Effects of dry spells on soil moisture and yield anomalies at a montane managed grassland site: A lysimeter climate experiment

The frequency and severity of droughts in the Alps are expected to increase due to rising air temperatures and changes in precipitation regimes. Although biomass production in humid mountain areas tends to be energy limited rather than water limited, an increase in droughts may have negative impacts on the water availability and thus agricultural yields. This study aimed to analyse the impacts of dry spells on soil moisture and yield anomalies at a montane permanent grassland site in Austria. Dry spells in the time period from 2018 to 2020 were identified using the Standardized Precipitation Index, Palmer Drought Severity Index and the Soil Moisture Anomaly Index. Data from a lysimeter climate experiment were used to evaluate drought impacts on soil water storage and grassland yield under ambient and manipulated conditions. The results indicated the occurrence of three extreme droughts between 2018 and 2020. Although the studied grassland is generally considered a nonwater‐limited ecosystem, the most extreme drought in summer 2019 caused severe and extreme yield anomalies under ambient and heated conditions. Only mild yield anomalies were observed on plots with elevated atmospheric carbon dioxide concentration. This drought‐mitigating effect was attributed to the water savings enabled by partial stomatal closure under elevated CO2. The shorter dry spells in spring and late summer 2018 led to more diverse effects; mildly to moderately negative yield anomalies were found on the heated plots, whereas the anomalies tended to be less negative or even positive on plots under ambient temperature. In contrast, some time periods without water stress showed positive effects of heating on yield. These findings suggest that drought impacts on a humid montane grassland depend on both water availability and air temperature. Higher air temperature can have positive effects on yield if the ecosystem is energy limited. However, global warming suggests a tendency from energy to water limitation, in which the increased evaporative demand of the atmosphere aggravates soil moisture droughts and thus has potentially negative effects on yield.

cultural yields. This study aimed to analyse the impacts of dry spells on soil moisture and yield anomalies at a montane permanent grassland site in Austria. Dry spells in the time period from 2018 to 2020 were identified using the Standardized Precipitation Index, Palmer Drought Severity Index and the Soil Moisture Anomaly Index. Data from a lysimeter climate experiment were used to evaluate drought impacts on soil water storage and grassland yield under ambient and manipulated conditions. The results indicated the occurrence of three extreme droughts between 2018 and 2020.
Although the studied grassland is generally considered a nonwater-limited ecosystem, the most extreme drought in summer 2019 caused severe and extreme yield anomalies under ambient and heated conditions. Only mild yield anomalies were observed on plots with elevated atmospheric carbon dioxide concentration. This drought-mitigating effect was attributed to the water savings enabled by partial stomatal closure under elevated CO 2 . The shorter dry spells in spring and late summer 2018 led to more diverse effects; mildly to moderately negative yield anomalies were found on the heated plots, whereas the anomalies tended to be less negative or even positive on plots under ambient temperature. In contrast, some time periods without water stress showed positive effects of heating on yield. These findings suggest that drought impacts on a humid montane grassland depend on both water availability and air temperature. Higher air temperature can have positive effects on yield if the ecosystem is energy limited. However, global warming suggests a tendency from energy to water limitation, in which the increased evaporative demand of the atmosphere aggravates soil moisture droughts and thus has potentially negative effects on yield.
K E Y W O R D S climate change, drought, lysimeter, montane grassland, soil moisture, yield anomalies

| INTRODUCTION
Mountains are called 'water towers', because they are commonly more humid than the adjacent areas downstream (Viviroli et al., 2007).
As a result of the water abundance, plant growth in mountain areas is often limited by low temperatures rather than by water availability.
Thus, warming due to climate change is generally expected to lengthen the growing season in these areas (Menzel & Fabian, 1999;Schuchardt et al., 2021) such that agricultural yields possibly benefit.
The Alps provide an example of a mountain range that experienced an increase of temperature in the 20th century. The temperature increase in the Alps evolved stepwise with a first peak near 1950 (+1.2 C) and a second increase (+1.3 C) in the 1970s. Centennial and decadal scale temperature trends were identical for all subregions, whereas precipitation showed the most significant regional and seasonal differences, with opposite evolution for the Northwest (9% increase) versus the Southeast (9% decrease) in the 20th century (Auer et al., 2007). In western Austria, an increase in annual precipitation of about 10% to 15% was recorded, whereas in the southeast a decrease of a similar magnitude was observed (Austrian Panel on Climate Change, 2014). Under the current humid climate in (north-western) Austria, grassland yields thus are expected to benefit from warming and the associated extension of the growing season (Schaumberger et al., 2019;Schuchardt et al., 2021). The prolonged growing season and shifts in temperature and precipitation likely also affect plant phenology and growth in alpine shrublands and grasslands of the Swiss Alps (Rammig et al., 2010). These ecosystems were shaped by extreme climatic conditions, such as long-lasting snow cover and short vegetation period. As projected to occur by the end of the century, earlier melt-out and onset of growth are expected to increase plant height and biomass production. Likewise, Volk et al. (2021) found that moderate warming (+1.8 C) led to increased productivity of subalpine grassland in Switzerland, whereas the yield increase was lower under higher temperature increase (+3.0 C). The positive response to warming suggests that despite growing soil moisture deficits, productivity will increase with continued warming in the near future (Volk et al., 2021).
However, if warming is accompanied by drought, critical water deficit thresholds may be crossed, potentially causing negative yield anomalies (Schuchardt et al., 2021). This is illustrated by photosynthetic activity estimates derived by remote sensing across the Alps, which revealed shorter effective growing seasons and suppressed vegetation growth at low elevation, but longer growing seasons and enhanced growth at high elevation in response to the 2003 heat wave (Jolly et al., 2005). While the lengthened snow-free period benefited plant growth at high elevation, the growth decrease at low elevation was attributed to the increased evaporative demand and associated water stress. Jolly et al. (2005) thus concluded that there is a need to evaluate how extreme weather conditions affect biological processes.
Most continents worldwide have experienced frequent droughts in the last three decades (Mishra & Singh, 2010). For central Europe, rising air temperatures and variabilities in precipitation will increase the probability of drought periods in the Alps, where the water supply has been sufficient in most areas so far (Frenck et al., 2018). The European drought in 2018 was an example of particular interest due to extremely low precipitation and high temperatures, resulting in substantial yield losses (Buras et al., 2020;Kleine et al., 2020;Reinermann et al., 2019). In April 2018, a high-pressure system established over central Europe and persisted almost continuously until the middle of October. This caused a long-lasting dry spell and record temperatures in central and northern Europe (Buras et al., 2020). When the European drought in 2018 caused annual yield losses in northwest Europe, the annual forage yields showed a strong correlation with water deficits (Emadodin et al., 2021). Furthermore, the 2018 drought led to ongoing water deficits through the year 2019 and into 2020 (Kleine et al., 2021). Recent studies in nonwater-limited ecosystems showed that under heated (+3 C) compared to ambient conditions, soil water storage and aboveground biomass production tended to decrease in the drought year 2018 (Forstner et al., 2021). Therefore, understanding how nonwater-limited ecosystems in a humid montane environment will be affected by more frequent and severe droughts under the current climate and future conditions is important to maintain grassland agriculture's economic and ecological benefits.
Drought impacts on plant growth strongly depend on the propagation of precipitation deficits into the soil. Changes in soil moisture caused by future shifts in climate may have far-reaching implications for the structure and dynamics of managed and unmanaged ecosystems (Jasper et al., 2006). Critical soil moisture contents can serve as an indicator to mark the transition from energy-limited to water-limited conditions and the potential negative impact of drought on biomass production (Buitink et al., 2020). Therefore, detailed knowledge of soil moisture variability is essential for understanding climate-induced changes in seasonal/ temporal vegetation patterns and agricultural yield (Jasper et al., 2006). This work thus aims to identify the effects of droughts on soil moisture and grassland yields under the current climate and under manipulated conditions in a nonwater-limited grassland ecosystem.
To this end, hydro-meteorological and biomass data from a climate lysimeter experiment were standardized and analysed using various drought and anomaly indices.
The main objectives are to a. identify dry spells from meteorological and soil moisture data; b. assess the impacts of dry spells on soil moisture under ambient conditions, elevated temperature and/or elevated atmospheric carbon dioxide content; and c. assess if and how grassland yields respond to dry spells under ambient conditions, elevated temperature and/or elevated atmospheric carbon dioxide content.

| Experimental test site
A multifactorial climate experiment (Lysi-T-FACE concept) has been carried out at an experimental grassland site since 2014 (Herndl et al., 2011). The site is operated by AREC Raumberg-Gumpenstein and located in the northern Enns Valley of Styria (Austria) at 707 m.a.
s. The Lysi-T-FACE concept involves the exposure of grassland plots to elevated temperatures using an infrared heating system (Kimball et al., 2008) and elevated CO 2 concentrations using a mini-FACE system (Miglietta et al., 2001) in different factor combinations based on a surface response approach (Piepho et al., 2017).
According to the Köppen climate classification, the study area is located in the subpolar oceanic climate zone of the montane belt of the Alps (Rubel et al., 2017). The mean annual air temperature and mean annual precipitation for the period 1991 to 2020 were 8.5 C and 1009 mm, respectively. The geological zone is the Eastern Grauwackenzone; the grassland soil type is classified as a Stagnic Cambisol, having a loamy sand texture. The grassland consists mainly of C 3 grasses Arrhenatherum elatius and Festuca pratensis and leguminous species Lotus corniculatus and Trifolium pratense (Herndl et al., 2011).
The site has been managed with three cuts per year (Table S1). After each cut, the harvest from each plot is analysed for fresh and dry matter, and the plots are fertilized with mineral fertilizer (Herndl et al., 2011). The site may be considered as a representative permanent grassland of the Alps (Schaumberger, 2011).
The grassland test site is fully operational only during the growing seasons. In winter, CO 2 enrichment is switched off when the soil temperature is below 3 C, while heating stops when the snow cover is above 10 cm. Therefore, only the growing seasons were considered for the analyses of treatment effects. The growing season was defined according to Ernst and Loeper (Ernst & Loeper, 1976), where the start of the growing season was identified by summing all positive mean daily air temperatures since 1 January with specific weighting factors for each month (0.5 for January, 0.75 for February and 1 for March).
Once a total temperature sum of 200 C was reached, the growing season was expected to start. To determine the end of the growing season, the weighted (0.5 for December, 0.75 for November and 1 for October) daily air temperatures were added backwards from 31 December to a threshold of 200 C (Table S2).
This study used six plots at the experimental site equipped with high precision weighable lysimeters (METER Group AG, Munich, Germany). One lysimeter is operated under ambient conditions (C0T0), two lysimeters are treated with higher temperatures (+3 C; C0T2), two lysimeters are operated under elevated CO 2 concentrations (+300 ppm; C2T0) and one is a combination of both, elevated temperature and elevated CO 2 concentration (+3 C and +300 ppm; C2T2) ( Figure 1). Each lysimeter has a surface area of 1 m 2 and a depth of 140 cm. Lysimeter and seepage weights are measured at 10 (0.01 mm) and 1 g (0.001 mm) resolution.
The lysimeters are equipped with time-domain reflectory (TDR) probes (METER Group AG, Munich, Germany) measuring volumetric soil water content at depths of 10, 30 and 50 cm. Furthermore, temperature combined tensiometers (T08-M; METER Group AG, Munich, Germany) are installed at depths of 10, 30, 50 and 140 cm for matric potential measurements. The lysimeter system is operated with a controlled bottom boundary for an upward and downward directed water flux across the lysimeter bottom to ensure that the lysimeter water fluxes can be adjusted to matric potentials measured in the undisturbed soil (Groh et al., 2016).
The weather station at Irdning-Gumpenstein operated by the Austrian meteorological service (Zentralanstalt für Meteorologie und Geodynamik, ZAMG) was used to obtain meteorological data such as precipitation, air temperature, air humidity, solar radiation and wind speed, needed to calculate grass-reference evapotranspiration (ET 0 ) according to Allen et al. (2006).

| Lysimeter data analysis
The lysimeter weight and the seepage weight provide information about the soil water balance components precipitation (P), actual evapotranspiration (ET a ) and seepage rate (NetQ), that is, the water flux across the lysimeter bottom. To quantify P, ET a and NetQ, data processing with manual and technical correction was necessary to compensate disturbances due to external forces on the lysimeter such as wind, manual operations (cutting events, maintenance) or animals (Forstner et al., 2021). The changes in soil water storage (ΔS) obtained from the lysimeter water budget (Equation 1) were used to determine the amount of available moisture within a plant's root zone: Assuming that evaporation is zero when it rains, P and ET a can be directly obtained from the weights of the lysimeter and seepage. In case of missing and/or unrealistic data, however, the precipitation from the weather station and the grass-reference evapotranspiration (ET 0 ) were used to fill the gaps. Changes in soil water storage were further aggregated into hourly, daily, monthly and yearly time steps.
Lysimeter plots with the same factor combination (C0T2 and C2T0) were averaged and displayed with their standard deviation. Due to F I G U R E 1 Lysimeter plots at the experimental grassland test site (Lysi-T-FACE) with their different factor combinations: ambient (C0T0), +3 C (C0T2), +300 ppm CO 2 (C2T0) and +3 C combined with +300 ppm CO 2 (C2T2) technical problems, water balance data for the treatment C2T0 were only available for one lysimeter in 2018.
The volumetric soil water contents (SWC in % vol) obtained from the TDR probes were used to determine soil moisture anomalies at 30 cm depth for each lysimeter. This depth was chosen based on the consideration that effects of dry spells on soil water content are most obvious in soil layers with active plant roots (Zhang et al., 2017). For the grassland considered here, active plant roots are present at least to a depth of a few decimetres, allowing water to be drawn from deeper layers in the event of drought. Measurements of soil water content at 30 cm depth were thus preferred to those at 10 cm, as the identification of dry spells from a sensor at 10 cm potentially overestimates the effects of drought. Soil moisture anomalies at 10 and 50 cm depth show similar dynamics to those at 30 cm depth and are illustrated in Figures S1 and S2.
Furthermore, tensiometer data for each lysimeter were used to compare matric potentials (Ψm in hPa) under different treatments and at different depths. The matric potential and soil water content data were subject to data correction, where technical thresholds from the systems were automatically flagged and deleted. Soil water fluxes and soil moisture conditions were analysed over three growing seasons between 2018 and 2020.

| Identification of dry spells
Different drought indices were used to identify and describe droughts from 2018 to 2020. This work focused on meteorological drought, which is defined as a lack of precipitation over a region for a period of time, as well as on agricultural drought, which refers to a period with declining soil moisture and consequent crop failure without any reference to surface water resources (Mishra & Singh, 2010). In agricultural areas, soil moisture is particularly useful, as it reflects the water content in the upper part of the soil where crops grow (Sgroi et al., 2021).
In this work, we combined drought indices based on local meteorological data with the analysis of soil water anomalies inferred from soil moisture measurements.
The Standardized Precipitation Index (SPI; McKee et al., 1993) is one of the most widely used drought indices and is based on the consideration that each component of a water resources system reacts to a deficit in precipitation on different time scales (Tsakiris et al., 2007).
For agricultural interests, short-term durations in the order of months may be important (Guttman, 1998). Therefore, monthly SPI, obtained with precipitation from the weather station Irdning-Gumpenstein (ZAMG), was used to analyse drought anomalies at the grassland site. and Yield Anomaly Index [YAI]).
The PDSI is obtained from hydro-meteorological data and is one of the most commonly used regional drought indices for drought (Liu et al., 2012;Mishra & Singh, 2010;Palmer, 1965). In addition to precipitation, the PDSI also uses monthly potential evapotranspiration and a two-layer bucket-type hydrological model to include the effects of water availability in the root zone (Liu et al., 2012). In our work, monthly values of the PDSI were calculated using precipitation and the grass-reference (Allen et al., 2006) obtained from the weather station Irdning-Gumpenstein (ZAMG). The available water capacity (≈14 cm) required to calculate PDSI was determined based on laboratory measurements from soil samples collected in the field.
The SMAI is defined as the deviation of the soil moisture from its long-term mean (Jiménez-Donaire et al., 2020). For each lysimeter, daily SMAI was calculated from the volumetric soil water content at 30 cm depth measured inside the lysimeters (2): where SWC (%) represents the volumetric soil water content at 30 cm depth, SWCμ (%) its mean value and SWCσ (%) the standard deviation in the time period 2015 to 2020. SMAI values >0 correspond to wet conditions, whereas SMAI values <0 correspond to dry conditions.
The SMAI thus reflects the degree of soil dryness or saturation compared to normal conditions at a depth of 30 cm, whereas the abovementioned change in soil water storage from the lysimeter water budget characterizes the changes in the bulk soil water content.
The Budyko framework was used to characterize the observed ecosystem's hydrological status (Budyko & Miller, 1974). The aridity index (AI) was calculated as the ratio of the annual grass-reference evapotranspiration (ET 0 ) to the annual precipitation (P) measured at the aforementioned weather station. Likewise, the evaporative index was estimated by dividing the actual annual evapotranspiration (ET a ) through the annual P. If AI < 1 evapotranspiration is limited by its demand for energy (energy limited), whereas for AI > 1 water is the limiting factor (water limited) (Giraud et al., 2021).

| Yield anomalies
At each cutting event, fresh aboveground biomass was manually cut to 7 cm crop height. The dry matter yield was determined using a method adapted from the Association of German Agricultural Inspection and Research Institutes (VDLUFA, 2013). Three samples of fresh aboveground biomass were taken for each lysimeter (12 samples in total for six lysimeters) and each biomass sample ($600 g) was filled where AGB (kg/ha) is the yield of the time period considered, AGBμ (kg/ha) is the mean annual yield for the time period from 2015 to 2020 and AGBσ (kg/ha) represents the standard deviation for each lysimeter treatment.

| Statistical methods
We used a linear model to test the significance of the regression slope of SPI, PDSI and temperature over time, using R version 4.  The PDSI displayed monthly values of less than À2 (extreme drought) in 1991, 1992, 1994, 1995, 1996, 1999, 2003, 2018 and 2019, while the monthly SPI showed the occurrence of extreme drought events for 1992, 1994, 1995, 1997, 2007, 2015 and 2019. The trend for SPI and PDSI from 1990 to 2020 showed a slight increase, which is statistically not significant (p > 0.05). The mean annual air temperatures (T avg ) from 1990 to 2020 showed a statistically significant increase (May, June, August and September) and 3 months in 2019 (June, July and August). Like the SPI and PDSI, SMAI did not detect any days with extreme drought in 2020 (Table 2, Supplement Table S 3).  (Table S4).

| Impacts of dry spells on soil moisture change
Similar to ΔS, the Soil Moisture Anomaly Index showed strong decreases (with values of less than À2) in spring 2018 for ambient conditions, in late summer 2018 and summer 2019 for all climate treatments and in spring 2020 for heated conditions (Figure 3 and Table S3).
For the 2018 and 2019 extreme drought events, the treatment effects (C2T0, C0T2 and C2T2) on SMAI differed from those observed for ΔS.
SMAI exhibited the strongest decreases for C0T0, whereas ΔS showed the highest decreases for C0T2. The lowest SMAI value (À3.7) in the whole observational period was obtained for C0T0 on 26 July 2019. In spring 2020, C0T2 showed the lowest SMAI value of all plots (À2.4 on the 24 May), thus behaving similarly to ΔS (Figure 3).
The soil moisture dynamics represented by ΔS and SMAI followed the precipitation obtained from the weather station Irdning-Gumpenstein (ZAMG) (Figure 3). During the extreme drought period detected in spring 2018, that is, from the beginning of the growing season to the first cut in 2018, the average daily P was only 2 mm, which is lower than in the subsequent growth periods (4 and 3 mm) (Table S1). During the extreme drought period immediately after the second cut of 2018, time periods with low precipitation were recorded (8.8 mm between 9 August and 23 August). In summer 2019, from the end of June (28 June) to the end of July (28 July), the daily precipitation did not exceed 3 mm for 23 days. During this time period, SMAI reached its lowest values (<À3) for all treatments ( Figure 3). During the dry spell in spring 2020 (first growth period), F I G U R E 3 Change of the soil water storage (ΔS) relative to the beginning of the growing season and Soil Moisture Anomaly Index (SMAI) at 30 cm obtained from the lysimeters operated under ambient conditions (C0T0), +3 C (C0T2), +300 ppm CO 2 (C2T0) and +3 C combined with +300 ppm CO 2 (C2T2), precipitation (P) and mean average temperature (T avg ) measured from the weather station data at Irdning-Gumpenstein (ZAMG) during the growing seasons from 2018 to 2020 F I G U R E 4 Matric potential (Ψm) at 10, 30 and 50 cm measured from the lysimeters operated under ambient conditions (C0T0), +3 C (C0T2), +300 ppm CO 2 (C2T0) and +3 C combined with +300 ppm CO 2 (C2T2) during the growing seasons from 2018 to 2020 the average daily precipitation of 2 mm was lower than that in the subsequent growth periods (4 and 5 mm) (Table S1). Precipitation during the 2018 growing season reached 675.8 mm, slightly higher than that of 2019 (572.1 mm), but lower than in 2020 (714.5 mm).
The dynamics of the matric potential (Ψm) (Figure 4)   Tables S1 and S5) of the years 2018 to 2020. for the first cut in 2020 compared to that in 2018 and 2019. Under heated conditions, the AGB for the first cut in 2018 was lower than 2020 and 2019 ( Figure 5 and Table S5). The third cut in 2018, which was associated with the extreme dry spell of the late summer of 2018, displayed negative YAI (mild anomaly) values for all plots; the largest (most negative) anomalies were found for C0T2 and C0T0.

| Yield anomalies
The smallest anomalies were found for C2T0 and C2T2. During the extreme drought period in summer 2019, which was associated with the second growth period, negative YAI values (from mild to extreme) were observed for all plots (second cut 2019 in Figure 5). Precipitation (105 mm) was lower and mean air temperature (24 C) higher during this time as compared to the second growth period in 2018 and 2020 (242 and 252 mm; 17 C and 16 C) ( Table S1). The AGB of the ambient plot was only 1818 kg/ha for the second cut 2019 compared to F I G U R E 5 Precipitation (P) and average air temperatures (T avg ) measured from the weather station Irdning-Gumpenstein (ZAMG), aboveground dry biomass production (AGB) and Yield Anomaly Index (YAI) under ambient conditions (C0T0), +3 C (C0T2), +300 ppm CO 2 (C2T0) and +3 C combined with +300 ppm CO 2 (C2T2) for each cutting event from 2018 to 2020 2901 (second cut 2018) and 2811 kg/ha (second cut 2020) in the corresponding growth periods of the other years ( Figure 5 and Table S5).
Similarly, the treatments showed the lowest AGB of the second cut in 2019. An exception was C2T2, which showed a slightly larger AGB (1761 kg/ha) for the second cut 2019 than for that in 2020 ( indicating extreme drought events in these years. The number of extreme drought events indicated by PDSI was higher than that by SPI ( Figure 2 and Table 2). Comparisons of SPI and PDSI and their individual strengths and limitations are frequently reported in the (Mishra & Singh, 2010). Generally, PDSI is considered more suitable for agricultural impacts than for hydrological droughts due to its inherent memory (Guttman, 1998). As compared to SPI, PDSI considers temperature in addition to precipitation. While precipitation anomalies dominate PDSI in the cold season, temperature gains importance in the warm season (Mishra & Singh, 2010). This suggests that the higher number of dry spells in the growing seasons 2018 and 2019 identified by PDSI are related to the high air temperatures in those time periods.
The number of months in which SMAI indicates extreme droughts lies between those of SPI and PDSI but appears closer to that of PDSI (Table 2). Thus, the soil moisture-derived droughts indicated by SMAI were more successfully identified from hydro-meteorological data using PDSI than SPI. This contrasts with findings from three regions of the United States where SPI was found to be a better indicator for soil moisture variation than PDSI (Sims et al., 2002). This suggests that the interpretation of drought indices should consider the local environmental setting and research context. Thus, caution is needed when generalizing results about the performance of drought indices. and from 26 June to 31 July 2019 (second growth/beginning of third growth; summer 2019) were considered as extreme drought periods (under ambient conditions) for the following discussion (Table S3).

| Soil moisture and yield anomalies
This section focuses on the impacts of the three above-mentioned  Figure 5) yield anomalies on the ambient plots. The finding that only the summer drought 2019, that is, the only extreme drought event identified by SPI, resulted in severe yield anomalies agrees with a study from the Canadian Prarie where SPI was superior to PDSI in predicting crop yields (Quiring & Papakryiakou, 2003). Based on that study, Mishra and Singh (2010)  A more comprehensive understanding of the yield anomalies is obtained if the impact on biomass production of factors other than water availability is considered. As the site is located in a humid region (e.g., as indicated by the AI), biomass production generally is expected to be energy limited rather than water limited. The AGB and yield anomalies of the first cuts generally agree with this expectation. For example, the AGB and YAI of the ambient plot follow the air temperature. In particular, the low yield along with the low air temperature in spring 2020 is striking ( Figure 5). The positive YAI of the heated plots in spring 2020 also confirm that biomass production benefits from higher air temperature at this site if the water availability is not  Figure 5).
The depletion of the soil water storage throughout the 2018 to 2020 growing seasons was generally lower in the plot enriched with atmospheric CO 2 (C2T0) compared to the other plots; likewise, the soil water storage of C2T2 remained higher than that of C0T2. This can be attributed to the stomatal effect of CO 2 , which reduces transpiration and thus results in soil water savings (Field et al., 1995;Vremec et al., 2022). The weaker yield anomaly for the second growth period 2019 ( Figure 5) for the plots enriched with CO 2 (C2T0 and C2T2; YAI > À1) relative to the other plots (C0T0 and C0T2; YAI < À1.5) suggests that the water savings resulting from elevated CO 2 mitigated the adverse effect of the extreme drought on yield.
This is also apparent in the third growth period 2018, which was affected by the extreme drought period in the late summer. Here, the plots with elevated CO 2 were less depleted in soil water storage ( Figure 3) and tended to have higher matric potentials (Figure 4) than the other plots at the beginning of the third growth period 2018. Mitigation of drought effects on yield resulting from these water savings is indicated by the weaker negative anomalies on the plots with CO 2 enrichment (C2T0 and C2T2) in the third growth period (third cut 2018 in Figure 5). Yet the effect of elevated CO 2 appears to be less pronounced in the late summer drought 2018, which resulted in only mild yield anomalies under ambient conditions, whereas the mitigation of water stress by elevated CO 2 is more clearly indicated in the more extreme drought of 2019 by the shift from severe yield anomalies under ambient conditions to only mild anomalies (second cut 2019; third cut 2018 in Figure 5).
Beside yield anomalies, also shifts in functional groups as grasses, herbs and legumes are expected as a result of drought conditions (Tello-García et al., 2020). Generally, grasses are responsible for the yield stability and yield certainty, herbs and legumes are more responsible for mineral amounts and nutrient fixation (Buchgraber & Gindl, 2004).
The manipulated lysimeters (C0T2, C2T0 and C2T2) showed a different composition of the functional groups compared to ambient conditions (C0T0) ( Table S6). Under ambient conditions, the highest amounts of grasses with the lowest amounts of herbs were observed, matching the highest AGB in Figure 5.
Comparing the different climate treatments (C0T2, C2T2 and The shifts of functional groups have an influence on yield, as exemplified by the above-mentioned highest percentage of grasses coinciding with the highest AGB of the ambient lysimeter. Likewise, effects on the yield anomalies during dry spells might be expected.
These, however, appear to be less evident here compared to the other effects discussed above, such as the water savings resulting from the stomatal effects of CO 2 . Thus, more comprehensive, systematic investigations will be required to identify effects of shifts in functional groups on yield anomalies.
The above findings demonstrate that inferring impacts on soil moisture and yield from drought indices is not straightforward, because these variables depend on factors not considered by the indices. This is particularly relevant to drought impacts at humid montane sites, where the shifts between energy-limited and water-limited conditions are controlled by both water availability and air temperature.
Thus, at these sites, effects of global warming are expected to be twofold. Firstly, the generally higher air temperatures suggest a tendency from energy-limited to water-limited conditions, that is, time periods where biomass production is not hampered by low air temperature, are expected to become more frequent. Secondly, higher air temperature results in increased evaporative demand of the atmosphere, which aggravates soil moisture drought. However, during intense droughts, elevated atmospheric CO 2 concentrations may mitigate the effects of water stress on biomass production.
If global warming tends towards generally lower water availability, the drought-related anomalies may appear weaker. This is illustrated by the different behaviour of the soil water storage and the Soil Moisture Anomaly Index in Figure 3. The depletion of the soil water storage generally was strongest on the heated plots (C0T2), but this was not reflected by the anomalies of the soil moisture. The largest soil moisture anomalies were observed under ambient conditions. Likewise, in times of longstanding dry spells, elevated CO 2 resulted in SMAI values similar or even lower than those of the heated plots.
Thus, despite the above-mentioned water savings, the anomalies under elevated CO 2 appeared to be larger than those under heated conditions.
A higher relative reduction of soil moisture at a wetter site (Stubai in the Austrian Alps) compared to a drier site (Lautaret in the French Alps) was also found in the study by Leitinger et al. (2015).
Also, the seepage rate under drought was more reduced at the wetter site than at the drier site. Likewise, Rahmati et al. (2020) found that increasing dryness led to more temporal variability in soil water content at an energy-limited site compared to a water-limited one. Thus, caution is needed when drought or anomaly indices are employed and interpreted in a changing environment, particularly if the ecohydrological controls are fundamentally altered because of a shift from energy-limited to waterlimited conditions.

| CONCLUSIONS
In this study, we analysed the relationship between soil moisture and yield anomalies and its response to drought under ambient and manipulated climate conditions at an experimental montane grassland site.
Drought indices based on hydro-meteorological data (SPI, PDSI and SMAI) were used with the soil water storage obtained from lysimeter measurements to detect drought periods.
Even though the grassland site is considered a nonwater-limited ecosystem, all drought indices used in the study detected an extreme drought event in summer 2019. During this period, reduced water availability also led to severe and extreme yield anomalies under ambient and heated conditions. The lowest yield anomalies (mild) were observed on plots enriched in atmospheric CO 2 . This finding is attributed to the partial stomatal closure under elevated CO 2 , which enables water savings in the period preceding the drought, and was found to be most pronounced on the heated plots. This suggests that the drought-mitigating effect of elevated CO 2 gains importance under warming. PDSI and SMAI indicated the occurrence of extreme droughts also in the growing season 2018. However, these were not detected by SPI and resulted in only mild yield anomalies both under ambient conditions and elevated CO 2 , arguably because the drought events did not coincide with high temperatures and resulted in less severe and less persistent depletion of soil moisture as compared to 2019. During these (shorter) drought events in spring and late summer 2018, the heated plots showed more pronounced (moderate and mild) yield anomalies than the other treatments. This suggests that warming can aggravate water stress during droughts. If the water availability is not limiting, higher air temperatures facilitate biomass production, as indicated by positive yield anomalies on the heated plots in spring 2020.
Our results show that the yield of a grassland ecosystem in a humid montane environment that is generally nonwater-limited can be reduced due to the temporary depletion of the soil water storage under extreme drought. However, our findings demonstrate that extreme drought does not always lead to significant yield reductions in a humid montane environment. Temporary depletions of soil moisture may have no or little impact on yield, particularly when low air temperatures, for example, in spring, limit plant growth. Other potential controls, such as the timing and duration of the drought, also merit further investigation.