Drivers of soil respiration across a management intensity gradient in temperate grasslands under drought

Soil respiration is an important pathway of soil organic carbon losses in temperate grasslands; however, it is rarely studied across broad management intensity gradients in a landscape. Using the soda-lime method, we measured in-situ soil CO2 efflux with single measurements of long exposure time (i.e. 3 day long) in 150 grasslands in three German regions in early summer 2018 and 2019. The grasslands ranged from unfertilized and grazed grasslands to intensively fertilized and frequently harvested ones. To assess effects of grazing and fertilization intensities and plant diversity on soil CO2 efflux, we used Structural Equation Modeling to account for direct effects and indirect effects through soil and plant organic matter quantity and quality. Soil CO2 efflux was suppressed by limited water availability caused by naturally occurring droughts in both study years. Under the prevailing environmental conditions, grazing intensity, plant biomass and plant C:N ratio were not related to soil CO2 efflux. In contrast, fertilization intensity was positively associated with soil CO2 efflux (standardized coefficient of net effect: + 0.04 in 2018 and + 0.03 in 2019). This was because fertilization led to lower plant species richness and, thus, to lower C:N ratios in soils, which were associated with higher soil CO2 efflux (plant species richness net effect: −0.09 in 2018 and −0.18 in 2019; soil C:N ratio direct effect: −0.23 in 2018 and −0.33 in 2019). Intensively managed grasslands have higher soil respiration than extensively managed, plant species-rich grasslands even under the extreme conditions of natural droughts.


Introduction
Temperate grasslands store large amounts of carbon (C) in soils (Jobbagy and Jackson, 2000), and the fate of this C is determined by the balance between C inputs and losses due to decomposition. Grassland management intensification potentially threatens soil organic carbon (OC) stocks (Jacobs et al. 2020;Conant et al. 2017), since it can alter carbon fluxes, and especially soil respiration. Improving our understanding on the interactions among management, biotic drivers and soil respiration, while accounting for abiotic and environmental drivers (i.e. soil properties and climate), can provide important insights for sustainable grassland management and climate change mitigation (FAO 2010).
Intensification of grassland management, which mainly includes fertilization, mowing and grazing, increases soil respiration rates globally (Raich and Tufekcioglu 2000; Wang and Fang 2009). Fertilization, in general, increases soil respiration, but its effect also depends on the amounts and type of nitrogen (N) applied (Jones et al. 2006;Luo et al. 2016). Both mowing and grazing reduce soil respiration presumably by removing plant biomass (Bahn et al. 2008). However, most of the studies investigating management effects on soil respiration do not consider broad management gradients, and they include only a small number of replicated plots, hampering our understanding of management effects on grassland soil respiration. Since the temporal patterns of soil respiration remain relatively constant during the growing season (Johnson et al. 2008), short-term measurement campaigns can be used to sample a high number of replicated plots (Dias et al. 2010) and cover broad management gradients.
Grassland management can influence soil, plant and microbial properties to various degrees (Boeddinghaus et al. 2019;Conant et al. 2017;Herold et al. 2014a;Herold et al. 2014b;Kleinebecker et al. 2014), which might further affect soil respiration. For example, fertilization increases plant productivity  and the associated C inputs into the soil (da Silva et al. 2022), which increases the substrate for soil fauna and microbial activity. In addition, fertilization decreases soil C:N ratio and increases enzymatic activities (Herold et al. 2014a), possibly leading to increased C losses via decomposition. However, it is still not clear whether the quantity or the quality of organic matter, which are both influenced by management, mainly control soil respiration in temperate grasslands.
Fertilization is a main driver of plant diversity loss in grasslands (Blüthgen et al. 2012;Midolo et al. 2018), which might have an adverse effect on soil OC since recent evidence suggests that plant diversity increases C storage and decreases losses (Lange et al. 2015;Steinbeiss et al. 2008). Furthermore, fertilization of grasslands shifts soil microbial composition towards bacteria-dominated communities (de Vries et al. 2006), but other studies identified soil properties as more important drivers of microbial properties than management (Herold et al. 2014a). In conclusion, responses of soil respiration to grassland management should not only be reported as direct effects, but also account for changes that management causes to the soil, plant and microbial systems Koncz et al. 2015).
When investigating the effects of plant diversity on soil respiration both positive (Chen and Chen 2019;Chen et al. 2014;Dias et al. 2010) and neutral (Dietrich et al. 2017) effects have been reported. For instance, plant diversity can increase plant biomass productivity, and exert a positive effect on soil respiration (Craine et al. 2001;Dias et al. 2010). In contrast, a higher N use efficiency found in species-rich grasslands (Kleinebecker et al. 2014) can lead to lower N concentrations in plant biomass and litter (Fargione et al. 2007) and, thus, to reduced decomposability. Dias et al. (2010), using Structural Equation Modeling, found that plant diversity increased soil respiration in experimental grasslands (1) by increasing aboveground plant biomass, and (2) by alleviating the negative relationship between plant nitrogen on plant biomass. However, fertilization increases plant productivity and decreases plant diversity Tilman et al. 2014), leading possibly to a decoupling of the aforementioned processes in managed grasslands. Most experimental studies examining grassland diversity effects on ecosystem functions, simulate plant species losses by random assemblages of plant communities at different diversity levels, and frequently do not consider interactive effects of management practices (Klaus et al. 2020a), which are, however, a main driving force of plant diversity and ecosystem functioning in real world ecosystems.
We obtained single soil CO 2 efflux measurements of long exposure time (i.e. 3 day long) from 150 grasslands in the Biodiversity Exploratories project, which cover a broad management intensity gradient. The measurements were taken in summer 2018 and 2019 with the soda-lime method that is particularly suitable to study a high number of sites. Two successive droughts occurred in central Europe in 2018 and 2019, which affected our study regions. Our hypotheses were: • Grassland soil respiration rates differ among the three study regions due to climatic and soil differences. • Fertilization and grazing intensification indirectly increase soil respiration due to increased quantity (i.e. higher plant productivity and soil organic matter) and quality (i.e. lower C:N ratios in soils and plant biomass) of organic matter. • Extensive, plant species-rich grasslands have lower soil respiration fluxes than intensively managed grasslands due to lower nutrient status.

Study regions
This study was conducted as part of the Biodiversity Exploratories project that includes three regions in Germany; the Schwäbische-Alb (ALB), the Hainich-Dün (HAI) and the Schorfheide-Chorin (SCH) (Fischer et al. 2010). The regions differ in their geology, climate and topology (Table S1). In ALB, soils developed mainly on Jurassic limestone and were clay-rich Leptosols or Cambisols (IUSS Working Group WRB 2014). In HAI, soils had a loamy or clayey texture due to the dominant geological substrate of loess over limestone and the main soil types were Cambisols, Stagnosols and Vertisols. Soils in SCH were drained Histosols with a loamy texture, but also Gleysols, Cambisols, Luvisols and Albeluvisols.
In each study region, we selected 50 agriculturally managed permanent grasslands and, within each grassland, we established a plot of 50 m × 50 m (Fischer et al. 2010). Information regarding the regional management practices as well as the basic plant, microbial and soil properties are given in Table 1. Management ranged from unfertilized to heavily fertilized and intensively grazed and/or mowed grasslands, covering a wide management gradient. Fertilization, expressed as N additions, was applied as organic and/or mineral fertilizer. Grazing, expressed as equivalent-livestock units per area multiplied with grazing days (i.e. the days the animals are on the plot; Blüthgen et al. 2012), included mainly cattle or sheep grazing. Due to the high correlation between fertilization and harvesting in our grasslands, mowing (expressed as cuts per year) was not included as an explanatory variable.

Plant properties
Vegetation data were collected in May 2018 and 2019 at peak standing biomass. Aboveground biomass was measured in eight 0.5 m × 0.5 m subplots in each grassland. The subplots were fenced to prevent livestock grazing or mowing prior harvest. Vegetation was clipped at 20 mm height, dried at 80 °C for 48 h, weighed and ground for chemical analysis. In a nearby 4 m × 4 m subplot, we recorded all vascular plant species to determine plant species richness. We further determined the C and N concentration of plant biomass with a Near Infrared Reflectance spectrometer (SpectraStar 2400, Unity Scientific, Columbia, MD, USA) Kleinebecker et al. 2011). The reflectance spectrum of each sample, averaged over 24 scans, was recorded from 1250 to 2350 nm with 1 nm intervals.

Soil properties
One composite soil sample per grassland was prepared from fourteen soil subsamples of the upper 0.10 m of the soil along two intersecting 20 m transects in May 2017. All soil samples were sieved to < 2 mm and air-dried, and an aliquot was ground for the elemental analysis. Soil pH was measured in the extraction solution of 10 g of soil with 25 mL of CaCl 2 (0.01 M) with a pH-meter and a glasselectrode (WTW pH meter 538, Gießen, Germany). Total C and total N were determined by dry combustion at 1100 °C with an elemental analyser Vari-oMax (Elemental, Hanau, Germany). Soil inorganic carbon concentration was determined with the same analyser after removing organic carbon (OC) by exposing 250 mg of soil to 450 °C for 16 h. Soil OC concentration was calculated as the difference between total and inorganic carbon. Soil texture and fine root (< 2 mm) biomass were determined in composite samples collected in May 2011 with the same sampling procedure (Solly et al. 2014). They used a combined sieving and sedimentation method (DIN ISO 11277, 2002) for soil texture determination and weighed the isolated, cleaned Table 1 Mean values and standard deviation of management practices and soil, plant and microbial properties of the three study regions Fertilization, grazing and mowing intensity data were obtained from Ostrowski et al. (2020) and averaged for the period 2015-18 and 2016-19. Plant properties were measured both in 2018 and 2019. Microbial and soil properties were measured in 2017. Fine root biomass and soil texture were measured in 2011. Lower-case letters indicate differences among the three study regions in a given period. Asterisks indicate significant differences between 2018 and 2019 in a given study region 6.3 ± 0.6b 6.9 ± 0.5a 6.5 ± 0.9b Soil organic carbon (g C kg −1 ) 69.8 ± 14.3b 49.9 ± 12.1b 95.5 ± 90.7a Soil nitrogen (g N kg −1 ) 6.8 ± 1.5b 4.9 ± 1.2b 10.4 ± 9.0a Soil C:N ratio 10.4 ± 0.7a 10.2 ± 0.5a 9.4 ± 1.4b and oven-dried roots (at 40 °C for two days) for fine root determination.

Microbial properties
Aliquots of sieved soil (from the composite samples collected in May 2017, stored at -20 °C) were used for microbial analysis. Microbial C and N concentration was determined with the chloroform-fumigationextraction method (CFE), according to Vance et al. (1987) and Keil et al. (2011). Fumigated sample replicates were incubated with ethanol-free CHCl 3 for 24 h. C and N were extracted from each fumigated and non-fumigated replicate (5 g each) with 40 ml 0.5 M K 2 SO 4 . The suspension was horizontally shaken (30 min, 150 rpm) and centrifuged (30 min, 4400 × g Soil temperature and volumetric water content Soil temperature (°C) and volumetric water content (%) were recorded in 30-min intervals in each grassland since 2008 using the ADL-MX Data Logger System (Meier-NT GmbH, Zwönitz, Germany). Instruments were installed in fenced areas (2 m × 2 m) within the managed grasslands. Soil temperature was monitored at 0.05 m depth below the surface of the mineral soil with the MNT-FExtension and water content at 0.10 m with the Delta-T ML2X Soil Humidity Probe (Delta-T Ltd, Cambridge, United Kingdom). Sub-hourly soil temperature and volumetric water content data were averaged with respect to soil respiration measurement dates, for each grassland site and study year.
Soil CO 2 efflux Soil CO 2 efflux was measured with an absorption method using soda-lime as the absorption material along with an open and static chamber. The chamber design was based on previous work of Bierbaß et al. (2015) and Näthe et al. (2018). The chamber consisted of a PVC ring (internal diameter = 0.10 m and height = 0.12 m), a PVC lid and an O-ring to ensure air-tightness (Fig. S1). A plastic tube passing through a hole on the PVC lid allowed pressure equilibrium between the chamber and the ambient air. The outer ending of the tube was connected to a syringe containing 10 g of soda-lime to filter incoming ambient air CO 2 . Inside the chamber, soda-lime was contained in a 50 mL syringe with 64 1-mm-holes held on the PVC lid. Non-hygroscopic soda-lime with a diameter range of 2.4-5.0 mm and a saturation point of about 28% was acquired from the Fisher Scientific GmbH, Germany. Soda-lime is mainly Ca(OH) 2 and NaOH and reacts with CO 2 to form CaCO 3 . Dried soda-lime mass (105 °C for 48 h) was weighed with an accuracy of 0.1 mg before and after the exposure to soil CO 2 efflux. Each 50 mL syringe containing soda-lime was sealed in a sampling bag, while syringes from each grassland site were stored in CO 2 -free sampling bags until field installation. To correct for atmospheric CO 2 adsorbed during the laboratory and field work, bottom-sealed chambers following the same design were used as controls. The installation time was about three days (i.e. 65 to 70 h). For each measurement, we used 12 g of dried soda-lime per installation day. We determined the absorbed CO 2 mass by the difference between the dried soda-lime mass before the field measurement and the dried soda-lime mass after the field measurement (i.e. weight gain, WG). The soil CO 2 efflux was calculated by the equation (Keith and Wong 2006): where WG is the weight gain [g], CA is the chamber basal area in [m −2 ], T is the installation time in [h] and the factor 1.69 compensates for the H 2 O formed during CO 2 sorption and lost during drying (Grogan 1998).
Soil CO 2 efflux measurements were taken from June to July in 2018 and 2019 following the order HAI (from 28 of May to 20 of June), SCH (from 21 of June to 12 of July) and ALB (from 14 to 30 of July). Four chambers, forming a square of 10 m side length, were installed for the measurement of soil CO 2 efflux in each grassland site and one bottomsealed trap served as control. Aboveground vegetation was clipped and removed from the installation area and PVC rings were plugged down to 10-20 mm soil depth. Measurements (i.e. installation of sodalime and PVC lids) started five to six days after vegetation clipping. Just before installation, we applied deionized water to the dried soda-lime to compensate for the initial moisture content (about 18%) that was lost during drying, since CO 2 needs to be hydrated before reacting with the soda-lime. Over the two campaigns, we installed 1200 traps and 300 controls for the determination of soil CO 2 effluxes. In our study, the soda-lime mass gain was lower than the proposed limit of 10% per dry mass (given a saturation point of about 28% per dry mass, Janssens et al. 2000) to avoid saturation of soda-lime (Table S2). In addition, we tested the performance of the soda-lime method against an infra-red gas analyser (Li8100A, LI-COR Biosciences) in a grassland with soil pH = 7.5 over a period of three days. Based on a one-way Analysis of Variance (ANOVA) model, we did not detect significant differences between measurements taken with the two methods (Fig. S2).

Data analysis
Using ANOVA models and the Tukey's Honest Significant Difference (HSD) test, we tested for differences in in-situ soil CO 2 efflux, soil, plant and microbial properties and environmental conditions among the three study regions, and, when possible, between the two study years. Differences were considered significant at p-value < 0.05. We performed a variance partitioning to find the relative importance of (1) climate (i.e. soil temperature and moisture), (2) management (i.e. fertilization and grazing intensities), (3) plant properties (i.e. plant species richness, aboveground biomass, fine root biomass and plant C:N ratio), (4) microbial properties (microbial C and C:N ratio) and (5) soil properties (SOC, soil C:N ratio, pH and silt content) on soil CO 2 efflux. Then, we ran a (backward elimination) stepwise analysis to evaluate the importance of the individual variables on model performance based on the Akaike information criterion. Based on the selected variables from stepwise analysis, we examined the significance of these variables using Analysis of covariance (ANCOVA) models. ANCOVA models were performed assuming a type II sum of squares, which is not influenced by the order in which the explanatory variables are introduced in the model (Zuur et al. 2009). Diagnostic plots were applied to evaluate the assumptions of linearity, normality and homoscedasticity in the residuals, and to check for influential values. We applied logarithmic transformations on soil CO 2 efflux and plant species richness to tackle heteroscedasticity issues. In the selected set of independent variables for the ANCOVA models, we used the variance inflation ratio to test for collinearities. The (maximum) variance inflation ratios was ≤ 3.0 (in detail: 1.4 in 2018 and 1.6 in 2019) and our models were considered acceptable.
We used Structural Equation Modeling (SEM) to separate direct from indirect effects of grazing and fertilization intensities on soil respiration and, further, to test whether the indirect effects of grazing and fertilization are mediated by organic matter quantity and/or quality. For this, we developed an initial structure where fertilization intensity, grazing intensity and soil water content were included as independent variables, while plant species richness, soil organic carbon, soil C:N ratio, plant biomass and plant C:N ratio and soil CO 2 efflux were included as dependent variables (Fig. S3). However, plant biomass and plant C:N ratio were never associated with soil CO 2 efflux and, thus, they were not included in the final models. Using one-way ANOVA models, we removed the effects of study regions from each variable, and we used their residuals in the SEMs. Structures with p-values > 0.05 and root mean square errors of approximation (RMSEA) < 0.05 were considered acceptable. The high number of observations (i.e. 139 observations in 2018 and 147 observations in 2019 from 150 grasslands due to missing values) compared to estimated parameters provides stability against multivariate non-normality issues. In addition, we used bootstrapping (with 1000 bootstrap draws) for additional stability. Structural Equation Models represent associations between variables, and not necessarily causal relationships. However, we interpret these associations as evidence of management effects and, for simplicity, we often use terms such as 'effects' and 'drivers' hereafter.
Statistical analysis was performed with the R software (R Core Team 2020). ANOVA and ANCOVA models were performed with the lm function. Tukey's HSD test was performed with the TukeyHSD of the stats package. For the backward elimination stepwise analysis the step function from the stats package was used. Variance inflation ratio was calculated with the vif function of the faraway package (Faraway 2016). Variance partitioning was performed with the calc. relimp function using the lmg method of the relaimpo package (Groemping 2006). Structural Equation Modeling was performed with the sem function from the lavaan package (Rosseel 2012).

Soil temperature and water content
In the three study regions, soil temperature ranged from 14.5 to 22.8 °C in 2018 and from 14.6 to 24.2 °C in 2019. Soil temperature increased in the order SCH < HAI = ALB in 2018 and in the order HAI < SCH < ALB in 2019 (Fig. 1a). Soil temperature differed between the two study years in every study region, and was higher in 2019 than 2018 in ALB and SCH, but lower in HAI. Soil volumetric water content ranged from 7.4 to 51.5% in 2018 and from 7.0 to 43.6% in 2019. In 2018, soil water content did not differ among the three study regions, while in 2019 it increased in the order SCH < ALB < HAI (Fig. 1b). Between the two study years, soil water content differed only in HAI, and it was higher in 2019 than 2018. The ratio of soil water content over the water holding capacity increased in the order ALB = SCH < HAI both in 2018 and 2019 (Fig. S4). Compared to the 10-year average, soil temperature was exceptionally high, and soil water content low, during and before our sampling months, suggesting exceptionally dry conditions both in 2018 and 2019 (Fig. S5).

In-situ soil CO 2 efflux
In the three study regions, in-situ soil CO 2 efflux ranged from 0.5 to 6.9 in 2018 and 1.1 to 9.2 g CO 2 -C m −2 d −1 in 2019. In 2018, soil CO 2 efflux did not differ among the study regions, while in 2019 it increased in the order ALB = SCH < HAI (Fig. 1c). Between the two study years, soil CO 2 efflux differed only in HAI, and it was higher in 2019 than 2018, following the increase observed in soil volumetric water content (Fig. 1b). Soil properties (i.e. SOC, soil C:N ratio, pH and silt content) explained 7% in 2018 and 14% in 2019 of soil CO 2 efflux variance, which was higher than the variance explained by climate, management, plant or microbial parameters (Fig. 2). In contrast, management (i.e. fertilization and grazing intensities) had the lowest explanatory power (~ 1%) in both years. Microbial properties (i.e. microbial C and C:N ratio) accounted for only 1% in 2018 and 4% in 2019 of soil CO 2 efflux variance, and plant properties (i.e. plant species richness, aboveground biomass, fine root biomass and plant C:N ratio) explained 5-6% in both years. Climate parameters (i.e. soil temperature and water content) accounted for 4% in 2018 and 14% in 2019 showing the largest change between the two years.
ANCOVA models explained 18% of the variation in in-situ soil CO 2 efflux in 2018 and 46% in 2019, respectively (Table 2). Study regions were important predictors of soil CO 2 efflux in both study years. Fertilization and grazing intensities were rarely selected in the final models (only fertilization in 2019), and their effects were not significant. This indicates that management did not affect soil respiration at least directly. Soil volumetric water content had a strong positive effect on soil CO 2 efflux, indicating waterlimiting conditions in both study years. In 2018, plant species richness and soil C:N ratio were selected in our final models, and had a significant and a marginal negative effect on soil CO 2 efflux, respectively. In 2019, microbial C and silt content had significant positive effects on soil CO 2 efflux. Soil texture is important for water, nutrient and gas fluxes in soils, and the effect of silt on respiration is possibly related to this. As in 2018, soil C:N ratio had only a marginal effect on soil CO 2 efflux in 2019. Finally, grazing intensity, soil temperature, plant biomass and C:N ratio, fine root biomass, soil OC concentration and microbial C:N ratio were not selected as predictors of soil CO 2 efflux in neither of the study years. Fig. 2 Variance partitioning of in-situ soil respiration in 2018 and 2019 with five explanatory groups: (i) climate (soil temperature and volumetric water content), (ii) management (grazing and fertilization intensities), (iii) plant properties (plant species richness, aboveground plant biomass, plant C:N ratio and fine root biomass) (iv) microbial properties (microbial biomass C and C:N ratio) and (v) soil properties (soil OC concentration, soil C:N ratio, pH and silt content) Table 2 Linear models (according to ANCOVA type II) of insitu soil respiration in 2018 and 2019 with management intensities, soil water content, plant, microbial and soil properties Full models were stepwise reduced according to Akaike information criterion. Study region effects are given relative to the Schwäbische-Alb (ALB) region. Fertilization and grazing intensities refer to average management practices in the periods 2015-18 and 2016-19. Significant predictors are given in bold (p-value <0.05). Grazing intensity, soil temperature, fine root biomass, plant and microbial C:N ratios and soil organic carbon were never selected in the final models. Abbreviation: df = degrees of freedom; HAI = Hainich-Dün; SCH = Schorfheide-Chorin Indirect management intensity effects on in-situ soil CO 2 efflux We used Structural Equation Modeling to investigate indirect effects of fertilization and grazing intensities on in-situ soil CO 2 efflux through organic matter quantity (expressed as plant biomass and soil organic carbon concentration) and quality (expressed as C:N ratios in plant biomass and soils) (Fig. S3). Plant biomass and C:N ratio were not related to soil CO 2 efflux and, thus, not included in the final models. Soil water content had a positive effect on soil CO 2 efflux in both years (Fig. 3). In both years, soil C:N ratio was directly and negatively associated with soil CO 2 efflux, while soil OC concentration was positively associated with soil CO 2 efflux only in 2019 (Fig. 3c). Plant species richness was indirectly and negatively associated with soil CO 2 efflux due to its positive correlation with soil C:N ratios in 2018 and, in addition, due to its negative correlation with soil OC concentration in 2019 (Fig. 3b,  d). Grazing intensity remained unrelated to soil CO 2 fluxes in both study years. Finally, in both years, fertilization intensity was positively linked to soil CO 2 efflux by decreasing plant species richness and the associated decreased soil C:N ratio but, in 2019, this effect was offset by a positive indirect effect mediated by soil OC concentration.

Discussion
Grassland soil respiration is rarely studied over broad management gradients, partially due to logistical constraints associated with covering a high temporal and spatial scale. Since the temporal patterns of soil respiration remain relatively stable in the growing season (Johnson et al. 2008), short-term measurements can be used to measure in-situ soil respiration in a large number of grasslands (Dias et al. 2010), and, thus, to identify drivers of soil respiration at the regional scale. Here, using the soda-lime method, we determined in-situ soil CO 2 efflux with single measurements of long exposure time (i.e. 3 day long) in 150 temperate managed grasslands in three regions in Germany, to assess fertilization and grazing intensity effects on soil CO 2 efflux through changes in plant species richness and organic matter quantity and quality.
Grassland soil respiration was severely impacted by dry conditions In this study, in-situ soil respiration in summer averaged at 2.8 ± 1.1 gCO 2 -C m −2 d −1 in 2018 and 3.9 ± 1.3 gCO 2 -C m −2 d −1 in 2019. Soil respiration in temperate grasslands peaks in early summer time and, frequently, reaches values higher than 5 gCO 2 -C m −2 d −1 (Bahn et al. 2008;Burri et al. 2014;Koncz et al. 2015;Moinet et al. 2019, after unit conversion). Thus, compared to the literature, our summer soil respiration fluxes were relatively low. This, together with the strong correlation between soil respiration and soil water content (Table 2), indicates suppressed soil respiration because of the limiting water conditions in summers 2018 and 2019 ( Fig. S5; Apostolakis 2022). In our first hypothesis, we expected that soil respiration would differ among the three study regions due to their differences in climatic, soil and ecological conditions (Table 1 and S1). Indeed, we observed differences among the three study regions in 2019, when soil respiration in HAI was higher than in ALB and SCH, but not in 2018 (Fig. 1c). This discrepancy between the two study years was possibly caused by a slightly less severe drought in 2019 compared to 2018 (Fig. S5), a condition that was more profound in HAI (Fig. 1b). Thus, the particularly dry conditions in 2018 most likely suppressed soil respiration, and masked differences among the three study regions. Apart from masking study region effects on soil respiration, drought also altered the dependency of in-situ soil respiration from soil properties, like the soil OC. In more detail, soil respiration did not correlate with soil OC concentration in 2018, but it did in 2019 (Fig. 3). Similarly, Dietrich et al. (2017) found that microbial basal respiration did not correlate with soil OC under a naturally occurring summer drought, while they correlated in the summer of the next year that was wetter. Some studies have suggested that droughts suppress heterotrophic respiration more than autotrophic respiration both in forests and grasslands (Apostolakis 2022;Moinet et al. 2019), but such observations are frequently reported together with strong correlations between soil respiration and root biomass. Here, using fine root biomass data measured in 2011, we did not observe a correlation between soil respiration and root biomass. Grazing and fertilization intensities were not important predictors of in-situ soil respiration (Table 2), but, when we accounted for indirect effects (with Structural Equation Modeling), we found that fertilization was indirectly and mostly positively linked with soil respiration (Fig. 3a, c). Fertilization-induced effects on soil respiration were stronger in the drier (2018) than in the wetter (2019) year (Fig. 3b, d), which is in contrast with previous findings for fertilization effects on microbial basal respiration in temperate grasslands under dry and wet summers (Dietrich et al. 2017). The two successive years of drought, compared to the single dry year in the study of Dietrich et al. (2017), might have led to legacy effects in the plant and microbial communities, contributing to the discrepancy between the studies. In addition, organic fertilization can partly compensate for yield losses due to drought (Klaus et al. 2020b) which could lead to more pronounced effects from fertilization on soil respiration under drier conditions. Studies investigating fertilization effects on soil respiration in grasslands often report contrasting results including both positive (Jia et al. 2012;Jones et al. 2005) and negative responses (Ward et al. 2017). Fertilization can influence soil respiration through many mechanisms, i.e. by causing changes in soil and plant properties. For instance, fertilization effects on plant properties involve (1) increases in aboveground productivity ), (2) reductions in belowground productivity (Bardgett et al. 1999), and (3) plant species loss (Isbel et al. 2013;Midolo et al. 2018). In addition, fertilization alters the nutrient status of soils and plants Blüthgen et al. 2012;Dietrich et al. 2017;Herold et al. 2014a), which can further alter soil respiration, and its autotrophic and heterotrophic components (Zechmeister-Boltenstern et al. 2015;Zhang et al. 2014). Environmental conditions, fertilization type and time-length of the experiments (short-vs long-term) possibly determine the relevance of the different mechanisms involved and might obscure comparison between studies. Partially in line with our second hypothesis, we found that fertilization intensity indirectly increased soil respiration by altering both the organic matter quantity and quality of soils, but not of plant biomass. The lack of an effect from plant biomass and quality on soil respiration might be because of the droughts: (1) droughts occurred in June and lasted until autumn in 2018 and in 2019 affecting our in-situ soil respiration measurements (measured in June and July in both years), but not the plant properties that were measured earlier (in mid-May in both years) and/or (2) plant residues possibly occurred in upper soil depths that were drier due to low rainfall and higher evapotranspiration than deeper soil depths, which inhibits microbial activity and decomposition.
Fertilization intensification was positively associated with soil respiration even though indirectly (Fig. 3). In detail, fertilization intensification was negatively associated with plant species richness, while plant species richness was positively linked to soil C:N ratio and negatively linked to soil OC concentration. On their turn, soil C:N ratio was negatively linked to in-situ soil respiration both in 2018 and in 2019, and soil OC was positively linked to in-situ soil respiration only in 2019. However, in 2019, these effects were offset by a negative effect from fertilization on soil OC concentration. Thus, in our grasslands, both soil organic matter quantity and quality are potentially important drivers of soil respiration, but soil organic matter quality is a stronger driver than quantity and its effects vary less inter-annually (i.e. depend less on the environmental conditions).
Grazed grasslands are associated with higher soil CO 2 efflux and soil microbial respiration compared to mowed grasslands, possibly due to a higher microbial biomass and more diverse and degradable C inputs into the soil as animal excretions in the grazed grasslands (Gilmullina et al. 2020;Ma et al. 2021;Owensby and Auen 2020). In this work, grazing intensity was not associated with in-situ soil respiration neither directly nor indirectly via soil organic matter quantity and quality (Fig. 3). Similarly, in the same grasslands, grazing was found to affect nutrient cycling mostly via plant properties, and not via soil properties . In this study, plant properties effects on soil respiration were presumably dampened by the severity or the timing of the droughts.
Extensive, plant species-rich grasslands associate with lower in-situ soil respiration compared to intensive, species-poor grasslands Plant diversity effects on soil respiration have been found to be positive (Chen and Chen 2019;Dias et al. 2010) or neutral (Dietrich et al. 2017), and they are frequently reported as direct effects (Spehn et al. 2000). In contrast to most studies, we found that plant species richness indirectly reduced in-situ soil respiration due to decreased soil OC concentration and increased soil C:N ratio (Fig. 3). However, plant diversity effects on soil respiration were ultimately driven by fertilization, since fertilization intensity is a main driver of plant species richness in managed grasslands. As fertilization increases grassland productivity and reduces its biodiversity, fertilization intensity reverses the positive link between plant biomass and productivity observed in biodiversity experiments (Oelmann et al. 2021).
Extensively managed grasslands host richer plant communities compared to intensively managed ones (Blüthgen et al. 2012). Species-rich grasslands are characterized by conservative, slow growing plant communities with a high N use efficiency Kleinebecker et al. 2014). This leads to lower N concentrations in plant biomass and litter and, as a consequence, slower litter decomposition (Chen et al. 2017;Fargione et al. 2007;Lavorel et al. 2011). Apart from plant chemistry, plant species richness was also negatively associated with (aboveground) plant productivity in our grasslands . Plant productivity can increase soil respiration, possibly because of increased C inputs into the soil (Craine et al. 2001;Dias et al. 2010;Malhi et al. 2010). In our grasslands, the negative plant species richness gradient driven by fertilization intensification was positively associated with plant biomass and negatively associated with plant C:N ratio (Apostolakis et al. 2022), but these changes did not influence soil respiration (Fig. S3) in neither of the study years, presumably because of the droughts. In line with our third hypothesis, we suggest that the aforementioned characteristics of extensive, plant species-rich grasslands lead to slower C cycling compared to intensive, species-poor grasslands, which has also been proposed for root turnover and N and P cycling Oelmann et al. 2021;Solly et al. 2014) and, thus, to lower soil respiration fluxes.

Conclusions
Management intensification is a main driver of ecosystem degradation in temperate grasslands, driving plant species loss and alterations in C cycling. Under the prevailing dry conditions of this study, fertilization intensity indirectly increased soil respiration, and its effects were accompanied by a plant species richness loss and both soil organic matter quantity and quality. Soil organic matter quality increasing effects on soil respiration were observed in both study years, while those of soil organic matter quantity only in one of the years. This suggests that organic matter quality has a stronger and more stable effect on soil respiration than organic matter quantity. Intensively managed, plant species-poor grasslands have a faster C cycling than extensive, species-rich grasslands even under the extreme conditions of a natural drought. Since C cycling is tightly coupled with nutrient cycling, grassland management and plant diversity effects on litter and soil organic matter decomposition will, at least partially, determine the availability of other nutrients for plant uptake or leaching. Grassland management guidelines that consider the protection of biodiversity are needed in order to cease, if not to reverse, plant species loss and to ensure ecosystem functioning. environmental offices of Baden-Württemberg, Thüringen, and Brandenburg.
Authors' contributions Antonios Apostolakis designed and carried out the field and laboratory work related to soil respiration measurements, analysed and discussed the data, and was responsible for writing this manuscript. Ingo Schöning, Beate Michalzik and Marion Schrumpf designed the experiment and reviewed and edited drafts of the manuscript. Runa S. Boeddinghaus, Ellen Kandeler and Sven Marhan contributed data related to microbial properties. Valentin H. Klaus, Ralph Bolliger, Markus Fischer, Daniel Prati, Norbert Hölzel, Till Kleinebecker contributed data related to plant properties. Falk Hänsel, Thomas Nauss contributed soil temperature and water content data. All authors contributed to the manuscript and approved the final version.
Funding Open Access funding enabled and organized by Projekt DEAL. The work has been funded by the DFG Priority Program 1374 "Biodiversity-Exploratories" (TR 1054/4-3 and SCHR 1181/2-3) and the Max Planck Society.
Code availability Not applicable.

Conflicts of interest
The authors declare that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.