The climate change experiment Lysi-T-FACE is an experimental concept that has been designed to enable the warming of grassland plots, using an infrared heating system (Kimball et al., 2008), and enriching the CO2 content of the air using a Mini-FACE system (T-FACE; Miglietta et al., 2001). This experimental set-up was implemented at an alpine grassland site in 2010 and 2011 (Herndl et al., 2011). The overall experimental design at this site is based on a surface response approach (Piepho et al., 2017) and includes factor combinations of two elevated temperatures and two elevated CO2 concentrations at 24 grassland plots under open-field conditions. The Lysi-T-FACE approach at weighable lysimeters (Fank and Unold, 2007) is implemented at six of these plots. The proper functioning of the T-FACE performance has been repeatedly tested and has been fully operational since May 2014 at the experimental site.

The observational approach for quantifying climate change impacts in the soil–plant system has been implemented in the TERENO-SOILCan lysimeter network since 2010–2011 (Pütz et al., 2016, 2018). Intact soil monoliths were transferred within and between TERENO observatories to expose them, apart from the observations at their original site, to other climatic conditions (space-for-time substitution; see details in Pütz et al., 2016; Groh et al., 2020). The concept of the space-for-time substitution means that soils were translocated in space instead of waiting at the same location for changes in climatic conditions in time. The change in the climate regime by the transfer of the lysimeters was abrupt, which implies that we are not able to detect gradual changes in the grassland ecosystem over time, as suggested in standard space-for-time approach, but we can account for unsuspected effects from the past (Groh et al., 2020).

The study was conducted at three test sites (Table 1). The Lysi-T-FACE test site is located at the Agricultural Research Centre Raumberg-Gumpenstein (GS) in Austria. The experimental site is located at an altitude of 707 m a.s.l. (above sea level) within the Enns valley of the Austrian Alps. The mean air temperature at the site is 7.2 ∘C, mean annual P is 1000 mm, and the soil is a Cambisol. Thus, the site may be considered as representative of permanent grassland in the Alps (Schaumberger, 2011). The grasses Arrenatherum elatius and Festuca pratensis and the leguminous species Lotus corniculatus, and Trifolium pratense dominate the grassland established at the Lysi-T-FACE site. The grassland was mowed three times per year (see the Supplement; Table S1), each followed by mineral fertilisation (Herndl et al., 2010), and the average length of the growing season over the observation period from 2015 to 2018 was 197 d (for more details on the method, see Sect. 2.3 and Table S2).

At GS, six lysimeters were installed; one lysimeter was operated under ambient conditions (C0T0), two under a constant warming of 3 ∘C of grassland (C0T2) relative to the ambient surface temperature, two under a constant elevated CO2 concentration (300 ppm – parts per million) relative to the ambient atmospheric CO2 concentration (C2T0), and one with a combination of elevated temperature and elevated CO2 (C2T2). The used abbreviations of C and T within the treatments stand for CO2 and temperature, and the number 0 is for ambient conditions and is 2 for a specific level of the elevated conditions. A nearby weather station was used to obtain meteorological observations of reference P (tipping-bucket method; Young), air temperature, air humidity, solar radiation, net radiation, and wind speed at a height of 2 m above ground. Data from the weather station were used to calculate grass reference evapotranspiration (ET0) according to Penman–Monteith (Allen et al., 2006).

The two other test sites are part of the TERENO-SOILCan lysimeter network, located in the northwest of Germany, in Rollesbroich (RO) and Selhausen (SE). Both sites have a humid, temperate climate, with an average annual air temperature of 8 and 10 ∘C and an average annual P of 1150 and 720 mm for Rollesbroich and Selhausen, respectively. The plant community consists mainly of Lolium perenne and Trifolium repens. The grassland lysimeters at both sites were subject to management (cutting and fertiliser), according to the local agricultural management of the surrounding field at Rollesbroich. This includes three to four cuts per growing season (Table S1) and the application of liquid manure or mineral fertiliser three to four times per year (Pütz et al., 2016). Mean length of the growing season over the observation period was 208 and 243 d in RO and SE, respectively.

At RO (C0CL0; CL stands for climate) six lysimeters were installed to quantify the soil water budget and AGB under ambient conditions (RO – C0CL0). At SE (C0CL2), three lysimeter were installed to quantify soil water fluxes and AGB under altered climatic conditions (SE – C0CL2; less precipitation; higher ET0). A weather station (WXT510; Vaisala Oyj) was installed at both sites, logging the same meteorological parameters as at GS. The reference P was measured with a weighing rain gauge (OTT Pluvio2; OTT HydroMet GmbH), and a net radiation sensor (LP Net07; Delta OHM S.r.l.) was installed above one lysimeter at each site. In addition, the vegetation height observations were obtained for calculating ET0 with Penman–Monteith model (Allen et al., 2006).

Weighable high-precision lysimeter systems provide measurements of the components of the soil water balance equation, P, NRW, ETa, and the vertical net flux above the lysimeter bottom (NetQ). The NetQ component comprises water flow out of (Q; seepage) and into the lysimeter at the bottom; note that the latter represents the upward flow by capillary rise. The change in soil water storage (ΔS), which affects water availability in the soil ecosystem, was calculated as follows: 1ΔS=P+NRW-ETa-NetQ. The water balance components P, NRW, and ETa were obtained at each site from the highly resolved (1 min) and precise (0.01 mm resolution) lysimeter observations over a period of 4 consecutive years (1 January 2015 until 31 December 2018). NetQ was obtained at the same time interval from a weighable water tank (0.001 mm resolution).

Lysimeter mass changes are prone to external disturbances like management operations or wind. Thus, the lysimeter data of mass changes have been processed by a pre- and post-processing routine to prevent external errors and noise affecting the determination and separation of water fluxes across the land surface (P, NRW, and ETa). The procedure included in a first step a visual data quality check of the 1 min lysimeter data. In a second step, the adaptive window and adaptive threshold filter (AWAT; Peters et al., 2017) was applied, which has been shown to allow the quantification of small water fluxes like dew, water vapour adsorption, or nighttime ETa (Groh et al., 2019; Kohfahl et al., 2019). Water fluxes across the land surface and bottom boundary were aggregated to 10 min time intervals for detailed analysis of P, especially for the identification of NRW (Groh et al., 2018) and ETa. Gaps in the time series of measured P and ETa were filled with observations from parallel lysimeter observations. In the case of missing parallel lysimeter observations, gaps in the hourly time series were filled using linear regression between observed water fluxes and external observations or calculations (i.e. a rain gauge and ET0). Finally, all water balance components were aggregated on a daily timescale and were averaged over the number of available repetitions for each treatment. At GS, the water balance components for the treatment C2T0 in 2017 and 2018 were taken only from one lysimeter due to technical problems.

The component NRW, which includes fog, dew, and hoar frost formation, was determined by lysimeter mass increases between sunset and sunrise, when the corresponding rain gauge from the meteorological station did not detect P during the corresponding 10 min time step (Groh et al., 2018). The treatment of the Lysi-T-FACE plots at GS, i.e. free air carbon enrichment together with infrared heating, was active only during the growing season when plant growth occurs. In winter periods, CO2 enrichment was out of operation. The CO2 enrichment was also deactivated when the soil temperature at a depth of 10 cm was below 3 ∘C. This condition was used in the set-up at GS to define the non-growing season. However, in this work, the non-growing season was defined in a different way to that described next for comparing the manipulative approach with the observational approach. In contrast to the CO2 enrichment, the heating was not generally out of operation in winter periods but only turned off if the snow cover was higher than 10 cm. Thus, a comparison of the two treatments, i.e. free air carbon enrichment and infrared heating, was possible only in the growing season. Hence, the water balance components of the different soils at the specific sites under the original and climate change conditions were compared separately for the growing season and non-growing season.

The mean error representing the average deviation between the daily values obtained under changed conditions and those of the ambient reference was calculated using the R software (R Core Team, 2016) and the function me of the package hydroGOF (Zambrano-Bigiarini, 2017) as follows: 2Mean error=1N∑i=1NObsTi-Obsrefi, where n is the number of samples, and ObsTi and Obsrefi are the daily value on the corresponding water balance term from the treatment (subscript Ti) and reference (subscript refi). The AGB production of the grassland ecosystem were analysed solely for the growing season. A thermal-based definition of the growing season and non-growing season, proposed by Ernst and Loeper (1976), was employed in this study to identify the beginning and end of the growing season and non-growing season at the corresponding site. In this approach, the beginning of the growing season can be obtained by adding up all positive average daily air temperatures from 1 January and considering specific weight factors for each month. The daily means of air temperatures were multiplied by weighting factors of 0.5 (January), 0.75 (February), and 1 (March) and summed up. The beginning of the growing season in spring was defined as the day on which the cumulative temperature sum exceeded a threshold of 200 ∘C. The same approach was used to obtain the end of the growing season; this was achieved by starting the temperature sums backward from 31 December, with weighting factors of 0.5 and 0.75 for December and November, respectively.

The Budyko framework (Budyko and Miller, 1974) was used to obtain information about the hydrological status of the corresponding soil ecosystem. The Budyko plot characterises how the aridity index (AI = ET0/P) controls the fraction of P into evapotranspiration (evaporative index is EI = ETa/P) and runoff (Berghuijs et al., 2020). In case of values of AI <1, the system can be described as a demand (i.e. energy limited) and in case of AI >1 as a supply (i.e. water limited). Please note that, instead of considering P alone in the Budyko framework, we have also used NRW inputs here.

The dry matter AGB was gravimetrically determined with a precision balance (Gumpenstein – EA 6DCE-I, Sartorius; Selhausen and Rollesbroich – EMS 6K0.1, KERN). At GS, AGBs were dried at 55 ∘C for 48 h, and AGBs were dried at RO and SE at 60 ∘C for 24 h. The AGBs from the different cuts were summed over the growing season to the annual values and averaged over the replicated treatments (see Table 1).

The crop WUE, defined here as the amount of dry AGB produced per unit of water used by a plant (Tello-García et al., 2020), was estimated as follows: 3WUE=AGBETa, where AGB represents the dry matter biomass production (grams per square metre; hereafter g m-2) and ETa the actual evapotranspiration (millimetres) during the growing season of the corresponding year and treatment. The annual PUE (grams per square metre per millimetre; hereafter g m-2 mm-1), which are defined as the ratio of AGB and mean annual P (Zhou et al., 2020), and NRW were calculated as follows: 4PUE=AGBP+NRW. The PUE is a key indicator that explains the response of the ecosystem productivity to P (Wang et al., 2019) and NRW. Hence, PUE is used here to explore how the relationship between water balance and the crop components in the carbon cycle reacts to changes in the environmental conditions (Zhou et al., 2020).

The aridity index (–) and the ratio of seepage (Q) to P and NRW (–) were determined by Eqs. (5) and (6) as follows: 5AI=ET0P+NRW,6QP=QP+NRW. The QP ratio is a dimensionless indicator that describes the portion of P and NRW that becomes seepage and will eventually contribute to groundwater recharge. Here, we used the QP ratio as an indicator to assess how changing climatic conditions affect the hydrological functioning of the soil ecosystem. It should be noted that, for QP, AI, and PUE, normally only P is used, as quantitative information on NRW is often not available. Linear correlations between AGB, ETa, ET0, WUE, GSP, PUE, and AI were generated by the use of the package lm (R Core Team, 2016) to determine the responses of the corresponding variable/indicator under changing climate and the reference conditions of the manipulative or observational approach. The significance level (p<0.05) was used to indicate if the relationship between the variables are statistically significant.

The quantitative contribution of NRW to the soil water balance is of ecological relevance (e.g. by foliar water uptake; Dawson and Goldsmith, 2018) and can become important for crop production, especially under water-limiting conditions. This is demonstrated by the impact of the extreme weather conditions in 2018, which were associated with a decrease in the northern European crop production (i.e. drought and heatwave; Beillouin et al., 2020). The results from both experimental approaches show that the amount of NRW formed in the non-growing season is higher than that of the growing season. This agrees well with previous findings on the seasonality of dew formation across different climate zones (Zhang et al., 2019; Groh et al., 2019; Atashi et al., 2019). The results from the treatments with elevated temperature demonstrate an impact of warming on the incoming atmospheric water, i.e. NRW and P. Under these treatments, the amount of water reaching the soil ecosystem in the growing season decreases by up to 4 % for P and by up to 24 % for NRW. The observed decrease in NRW agrees well with Feng et al. (2021), who showed that heating with an infrared heater system reduced dew formation for an alpine grassland ecosystem of the Tibetan Plateau by up to 91 %. Surprisingly, not only warming but also elevated CO2 was found to alter the amount of NRW. One possible reason for the observed increase in NRW might be related to the plants themselves. It is frequently assumed that an increase in atmospheric CO2 concentration leads to a decrease in stomatal conductivity, which increases leaf temperature (less evaporative cooling; Kirschbaum and McMillan, 2018), but simulation at the leaf level showed that leaf temperature is far more affected by leaf size or wind speed than by the indirect effects of elevated atmospheric CO2 (Konrad et al., 2021). However, the effects might be more complex; a recent investigation by Habermann et al. (2019) demonstrated a clear impact of elevated atmospheric CO2 on the leaf anatomy and physiology of the C4 forage species Panicum maximum. Thus, the anatomical and physiological changes in the plant leaves might also be a possible explanation of higher dew amounts because surface properties (i.e. radiative and wetting) are of crucial importance for the formation of dew and dew yield (Trosseille et al., 2021).

As mentioned above, both experimental approaches show that the formation of NRW declines if temperature increases. This agrees well with Tomaszkiewicz et al. (2016), who predicted a decline in dew formation for forecasted trends of increasing temperature and relative humidity under future climatic scenarios for the Mediterranean region. As opposed to the changes in the absolute values of NRW, the percentage of NRW relative to P was found to be lower only under the elevated temperature of the manipulative approach, whereas this percentage increased under the warmer conditions in the observational approach. This is related to an important difference in the two approaches, namely the controlled change of individual factors in the manipulative experiment as opposed to the concurrent change of multiple drivers in the observational approach. More specifically, in the observational approach, the site with elevated temperature receives less P than the higher elevated, colder site. Thus, the relative contribution of NRW appears to be more affected by the changes in P than by the effect of warming.

A direct comparison of effects resulting from changes in P in the two experiments is difficult because only the observational approach explicitly includes a change in P within its design. Including a change in P, especially during the growing season, is important for climate change studies as the interannual variability in the ecosystem productivity (i.e. AGB), at least under water-limited conditions, is strongly correlated (Knapp et al., 2018). Hence, climate change experiments that contain information on key interacting variables such as P are indispensable to provide observations that enable a more mechanistic understanding of the ecosystem response (Hanson and Walker, 2020). The manipulative experiment shows a relatively low interannual variability in AGB (Fig. 3a), although the observation period includes the drought of the year 2018. According to the Budyko framework, the hydrological status at the experimental site was energy limited even in 2018 (Fig. 2). Thus, the interannual changes in P were not sufficiently strong to cause a shift to water-limited conditions. Assessing impacts of water stress at the site of the manipulative experiment thus requires including P as an additional treatment in the experimental design. Despite no active control on P within the manipulative approach, treatments with an increased surface temperature affected not only the formation of NRW but, to some extent, also the amount of P. Apparently, the effect of local warming on the relative humidity of the air within and above the canopy plays a role here. The local decrease in relative humidity might enable some ETa during P events, thus violating the assumption of zero ETa during rainfall, underlying the analysis of the lysimeter data. Such local effects on NRW and P in manipulative climate change experiments need further investigation, as they may lead to an under- or overestimation of ecosystem responses.

For ETa, treatments at GS with elevated CO2 and an increased temperature resulted in contrasting impacts due to their opposite individual effects on transpiration (Sorokin et al., 2017). Plants respond to elevated atmospheric CO2 concentrations with a reduced stomatal opening and an increase in photosynthesis (Kruijt et al., 2008; Ainsworth and Rogers, 2007). This is expected to lead to lower ETa and higher AGB and, thus, an enhanced plant WUE (Hovenden et al., 2017) compared to other treatments. Our results confirm the water saving effect of CO2, whereas an effect on AGB was not observed.

In contrast to elevated CO2, elevated temperatures lead to an increase in the vapour pressure deficit and, thus, enhanced evapotranspiration (Kirschbaum and McMillan, 2018). Kirschbaum and McMillan (2018) suggested that, for a range of locations from tropical to boreal forest, the transpiration-depressing effect of elevated CO2 was stronger than the opposite effect of elevated temperatures. Lenka et al. (2020) showed that treatments with an elevated CO2 concentration and a combined increase in CO2 and temperature reduced the stomatal conductance, lowered the ETa, and, consequently, improved the WUE of soybeans. In our experiments, the ETa of C2T2 was higher compared to the reference but clearly below values of ETa from C0T2. This suggests that the effect of elevated CO2 in reducing ETa only partially compensated for the effect of elevated temperature in the combined C2T2 treatment.

Kirschbaum (2004) suggested that the effect of CO2 on ETa is more pronounced for C3 plants under water-limited conditions and at higher temperatures (Kirschbaum, 2004) because it enhances the WUE of the plant (Kirschbaum and McMillan, 2018). However, its effect under non-water-limited but temperature-limited growth conditions on plants is still unclear. The low ETa of the treatment C2T0 during the dry year of 2018 indicates that elevated CO2 can mitigate the effects of summer droughts on alpine grasslands, which agrees well with Inauen et al. (2013). They showed that elevated CO2 reduced ETa by up to 7 % across a range of different grassland types in the central Swiss Alps.

Interestingly, the ETa response to the drought conditions in 2018 differed between the manipulative and observational approach. In the manipulative experiment, ETa was found to be mainly energy limited (Fig. 2) and, thus, increased under conditions with elevated temperatures (Fig. 1j). In contrast, the observational approach showed a clear decrease in ETa under conditions with elevated temperatures, as the soil ecosystem was shifted from an energy-limited to a water-limited regime (Fig. 2) due to the transfer from RO to SE (Rahmati et al., 2020). A similar reduction in ETa across different ecosystem types was observed for sites in Europe affected by the drought in 2018 (i.e. forests, grasslands, croplands, and peatlands; Graf et al., 2020). Our results revealed that the limitation of the water supply at site SE increased during the drought, and the heatwave intensified the water stress (AI in 2018 ≥1.6) and, thus, significantly reduced grassland ETa.

The above discussion shows that the manipulative experiment mainly provides insights into effects of climatic changes under energy-limited conditions, whereas the observational approach enables investigations under both energy-limited and water-limited conditions. Despite this advantage in the design of the observational approach, the influence of elevated atmospheric CO2 on the ETa response of the soil ecosystem is missing, which is an important aspect for assessment of future climate change impacts.

Another question was how climatic changes affected the drainage behaviour (NetQ) and water storage of the soil ecosystems. In the manipulative approach, elevated CO2 resulted in a significant increase in NetQ, whereas elevated temperatures (C0T2 and C2T2) significantly reduced NetQ. Ultimately, this is a result of the above-described changes in grassland ETa, which agrees well with the findings of Mastrotheodoros et al. (2020) for the European Alps. A previous investigation for a grassland at the Swiss Alps showed only a slight increase in the seepage under elevated atmospheric CO2 (Inauen et al., 2013). These changes in seasonal seepage potentially affect the catchment runoff, which is important for hydropower productivity and profitability in Alpine regions (Anghileri et al., 2018).

The observational approach also showed the tendency for NetQ to decrease under elevated temperature but in a much more pronounced way and for different reasons. At the water-limited site of SE, the non-growing season P is mainly used to replenish the large decrease in ΔS observed during the growing season. This partly explains the low impact of the dry year of 2018 on NetQ at the water-limited site of SE. However, for the energy-limited ecosystem, a strong decline in growing season NetQ (159 %) was observed in 2018 compared to the average for 2015 to 2017, which, eventually, is expected to affect streamflow. Although streamflows are buffered by groundwater, reduced recharge will deplete regional water storage reserves, as shown by Fennell et al. (2020) for a catchment in 2018 in Scotland. Long-lasting droughts can affect the soil water storage even in the following year if the winter P is not sufficient to fully replenish the depleted soil water storage (Riedel and Weber, 2020). The decrease in NetQ underlines the fact that the physical and biological response to changing climatic conditions, as well as the hydrological status of the ecosystem, controls water fluxes. Warming of the land surface, reduced P, increased atmospheric concentrations of CO2, and a higher atmospheric demand for evaporation reduce seepage and, thus, groundwater replenishment.

The partially sharp decline in AGB production under changing climatic conditions observed in both experimental approaches can be attributed to very distinct reasons. The AGB decrease for the manipulative approach might be related to an increase in heat stress due to the treatments, which is either directly induced by an elevated canopy temperature or, in case of elevated CO2, indirectly because elevated CO2 probably reduces evaporative cooling of plants (Obermeier et al., 2018). The results are in contrast to the widely expected positive effects of increasing CO2 on productivity of agricultural land (Amthor, 2001; Degener, 2015; Zheng et al., 2018). The positive effect of CO2 fertilisation on AGB was only visible for the treatment C2T0 under the conditions with less P and higher temperatures in the year 2018. This agrees well with previous studies (see, e.g., Morgan et al., 2004; Ainsworth and Rogers, 2007). In addition, the vegetation composition might have changed between the treatments and reference, as drought- and heat-induced shifts at the community and plant functional group level impact the AGB production of alpine grasslands (Tello-García et al., 2020).

The decline of AGB under a drier and warmer climate of C0CL2 in the observational approach is closely related to the hydrological status of the soil ecosystem. The combination of lower incoming atmospheric water (i.e. P and NRW) coupled with higher evaporative demand and a longer growing season results in greater pressure on the soil water resources under water-limited rather than under energy-limited conditions. The AGB for C0CL2 was reduced during the heatwaves in 2016 and 2018. In the summer of 2018, the soil ecosystem was exposed to an increasing drought intensity, and the plants turned slightly dry and brown and were visibly affected by drought stress (Rahmati et al., 2020).

In 2015, the AGB of C0CL2 was higher than that of C0CL0, which differs from the other years. This might be related to a moderate drought in 2015 (Ionita et al., 2017) combined with higher temperatures and solar radiation. For the first and most important cut in the season, C0CL0 achieved a below-average biomass, which can be explained by the exceptionally dry conditions in May in this region (i.e. second-driest month since 1950; Ionita et al., 2017). The soil ecosystem C0CL2 in SE, however, was less affected by this drought, as biomass production was able to develop much earlier in this season. The year 2015 had the longest growing season (250 d) in SE, and the warm conditions before the drought in May, and a sufficient water availability after the winter, led to similar AGB as in 2016. However, here a change in plant community composition might also be the reason for the altered AGB production. Observations on the species abundance from the observational approach underlines this explanation, as the plant community of the transferred soil ecosystem changed under water-limited conditions by reducing the abundance of herbs at the cost of grass species (Jarvis et al., 2021). Ecohydrological simulation of the grassland at RO and SE by Jarvis et al. (2021) for the years 2013 to 2018 suggests that the plant community adapted to the changing climate conditions in SE by developing a deeper root system, with a greater proportion of assimilates being distributed below ground, while stomatal conductance also increased significantly. These results show the importance of the interplay between environmental conditions (i.e. temperature) and the seasonal development of the plants (e.g. duration of the vegetation period), which cannot be taken into account in the treatments of the manipulative approach and thus represents a further limitation.

As expected, the different treatments also affected the WUE of the alpine grassland. Among the different treatments, C2T0 achieved the highest WUE. It was also shown previously for a range of different agricultural ecosystems (Nendel et al., 2009; Roy et al., 2016) that the WUE increases, as plants increase their assimilation rates and simultaneously reduce their water loss by decreasing stomatal conductance under elevated CO2 (Lammertsma et al., 2011). A higher temperature, however, seems to increase the non-productive water losses, as evident from treatment C0T2 or C2T2, which led to less efficient crop water use as compared to the unheated plots (Fig. 3c). The small difference between the WUE of C0T2 and C2T2 suggests that the higher temperatures dominated the response of the grassland ecosystem under conditions with elevated temperature and atmospheric CO2 concentration (no compensation). The largest difference in WUE between C2T2 and C0T0 was seen in 2016 and 2018. This might be related to increasing heat stress under this treatment (Obermeier et al., 2018).

Similar to the manipulative approach, the observational approach showed lower WUE under elevated temperature. De Boeck et al. (2006) also showed that the warming of several grasslands in Belgium led to a decrease in biomass production and WUE. But the same study also implied that the WUE of individual species was affected differently by warming, which might have led to compositional changes in the ecosystem. One reason for the lower WUE under elevated temperature and/or water-limited conditions might be the divergent biomass partitioning to above and below the ground in grasslands. Experimental studies found that frequent soil drying and droughts enhance root growth and the production of belowground biomass production (e.g. Hofer et al., 2017; Nosalewicz et al., 2018; Padilla et al., 2013). At least for the observational approach, the simulation of crop growth and water fluxes with an ecohydrological model reveals a deeper root growth (i.e. more belowground biomass) under drier climatic conditions at SE (Jarvis et al., 2021).

Comparing the results of WUE and AGB between both approaches suggests that the impact of the drought in 2018 on the WUE and AGB under ambient conditions is much larger in the lowlands than at higher altitudes. Both approaches showed a decline in AGB and WUE under dry conditions. In addition to the possible plant compositional changes, an elevated heat stress might further explain the different response to AGB and WUE within the manipulative experiment, whereby, for the observational approach, additional water stress due low GSP and a more depleted soil water storage were the main drivers for the changes in AGB and WUE.

The PUE values obtained are within the range (0.05 to 1.81 g m-2 mm-1) reported by Le Houérou et al. (1988) and Jia et al. (2015). The observed effect of the elevated temperature on PUE was different among the approaches. The water-limited soil ecosystem C0CL2 consumed nearly all incoming atmospheric water (i.e., P and NRW) for plant growth and drastically reduced drainage (average QP<0.1), whereas, at the energy-limited ecosystem, water was not limiting plant productivity (QP range from 0.2 to 0.5). Our results further suggest that the formation of NRW should be part of the PUE calculations as, under dry conditions, water from dew or fog might be beneficial for plants and their biomass production. NRW can be either directly taken up by plants (Berry et al., 2019) or indirectly improve their growth conditions (Dawson and Goldsmith, 2018) by, for example, reducing leaf temperature, increasing the albedo, and decreasing the vapour pressure deficit (Gerlein-Safdi et al., 2018).

The observational approach demonstrates a clear pattern in the functional relationships. Decreasing incoming atmospheric water during the growing season (P and NRW) significantly reduced the biomass productivity of the grassland, which is consistent with other studies (Zhang et al., 2020; Hossain and Beierkuhnlein, 2018; Bernhardt-Römermann et al., 2011). Our results are in contrast to the findings of Knapp et al. (2018), as the slope of the relationship between AGB and the incoming atmospheric water in the growing season was steeper in the observational compared to the manipulative approach (Fig. 5a). The lower sensitivity of AGB to P and NRW in the manipulative approach suggests that, under non-water-limited conditions, other climatic variables, such as temperature, or additional factors, such as the frequency of disturbance and fertilisation, may be more important for AGB productivity (Bradford et al., 2006). This is consistent with the low interannual variability in the growing season ETa. Even between years with contrasting P and NRW (e.g. average reduction of 23 % from 2017 to 2018), ETa was only slightly reduced (e.g. average reduction of 9 % from 2017 to 2018). Thus, water availability generally is not the limiting factor for ETa at the alpine site, although water stress may occur temporarily under drought. Our results agree well with a previous study by Wieser et al. (2008), which showed that, for 16 sites across different altitudes (580 and 2550 m a.s.l.), grassland ecosystems in the Austrian Alps seem not to suffer from water stress even in drier years.

The functional relationship of AGB, WUE, or PUE with AI also reflects the importance of the hydrological status (i.e. water limitation) on ecosystem productivity and efficiency. These functional relationships were significantly negative under the water-limited conditions of the observational approach, i.e. increasing aridity caused decreases in these indicators. A negative significant relationship also found for the energy-limited soil ecosystem (C0CL0) is mainly related to the exceptionally dry year of 2018. In other years, AGB and WUE varied considerably at the energy-limited sites both in the observational and the manipulative approach, while AI did not change (Fig. 5d–f).

The analysis of the functional relationships between hydroclimatological and ecohydrological indicators reveals that climate change experiments need to include sufficiently extreme conditions to resolve the key question of how changes in the climate regime will affect and alter the ecosystem function in the future (Knapp et al., 2018). Ecosystem responses to changes in P are of crucial importance to assess future changes in the carbon and water cycle (Paschalis et al., 2020). More distinct changes in the P conditions, for example, by using rain shelter in the manipulative approach would, thus, allow us to better capture the response of the ecosystem to future climate change.