, and root-mean-square difference (RMSD) of 1.8 mm d−1). Instead, in the subsequent IIP, sensible heat flux observed at both sites became high and similar in magnitude (however, still higher at the mix-shallow site) due to the high surface temperatures reached in both sites (correlation coefficient of 0.49, and RMSD of 1.48 mm d−1). Latent heat flux was higher in the grass-deep site than in the mix-shallow site during May 2005, because both LAI was higher (grass was still green) and the soil was wetter in the grass-deep field. Hence, the greater water use in the grassland was due to its more intensive water use strategy and the wetter soils in the valley when compared with the uplands. It is also likely that the valley position of the grass-deep field led to wetter conditions from convergent subsurface flow paths. In IIP, LE is shown to increase in response to a rain event (DOY 175; Figure 5) when soil evaporation contributed significantly due to the wetter soil conditions (Figure 4). After this event in IIP, LE of both the Nurri and Orroli sites converge to similar low values due to the very dry soil conditions typical of the Sardinian summer (correlation coefficient of 0.73, and RMSD of 0.55 mm d−1).

Figure 4.

Comparison of (a) sensible heat flux and (b) latent heat flux observed at the Nurri (grass-deep) and Orroli (mix-shallow) sites. The IP and IIP periods are indicated.

Figure 5.

Observed and modeled soil moisture for the four configurations during the observation period. In the secondary y axis, the precipitation time series is reported and the IP and IIP periods are indicated. The shaded gray area around the model calibrated values at the grass-deep site indicates the part of the ensemble of the θ predictions generated with the Monte Carlo simulation framework between the 5th and 95th percentiles.

[29] These results are partially in contrast with Baldocchi et al. [2004] findings. We observed larger differences between grassland and woodland ET in spring, and negligible differences in the following period, whereas Baldocchi et al. [2004] observed higher ET in the grassland during the spring and higher ET in the woodland during the summer of the California site (negligible differences during the other seasons). This could possibly be due to the differences in soil properties and depth and position (upland versus valley) between Orroli and Nurri sites (in contrast with the similar soils of the Californian sites) and the difference in the fraction of woody vegetation, which was twice as large at the California site when compared with the mix-shallow site.

[30] Interestingly, we note that similar patterns on H and Rn (both higher at the forest when compared with the grasslands during the summer) were found by Teuling et al. [2010] during the 2006 heat waves in Central and Western European, although they investigated sites in a traditionally more humid zone than our water-limited Mediterranean sites. In the present study, there is a reduced Rn (due to the higher albedo) and, therefore, lower energy availability at the grass-deep site, which led to reduced H when compared with the mix-shallow site. This was true despite the dry conditions and high surface temperature of the soil over the dead grass.

4.2. CO2 Exchange and Plant WUE

[31] In comparing Fc observations at the two sites, we note that at the beginning of IP, Fc values at the mix-shallow and grass-deep sites were similar (DOY 138–140; Figure 6a) in response to a small rain event (Figure 5). The rapid drying of the soil at the mix-shallow site led to a drop of Fc to constant values, most likely due to continued activity of woody vegetation (with wider ranging roots) that remained green through the summer. The grass-deep site, in contrast, with its deeper root zone saw a slower decay of Fc as it slowly drained the larger reservoir of stored water from the spring rains. Hence, in the grass-deep site, the grass was still green and CO2 assimilation was higher than in the mix-shallow site during IP (correlation coefficient of 0.49, and RMSD of 9.44 g C m−2 d−1; Figure 6a). In IIP, Fc observations at the grass-deep site become positive due to the soil respiration component that becomes predominant (correlation coefficient of 0.49, and RMSD of 9.44 g C m−2 d−1). Note that daily energy and carbon fluxes are computed for daylight hours, when heterotrophic respiration is expected to be a minority contributor to the net carbon flux.

Figure 6.

Comparison of (a) CO2 exchanges and (b) water use efficiency observed at the Nurri (grass-deep) and Orroli (mix-shallow) sites. The IP and IIP periods are also indicated.

[32] Ecosystem WUE observed at the two sites (Figure 6b) were similar during IP as grass at both sites had reasonable access to water; however, in IIP, WUE at the mix-shallow site was higher due to the drought-tolerant trees. At the grass-deep site, the concept of ecosystem WUE becomes meaningless as the grass died and dry bare soil becomes the predominant land cover.

[33] Through the investigation of the ecosystem WUE relationship with vapor pressure deficit (VPD; Figure 7), we note that WUE at the mix-shallow site tends to be higher than at the grass-deep site over the full range of observed VPD, although a large scatter of WUE is observed for both sites. This result confirms that the mix-shallow ecosystem is the most efficient (gaining carbon while conserving water) as attributed to the drought-tolerant wild olive trees.

Figure 7.

WUE plotted as a function of VPD for the two sites.

[34] To evaluate the effect of the vegetation cover type on the soil water budget, the relationships between energy and carbon fluxes and θ must be estimated. However, θ is available only at the Orroli site during the May 2005 to September 2005 comparison period. We use the ecohydrologic model, calibrated for the 2003–2004 period, to estimate θ for the comparison period at Nurri. The model also allows us to evaluate the role of soil type and depth on the energy and CO2 balance of the two ecosystems. Indeed, for instance, the soil at the grass-deep site, which is deeper and with higher retention capacity, allows for an extended green phase of the grass through the end of the spring season, with significant implications on land surface flux in contrast with the mix-shallow site. Hence, soil influences on land surface fluxes of the two ecosystems are evaluated below.

5. Modeling Results

[35] First, we test the ecohydrologic model for the two sites to understand the model performance and to estimate the time series of θ at the grass-deep site for the comparison period. Then, we investigate the impact of the soil properties and depth differences on the land surface fluxes and vegetation cover type.

5.1. Model Validation

[36] The Orroli heterogeneous field includes three land cover components: grass, woody vegetation, and bare soil. The model was already tested successfully for the Orroli mix-shallow site in Montaldo et al. [2008] (see Table 1 for model parameters). As demonstrated in Montaldo et al. [2008], the model predicts very well the soil moisture time series at the mix-shallow site. The new Fc simulations match well with the observations for the full observation period (2003–2006, RMSE of 6.04 g C m−2 d−1 and correlation coefficient of 0.43; Figure 8), confirming the robustness of the coupled VDM-LSM model for simulating vegetation and soil water balance dynamics. The comparison of cumulative values of modeled and observed Fc (Figure 8b) confirms the high model accuracy (difference of 4.7% of the total Fc).

Figure 8.

(a) Comparison of mean daily modeled and observed daily CO2 exchanges at the Orroli (mix-shallow) site (the thick black line indicates the comparison period) and (b) their cumulative values.

[37] At the grass-deep site, only two land cover components are present, grass and bare soil. Past and present management (e.g., clearing, fires, grass cuts, and planting) of the site largely control the land cover and the LAI evolution during the year. Root zone depth of the model at the grass-deep site is 0.6 m based on field observations (Table 1). The accuracy of model predictions for the grass-deep site is demonstrated in Figure 1, where soil moisture and ET observations and predictions are compared. The whole period of observation (2003–2005) is considered for a better evaluation of the model. Unfortunately, the soil moisture was monitored only during the 2003–2004 period, and thus, the model has been tested successfully for that period only. However, as the model is also able to well predict soil moisture at the close Orroli site, the soil moisture model is appropriated. Note that ET observations are estimated here as the residual of the energy balance (= RnHG, see section 2.2) for the November 2003 to August 2004 period and by using the eddy covariance technique for the May 2005 to September 2005 period. Statistical performance results for the model are as follows: for ET (only for the comparison period with eddy covariance data), the RMSE is 0.78 mm d−1 and the correlation coefficient of 0.87; and for the soil moisture (only for the period with observed data), the RMSE is 0.05 and the correlation coefficient is 0.91. For the comparison period (May 2005 to September 2005), the model also predicts well the Fc dynamics (Figure 9), correctly reproducing the CO2 assimilation rate decline into the dry summer period (the RMSE is 3.84 g C m−2 d−1, and the correlation coefficient is 0.824). From July 2005, the grass is absent, and therefore, only soil respiration is estimated. The comparison of cumulative values of modeled and observed Fc (Figure 9b) confirms the model accuracy (difference of 7.5% of the total Fc).

Figure 9.

(a) Comparison of mean daily modeled and observed daily CO2 exchanges at the Nurri (grass-deep) site and (b) their cumulative values. The shaded gray area around the model calibrated values at the grass-deep site indicates the part of the prediction ensemble generated with the Monte Carlo simulation framework between the 5th and 95th percentiles.

[38] Note that due to the higher transpiration rates (i.e., higher canopy conductance) of the grass-deep site during the relatively wet IP (Figure 4b), the minimum stomatal resistance model parameter, rs,min, is less than half the value at the mix-shallow site model (Table 1). On the contrary, for predicting the high drought-tolerant capacity of the woody vegetation at the mix-shallow site, the wilting point is set equal to 0 (Table 1 and Montaldo et al. [2008]).

[39] With focus on the comparison period (May 2005 to September 2005), we see that θ of the grass-deep and mix-shallow sites are largely different, especially during IP (Figure 5). The decay of θ at the grass-deep site is slower due to the larger depth and water retention capability of that soil. However, due to the lack of observed θ values during the comparison period (May 2005 to September 2005), we generate uncertainty estimates in the soil water model forθ predictions and estimate the resulting impacts on ET and Fc predictions. Using a global multivariate approach [Franks et al., 1997; Montaldo et al., 2003] based on a Monte Carlo simulation framework, we generated 3000 simulations varying simultaneously and independently the key parameters of the soil water model (Table 1) over a range of realistic values (ks = 10−7 to 10−5 m s−1, Ψs = 0.2–0.7 m, b = 7–10, θs = 0.4–0.65). The 5th and 95th percentiles of the ensemble of the θ predictions are plotted with the calibrated θ in Figures 1 and 5. The spread of the ensemble decreases for lower soil moisture values whereas increases for wetter conditions due to the higher influence of soil parameters. Finally, the influence of θ uncertainty on ET and Fc predictions is negligible, as demonstrated in Figure 1a (for ET) and Figure 9 (for Fc).

5.2. Relative Influences of Soil and Vegetation Types on the Land Surface Fluxes

[40] As suggested above, it is not only vegetation type differences but also differences of soil properties and depth that may impact land surface fluxes. We examine here the relative influences of vegetation and soil on land surface fluxes through an analysis of model estimated land surface fluxes for the vegetation and soils of the two sites. We use the model to estimate the hypothetical energy, water, and vegetation dynamics for the Orroli vegetation on the soil properties (i.e., soil model parameters of Table 1: ks, Ψs, b, θs, and drz) of the Nurri site (hereafter “mix-deep” configuration). This is to estimate how the natural mixed Mediterranean vegetation (e.g., of Orroli) would function if on the typical alluvial valleys of the Flumendosa basin (e.g., Nurri soils). Hence, this is a numerical transplant experiment that allows for the estimation of land surface fluxes of the supposed natural environment (i.e., natural mixed vegetation) in the alluvial valley before the human-induced land use changes. For a complete analysis, the case of grass vegetation on the Orroli soil (hereafter “grass-shallow” configuration) is also examined. This numerical transplant experiment further depicts the effect of deforestation along the shallow soils on the uplands. We compare these hypothetical results with the data from the two sites.

[41] The effect of the soil type on the soil moisture time series is depicted in Figure 5. When the natural mixed vegetation is on the valley soils (mix-deep), we see markedly greater θ values (more than double) in IP than when on the upland soils (mix-shallow). In the valley, the natural vegetation would not reach the low summer levels of θ until late July as opposed to early May for the upland case. In contrast, the second numerical transplant experiment (grass-shallow) confirms the significant effect of the soil, which dramatically decreases its water content when grass is moved from the valley to the uplands. Furthermore, note that differences of θ between the mix-shallow and grass-shallow configurations are negligible, confirming the key role of the soil.

[42] Hereafter, we investigate the impact of the soil on the vegetation components of ET (i.e., transpiration, inline image) and total net CO2 flux (CO2 assimilation, inline image). In Figure 10, we report for the four configurations the comparison of cumulative Tv and Fcv (Figures 10a and 10b, respectively) and, for better summarizing the results, the comparison of the total Tv and Fcv separating IP and IIP contributions (Figures 10c and 10d, respectively). The soil and vegetation effects can be isolated as follows:

[43] (a) The vegetation effect is revealed through comparing the grasslands with the natural mixed vegetation on identical soils (e.g., grass-deep versus mix-deep or grass-shallow versus mix-shallow).

[44] (b) The soil effect is revealed from the contrast between the same vegetation on the upland soils and on the valley soils (e.g., mix-shallow versus mix-deep or grass-shallow versus grass-deep).

Figure 10.

(top) The comparison of predicted cumulative values of (a) transpiration and (b) CO2 assimilation for the four configurations is shown. The shaded gray areas around the model calibrated values at the grass-deep and mix-deep sites indicate the parts of the prediction ensembles generated with the Monte Carlo simulation framework between the 5th and 95th percentiles. (bottom) The corresponding total (c) transpiration and (d) CO2 assimilation in IP and IIP periods for the four configurations are shown (dotted lines indicate the 5th and 95th percentiles of the predicted ensembles).

[45] For transpiration (Figures 10a and 10c) in the springtime (IP), it is clear that the soil effect is dominant as the natural vegetation on the shallow soils (mix-shallow) has a much lower cumulative transpiration (52% less) than when on the deeper valley soils (mix-deep) and similarly for the grass vegetation. Furthermore, the natural vegetation on the valley soils has a similar spring time transpiration as the managed grass. However, in the summer (IIP), the vegetation effect is dominant as the natural vegetation, whether on the upland or the deeper valley soils, has a continued but modest rate of transpiration due to the drought-tolerant trees; however, grass died for both soils with a significant decrease of Tv (grassland Tv is around 25% of natural mixture Tv). In total, the natural vegetation in the uplands transpires 20% less than the grasslands in the valley; however, where the natural vegetation to have been located in the valley, we would expect it to have consumed 37% more water than a similarly located grassland (Figure 10a).

[46] In addition, the CO2 fluxes (Figures 10b and 10d) in the springtime (IP) are dominated by the soil effects with Fcv of the mixed vegetation on the valley soil being higher (51% more) than those over the upland soils. In the summer (IIP), however, the grass went into senescence at both sites, and the dominant control of the magnitude of Fcv was the vegetation type. Note that for the deep soil configurations (mix-deep and grass-deep), we account for uncertainty in the soil model parameters (see section 5.1); however, the effects are negligible on transpiration and carbon assimilation, and thus, the above described results are concluded to be robust.

[47] Interestingly, although the effect of changing vegetation (from C3 grass to mixed vegetation) on the valley soil of Nurri site produces a total increase of 37% for transpiration consumption, this is greatly outweighed by the corresponding 83% increase in carbon gain.

6. Conclusions

[48] The concurrent micrometeorological observations at the Nurri and Orroli sites highlight differences in land surface flux dynamics at sites with contrasting vegetation cover and soils. As is typical of many Mediterranean settings (and in general of semiarid climates), the low-lying valley site is grass covered with deep soils and the upland site is a natural mixed vegetation (trees and grass) on a shallow soil. Although the two sites started the observation period with similar soil moisture states, the deeper soils and the propitious position of the valley site, where convergent flow paths provide an additional input of water, allowed for prolonged persistence in soil moisture into the summer when compared with the more shallow upland site. The Nurri site (grass-deep) with its deeper valley soil had an extended wet condition until the beginning of the summer and supported higher ET rates when compared with those of the thinner soil at the upland Orroli site (mix-shallow). Later, during the typical dry summer 2005, the wild olive trees of the Orroli sites continued to function, capturing carbon at high rates and showing high WUE. In contrast, at the managed grassland in the valley, the carbon assimilation shutdown due to the dry summer conditions and the CO2 fluxes were largely due to soil respiration.

[49] An extended version of the ecohydrologic model of Montaldo et al. [2008] was used to assist in the isolation of the relative effects of vegetation and soil on the water and carbon fluxes, in a form of numerical transplant experiment. The ecohydrologic model allowed for estimation of the effect of the soil depth on the soil water balance, ET, and carbon exchange, demonstrating a marked potential increase of ET and carbon assimilation at the Nurri site (grass-deep) if the vegetation were still natural mixture vegetation (woody and grass vegetation). In other words, the land use change of deforestation in the valley has led to a moderate decrease in growing season transpiration but a much greater decrease in carbon gain (i.e., ecosystem productivity). The effect is more significant during the end of the spring season and the start of the summer season, which is a key period for the water resources and forest planning of the Sardinia region.

[50] The results support a final conclusion that the variability in water and carbon fluxes in these ecosystems is more controlled by soil differences during the late spring, when differences in soil depth and in the position along the hillslope lead to important differences in moisture availability and switches to vegetation control in the summer as the presence or lack of drought-tolerant trees has a strong imprint on continued transpiration and photosynthesis.

[51] Finally, we note that limitations of the transplant experiment are mainly due to the assumptions on the soil moisture and root system vertical distributions, as the root zone was modeled in bulk without considering particular soil moisture and root biomass vertical stratifications. This assumption may limit the representation of the root-soil-water interactions in the root zone, which can exhibit variability along the soil profile when the soil is deeper (such as for the Nurri field). Future developments will investigate the importance of including plant root vertical distribution model.


[52] This work was supported by the RegioneSardegna LR 7/2007 through grant CRP2_708 and by the National Science Foundation's (NSF) Hydrologic Science Program through grant NSF-EAR-08-38301. The authors thank Marco Mancini for the support in the preliminary efforts on the tower setup. They acknowledge the local authorities of Orroli and Nurri for their support. The authors also thank Christopher A. Williams and two anonymous reviewers for their useful comments.