For the present study, measurements were taken at three study sites on the Tibetan Plateau. Details are given in Table . For the experimental activities at the sites see Sect. 2.5.

Xinghai: The experimental site is located in Qinghai province in the northeastern Tibetan Plateau, approximately 200 km southwest of Xining, and about 15 km south of Xinghai city. The montane grassland has developed on a loess-covered (1.2 m) terrace of the Huang He River. The grassland is used as a winter pasture for yaks and sheep for 6–7 months of the year (Miehe et al., 2008b; Unteregelsbacher et al., 2012). About 20 % of the pasture at the experiment site is completely covered with blue-green algae and crustose lichens.

Kema: The “Kobresia pygmaea Research Station Kema”, established in 2007, is located in the core area of alpine Kobresia pygmaea pasture. All measurements were established either within or in the close surroundings of an area of 100 m by 250 m, fenced in 2009, on a pasture where grazing was restricted to a few months during winter and spring. The growing season strongly depends on the availability of water, and usually starts at the end of May with the onset of the monsoon and ends with longer frosts by the end of August or September. Kobresia pygmaea has an average vegetation grazed height of 1–2 cm (Miehe et al., 2008b) and forms a very tough felty root turf of living and dead Kobresia roots, leaf bases and soil organic matter (Kaiser et al., 2008). It is designated as Kobresia root mat throughout this study and attains a thickness of 14 cm.

The site is covered with Kobresia pygmaea (Cyperaceae), accompanied by other monocotyledons (Carex ivanoviae, Carex spp., Festuca spec., Kobresia pusilla, Poa spp., Stipa purpurea) and to a minor degree by perennial herbs. For more details on the species diversity see Biermann et al. (2011, 2013).

Nam Co: The “Nam Co Monitoring and Research Station for Multisphere Interactions” (NAMORS) of the Institute of TibetanPlateauResearch of the Chinese Academy of Science (Ma et al., 2008) is located within an intramontane basin, 1 km SE of Lake Nam Co and in approximately 10 km distance NNW of the foot of the Nyainqentanglha mountain range. The zonal vegetation comprises mosaics of Kobresia turfs and open alpine steppe; water surplus sites have degraded Cyperaceae swamps (Mügler et al., 2010; Wei et al., 2012; Miehe et al., 2014).

At the Kema site a patchy structure of different degradation stages exists, which were classified according to the following classes (Fig. ): intact root Mat (IM), degraded root Mat (DM) and bare soil (BS).

We conducted model experiments in order to estimate the impact of the defined degradation classes on water and carbon fluxes, including feedback on atmospheric circulation. Therefore three 1-D soil–vegetation–atmosphere transfer models were utilised to examine evapotranspiration (Sect. ), carbon fluxes (Sect. ), and surface feedbacks (Sect. ). While the first two models were driven by measured standard meteorological forcing data, the latter is fully coupled to the atmosphere, which allows for feedbacks of land surface exchange to the atmosphere.

Experimental investigations on the Tibetan Plateau are not comparable with typical meteorological and ecological experiments. Not only do the high altitude and the remoteness of the area impose limitations, but also unforeseeable administrative regulations challenge the organisation of experiments with different groups and large equipment. It was initially planned to investigate small degraded plots with chambers and micro-lysimeters and to use a larger plot, in the size of the eddy-covariance footprint, as a reference area to investigate the daily fluctuations of the evaporation and carbon dioxide flux. Due to customs and permit problems, this was unfortunately only partly possible at Kema site in 2010, and not at all during the main chamber experiment in 2012.

Therefore, model-specific parameters were investigated in 2012 and the models were adapted to the specific Tibetan conditions with the chamber data. These model versions were then tested with the eddy-covariance data in 2010 at the Kema site with nearly intact Kobresia cover. Forced with measured atmospheric conditions, these simulations are used to examine the differences among degradation classes in carbon and water exchange between surface and atmosphere. The 13C labelling studies enabled us to relate the differences in carbon exchange to the specific vegetation and soil compartments. Finally, a surface scheme coupled with a meso-scale atmospheric model served to estimate feedbacks of surface forcing on the atmosphere. A summary of the experimental setup according to measurement technique is given in Table .

In accordance with this concept, we adapted both SEWAB and SVAT-CN to the Kema site using the vegetation and soil parameters elaborated in 2012, and chamber measurements from 2012 for calibration. Two parameter sets were established: one for surfaces with root mat (Kema RM: IM and DM differ only in vegetation fraction), and one for BS conditions (Kema BS). Simulations with in situ measured atmospheric forcing data were performed specifically for each of the degradation classes SIM, SDM and SBS according to the definition in Table . These model runs serve to expand the chamber data beyond their measurement period, and we are now able to compare the class-specific fluxes over a 46-day period (12 July to 26 August 2012).

Furthermore, we compared the adapted model versions with eddy-covariance data from 2010 using the respective forcing data measured in situ in 2010. The eddy-covariance measurements integrate the fluxes from a source area ranging 50–200 m around the instrument (for detailed footprint analysis see Biermann et al., 2011, 2013), and therefore represent H2O and CO2 fluxes from IM, DM and BS according to their proportion of total surface area in Table . In order to ensure comparability, we reproduce this composition with the simulations as well using the tile approach (SRefEC). An overview of model scenarios conducted at the Kema site is given in Table .

The differences in flux simulations among the degradation stages were controlled by the variation of the vegetation fraction and soil properties. A consistent parameter set for several experiments and multiple target variables (evapotranspiration, net/gross ecosystem exchange, ecosystem respiration) is a necessary pre-condition to ensure that the model physics implemented reflect these changes in a realistic manner. Therefore we abstained from optimising the parameter space, but used parameter estimates from field and laboratory measurements as far as possible (Appendix A), and inevitable calibration has been done for SVAT-CN by scaling the leaf area index with a single factor as well as a complete set of leaf physiology parameters.

For the investigation of the impact of surface degradation on the atmosphere, it was decided to run a relatively simple numerical experiment prescribing a symmetric, two-dimensional Tibetan valley with 150 km width, and surrounded by Gaussian hills with 1000 m altitude. A sounding taken at Nam Co on 17 July 2012 was used as the initial profile. The setup is comparable to Gerken et al. (2013, 2014). A total of four cases were chosen for this preliminary analysis. A dry scenario with initial soil moisture of 0.5× field capacity and a wet scenario with soil moisture at field capacity, as might be the case during the monsoon season, were used. For both surface states, simulations were performed with a vegetation cover of 25 and 75 % corresponding to a degraded and intact soil-mat scenario.

The study is limited by conceptual restrictions, which are mainly due to the scale problem in the different compartments (Foken et al., 2012b, see Appendix of this paper) and the working conditions in remote and high altitudes. Only one more-or-less uniform type of degradation has been investigated within the footprint area of the eddy-covariance measurements (Göckede et al., 2006) of 50–200 m extent, which is, in the case of this study, an almost non-degraded Kobresia pasture. The other types could only be found on much smaller plots, and had no significant influence on the whole footprint area, even when the non-linear influence of the different land-cover areas on the fluxes of the larger area is considered (Mölders, 2012). However, the investigation of degraded stages could only be done with small-scale measurements, such as those obtained with chambers and micro-lysimeters.

In order to test the performance of evapotranspiration (ET) with SEWAB and net ecosystem exchange (NEE) with SVAT-CN, we compared the model results for Kema with the eddy-covariance measurements from 2010 (Sect. 2.5). The results show that SEWAB simulations represent the half-hourly measured turbulent fluxes at Kema generally well (Table , see scatter plots and diurnal cycles in the Appendix, Figs. –). Model performance at Nam Co for the measurements in 2009 was very similar, as well as the magnitude of the fluxes (Table , from Biermann et al., 2014). Measured hourly medians (from an ensemble diurnal cycle over the entire period) of NEE at Kema ranged between -2.8 and 1.5 gCm-2d-1 over the course of the day, whereas modelled medians reached a minimum -3.0 and a maximum of 1.7 gCm-2d-1. Although the model overestimated the CO2 uptake, especially in the midday hours, the correlation between hourly medians of model output and measured NEE was generally realistic (Table ). Compared to Kema data, mean diurnal patterns of measured and modelled NEE at Nam Co site showed smaller fluxes and less variation. Measured hourly medians of NEE ranged between -2.3 and 1.0 gCm-2d-1 over the course of the day, and modelled medians between -2.7 and 1.0 gCm-2d-1 (Table ).

Daily evapotranspiration (ET) of the Kobresia pygmaea ecosystem was about 2 mmd-1 during dry periods and increased to 6 mmd-1 after sufficient precipitation (not shown). This was confirmed with small weighable micro-lysimeters giving a direct measure of ET from small soil columns over several days and SEWAB simulations. For a 33-day period at Kema 2010, ET for both micro-lysimeter and simulations varied around 1.9 mmd-1, reflecting drier conditions, while in 2012 the micro-lysimeter showed a maximum ET of 2.7 mmd-1 at BS, and the simulations 3.5 mmd-1 at IM (Fig. ). In both periods, the lysimeter measurements do not differ significantly between IM and BS (two-sided Wilcoxon rank sum test, n=4). The model results support this finding in general, as they are within the 95 % confidence interval (1.96 × standard error) of the lysimeter measurements in three cases; however they differ significantly from the lysimeter measurements for IM in 2012. The model results suggest that even for dense vegetation cover (IM), a considerable part of ET stems from evaporation. At DM and BS, transpiration of the small aboveground part of Kobresia is lower, but it is compensated by evaporation. Therefore, the water balance is mainly driven by physical factors, i.e. atmospheric evaporative demand and soil water content.

During the Kema 2012 campaign, the carbon fluxes for different degradation levels were investigated with chamber-based gas exchange measurements. Parallel measurements could not be established due to instrumental limitations, therefore the SVAT-CN model is utilised to compare the degradation classes over the whole period. In order to adapt SVAT-CN to the chamber measurements, the parameters of leaf physiology and soil respiration have been set to values that accommodate the different vegetation types and cover of the plots (Appendix A, Table ).

Daily sums of ecosystem respiration (Reco) over IM were overestimated by the model during period 1, but underestimated during the second setup over IM (period 4); see Fig. . This might be attributable to a difference in leaf area index (LAI) between the rings for period 1 and period 4, as they differed in biomass content at the end of the measurement campaign (Ring P1, NEE chamber: 3.1 g and P4, NEE chamber: 4.5 g). The model has been adapted to both periods with one parameter set in order to reflect average conditions. Overall, the model predicted a mean Reco of 2.37 gCm-2d-1 for IM, whereas the mean of the chamber data yielded 2.31 gCm-2d-1. For the chamber setup over bare soil (BS, period 2), Reco were, on average, represented well by the model (on average 0.77 gC m-2s-1) as compared to the data average of 0.81 gCm-2d-1. Similarly, for DM (period 3) modelled (1.81 gCm-2d-1) and measured (1.69 gCm-2d-1) average Reco compared well. Analogous patterns were found for daily sums of gross ecosystem exchange (GEE=NEE-Reco): under- and overestimations of the daily sums characterised the setups over IM (period 1 and 4), but were compensated to some extent when analysing period 1 and 4 together (modelled average GEE -5.39 gCm-2d-1, measured average GEE -4.96 gCm-2s-1). Average modelled GEE over BS with -0.89 gC m-2d-1 compared well to measured GEE for period 2 (-0.69 gCm-2d-1). Over DM, the average modelled GEE was -1.64 gCm-2d-1, and measured GEE showed an average of -1.94 gCm-2d-1. The model performance with respect to 30 min NEE is shown in Table , scatter plots of the regression are given in a supplement.

The mean carbon fluxes derived from SVAT-CN simulations for the different degradation classes over the vegetation period are shown in Fig. . A noticeable carbon uptake of -2.89 gCm-2d-1 for IM reduces to -0.09 for BS and even shifts to a weak release of 0.2 at DM. This is mainly related to a drop in GEE by 83 % for BS and 64 % for DM, compared to IM (100 %). While Reco for BS is reduced by 66 %, it only reduces by 12 % for DM, leading to the small net release already mentioned.

Cumulative NEE was calculated applying the four different model setups previously described: IM; DM and BS stages of Kobresia pastures at Kema; and alpine steppe (AS) ecosystem at Nam Co (Fig. ). The simulation period ranged from the period 12 July to 26 August 2012. For this period, only the IM stage showed significant carbon uptake of -133 gCm-2. DM and BS ecosystems were more or less carbon neutral (-4 gCm-2 uptake at BS, and 9 gCm-2 release at DM). The model for AS resulted in a carbon loss of 24 gCm-2 for the investigated period.

The results from two 13CO2 pulse labelling experiments at Xinghai 2009 (Hafner et al., 2012) and Kema 2010 (Ingrisch et al., 2014) show the distribution of assimilated carbon (C) in a montane and alpine Kobresia pasture (Fig. ). The study in Xinghai showed that C translocation was different on plots where vegetation had changed from Cyperaceae to Poaceae dominance, induced by grazing cessation. Less assimilated C was stored in belowground pools. The study in Kema showed that roots within the turf layer act as the main sink for recently assimilated C (65 %) and as the most dynamic part of the ecosystem in terms of C turnover. This is also the main difference between the experiments on the two sites as in the case of the alpine pasture (Kema) more C was allocated belowground than in montane pasture, where such a turf layer does not exist. However, as the experiments were conducted under different conditions and in consecutive years, a comparison of absolute values is not possible as the determined C fraction varies also throughout the growing season (Swinnen et al., 1994; Kuzyakov and Domanski, 2000).

At Kema, the 13CO2 labelling was furthermore coupled with eddy-covariance measurements to determine the absolute values of the carbon distribution in the plants, roots and the soil following a method developed by Riederer (2014): The relative C distribution within the various pools of the ecosystem, at the end of the allocation period (i.e. when the 13C fixing reaches a steady state, in our case 15 days after the labelling) was multiplied with a nearly steady-state daily carbon uptake measured with the eddy-covariance method. Besides the determination of absolute values, the continuous observation of the exchange regime with the EC confirms that the pulse labelling was conducted under atmospheric conditions similar to those of the whole allocation period. This leads to more representativeness of the result of the 13CO2 labelling experiment, which could not be repeated due to the short vegetation period and restricted access to this remote area. Please note that repetitions have been carried out, leading to standard errors as depicted in Fig. .

For investigating the influence of degradation on the development of convection and precipitation, the ATHAM model was applied for 25 % (V25) and 75 % (V75) plant cover at the Nam Co basin, with each of these in a dry and a wet scenario. From Fig.  it becomes immediately apparent that wet surface conditions are associated with higher deposited precipitation. At the same time, near-surface relative humidities are higher (not shown). For both the dry and wet cases an earlier cloud and convection development is observed for the less vegetated surface: simulations produce higher cloud cover and more convection from 10:00 local mean time (LMT) onward. At Nam Co we observed the frequent development of locally generated convective systems at similar hours in the field. Thus clouds block more incoming solar radiation between 10:00 and 14:00 LMT, the time with the potentially highest short-wave radiation forcing, for the less vegetated system compared to the intact vegetation scenario. Consequently, simulated surface temperatures were higher for the V75 scenario, leading to higher surface fluxes and a stronger simulated convection development over the day as a whole. A potential albedo effect can be excluded since the observed albedo of the vegetated surface is similar to that of the bare surface and surface temperatures remain virtually identical until convection develops.

The mechanism for this process is presumably that the vegetation cover reduces bare soil evaporation. At the same time, higher surface temperatures due to higher radiation input result in both larger sensible and latent heat fluxes in the afternoon hours, while the plant cover is able to access water that is not available for surface evaporation.

This hypothesis obviously needs to be investigated more thoroughly with field observations and simulations, but the findings indicate that changes in surface conditions can affect convective dynamics and local weather. This preliminary investigation of vegetation–atmosphere feedbacks did not take into account any spatial patterns in surface degradation that may result in larger patches with different surface conditions that may then affect circulation. However, such circulation effects are typically found in modelling studies using patch sizes with length scales that are several times the boundary-layer height.

The results for the different degradation states allow the simulation of the NEE and evapotranspiration for a gradual change from IM to BS using a tile approach of the fluxes (Avissar and Pielke, 1989). Such a tile approach is exemplarily shown for different percentages of the ecosystem types IM and BS for a 46 days period in July and August 2012 at Kema site, with simulated NEE (Fig. a) and evapotranspiration (Fig. b). As expected from the cumulative carbon gains for SIM and SBS shown in Fig. , SIM developed the largest carbon sink over the investigated summer period, whereas SBS is nearly carbon neutral in summer and a source for longer periods. The intermediate stages showed decreasing average carbon uptake with increasing amount of bare soil. Diurnal variability is largest for 100 % SIM and smallest for 100 % SBS in the ecosystem, as indicated by the interquartile ranges in the box plot.

Evapotranspiration decreases from SIM to SBS in this model degradation experiment (Fig. b), but this reduction is small compared to the overall day-to-day variability and is not supported by the lysimeter measurements (Fig. ). Therefore a change in mean ET due to degradation cannot be confirmed in this study. The day-to-day variability, however, increases from SIM to SBS. This is connected to a larger variability of simulated soil moisture in the uppermost layer, as the turf layer retains more water due to its higher field capacity and lower soil hydraulic conductivity, and the roots can extract water for transpiration from lower soil layers as well.