Ecohydrological effects of management on subalpine grasslands: From local to catchment scale



[1] Grassland and pastures are important land uses in subalpine and alpine environments. They are typically subjected to management practices that can change the biophysical structure of the canopy through defoliation and can alter soil hydraulic properties. These modifications have the potential to impact hydrological and energy fluxes as well as the primary productivity of grasslands. We investigate how a series of management practices, such as grass cut, grazing, and the consequent soil compaction due to treading by animals are affecting water resources, flood generation, and grassland productivity in a subalpine region. Results are obtained using a mechanistic ecohydrological model, Tethys-Chloris. The model is first confirmed using energy, water, and carbon fluxes measured at three eddy covariance stations over grasslands in Switzerland and discharge measured in a small experimental catchment. A series of virtual experiments are then designed to elucidate the importance of various management scenarios at the plot and catchment scales. Results show that only severe management actions such as low grass cuts or heavy grazing are able to influence considerably the long-term hydrological behavior. Moderate management practices are typically unable to modify the system response in terms of energy and water fluxes. An important short-term effect is represented by animal-induced soil compaction that can reduce infiltration capacity leading to peak flow considerably higher than in undisturbed conditions. The productivity of vegetation in absence of nutrient limitation is considerably affected by the different management scenarios with tolerable disturbances that lead to higher aboveground net primary production.

1. Introduction

[2] Grassland is one of the principal land covers of the Earth's surface [Hansen et al., 2000] and it occurs in a wide range of climates. In Switzerland, for instance, it occupies about 25% of the entire territory [Leifeld et al., 2005; Zeeman et al., 2010]. Assessing the dynamics of grassland ecosystems is therefore important to study energy, water, and carbon fluxes at the Earth surface [Nouvellon et al., 2000; Wever et al., 2002; Novick et al., 2004; Asner et al., 2004; Baldocchi et al., 2004; Baldocchi, 2008]. European alpine grasslands are typically managed ecosystems, sometimes heavily managed, to support the traditional grassland farming system, “Alpwirtschaft” [Zeeman et al., 2010]. During spring and summer periods, cattle grazing on higher-elevation pastures contribute to the removal of a significant portion of aboveground biomass. Grass that is not grazed in European subalpine and alpine meadows typically undergoes several cuts during the growing season to be used as a source of forage (in the form of hay or silage). In both cases, this leads to a severe disturbance of the ecosystem natural conditions.

[3] Grass cuts mainly affect hydrology through the biophysical effect of defoliation, whereas animal grazing has a double effect because it adds to defoliation a compaction of soil structure that generally reduces the infiltration capacity [Tate et al., 2004; Bilotta et al., 2007]. Grazing, as well as grass cutting, reduces the aboveground biomass, thus impacting the green and dead Leaf Area Index (LAI), which in turn controls energy fluxes and consequently transpiration and soil moisture. Effects of management on vegetation height and proportion of alive and dead biomass also have the capability to affect surface roughness and albedo, modifying the land surface turbulent exchanges and radiation balance.

[4] Several studies attempted to quantify the impact of grassland management on carbon, nitrogen fluxes, soil carbon storage, and species diversity [Collins et al., 1998; Asner et al., 2004; Leahy et al., 2004; Ammann et al., 2007; Wohlfahrt et al., 2008a, 2008b; Ammann et al., 2009; Lazzarotto et al., 2009; Zeeman et al., 2010]. However, considerably less research has investigated the effect of grassland management in terms of hydrological response, looking at quantities such as energy fluxes, transpiration, evaporation, soil moisture, or discharge, mostly due to difficulties in the observations and experimental constraints. Former studies mainly analyzed the impact of grazing animals on soil properties, highlighting how intense grazing activities may increase runoff rates, through increasing soil density and decreasing preferential flow and soil hydraulic conductivity [Gifford and Hawkins, 1978; Greenwood et al., 1997; Dreccer and Lavado, 1993; Mwendera and Mohamed Saleem, 1997; Kohl and Markart, 2002; Schneider et al., 2008; Leitinger et al., 2010; Stavi et al., 2011; Wine et al., 2012]. The results of these studies, which are typically based on empirical experiments where a small part of a pasture was exposed to sprinkling rainfall, or few hectares were fenced to impose different stocking densities, differ considerably in the extent to which grazing activity affects runoff production. Furthermore, few attempts were made to infer the effect of grazing on the long-term water availability and most studies pertained to semiarid grasslands [Bremer et al., 2001; Frank, 2003; Miao et al., 2009; Zhao et al., 2010; Wang et al., 2012] but see Inauen et al. [2013] for an analysis in the alpine area.

[5] The present study aims to fill this information gap by providing answers to the following questions: are different management scenarios in terms of grazing, summarized by stocking density (number of livestock units per hectare), and grass cut at different heights, controlling (a) water balance and water yield in subalpine grasslands, (b) flood generation at the catchment scale, and (c) vegetation productivity? The study is thought to provide quantitative answers to the above questions using a mechanistic modeling approach and to overcome the space and time constraints of experimental studies. In this regard, we expect to provide a direct link between the alteration of the biophysical structure of grass due to management practices and energy, water, and carbon fluxes, which can confirm and integrate knowledge derived from empirical studies. For the case of subalpine and alpine grasslands, there is also a practical interest in quantifying the effects of management practices due to the progressive cessation of grazing or cutting of mountain pastures (abandoning) due to economic unattractiveness [Inauen et al., 2013].

[6] In order to achieve these goals, the mechanistic ecohydrological model Tethys-Chloris [Fatichi et al., 2012a, 2012b] was modified from its original formulation and confirmed at three grassland sites with flux towers (“Fluxnet network”) in Switzerland: Oensingen, Chamau, and Fruebuel, where energy and carbon fluxes were measured using the eddy covariance approach [Baldocchi, 2003; Aubinet et al., 2012]. The three locations have different elevations and intensities of grassland management. The hydrological performance of the model was further confirmed in a small experimental catchment, Rietholzbach, typical of the prealpine setting and mostly covered by meadows.

[7] A sensitivity analysis was carried out with the ecohydrological model at the plot scale and for the Rietholzbach watershed to investigate the consequences of management activities on the components of the hydrological cycle and on vegetation productivity. The sensitivity analysis was structured through a series of synthetic experiments. We first investigated the biophysical effect of grass cutting and grazing imposing several types of disturbances that resemble typical management practices. In the case of grazing by livestock, we successively superposed an alteration of soil properties to the biophysical effects of grass canopy removal.

[8] We adopted a modeling approach because testing experimentally the long-term effects of various management practices in different locations is simply unfeasible due to time, cost, and logistic constraints. Furthermore, a distributed analysis, essential to understanding impacts at the watershed scale, would have been impracticable given the difficulties in identifying pairs of catchments with similar properties and in imposing specific management practices. Numerical modeling appears to be, so far, the only way to investigate the presented research questions. New generation mechanistic ecohydrological models [Ivanov et al., 2008, 2010, 2012; Fatichi et al., 2012a; He et al., 2014] make this feasible due to their capabilities of (i) accounting for many important feedbacks between hydrology and vegetation dynamics, (ii) simulating the time evolution of vegetation functioning and structure, and (iii) having the flexibility of imposing different management scenarios.

2. Data Description

[9] The data set and corresponding locations used for model confirmation and subsequent simulations have been described in previously published research. Therefore, only a brief summary is provided here. The interested reader can refer to the cited literature and supporting information Text S1.

2.1. Chamau, Oensingen, and Früebüel

[10] Eddy covariance measurements together with meteorological and additional hydrological variables were collected in the three experimental facilities Chamau, Oensingen, and Früebüel for 3 (2006–2008), 2 (2002–2003), and 3 (2006–2008) years, respectively (Figure 1). The locations differ in elevation and in mean climatic properties with Chamau and Oensingen more representative of the Swiss Plateau and Früebüel of the alpine region (Table 1). Soil can be classified as loam and clay loam according to the soil texture of the three locations (Table 1). The principal vegetation type for Chamau is a mixture of Italian ryegrass (Lolium multiflorum), white clover (Trifolium repens), and smooth meadowgrass (Poa pratensis L.). Oensingen has a complex mixture of over 30 grass, clover, and herb species [Ammann et al., 2007], and vegetation in Früebüel is dominated by a mixture of species consisting of meadow foxtail (Alopecurus pratensis), cocksfoot grass (Dactylis glomerata), ryegrass (Lolium sp.), dandelion (Taraxacum officinale), buttercup (Ranunculus sp.), and white clover (Trifolium repens) [Sautier, 2007; Hartmann and Niklaus, 2012]. Grass was cut six/seven times per year with the first cut occurring at the beginning of May in Chamau [Zeeman et al., 2010], while cuts were typically applied three times per year with the first cut not taking place before 1 June in Oensingen [Lazzarotto et al., 2009]. In Früebüel, grass was cut three/four times per year with the first cut occurring in mid June [Zeeman et al., 2010; Hartmann and Niklaus, 2012].

Figure 1.

Map of Switzerland topography with the geographical identification of the locations used in this study.

Table 1. Description of the Principal Characteristics of the Locations in Switzerland Used for the Confirmation of Model Results and Simulationsa
  1. a

    Chamau, Früebüel, and Oensingen are the three experimental facilities equipped with meteorological stations and eddy covariance towers, Rietholzbach is an experimental catchment with meteorological and discharge measurements, Sion has only meteorological data.

Coordinate47.21°N, 8.41°E47.11°N, 8.53°E47.28°N, 7.73°E47.37°N, 8.99°E46.22°N, 7.33°E
Soil sand/silt/clay fractions35/40/2530/40/2025/32/44Not availableNot available
Number of cuts per year6/743Not availableNot available
Mean precipitation obs. period (mm yr−1)1156169011871459597
Mean air temperature obs. period (°C)

[11] For a complete description of the three locations and instrumental equipment, the reader can refer to Ammann et al. [2007, 2009]; Lazzarotto et al. [2009]; Gilgen [2009]; Gilgen and Buchmann [2009]; and Zeeman et al. [2010].

2.2. Rietholzbach and Sion

[12] Rietholzbach is a small experimental catchment located in northeastern Switzerland (47.37°N, 8.99°E; elevation: 679–924 m a.s.l., Figure 1), that drains an area of 3.31 km2 dominated by grass (74.5%) with a relatively minor cover of evergreen trees (2%), deciduous trees (5%), and mixed evergreen/deciduous trees (18.5%) (Figures 2a and 2b). Hourly observations of standard meteorological variables and runoff in the catchment started in 1975, and they were successively complemented with soil moisture and water table depths. A weighing lysimeter was also installed within the catchment to measure storage of water and long-term evapotranspiration [Calanca, 2004; Teuling et al., 2009]. The catchment is located in a prealpine setting with a range of elevation of about 250 m (Figure 2). Mean precipitation and air temperature at the meteorological station during the observational period of 1976–2005 are 1459 mm yr−1 and 7.0°C, respectively. Snow cover is rather common during winter. Precipitation integrated at the watershed scale using a gridded precipitation product [Wüest et al., 2010] is 1570 mm yr−1. Soil depth is not uniform and ranges from <0.5 m on the steeper slopes up to >2 m in the valley bottom [Germann, 1981]. More information on the catchment and observations are available through Teuling et al. [2010] and Seneviratne et al. [2012].

Figure 2.

A representation of topographic and vegetation attributes of the Rietholzbach experimental catchment, (left) Digital Elevation Model 50 × 50 m2, and (right) vegetation map. The circle in the left figure indicates the locations of the meteorological station. “Mixed.” refers to a mixed deciduous and evergreen forest.

[13] Meteorological forcing and boundary conditions for an additional plot-scale location, Sion (46.22°N, 7.33°E; elevation: 482 m a.s.l.), in southwest Switzerland (Figure 1) were selected to carry out further numerical experiments. The average air temperature and annual precipitation of Sion in the period of 1988–2011 are 10.1°C and 597 mm yr−1 (Table 1). Although data were not available for confirming model performance for this specific location, the drier climate conditions when compared to the other sites (the driest conditions in Switzerland) were considered important to broaden the results of the study.

3. Model Description

[14] The mechanistic ecohydrological model Tethys-Chloris (T&C) is designed to simulate coupled dynamics of energy-water-vegetation in different environments and climates [Fatichi, 2010; Fatichi et al., 2012a, 2012b; Fatichi and Leuzinger, 2013]. All the principal components of the hydrological cycle, such as precipitation interception, transpiration, ground evaporation, infiltration, and surface and subsurface water fluxes are accounted for. The model solves the ecohydrological dynamics over complex topography of a watershed, explicitly considering spatial variability of meteorological fields and the role of topography in controlling incoming radiation and transferring water laterally through the surface and subsurface. Heterogeneity in soil properties and vegetation are accounted for. The basic computational elements are represented using cells of a regular grid with a typical area of 25–5000 m2.

3.1. Hydrological Component

[15] Incoming energy in the form of shortwave and longwave radiation is explicitly transferred through the vegetation [Ivanov et al., 2008]. The energy, water, and carbon exchanges between the surface and the atmospheric surface layer are solved accounting for aerodynamic, undercanopy, and leaf boundary layer resistances, as well as for stomatal and soil resistances [Sellers et al., 1997]. In each element, vegetation can occupy two vertical layers, in a way to accommodate for the coexistence of tree and grasses. Horizontal composition of vegetation is also possible since each element can account for multiple species or plant functional types [Fatichi et al., 2012a; Fatichi and Leuzinger, 2013]. Dynamics of water content in the soil profile are solved using the one-dimensional (1-D) Richards equation for vertical flow and the kinematic wave equation for lateral subsurface flow, making the approach quasi 3-D. Saturated and unsaturated parts of the soil column are explicitly identified. Surface overland and channel flow are also solved through the kinematic equation. Snowpack dynamics are accounted for by solving the energy balance. Snow can be intercepted by the vegetation or it falls to the ground, where it accumulates and successively melts. Runoff generation is made possible via saturation excess and infiltration excess mechanisms and depends on lateral moisture fluxes in the unsaturated and saturated zones as well as in runon of overland flow [Loague et al., 2010]. With respect to the original version of the model [Fatichi et al., 2012a, 2012b], the soil heat flux is computed solving the heat diffusion equation [Hillel, 1998] using the method of lines [Lee et al., 2004]. This allows the calculation of soil temperature for each layer of the soil column, and offers an additional metric for model confirmation. Another change introduced in this study concerns the possibility for water to pond at the surface until a microroughness depth threshold is exceeded. The presence of surface ponded water implies different albedo and thermal properties for the surface, allows direct water evaporation and impedes ground evaporation.

3.2. Vegetational Component

[16] Photosynthesis is simulated with a biochemical model [Farquhar et al., 1980; Collatz et al., 1991, 1992] that for this study has been modified following Bonan et al. [2011]. The original “big-leaf” assumption made in the model has been replaced by a “two big leaves” scheme, where sunlit and shaded leaves are treated separately when computing net assimilation and stomatal resistance [de Pury and Farquhar, 1997; Wang and Leuning, 1998; Dai et al., 2004]. An exponential decay of photosynthetic capacity is used to upscale photosynthesis from leaf to plant scale [Ivanov et al., 2008; Bonan et al., 2011]. Stomatal resistance is parameterized as a function of assimilation rate and environmental conditions [Leuning, 1990, 1995; Tuzet et al., 2003]. The dynamics of five carbon pools are explicitly simulated in the model and include green aboveground biomass, living sapwood (woody plants only), fine roots, carbohydrate reserve (nonstructural carbohydrates), and standing dead biomass (necromass). The latter carbon pool is introduced in this study since in grassland a considerable fraction of aboveground biomass can remain after management (i.e., as standing stubs, or grass clippings left after harvest) or senescence before transformation into ground litter occur. Such a standing dead biomass can modify the albedo, the roughness, and the partition of incoming energy [Rosset et al., 2001]. In the model, the green biomass that dies is transferred to the standing dead biomass pool and only successively is turned over according to a rate function of the air temperature [Lazzarotto et al., 2009]. The specific leaf area index of dead biomass is assumed equal to the green biomass and its optical properties are parameterized from literature [Asner et al., 1998; Oleson et al., 2010].

[17] The carbon assimilated through photosynthetic activity is used for growth and reproduction and is lost in the process of maintenance and growth respiration and tissue turnover. Carbon allocation is a dynamic process that accounts for resource availability (light and water) and allometric constraints [Friedlingstein et al., 1998; Krinner et al., 2005], e.g., a minimum ratio of fine root carbon to foliage carbon; and an upper limit for the storage of carbohydrate reserve [Kozlowski and Pallardy, 1997; Friend et al., 1997; Bonan et al., 2003; Krinner et al., 2005]. Carbon from reserves can be translocated to favor leaf expansion at the beginning of the growing season or after a severe disturbance [Chapin et al., 1990; Gough et al., 2009, 2010; Fatichi et al., 2012a]. Organic matter turnover of the different carbon pools is a function of tissue longevity and environmental stresses, i.e., drought and low temperatures [Sitch et al., 2003; Bonan et al., 2003; Arora and Boer, 2005; Ivanov et al., 2008; Fatichi et al., 2012a]. Phenology is tracked considering four states [Arora and Boer, 2005]: dormant, maximum growth, normal growth, and senescence that control patterns of plant allocation [Fatichi et al., 2012a]. With comparison to Fatichi et al. [2012a], a threshold on the photoperiod length is added to the already existing controls of soil temperature and soil moisture to simulate the beginning of the growing season. This helps to better reproduce phenology of species the leaf onset of which is related to day length [Körner and Basler, 2010; Polgar and Primack, 2011]. The normal phenological cycle can be modified after severe management disturbances. When cuts or grazing activity reduce the LAI of grass below a critical threshold, grass starts a new growing season provided that soil moisture, photoperiod length, and temperature conditions are still favorable. Nutrient dynamics and forest stand growth are neglected in the model, which thus always considers a mature vegetation and a nonlimited nutrient soil. These assumptions can be limitative in many real cases.

[18] Further details of model computational setup, structure, and description of process parameterizations are presented in Fatichi et al. [2012a].

4. Model Confirmation

[19] The ecohydrological model was tested to reproduce energy and carbon fluxes at the plot scale at the three locations of Oensingen, Chamau, and Früebüel which were equipped with eddy covariance towers. Performance in reproducing hydrological fluxes and the overall water budget at the watershed scale were tested against the observations of Rietholzbach for the 23 year period of 1983–2005. A list of the parameters used in the model simulations for grassland is presented in Table S1 (supporting information). Results were obtained without significant calibration once literature parameters for specific vegetation types were chosen. Specifically, the parameters for grassland were kept constant for all of the locations with the exception of the maximum Rubisco capacity (see Table S1 in the supporting information). Regular grass cut was imposed for the three locations during the growing season, according to the available data [Lazzarotto et al., 2009; Zeeman et al., 2010]. Specifically, six grass cuts from day of the year (doy) 125 to 278 (average intercut distance of 30.6 days) were imposed for Chamau, three grass cuts from doy 125 to 245 (average intercut distance of 50 days) for Oensingen, and four grass cuts from doy 170 to 265 (average intercut distance of 31.6 days) for Früebüel. In typically managed grassland (Chamau and Früebüel), the number of grass cuts and the period between the first and last cuts are mainly a function of elevation [Zeeman et al., 2010]. Grass was cut at the conventional height of 7 cm [Zeeman et al., 2010] and is converted into Leaf Area Index, LAI, using the Allen et al. [1989, 1998] equation, LAI = 24 Hveg, for grass height, Hveg (m), of less than 15 cm. Different empirical equations provided similar LAI for heights of less than 20 cm [Pocock et al., 2010].

4.1. Plot-Scale Confirmation

[20] In the simulation experiments at the plot scale, a flat domain is assumed consistently with the experimental setup at the flux towers, i.e., no lateral effects such as surface/subsurface inflows or local and remote obstructions of shortwave radiation are considered. The soil moisture and vegetation carbon pool initial conditions are obtained after a spin-up of the model with a simulation of 7 years created repeating sequentially the observations. Soil hydraulic and thermal properties (supporting information Table S1) were estimated through pedotransfer functions [Saxton and Rawls, 2006; Oleson et al., 2010] and are site specific. A free drainage condition is assumed at the bottom of a 1.3 m soil column for all of the locations. Vegetation cover is assumed to occupy the entire patch in all of the simulations. This means that LAI is spatially homogenous within a patch, i.e., an intermediate LAI is not given by a subfraction of the patch with bare soil and a subfraction with vegetation with high LAI but by an intermediate uniform LAI.

[21] The capability of the model to reproduce the daily cycles of mean and standard deviation of net radiation, sensible heat and latent heat for the location of Chamau is shown in Figure 3. The correlation between hourly time scale observations and simulations are satisfactory, although simulated standard deviations seem to be larger than the ones inferred from observations during the day time. The good performance in reproducing net radiation and the high correlations question whether the tendency of the model to overestimate latent and especially sensible heat is a model shortcoming or not. This is difficult to assert because eddy covariance observations suffer of a common problem of many flux towers worldwide [Wilson et al., 2002; Leuning et al., 2012], that is they do not close the energy budget by about 10%, conversely to model simulations. The results maintain their consistency even when they are disaggregated for the different months (not shown). Similar good performances in terms of simulation of energy fluxes are also obtained for Oensingen and Früebüel, with only significant deviations in Oensingen during the extreme heat wave of 2003, the impact of which in terms of plant water stress is overestimated by the model (see Figures S1 and S2 in the supporting information).

Figure 3.

A comparison between the observed (OBS) and simulated (SIM) average daily cycles of net radiation, latent heat, and sensible heat for the location of Chamau. The triangles represent the standard deviations. Scatter plots with the determination coefficients R2 are also shown.

[22] The overall realistic simulation of energy and hydrological dynamics is evident from the relatively good agreement between simulated and observed soil moisture at three different depths (15, 25, and 40 cm) and soil temperature at four different depths (10, 15, 25, and 40 cm) in the years 2006 and 2007, especially considering the uncertainty in defining hydraulic and thermal properties from pedotransfer functions (see Figures S3 and S4 in the supporting information). At high water contents soil moisture is slightly underestimated and soil temperature daily fluctuations overestimated. Nonetheless, the determination coefficients, R2 and the Root Mean Square Error, RMSE, between observed and simulated soil moisture θ, and soil temperature Ts, at the different depths are: R2 = 0.53 − 0.54 for θ and 0.95–0.96 for Ts, and RMSE = 0.025 − 0.058 − for θ and 2.14–2.28°C for Ts, respectively, demonstrating the capability of the model to simulate realistically these quantities.

[23] Carbon fluxes expressed through Gross Primary Production (GPP) are also well simulated at different time aggregations for the location of Chamau (Figures 4a and 4b). Simulated values of GPP at daily scale have a determination coefficient, R2, of 0.59. The capability of reproducing seasonality and interannual variability of GPP is also satisfactory. Worse and better performances are obtained for the other two locations, Oensingen and Früebüel, where the R2 for daily GPP are 0.50 and 0.76, respectively (supporting information Figures S5 and S6). The performance at Oensingen is strongly influenced by the summer of 2003 where a prolonged dry period [Granier et al., 2007; Reichstein et al., 2007] significantly reduces GPP in both observations and simulations. However, simulations tend to overestimate the effect of water stress in grassland with GPP almost equal to zero. Despite this mismatch the annual values of GPP and their variability are captured reasonably well by the model for Chamau (Figure 4c), as well as for Oensingen and Früebüel (supporting information Figures S5 and S6). Note that these values are among the highest observed GPP for terrestrial ecosystems [Baldocchi, 2008].

Figure 4.

A comparison between the observed (OBS) and simulated (SIM) (a) daily and (c) annual GPP for the location of Chamau. A scatter plot with the determination coefficient R2 is shown for (b) daily GPP.

4.2. Distributed Confirmation

[24] The ecohydrological dynamics of Rietholzbach were simulated for the period of 1983–2005 for a total duration of 23 years, for which input meteorological forcings were continuous. The imposed grid resolution, a compromise between detailed representation of the domain and computational efficiency, was 50 × 50 m2 that led to 1284 basic computational elements. A uniform 60 cm soil column and impermeable bedrock were assumed for all of the computational elements of the grid, because no reliable distributed soil depth map was available. The soil column depth was assumed to be representative of the average of the Rietholzbach watershed. The vertical mesh was composed of 10 soil layers of increasing thickness from the surface (10 mm) to a 60 cm depth (100 mm). Vertical homogeneity of soil hydraulic properties was assumed because of the lack of information. A horizontal anisotropy coefficient, aR = 100, was used for all of the grid elements to mimic preferential lateral flow paths [Germann, 1981]. The vegetation map was used to assign plant functional types to each element, i.e., evergreen, deciduous, mixed evergreen and deciduous, and grassland. Physiological and structural characteristics of plant functional type different from grassland were inferred from literature and previous analyses [Fatichi et al., 2012a]. Grassland was parameterized following the plot-scale confirmation at Chamau, Oensingen and Früebüel (Table S1). Considering that at Rietholzbach elevation and incoming solar radiation are similar to Früebüel, we assumed a similar frequency and number of grass cuts. We therefore imposed four grass cuts from doy 170 to 265 (average intercut distance of 31.6 days).

[25] Streamflow measured at the outlet of the Rietholzbach watershed was compared with model simulations (Figure 5). Multiple metrics of performance were evaluated at different aggregation times. The T&C model simulates well the discharge in Rietholzbach for hourly, daily, monthly, and yearly aggregation times, the hourly variability of streamflow is well simulated as shown, for example, in a 200 h period (Figure 5a). The coefficient of determination of simulated versus observed hourly discharge for the entire 23 year period is R2 = 0.71 with a RMSE of 0.108 mm h−1. The coefficient of determination improves to R2 = 0.83, R2 = 0.89, and R2 = 0.80, when daily, monthly, and yearly aggregation times are considered. The results in terms of monthly discharge also show a good agreement (Figure 5b). Another metric of performance, typically used in hydrology, the Nash-Sutcliffe efficiency [Nash and Sutcliffe, 1970], math formula, is 0.64 at the hourly scale and 0.81 at the daily scale, underlining how the overall streamflow dynamics are essentially reproduced by the model. The simulated average annual runoff agrees closely with the observed value, 1007 versus 991 mm yr−1. Differences of a few mm yr−1 are below the measurement uncertainty of both streamflow and distributed precipitation. The model tends to underestimate the discharge only for very dry periods as shown in the duration curve for durations above 310 days (Figure 5c). This shortcoming is likely related to the assumption of a uniform soil depth which does not allow to properly represent deep storages that can feed the stream during prolonged dry periods. Nonetheless, flow durations are reproduced well across 4 orders of magnitude indicating that the model setup is robust enough to perform virtual experiments.

Figure 5.

A comparison between the observed (OBS) and simulated (SIM) (a) hourly discharge for a selected period, (b) monthly discharge, and (c) flow duration curves for the Rietholzbach watershed. Qobs and Qsim denote the observed and simulated monthly discharge, respectively.

[26] The analysis of a few maps of ecohydrological metrics confirms the model consistency at the distributed scale (Figure 6). While high spatial resolution distributed observations were not available to evaluate the presented maps, their congruence with theoretical expectations is considered as a further corroboration of model simulations. The distributed pattern of total evapotranspiration integrated in the simulation period is mainly influenced by incoming shortwave radiation and soil moisture (Figure 6a). The south exposed slopes show higher values of evapotranspiration regardless of the vegetation type. The signature of vegetation type is more evident in the north exposed slopes where trees have larger evapotranspiration rates when compared to grass. The highest values of evapotranspiration are simulated in the convergent flat part near the river network where the combined presence of nearly saturated soil, larger temperatures and of a tree grove enhances plant moisture uptake. The effective saturation [Brutsaert, 2005, p. 258] in the entire soil profile is mainly governed by topographic convergence with the areas near the river network that show the largest values close to saturation level (Figure 6b). Nonetheless, the entire catchment is fairly wet, since even the cells near the divide have an average effective saturation of 0.65. Seemingly, the north exposed slopes are wetter than the slopes exposed to the south. The mean value of LAI in each cell correlates very well with the vegetation type, being larger for evergreen, followed by mixed deciduous/evergreen and deciduous only areas (Figure 6c). Grass that underwent several cuts shows a average LAI of about 2. Finally, vegetation productivity expressed as GPP exhibits an opposite pattern when compared to LAI in trees and grass, with cells covered by grass that show the highest productivity followed by cells with trees (Figure 6d). Differences are also imposed by environmental controls such as soil moisture, incoming radiation, and air temperature.

Figure 6.

The results of spatially distributed ecohydrological simulations averaged over the simulation period (1983 through 2005) for the Rietholzbach watershed. (a) Mean total evapotranspiration flux, (b) mean effective saturation in the 60 cm soil profile, (c) mean leaf area index, and (d) mean gross primary production.

5. Design of the Experiment

[27] A series of numerical experiments was designed to investigate the effects of grassland management types on hydrological fluxes and vegetation productivity (Table S2 in the supporting information). Simulations were carried out at the plot and catchment scales.

5.1. Plot-Scale Experiment

[28] Simulations at the plot scale were carried out using, as meteorological forcing, the 23 year time series of Rietholzbach and Sion. The climate recorded at Rietholzbach is slightly colder and wetter than that observed at Chamau and Oensingen, while the one of Sion is slightly warmer and considerably drier. Despite these differences the availability of longer time series was considered important to make robust inferences independent of interannual variability of meteorological conditions. Analogous simulations were performed using also the 2–3 year long meteorological time series of Oensingen, Chamau, and Früebüel, to further test the dependence of the obtained results to different climate conditions.

[29] First, the sensitivity to vegetation cut only was analyzed imposing three different cutting heights, 15, 7, and 2 cm. These heights were converted into LAI of 3.60, 1.68, and 0.48, respectively [Allen et al., 1989]. We acknowledge that the conversion of grass height into LAI (the variable used by T&C to impose the cut) remains rather uncertain, therefore cutting heights should be regarded as a reference rather than an exact value, where deviations of 3–4 cm from the reference value appear possible for a given LAI. The same frequency and period of grass cut used in the model confirmation for Rietholzbach, Oensingen, Chamau, and Früebüel were replicated changing only the cut height. We assumed for Sion the model parameters and the cutting frequency of Oensingen.

[30] Successively, the effects of grazing were tested imposing seven different stocking densities. Specifically, stocking densities of 1, 2, 3, 4, 5, 7, and 10 livestock units (LSU) per hectare were simulated in the model (Table S1 supporting information). According to the European Commission definition, 1 LSU corresponds to one adult dairy cow producing 3000 kg of milk annually. This metric facilitates the aggregation of livestock from various species and ages. The range of stocking densities covers normal up to intense grazing pressures over the entire stocking season. In Switzerland, according to the Swiss Federal Office of Statistics, a typical stocking density averaged at a regional level is on the order of 0.5–2 LSU per hectare. However, cattle tend to choose preferential paths, they have favorite meadows and overall tend to cluster. For these reasons, the local grazing pressure can be significantly higher. The stocking period was assumed to be from doy 140 to 280 for the locations at lower elevations (Chamau, Oensingen, Sion) and from doy 160 to 270 for Rietholzbach and Früebüel. These stocking periods correspond with the periods of possible outdoor activities of sheep and cows. Stocking densities were converted into consumption rates of aboveground biomass considering that a cow (1 LSU) typically consumes 15 kgDM d−1 of biomass that is equivalent to about 6.7 kgC d−1 [Minonzio et al., 1998; Bilotta et al., 2007; Zeeman et al., 2010]. Note that the assumed consumption for 1 LSU is equivalent to about 1.7 Forage Intake Units according to international standards [Allen et al., 2011]. Since the consumption rate of aboveground biomass in gC m−2 d−1 is used in the model, the LSU is just a relative measure and can be easily converted to different specific animals; for instance, a sheep typically consumes 2 kgDM d−1, a calf 7 kgDM d−1, and a lamb 1.5 kgDM d−1 [Bilotta et al., 2007; Allen et al., 2011]. Experiments at Chamau and Früebüel [Zeeman et al., 2010] provided specific consumption rates of 1.2–1.3 kgC d−1 for sheep and 2–5 kgC d−1 for cattle, overall confirming the magnitude of literature values.

[31] Uncertainties in the consumption rates of animals and in the number of cuts per year can be significant but we assume they do not represent a limitation of the study because of the wide range of cutting heights and stocking densities that we considered. Nonetheless, many times grazing pressure is the result of a very high stocking density during a shorter period rather than that of a few LSU per hectare over the entire growing season. Furthermore, specific farmer decisions can lead to a more or less frequent cutting than assumed in this study. For this reason, we also considered cases where a larger number of animals were allowed to graze for a shorter period and cases where cutting frequency rather than height was changed. These additional cases are described in the supporting information Text S2.

[32] The last series of virtual experiments was used to test the effect of soil compaction superposed on the biophysical effect of grazing (Table S2 supporting information). Specifically, three experiments were carried out by simulating a normal stock density (2 LSU per hectare) and reducing the hydraulic conductivity of the first 10 cm of the soil column by 50%, 80%, and 95%, respectively. There are indeed empirical evidences that the soil structural alteration resulting from treading by grazing animals mainly affects the upper part of the soil, near the surface [Drewry, 2006; Greenwood et al., 1997; Bilotta et al., 2007; Stavi et al., 2011]. The effect on hydrological parameters such as the hydraulic conductivity is more controversial among authors, who have observed reductions from 20% to almost 100% of infiltration capacity caused by stock treading [Gifford and Hawkins, 1978; Heathwaite et al., 1990; Mulholland and Fullen, 1991; Greenwood et al., 1997; Schneider et al., 2008; Leitinger et al., 2010]. Given these large uncertainties and the difficulty in mechanistically modeling soil deformation effects on structural and hydraulic properties without having local specific information, we tested three realistic reductions of the hydraulic conductivity (50, 80, and 95%) that may be considered as the consequences of different degrees of alteration of the soil structure [Taboada and Lavado, 1993; Assouline, 2006b; Alaoui et al., 2011; Gan et al., 2012]. Note that in this way, we neglected possible changes in soil porosity and in the water retention curve that are expected to take place but are very uncertain and difficult to parameterize [Assouline, 2006a; Alaoui et al., 2011]. Soil compaction can occur also as a consequence of farm machines used to cut grass. This possibility is, however, not explored here since it is expected to provide minor consequences compared to animal treading, being that machines are mostly used in dry periods.

[33] The effect of stock trampling by grazing animals on the destruction of a large amount of plant material and the consequent reduction of herbage growth/yield was also not investigated in this study. The imported water through animals is ignored because this quantity is estimated to be relatively small, on the order of 0.07 mm d−1 or about 13 mm during the entire growing season, even for a stocking density of 10 LSU per hectare. However, the fertilizing role of the manure produced by the animals is typically not negligible but it is not accounted for in this study that does not consider soil biogeochemical processes.

5.2. Catchment-Scale Experiment

[34] A subset of the virtual grassland management scenarios carried out at the plot scale was also tested at the catchment scale by simulating ecohydrological dynamics in Rietholzbach for a 5 year period (October 2000 to September 2005). The choice of a subset and of shorter time periods was pragmatical due to the elevated computational time of a model such as T&C for high-resolution distributed simulations. Specifically, two grass cuts (2 and 7 cm) and one stocking density (2 LSU per hectare) were simulated. The effect of soil compaction was superposed on the stocking density of 2 LSU per hectare during the entire growing season, reducing the hydraulic conductivity of the first 10 cm of the soil by 95%. In total, four management scenarios were evaluated at the catchment scale. Note that only the part of the watershed covered by grassland (≈75% in Rietholzbach) underwent simulated management practices.

[35] For all of the cases, a simulation that mimics natural conditions, i.e., without grass cuts or animal grazing was produced to serve as a benchmark against which the management activities could be compared. Being representative of nonmanaged conditions, this simulation is indicated in the following sections as “undisturbed” simulation.

6. Results: Plot Scale

[36] The results of the simulations for the different scenarios of grassland management are discussed only for the locations of Rietholzbach and Sion and are obtained averaging the 23 year long hourly time series. A 7 year spin-up period was used to initialize the grass carbon pools for the various management scenarios. Results for Oensingen, Chamau, and Früebüel do not provide additional information in comparison to the outcomes obtained using Rietholzbach and Sion climate and soil properties and are therefore not presented. Finally, since the results for short and very high (>10 LSU per hectare) stocking densities, as well as for increasing cut frequency are not particularly significant when compared to other treatments, they are only described in the supporting information (Text S2 and Figures S7 and S8).

6.1. Biophysical Effects on Hydrological Fluxes

[37] The simulated effects of different management practices on hydrological fluxes and states at Rietholzbach are shown in Figure 7. For most of the cases, the differences between natural and managed conditions are hardly distinguishable. The management noticeably affects the partition between soil evaporation, evaporation from interception, and transpiration only for 2 and 7 cm grass cut and when extreme stock densities of 7 and 10 LSU per hectare are considered. In these conditions, a large part of aboveground biomass is removed and ground evaporation tends to be three to six times larger than in natural conditions (Figure 7a). Correspondingly, evaporation from interception decreases because of the reduced LAI available to intercept water (Figure 7b). Transpiration shows minor changes and is mostly affected by stocking densities of 7–10 LSU per hectare or by a 2 cm cut which reduces transpiration by 18% with regards to undisturbed conditions (Figure 7c). Transpiration increases slightly (2.5%) when compared to undisturbed conditions for a 15 cm cut as a consequence of less evaporation from interception. The vertically integrated effective saturation and the leakage at the bottom of the soil column are influenced less significantly (+11% for a 2 cm cut) due to compensatory effects of ground evaporation, evaporation from interception, and transpiration (Figures 7d and 7e). A significant increase of deep leakage (about 25%) is only caused by unsustainable stocking rates that transform the entire meadow into bare soil a few weeks after the beginning of the grazing. Overland runoff is equal to zero in all of the considered cases due to soil properties and bottom-free drainage conditions (Figure 7f). Under the latter assumption no saturation or infiltration excess runoff is simulated at the plot scale.

Figure 7.

Sensitivity of hydrological fluxes and states to grass management practices, e.g., cut height of 2, 7, and 15 cm, and stocking densities of 1, 2, 3, 4, 5, 7, and 10 LSU per hectare for the plot-scale location of Rietholzbach over the simulation period of 1983 through 2005. “Undist.” represents the simulation in natural conditions. (a) Ground evaporation, (b) evaporation from interception, (c) transpiration, (d) effective saturation, (e) deep leakage, and (f) runoff.

[38] In the case of Sion, hydrological fluxes are strongly affected already at a stocking density of 2 LSU per hectare (Figure 8). The drier climate of Sion, which is less favorable for grass growth, leads to conditions where grass is no longer able to reestablish itself (total depletion of carbohydrate reserves stored in the roots) for grazing disturbances above 2 LSU per hectare and the meadow is permanently transformed into bare soil. The stocking density of 2 LSU per hectare can be considered a tipping point where the ecosystem passes from a behavior similar to undisturbed conditions to an unsustainable grazing. In the latter conditions, transpiration and evaporation from interception are equal to zero and deep leakage and ground evaporation significantly increase (270% and 450%, respectively) when compared to undisturbed conditions. The effects induced by grass cutting in Sion are qualitatively similar to Rietholzbach, with an increase in ground evaporation (+43%) and a decrease in transpiration (−5%) and evaporation from interception (−54%) appreciable for a 2 cm cut when compared to undisturbed conditions.

Figure 8.

Sensitivity of hydrological fluxes and states to grass management practices, e.g., cut height of 2, 7, and 15 cm, and stocking densities of 1, 2, 3, 4, 5, 7, and 10 LSU per hectare for the plot-scale location of Sion over the simulation period of 1988 through 2011. “Undist.” represents the simulation in natural conditions. (a) Ground evaporation, (b) evaporation from interception, (c) transpiration, (d) effective saturation, (e) deep leakage, and (f) runoff.

6.2. Biophysical Effects on Vegetation

[39] The impacts of management in vegetation productivity are more pronounced in most of the analyzed scenarios. Note that the effects of nutrient availability are not included in this analysis since nutrient limitations are neglected in T&C. In Rietholzbach, Gross Primary Production (GPP) and to a minor extent Net Primary Production (NPP) tend to decrease significantly when grass is cut to 7 and 2 cm and they also decrease when stocking density exceeds 5 LSU per hectare (Figures 9a and 9b). In the extreme grazing scenarios (for Rietholzbach > 5 LSU per hectare) grass growth rarely balances the consumption rate (the biomass is close to zero after the starting of stocking season and able to regrow only in the subsequent spring), clearly showing a situation where grazing is unsustainable. This threshold is significantly lower (2 LSU per hectare) for Sion which as stated above experiences a tipping point rather than a smooth transition (Figures 10a and 10b). The different thresholds for a sustainable management are related to the magnitude of GPP in undisturbed conditions, 1909 and 1117 gC m−2 yr−1 for Rietholzbach and Sion, respectively, which represent an indication of the grassland carrying capacity.

Figure 9.

Sensitivity of vegetation productivity, leaf area index, and albedo to grass management practices, e.g., cut height of 2, 7, and 15 cm, and stocking densities of 1, 2, 3, 4, 5, 7, and 10 LSU per hectare for the plot-scale location of Rietholzbach over the simulation period of 1983 through 2005. “Undist.” represents the simulation in natural conditions. (a) Gross primary production, (b) net primary production, (c) aboveground net primary production, (d) mean leaf area index of green biomass, (e) mean leaf area index of dead biomass, and (f) mean albedo during snow-free periods.

Figure 10.

Sensitivity of vegetation productivity, leaf area index, and albedo to grass management practices, e.g., cut height of 2, 7, and 15 cm, and stocking densities of 1, 2, 3, 4, 5, 7, and 10 LSU per hectare for the plot-scale location of Sion over the simulation period of 1988 through 2011. “Undist.” represents the simulation in natural conditions. (a) Gross primary production, (b) net primary production, (c) aboveground net primary production, (d) mean leaf area index of green biomass, (e) mean leaf area index of dead biomass, and (f) mean albedo during snow-free periods.

[40] The dynamics of Aboveground Net Primary Production (ANPP) are more complex. The model produces a larger amount of aboveground biomass when compared to the undisturbed case for moderate disturbances, e.g., a 7–15 cm cut, and even for 2 cm in Sion (Figure 10c) and for a stocking density up to 5 LSU per hectare in Rietholzbach (Figure 9c). These increments are on the order of 45–55% for Rietholzbach and 55–70% for Sion. The metric most affected by management activities is LAI of both alive and dead biomass (Figures 9d, 9e, 10d, and 10e). Green aboveground biomass decreases strongly when grass is cut and goes to zero (in Sion) or almost to zero (in Rietholzbach) for the most extreme stocking rates. Dead LAI also decreases when compared to undisturbed conditions because it is partially removed by cuts and because less biomass ages and dies. Finally, mean annual albedo of snow-free periods are very similar among different scenarios and tend to be slightly higher when the ratio of necromass to living biomass increases (Figure 9f) or lower when the meadow is transformed into bare soil (Figure 10f).

6.3. Soil Compaction Effects Superposed on Biophysical Effects

[41] The results of the simulations where the biophysical effects induced by defoliation are superposed to a decrease of hydraulic conductivity in the first 10 cm of the soil are shown in terms of hydrological and vegetation metrics for Rietholzbach only (Figure 11). Natural conditions, i.e., “undisturbed,” and 2 LSU per hectare grazing without soil compaction are also shown for comparison. The overall consequences of soil compaction on ground evaporation, evaporation from interception, effective saturation (not shown) as well as on transpiration (Figure 11a) are negligible. A reduction in soil hydraulic conductivity marginally affects runoff formation, with the exception of the extreme case of a 95% decrease. In this latter case, infiltration excess runoff is produced during intense precipitation events, with values on the order of 40 mm yr−1. This quantity is subtracted from the deep leakage but leaves unmodified all of the other hydrological fluxes. The implications of the soil compaction on vegetation productivity are also minor. The integrated GPP, NPP, and LAI are nearly identical whether the reduction in hydraulic conductivity is included or not (not shown). In all of the cases, there is an increase in ANPP with regards to natural conditions due to the stimulation that a moderate grazing has on allocating biomass to the aboveground compartment (Figure 11c).

Figure 11.

Sensitivity of transpiration, total runoff, and Aboveground Net Primary Production (ANPP) to the effect of soil compaction superposed to the biophysical effect of grazing. Specifically, a reduction of the hydraulic conductivity of the first 10 cm of the soil column by 50%, 80%, and 95% is superposed on a stocking density of 2 LSU per hectare for the plot-scale location of Rietholzbach over the simulation period of 1983 through 2005. “Undist.” represents the simulation in natural conditions. (a) Transpiration, (b) total runoff, and (c) ANPP.

7. Results: Distributed

7.1. Effects on Hydrological Fluxes

[42] Distributed simulations for the Rietholzbach catchment for a 5 year period (October 2000 to September 2005) at a resolution of 50 × 50 m2 were carried out to investigate the impact of grassland management at larger scales. Rietholzbach is a good case study because 75% of the watershed is covered by grassland. In agreement with the plot-scale simulations, the differences among management practices and natural conditions are minor in most of the cases (Figure 12). Only for a cut height of 2 cm, a reduction of total evapotranspiration appears to be an important feature (Figure 12a). This reduction (−12% or −69 mm yr−1) is mainly caused by a lower transpiration (−57 mm yr−1) (Figure 12b) and evaporation from interception (−81 mm yr−1) that are partially counterbalanced by a larger ground evaporation (+73 mm yr−1) (not shown). A lower evapotranspiration flux is reflected in a larger amount of saturation excess runoff (Figure 12c).

Figure 12.

Sensitivity of hydrological fluxes and states to grass management practices, e.g., cut height of 2 and 7 cm, stocking density of 2 LSU per hectare, and soil compaction (hydraulic conductivity of the first 10 cm of the soil column reduced by 95%) superposed on grazing for the Rietholzbach catchment over the simulation period of 2001 through 2005. “Undisturbed” represents the simulation in natural conditions. (a) Total evapotranspiration, (b) transpiration, (c) saturation excess runoff, (d) infiltration excess runoff, (e) percentage of saturated area, and (f) discharge.

[43] For all of the other management scenarios, saturation excess runoff is similar to undisturbed conditions with the exception of the extreme case of soil compaction, i.e., when the hydraulic conductivity is reduced to 95% of undisturbed conditions. In this case, a reduction of 30% of saturation excess runoff at the watershed scale occurs and infiltration excess runoff increases significantly (Figure 12d). Note that due to mechanisms of reinfiltration of ponded water and runon, the sum of saturation and infiltration excess runoff largely exceed the total discharge. The average percentage of saturated area (about 2.5%) and the total catchment discharge (about 930 mm yr−1) are nearly identical for all of the analyzed scenarios (Figures 12e and 12f). However, a tendency of the average fraction of saturated area (+6%) and total discharge (+7%) to increase for a grass cut of 2 cm can be detected as a consequence of the reduced evapotranspiration (−12%).

[44] In order to extend the analysis to the annual distribution of streamflow, discharge for given durations (i.e., for different levels of probability of exceedence expressed in days) were also analyzed. Significant differences can be identified at high and low durations for the most intense management practices, especially when compared to the minor effects detectable in total annual runoff (supporting information Figure S9). Specifically at high flows (very low durations of less than 1 day), the soil compaction exerts a strong effect increasing the discharge by about 36% and 5% corresponding to durations of 0.01 and 0.1 days. This has implications in terms of flood peaks. In other words, for specific events the reduced hydraulic conductivity might change the rainfall-runoff response of the watershed even though the annual effect is negligible. At low flows (long durations of more than 300 days), the major difference is produced by managing the grassland with a cut of 2 cm. This management practice leads to a reduction of evapotranspiration that sustains higher rates of streamflow (base flow) during the dry season. The effect of management practices for intermediate durations is negligible.

[45] At the scale of the single event the effects of management practices on the peak flow are rather complex. They depend on the interplay of initial soil moisture conditions and rainfall intensity. This is illustrated with two different precipitation events that produce streamflow of roughly the same magnitude in the Rietholzbach catchment (Figure 13). The first event has a more prolonged rainfall with maximum intensities of about 4 mm yr−1, the second event shows a more intense peak of rainfall, 8 mm yr−1, but a lower total amount. In these two events, the basin response to management practices is rather different. In the first event, the highest discharge peak is produced by the management of grass cut at 2 cm. This is an event where the antecedent soil moisture is fundamental because runoff is mostly generated through the saturation excess mechanism. In the second event, the highest discharge is produced by the management scenario where a 2 LSU per hectare stocking density and 95% reduction in soil hydraulic conductivity are imposed. In the case of a strongly compacted soil, the higher rainfall intensity of the second event leads to infiltration excess runoff. The streamflow peak increases by almost 50% suggesting that while grazing is unlikely to affect the water budget it might increase the flood risk because of soil compaction effects, at least for relatively small catchments such as Rietholzbach.

Figure 13.

Sensitivity of the hydrograph to grass management practices, e.g., cut height of 2 and 7 cm, stocking density of 2 LSU per hectare, and soil compaction (hydraulic conductivity of the first 10 cm of the soil column reduced by 95%) superposed on grazing for the Rietholzbach catchment for two specific rainfall-runoff events. “Undisturbed” represents the simulation in natural conditions.

7.2. Effects on Vegetation

[46] The effects of disturbances on nonnutrient limited vegetation productivity and LAI at the catchment scale (averaged only over the basic computational elements containing grass) are larger than for hydrological fluxes. ANPP tends to increase when compared to natural conditions for all of the simulated scenarios with the exception of the grass cut at a 2 cm height (supporting information Figure S10). A cut height of 7 cm is the most effective (+41%) in maximizing aboveground productivity among the analyzed scenarios. These levels of productivity are obtained even though the simulated average LAI are significantly lower (−50%) than those obtained for the undisturbed case (supporting information Figure S10). Grass cut management practices strongly influence LAI reducing its long-term average. Conversely, a moderate grazing of 2 LSU per hectare does not change significantly the average LAI, demonstrating how subalpine grasslands are able to provide forage for a certain number of animals without reducing the overall leaf cover. Soil compaction does not additionally affect vegetation productivity and LAI. This is mainly the result of the independence of long-term soil moisture content and total discharge from the hydraulic conductivity of shallow soil layers in the Rietholzbach catchment.

8. Discussion

8.1. Long-Term Effects on Hydrology and Energy Balance

[47] According to the result of the sensitivity analysis carried out without considering biogeochemistry dynamics, we found that for most of the management practices biophysical alterations induced by grazing, grass cut and soil compaction induced by livestock treading do not have the potential of significantly influencing long-term soil moisture and energy partition in subalpine grasslands. Only the most intense management scenarios, e.g., the lowest cut heights or overgrazing with stocking densities that are not sustainable (>5 LSU per hectare in Rietholzbach, ≥2 LSU per hectare in Sion), were found to be effective in considerably reducing evapotranspiration and increasing soil moisture and deep leakage. Note that given the uncertainty in relating grass height to LAI, the 2 and 7 cm reference values should be rather seen as 2–4 and 7–10 cm intervals and therefore also the lowest cuts are somehow realistic.

[48] Unsustainable stocking densities were shown to exhibit LSU thresholds which are a function of how favorable the environmental conditions are for grass to grow, which can be summarized in the GPP of the undisturbed cases. Considerably, lower thresholds might be expected in dry areas and at very high elevations.

[49] A low cut treatment would lead to an increase in leakage of 11% and 16% for Rietholzbach and Sion (plot scale) or discharge (+7%) for Rietholzbach catchment. These values are similar to the results of an experiment led in the Alps where grass was clipped at 4 cm to simulate sheep grazing during three seasons [Inauen et al., 2013]. Evapotranspiration and deep seepage were measured using weighing lysimeters and deep seepage was found to increase between 5% and 13% depending on the analyzed season. When runoff is used for hydropower production, a common case in the Swiss Alps, these changes were considered to have enough economic value to support the maintenance of high-elevation pastures [Inauen et al., 2013].

[50] The presented results agree also with previous observations that show an independence from LAI in the partition of the energy budget components of Swiss subalpine grasslands already at LAI larger than 1–1.5 [Rosset et al., 1997]. Although LAI of managed grasslands remains typically lower than in natural conditions, it is sufficient for herbaceous plants to intercept radiation, transpire, and photosynthesize at rates comparable with undisturbed conditions. Notably, even differences in LAI of 3–4 were found unable to significantly modify the hydrological budget and GPP of the grassland. The general insensitivity of ecohydrological metrics to grass LAI above a certain threshold is also confirmed by the analysis of Sion and by studies in more xeric environments [Bremer et al., 2001; Wang et al., 2012]. Note that this means that the typical correlation between annual GPP and LAI, which we also found, is rather due to LAI increasing as a consequence of larger assimilation rates and not of GPP increasing because of larger LAI. This result also highlights possible shortcomings in calculating evapotranspiration from dense grassland using equations which positively relate evapotranspiration to LAI, such as the Penmann-Monteith equation [Brutsaert, 2005].

[51] Further reasons behind the insensitivity of ecohydrological metrics to grassland management can be connected to the physiological properties and behavior of grass plants. Grass typically has a fast growing response, e.g., due to high values of Rubisco carboxylation and specific leaf area index, among other physiological parameters. Carbon is allocated only to three compartments: aboveground green biomass, fine roots, and carbohydrate reserves and when grass is grazed or cut, carbon allocation is mostly shift to aboveground biomass to reestablish a certain leaf to root ratio. All of these factors allow grass to quickly recover a significant value of LAI after disturbances.

[52] The consequences of abandoning grassland, denoted as undisturbed conditions in the presented analysis, have similarities and some difference with empirical results that compared abandoned and managed (1 cut per year) subalpine grasslands reported by Rosset et al. [2001]. Their study observed a decrease in albedo, an increase in net radiation and in the Bowen ratio in abandoned pastures as a consequence of a larger necromass, change in optical properties, and shift in species composition. However, these changes only slightly affected the total evapotranspiration by mostly decreasing ground evaporation in abandoned pastures. T&C simulated ratios of necromass to total standing biomass (20–30% in undisturbed conditions in summer for Rietholzbach) similar to observations of Rosset et al. [2001]. The simulated Bowen ratio is slightly larger for the case of the 7 cm cut compared to the undisturbed case but changes in net radiation are negligible. Simulated albedo increases in undisturbed conditions rather than decreases, due to more reflective dead biomass. Total evapotranspiration changes can be estimated at 1–6% for 7 cm and 5–12% for 2 cm cut and a decrease of ground evaporation is correctly simulated in undisturbed grasslands when compared with cut grass. The differences with observations mostly arise because changes in species composition and correlated biochemical and optical properties are not explicitly simulated and therefore are not reflected by the results. These types of properties have been shown to have a role in grassland behaviors [Wohlfahrt et al., 1999, 2001, 2003] and add another level of complexity to the system.

[53] However, given the good agreement of the model for the three different flux towers using the same parameter set, one can consider the obtained result, i.e., the relative small influence of biophysical effects on long-term hydrology and energy budget, a fairly general result, which is not confined to the subalpine area, as confirmed by findings in more xeric grasslands [Bremer et al., 2001; Frank, 2003; Miao et al., 2009; Wang et al., 2012]. These empirical studies in grasslands of Kansas, North Dakota, and Inner Mongolia, showed small differences in growing season evapotranspiration of 6% to 13% and only for heavy grazing, with the major effects on soil water content related to changes of albedo or soil hydraulic properties [Zhao et al., 2010; Gan et al., 2012; Wang et al., 2012].

[54] Finally, increasing cut frequency was found to be less significant in affecting transpiration and productivity than decreasing the cut height at least for Rietholzbach. Short periods (2 weeks) of very high stocking density (>10 LSU per hectare) can completely defoliate the surface, but in the long term have a lower impact than the equivalent grazing pressure (grazed biomass) distributed during the entire growing season (supporting information Figures S7 and S8). However, it is possible that such an overgrazing will favor establishment of invasive species (e.g., weeds) and therefore compromise the full recovery of edible grass in the long term.

8.2. Short-Term Effects on Discharge

[55] A strong reduction (95%) of soil hydraulic conductivity imposed as a consequence of soil compaction induced by grazing animals was found to affect the event-scale hydrological response, but to be negligible in changing evapotranspiration. Soil compaction can modify the mechanism of runoff generation passing from a saturation excess dominated generation to a combination of infiltration and saturation excess. Consequently, the shape of the hydrograph after intense rainfall events can be modified, leading, in the case of a less permeable soil, to discharge peaks even 50–80% higher than those simulated for nonmanaged or moderately managed conditions. This can have obvious consequences on flood risk of, at least, small catchments, where very high stocking densities took place. Note that trampling issues might become more significant at higher elevations and for steep slopes where farm machines cannot operate and grazing is the prevalent activity.

[56] In the presented plot-scale analyses, a threshold behavior is identified passing from 80% to 95% reduction in hydraulic conductivity. This threshold behavior and the overall net effect of soil compaction are likely to be strongly dependent on stocking density, soil texture properties, and local rainfall characteristics, making any generalization very difficult. This has important implications on the interpretation of direct measurements of runoff obtained through sprinkling experiments [Kohl and Markart, 2002; Leitinger et al., 2010] or gauging of small catchments [Wine et al., 2012], and might explain the disagreement of results obtained by empirical studies [Gifford and Hawkins, 1978]. Nonetheless, we demonstrated that the effect of soil compaction by animal treading can be considerable at the event scale for runoff generation particularly by modifying the partition of flow between the fast component (surface runoff) and the slow component (the one induced by deep percolation and lateral subsurface flows).

8.3. Effects on Aboveground Productivity

[57] Long-term vegetation productivity and vegetation carbon fluxes are generally more affected by management practices than energy and water fluxes. We reiterate that the focus of the presented study was on biophysical and soil alterations, without consideration of soil biochemistry and soil nutrient dynamics and shifts in species compositions. A very interesting numerical result, which confirms empirical ecological studies, is that moderate disturbances both in terms of cut and grazing have the capability to increase the aboveground productivity of grassland. This result is important because the aboveground biomass is the quantity of principal interest for farmers. Aboveground production gives an estimate of the total source of food for the livestock, either when grass is grazed or cut for winter storage.

[58] This result is confirmed by several empirical observations which show that plants are able to compensate for tissue removal up to some level, beyond which plant productivity begins to decline as the intensity of defoliation increases further [Vickery, 1972; McNaughton, 1983, 1993; Frank et al., 1998]. However, there are also counterexamples [e.g., Biondini et al., 1998] which point to geographical differences in grassland responses. The physiological explanation is given by a larger allocation of carbon to aboveground tissues when grass is disturbed. Subalpine grass in the model seems to cope very well with moderate disturbances shifting the allocation pattern from belowground fine roots to aboveground leaves and thus increasing the aboveground productivity. In Rietholzbach, average fine root biomass for a stocking density of 4 LSU per hectare is 15% less than the value obtained for undisturbed conditions. This might question whether larger yields might decrease the resilience of grassland to cope with severe disturbances such as prolonged droughts.

[59] Using the numerical model, we could give an estimate of the management threshold which maximizes aboveground productivity. If confirmed by more detailed empirical and numerical analyses, this behavior may help identify a range of best management practices to maximize the hay/silage production for specific conditions.

9. Conclusion

[60] We conclude that most of the tested grassland management activities have limited implications for long-term energy and water fluxes through LAI removal and soil compaction effects. Only 2–4 cm cut heights have the potential to decrease evapotranspiration and therefore increasing downstream water availability of about 5–10% as shown also by experimental studies [Inauen et al., 2013]. More significant effects can eventually occur because of changes in biochemical and optical properties due to shifts in species composition between abandon and managed grassland or through the effects of nutrient addition or depletion that are not considered in this study. However, we argue that overall changes in evapotranspiration and hydrological budget are likely to remain relatively small.

[61] Management practices were found to be much more important for vegetation productivity and overall for the carbon budget, as already demonstrated by other studies [e.g., Ammann et al., 2009; Zeeman et al., 2010]. Specific conditions might also occur in which the management is leading to an increase of runoff response at the event scale, especially when livestock treading considerably modifies the soil properties through compaction. All of the above results are derived for subalpine climatic, topographic and soil conditions. However, the insensitivity of long-term water and energy fluxes to a wide range of LAI in grasslands is confirmed by other studies [Rosset et al., 2001; Bremer et al., 2001; Wang et al., 2012] and suggests that this can be considered as a rather general result.


[62] Meteorological data for Switzerland were provided by MeteoSwiss, the Federal Office of Meteorology and Climatology. The digital elevation model was downloaded by the GeoVITe—ETH Geodata portal. We sincerely acknowledge all of the people involved in the maintenance of the instruments and collection of data at the Rietholzbach Experimental catchment for more than 30 years of existence. Flux towers observations were obtained from the network Swiss FluxNet (