Plant biomass tends to increase under nutrient addition and decrease under drought. Biotic and abiotic factors influence responses to both, making the combined impact of nutrient addition and drought difficult to predict. Using a globally distributed network of manipulative field experiments, we assessed grassland aboveground biomass response to both drought and increased nutrient availability at 26 sites across nine countries. Overall, drought reduced biomass by 19% and nutrient addition increased it by 24%, resulting in no net impact under combined drought and nutrient addition. Among the plant functional groups, only graminoids responded positively to nutrients during drought. However, these general responses depended on local conditions, especially aridity. Nutrient effects were stronger in arid grasslands and weaker in humid regions and nitrogen-rich soils, although nutrient addition alleviated drought effects the most in subhumid sites. Biomass responses were weaker with higher precipitation variability. Biomass increased more with increased nutrient availability and declined more with drought at high-diversity sites than at low-diversity sites. Our findings highlight the importance of local abiotic and biotic conditions in predicting grassland responses to anthropogenic nutrient and climate changes.
Nutrient inputs and extreme droughts are increasing in terrestrial ecosystems worldwide owing to global changes 1,2 , even in already colimited grasslands where plant growth is constrained by water and nutrients 3,4 . Resource supplies (for example, nutrient availability or soil moisture) often affect grasslands, causing increases (for example, nutrient addition) and reductions (for example, drought) in aboveground biomass [5][6][7][8] . Combined effects of drought and nutrient increases can yield a proportional impact, equivalent to the sum of their individual effects. However, non-proportional effects resulting in higher or lower responses than this sum can arise when one factor exacerbates or diminishes the effect of the other (for example, nutrient addition intensifying the impact of drought, and drought reducing nutrient-use efficiency) 9 . Nutrients can also buffer the impacts of drought, especially in colimited grasslands 10 . Variations in responses depend on soil water availability 11 , the plant community 12 and species-specific stoichiometric needs for water and nutrients 13 . Thus, nutrient addition can shift plant community interactions 14 , affecting drought sensitivity 15 , and drought can reduce productivity, diminishing nutrient sensitivity 16 . Understanding the mechanisms underlying these effects is crucial for predicting responses to climate-change-induced increases in drought frequency and nutrient availability.
Biotic factors such as plant richness and species abundance 17 , along with abiotic factors including water availability, interannual precipitation variability and soil texture, contribute to different responses to drought and nutrient addition 18 . Aridity critically modulates the responses of plant species to these factors 19 . In arid grasslands, water is the primary limiting factor, heightening drought sensitivity 6,20 , whereas subhumid grasslands are mainly colimited by nutrient and water availability, and humid grasslands are typically limited by nutrients or light 2 . High plant diversity and different functional groups may intersite variability in responses (Fig. 3 and Supplementary Figs. 1 and 2). The positive effects of nutrient addition and the negative effects of drought on biomass counteracted each other when treatments were applied together. This resulted in no significant difference compared with ambient conditions and produced non-proportional effects that were lower than the sum of those of the individual treatments (Fig. 3). At the local scale, individual site effects showed that nutrient addition significantly increased biomass in six sites (23%), and drought significantly decreased biomass in only four sites (15%), whereas the rest of the sites showed no significant treatment effects (Fig. 3, Supplementary Table 6 and Supplementary Fig. 2). The driest site (urat.cn; arid grassland located in China; Table 1 and Supplementary Table 1) experienced the greatest decline (80%) in biomass with drought (Fig. 3 and Supplementary Table 6). The combined treatments had significant effects in enable compensatory responses 21,22 . Communities with high graminoid abundance may experience drought-induced declines 23 , benefit from nutrient addition 24 or show no sensitivity 25 . This variability in graminoid responses may stem from differences in water and nutrient use, with faster-growing strategies and relatively shallower rooting systems compared with other resource-conservative plant functional groups 8,14,26 . However, community composition may shift under enhanced nutrient availability, driving increased dominance of certain functional groups and altering biomass production 21 .
The NPK Drought Network (NPK-D Net; https://www.bayceer. uni-bayreuth.de/npkd/index.php?lang=en) was established as a global experiment using a standardized design to assess the effects of nutrient addition and drought in grasslands across different environments. We aimed to examine the short-term effects of nutrient increases (through nutrient addition) and water limitation (through drought manipulation) on grassland biomass across environmental gradients and evaluate the influence of biotic and abiotic factors such as plant diversity and aridity 27 . We analysed 1-year biomass responses of grasslands and their main plant functional groups to drought and nutrient addition at 26 sites across nine countries on six continents (Fig. 1) and tested the following three hypotheses (Fig. 2):
(1) Arid grasslands are primarily limited by water and are thus sensitive to drought, which reduces biomass. By contrast, humid grasslands are primarily nutrient-limited, so nutrient addition notably increases biomass.
(2) Grasslands colimited by nutrients and water, which are mostly located in subhumid regions, can experience proportional (additive, equal to the sum of individual effects) or non-proportional effects of drought and nutrient addition; the non-proportional effects may be positive (>additive) or negative (<additive), depending on the influence of one factor over the other. (3) Graminoid species, owing to their having faster growth strategies and relatively shallow roots, respond more to nutrient addition and are more affected by drought than plant species from resource-conservative groups.
Across the 26 grasslands, drought reduced biomass by 19%, whereas nutrient addition increased it by 24%, although we observed high cedarsav.us ukulingadrt.za bange.cn nor.cn naqu.cn rijhn.nl yarradrt.au paike.ar bayrdrt.de cdptdrt.us MAP (mm) 0 >3,000 paike.ar marcdrt.ar chilcasdrt.ar ukulingadrt.za yarradrt.au llara.au sand.us cdpdrt.us cedarsav.us ayora.es freiburg.de rhijn.nl passogavia.it bayrdrt.de baddrt.de hulun.cn nor.cn xilin.cn urat.cn haibei.cn youyu.cn yanchi.cn hong.cn dan.cn bange.cn naqu.cn 0 -+ LRR biomass Nutrient addition Drought Nutrient addition + drought (additive effects) Nutrient addition + drought (>additive) Nutrient addition + drought (<additive) Nutrient limitation Water limitation Co limitation Fig. 2 | Conceptual model. Predicted effects of nutrient addition, drought and their combination (nutrient addition + drought) on biomass along water and nutrient availability gradients. Hypothetical responses are shown by the LRR of biomass (with confidence intervals), with positive values indicating increased biomass and negative values indicating biomass reduction. At the left extreme,
where water is the primary limiting factor, drought severely affects biomass. At the right extreme, where nutrients are highly limiting, their addition notably boosts biomass. The combined treatment effects are aligned with drought on the left and nutrient addition on the right, as one factor dominates at each extreme. The central portion represents colimited sites, where the combined effects are counteractive. Here, the response of biomass can be proportionally additive to both factors but in opposite directions or non-proportionally additive with a positive or negative effect if one factor influences the other. nine of 26 sites (35%). Of these, three sites showed higher values than the sum of individual effects, whereas most of them exhibited lower values (Fig. 3 and Supplementary Table 6). In the remaining 17 sites, no significant differences under nutrient addition + drought were found compared with ambient conditions (Fig. 3 and Supplementary Table 6). In arid sites, drought significantly reduced biomass by 28%, whereas nutrient addition increased it by 34% (Fig. 3). Biomass reduction under drought and increase with nutrient addition were higher at arid sites than at subhumid and humid sites, indicating stronger water and nutrient limitations at the arid sites (Fig. 3 and Supplementary Table 7). In subhumid sites, biomass increased by 20% with nutrient addition, decreased by 14% under drought and increased by 12% under nutrient addition + drought. In this arid category only, the combined treatment had a greater effect than the sum of the individual effects and enhanced biomass where nutrients stimulated a greater increase in biomass under drought than under ambient precipitation (Fig. 3 and Supplementary Table 7). Finally, biomass at the humid sites increased 13% in response to nutrient addition but was not significantly suppressed by drought (Fig. 3 and Supplementary Table 7). However, effects of the combined treatment were lower than expected at both extremes of the gradient (arid and humid).
The general response of our grassland sites to drought and nutrient addition was mostly driven by responses of graminoids, legumes and forbs (Fig. 4 and Supplementary Table 8). For graminoids, nutrient addition increased biomass by 30%, drought decreased it by 20%, and nutrient addition + drought increased it by a similar magnitude (20%), suggesting a non-proportional effect (greater than the sum of the individual effects, >additive) in which nutrients overcompensate for the effects of drought on graminoid biomass. Within the other groups, despite low abundance, legume biomass increased more strongly in response to nutrient addition under ambient precipitation than under drought (Fig. 4), whereas woody biomass showed a significant increase of more than 40% after nutrient addition (Fig. 4 and Supplementary Table 8).
Certain biotic and abiotic factors influenced the log response ratio (LRR) of biomass more strongly than others, and their effects interacted more with those of individual drought and nutrient addition treatments than with the effects of these treatments applied jointly (Fig. 5 and Supplementary Table 5). The effects of nutrient addition decreased with increases in mean annual precipitation (MAP) and estimated soil nitrogen (Fig. 5a,d). Interannual precipitation variability interacted with drought and with nutrient addition, resulting in a more negative impact on biomass sites with more variable precipitation (Fig. 5b,e and Supplementary Table 5). The negative effects of drought and the positive effects of nutrient addition on biomass were amplified in more diverse sites (Fig. 5c,f and Supplementary Table 5).
Our globally coordinated multisite study of drought and nutrient addition showed that both factors independently affected biomass after 1 year, with nutrient addition increasing grassland biomass and drought reducing it. However, when combined, these factors resulted in no overall impact. This general pattern can be largely explained by the strong response of a few grasslands within the global gradient, with most grasslands showing little to no response to drought or nutrient addition after 1 year. The lack of response in biomass to drought and nutrient addition combined might be due to a stronger aboveground biomass response to nutrient addition, potentially also changing water use by redistributing resources 28 . These effects may vary based on pre-existing resource conditions and site-specific characteristics such as climate, plant species composition and their idiosyncratic responses or soil nutrient availability 28 , highlighting the importance of considering aridity levels 29 . The combined effects of these factors specifically varied with aridity. In subhumid grasslands, they were significantly higher than the additive individual effects (that is, more than proportional to the sum of the individual effects). Here, nutrient addition mitigated drought-induced biomass reduction more effectively, resulting in higher biomass relative to ambient levels. On the other hand, the combined effects were lower than the sum of the individual effects (although not significantly) in arid and humid grasslands. These general responses in different aridity levels exhibited high intersite variability, with many sites lacking significant treatment effects and a few showing strong responses that could influence the overall outcome. Notwithstanding, this study offers a global perspective on short-term patterns, while acknowledging site-specific factors that can contribute to local variability. Focusing on a single year enhances the relevance of the study by capturing global geographical variability and emphasizing the need to analyse grasslands locally, considering factors such as aridity and water limitations that influence species composition and drought sensitivity 6,14 . This approach allows evaluation of short-term responses, which can differ greatly from long-term outcomes. Key processes such as soil nutrient availability, plant growth and resource allocation are highly sensitive on short time scales (for example, 1 year)
-0.4 -0.2 0 0.2 0.4 N D ND LRR biomass +24% -19% +2% a -1.0 -0.5 0.5 0 1.0 Arid sites (AI < 0.45) Subhumid sites (0.45 < AI < 0.75) Humid sites (0.75 < AI) LRR biomass -28% -2% +34% -14% +12% +20% -10% -3% +13% b c N D ND N D ND N D ND Nutrient addition (N) Drought (D) Nutrient addition + drought (ND) 4 2 -2 -4 ~200 mm yr -1
~1,000-1,200 mm yr -1 Arid grasslands Sites Humid grasslands urat.cn yanchi.cn dang.cn cdptdrt.us llara.au bange.cn xilin.cn youyu.cn nor.cn hulun.cn naqu.cn yarradrt.au haibei.cn baddrt.de marcdrt.ar cedarsav.us bayrdrt.de hong.cn sand.us rhijn.nl freiburg.de passogavia.it ukulingadrt.za ayora.es chilcasdrt.ar paike.ar 0 LRR biomass Arid (0.45 > AI) n = 7 Humid (0.75 < AI) n = 8 Drought Nutrient + drought Nutrient addition Subhumid (0.45 < AI < 0.75) n = 11 combination of both. Error bars represent 95% confidence intervals. Significant effects are indicated by error bars that do not overlap zero and by full circles. The aridity categories were defined according to the AI, estimated as the MAP divided by the PET.
and often trigger resource changes at individual and community levels that may lead to longer-term ecosystem shifts 30,31 . Further multiyear studies with an expanded number of sites are needed to fully analyse longer-term responses. Grasslands often experience colimitation from water and nutrients, affecting vegetation responses, where one factor may reduce the efficiency of the other or lessen its negative impact 2,31 . Many of the studied grasslands fell within the colimitation zone (Fig. 2), as water and nutrients influenced plant biomass 3,4 . Unlike ecosystems with extreme water (for example, hyperarid environments) or nutrient (for example, tropical rainforests) limitations, our grasslands did not encompass extremely severe deficiencies in either resource. Arid and subhumid sites exhibited important increases in aboveground biomass in response to nutrient addition, primarily driven by the pronounced responses of a few sites within each aridity level. However, some of these sites also showed negative effects of drought. Although subhumid grasslands experienced some water limitations, this was less severe than that in arid regions, where water is most strongly limiting for biomass accumulation 2,6,20 . However, subhumid sites exhibited substantial nutrient constraints, as shown by the high increases in aboveground biomass in response to nutrient addition, which also mitigated the negative impact of drought and compensated for drought-induced biomass reduction.
Previous studies found negative effects of drought and positive effects of nutrient addition on grasslands but no overall positive or negative effects after 1 year 15,[31][32][33] . In our studied arid sites, the negative impact of drought was the highest among all the grasslands analysed. However, the response to nutrient addition was greater than expected, as a higher sensitivity to drought rather than to nutrients had been anticipated 4,20 . Thus, the combined treatment was expected to exacerbate water scarcity, cause nutrient immobilization 34 and reduce nutrient-use efficiency 16 . Our results suggest that although nutrients may provide a buffering effect on biomass, drought simultaneously diminishes the ability of plants to utilize these nutrients 9 . In humid sites, where growth is mainly limited by nutrient-poor soils or light availability (when nutrients are highly available) and fast-growing species are more abundant, lower sensitivity to droughts was expected 4,20 . However, these sites showed no strong responses to nutrient addition or drought and a non-significant difference compared with ambient levels under combined treatments. This suggests that biomass regulation in humid grasslands is likely to be driven by factors such as light availability, which can be highly limited in grazing-excluded areas, with the effects of these factors surpassing those of nutrient addition or drought on plant competition 24,34,35 .
Graminoid biomass increased with additional nutrients but decreased under drought conditions, in line with previous studies 9, 15 . Graminoids typically benefited more from nutrient addition (in
Table 1 | Site characteristics Sites Elevation MAP MAT Aridity category AI CV of MAP Soil %N Percentage reduction for drought treatment ayora.es 1,050 446 12.6 Subhumid 0.49 0.26 0.20 40 baddrt.de 120 562 9.3 Subhumid 0.67 0.20 0.29 55 bange.cn 4,590 497 -0.3 Arid 0.45 0.24 0.16 40 bayrdrt.de 365 817 8.5 Humid 0.84 0.16 0.24 40 cedarsav.us 280 761 6.3 Humid 0.84 0.17 0.33 43 cdptdrt.us 965 471 9.6 Arid 0.40 0.22 0.13 30 chilcasdrt.ar 15 920 15.2 Subhumid 0.73 0.23 0.21 50 dang.cn 4,333 526 2.8 Arid 0.40 0.19 0.27 50 freiburg.de 238 974 11 Humid 1.03 0.16 0.26 30 haibei.cn 3,196 564 -1.9 Subhumid 0.66 0.12 0.31 50 hong.cn 3,500 788 2.0 Humid 0.86 0.14 0.69 50 hulun.cn 675 411 -2.2 Subhumid 0.51 0.25 0.29 50 llara.au 249 641 18.9 Arid 0.42 0.28 0.13 46 marcdrt.ar 7 879 14.3 Humid 0.75 0.19 0.33 50 naqu.cn 4,602 532 -1.6 Subhumid 0.57 0.25 0.30 50 nor.cn 145 479 6.1 Subhumid 0.49 0.25 0.19 41 passogavia.it 2,681 647 -2.8 Humid 3.76 0.16 0.33 43 paike.ar 150 321 6.6 Arid 0.29 0.36 0.15 54 rhijn.nl 8 792 9.7 Humid 1.16 0.15 0.24 51 sand.us 121 1,180 16.2 Humid 0.90 0.15 0.12 30 ukulingadrt.za 838 868 17.7 Subhumid 0.64 0.18 0.16 50 urat.cn 1,658 177 4.7 Arid 0.16 0.39 0.10 50 xilin.cn 1,263 375 0.7 Subhumid 0.46 0.39 0.19 50 yanchi.cn 1,523 371 8.2 Arid 0.32 0.23 0.08 50 yarradrt.au 24 863 17.3 Subhumid 0.66 0.27 0.20 65 youyu.cn 1,348 443 4.6 Subhumid 0.46 0.22 0.12 50 Elevation (in metres above sea level), climate (MAP in mm, mean annual temperature (MAT) in °C, AI, interannual precipitation variability (CV of MAP, calculated as the ratio of the standard deviation of precipitation to MAP), soil nitrogen percentage (%N) (estimated based on the model of Poggio et al. 53 ) and the percentage of precipitation reduction used to simulate drought. Climatic variables were estimated using the MSWEP database (https://www.gloh2o.org/mswep/) and assessed for accuracy by comparison with the WorldClim database (https://www. worldclim.org/data/worldclim21.html). Categories were defined as arid (AI < 0.45), subhumid (0.45 < AI < 0.75) and humid (AI > 0.75).
nutrient-poor soils) than other plant functional groups. Nonetheless, in the grasslands studied, graminoids showed sensitivity to both nutrient addition and drought, with considerable biomass increases and decreases, particularly in certain arid and subhumid grasslands. This response may be associated with the capacity of certain species to exploit resource pulses with a fast-growing strategy or promotion of annual graminoids 36 and associated concentration of roots in the upper soil horizons (unlike woody or other conservative-strategy species) 23 . Nutrient addition mitigated drought-induced biomass reductions in graminoids, which could be explained by increases in photosynthetic rates, leaf area and root density, allowing increased water-use efficiency 37 . Another possible explanation is that drought may indirectly reduce light limitation in humid grasslands, where light is most limiting, allowing graminoids to use nutrients more efficiently for growth 24 .
Legumes, although scarce, were highly sensitive to drought, and their biomass was reduced under combined treatments, probably owing to decreased nitrogen-fixing activity and drought stress 38 . By contrast, woody plants and forbs may have shown greater drought resistance, as species in these functional groups typically have more conservative growth strategies, often linked to long-lived perennials 37,39 . Woody biomass, although very low in abundance, unexpectedly increased with nutrients, possibly owing to pre-existence in nutrient-poor soils combined with above-average precipitation. The variability in drought and nutrient addition responses across all plant functional groups, with two of four showing no significant effects, was consistent with that reported by previous studies. These findings suggest that other factors, such as belowground traits (for example, extensive versus non-extensive rooting systems) or species origin (native versus exotic), may also have crucial roles in influencing resistance to changes in resource availability and should be considered in future studies 8,36 . c b a f e d Nutrient addition Drought 1.0 0.5 0 -0.5 LRR biomass LRR biomass -1.0 1.0 0.5 0 -0.5 LRR biomass -1.0 1.0 0.5 0 -0.5 LRR biomass -1.0 1.0 0.5 0 -0.5 LRR biomass -1.0 10 20 Plant richness 30 40 50 10 20 Plant richness 30 40 50 1.0 0.5 0 -0.5 LRR biomass -1.0 0.1 0.2 0.3 CV of MAP 0.4 0.1 0.2 0.3 CV of MAP 0.4 1.0 0.5 0 -0.5 -1.0 0.2 0.4 Soil %N 0.6 300 600 MAP (mm per year) 900 1,200 Fig. 5 | The relationships between biotic and abiotic factors and biomass response. a-d, Relationships between LRR of biomass and MAP (a), interannual variability (CV) of MAP (b), plant richness (number of species characteristic of each site; c) and estimated proportion of nitrogen in soil (%N; d) under nutrient addition conditions. e,f, Relationships between LRR of biomass and CV of MAP (e) and plant richness (f) under drought conditions. Lines and shaded areas indicate average predicted values and the corresponding 95% confidence intervals, respectively, according to linear mixed-effects modelling.
The observed 1-year biomass responses to nutrient addition and drought, if they persist or intensify over time, could alter the structure and composition of grasslands across various aridity levels; this, in turn, could potentially affect biodiversity and forage quality for livestock, which is mainly based on graminoid species 20 . Additional multiyear studies could complementarily assess how plant functional group responses may change over longer time scales. Biomass responses to drought and nutrient addition were mediated by interannual precipitation variability, plant species richness, water availability and soil nitrogen content. Drought and nutrient effects were strongly influenced by interannual precipitation variability, with more variable sites experiencing greater negative impacts. This is consistent with the findings of previous studies that sites with greater precipitation variability, typically the most arid, have similar responses to climatic extremes owing to their influence on plant community structure and growth strategies [40][41][42] . Plant species that typically dominate under these conditions lack the capacity to quickly acquire and utilize resource pulses but may persist during unfavourable periods 43,44 . However, large biomass reductions in arid ecosystems with high interannual precipitation variability may result from severe droughts, because extreme water reductions may exceed the ability of plants to tolerate the conditions 45 . Unexpectedly, high levels of species richness modulated treatment responses, amplifying both negative effects of drought and positive effects of nutrient addition. The impact of plant richness on drought effects could be linked to its relationship with resilience rather than resistance 46 , higher evapotranspiration rates 47 and intensified root competition 48 . The positive nutrient response with higher richness may have been due to niche complementarity, reduced competition and nutrient depletion 49 . Negative relationships were observed between biomass response to nutrient addition and increases in water availability and soil nitrogen content, along with unexpected positive biomass increases at some arid sites, suggesting that nutrients rather than water may be the limiting factor in arid grasslands with low soil nitrogen 2,4,34,50 .
This study addresses a critical gap in understanding how nutrient addition, drought and their combined effects affect worldwide grassland biomass production through experimental manipulations focusing on the short term. After 1 year, drought and nutrient addition showed mostly independent effects on biomass. However, in subhumid regions, effects greater than the sum of the individual effects were found under combined treatments. Our findings suggest that many grasslands are generally colimited by water and nutrients, especially in subhumid environments. In the short term, biomass increases with nutrient addition and decreases under drought, effects that are mainly explained by graminoid responses. Notably, graminoids can withstand drought when nutrient availability is high. In addition, as factors beyond water availability, such as interannual precipitation variability, plant species richness and soil nitrogen content, influence responses to drought and nutrient addition, it is essential to consider the local context when addressing the variability of these environmental factors. Further research should explore the mechanisms underlying the co-occurrence of nutrient addition and drought, as well as their effects over longer time scales, focusing on vegetation responses mediated by plant diversity, species composition, functional traits or provenance. This knowledge, in addition to that generated in the present work, will be essential for preserving grasslands and their valuable contributions to society in the context of global change.
NPK-D Net (https://www.bayceer.uni-bayreuth.de/npkd/index. php?lang=en) is a collaborative, globally distributed experimental network of researchers committed to understanding how terrestrial ecosystems respond to the combined effects of drought and nutrient addition. The network is composed of 26 sites located in grasslands across nine countries on six continents, encompassing a wide range of environmental conditions (Fig. 1, Table 1 and Supplementary Table 1). Sites follow standardized protocols for their experimental treatment and sampling regimes 6,35,51 . This study used data from the first year of the experiments.
All sites implemented four identical treatments: ambient (natural conditions with total annual rainfall); drought (rainfall reduced to an amount that simulates a one-in-100-year drought year based on local site conditions; Table 1; https://droughtnet.weebly.com/ (ref. 6)); nutrient addition, with nitrogen, phosphorus and potassium applied at 10 g m -2 per year by elemental mass (with the exception of two sites that were fertilized only with phosphorus); and nutrient addition + drought. Nutrients were added as slow-release urea for nitrogen, calcium superphosphate (P 2 O 5 ) for phosphorus and potassium sulfate (K 2 SO 4 ) for potassium. In addition, micronutrients (100 g m -2 per year of a mixture composed of 6% Ca, 3% Mg, 12% S, 0.1% B, 1% Cu, 17% Fe, 2.5% Mn, 0.05% Mo and 1% Zn) were applied once at the start of the experiment to avoid toxic levels from overapplication 51 . Drought was achieved by installing rainout shelters based on previous designs that passively remove a percentage of ambient rainfall 6,51 . Nutrient addition + drought was a joint rainfall reduction (drought) and nutrient addition treatment. There were between three and six replicate plots per site for each treatment, laid out in a randomized block design.
Abiotic factors. Climate data were sourced from the MSWEP database, a gauge-calibrated satellite estimate product that considers the period from 1979 to the present (https://www.gloh2o.org/mswep/). Daily rainfall estimates provided by MSWEP were used to calculate total rainfall over the 365 days preceding biomass harvest at each site. These estimates were also used to determine the MAP (in mm per year) based on data spanning from 1979 to the treatment year at each site. The interannual precipitation variability was quantified as the coefficient of variation (CV) of MAP, calculated as the ratio of the standard deviation to the MAP. Then, the WorldClim database (https://www.worldclim. org/data/worldclim21.html), which encompasses a time series from 1970 to 2000, was used to obtain the mean annual temperature at site level and MAP data for comparison with the MSWEP database. The aridity index (AI) and potential evapotranspiration (PET) for each site was obtained from gridded datasets of the Global Aridity Index and Potential Evapotranspiration Climate Database v.3 (https://csidotinfo. wordpress.com/2019/01/24/global-aridity-index-and-potential- evapotranspiration-climate-database-v3/), which uses the FAO-56 Penman-Monteith reference evapotranspiration equation and the ratio of precipitation and PET to estimate AI. The AI was classified following Le Houérou (1996) 52 into the following categories: arid and semiarid (AI < 0.45, denoted hereafter as 'arid' for simplicity), dry subhumid and subhumid (0.45 < AI < 0.75, denoted 'subhumid'); and humid (AI > 0.75, including category temperate grasslands). To estimate total soil nitrogen percentage at each site at a depth of 0-30 cm, we used models developed by Poggio et al. 53 . The modelled data were selected instead of actual site data to ensure that the applied metrics, calculation methodology and analysis depth were consistent and that data would encompass all study sites (Table 1).
Total live aboveground biomass (that is, the current year's growth) was measured in the first year of the experiment by clipping two quadrats of vegetation at peak biomass in each plot (quadrats ranged between 0.2 and 0.25 m 2 ). Clipped biomass from each plot was sorted by functional group. The proportion of each dominant functional group (that is graminoids, legumes, non-legume forbs or woody shrubs) in each plot, treatment and site was calculated by dividing the biomass of each functional group by the total biomass (for details of each study site's data, see Supplementary Table 2). The biomass was dried at 60 °C until a stable weight was achieved and then weighed to the nearest 0.01 g. Weights were expressed in grams per square metre. Species richness was quantified as the number of plant species in a permanent 1 m × 1 m subplot at peak biomass in each plot, when plant species cover data were assessed.
The LRR of aboveground live biomass (LRR biomass) was calculated in each plot for each treatment, in each site and each aridity category and functional group, according to the following formula: ln(biomass treatment /biomass ambient ). The corresponding confidence intervals (±95%) were also calculated. To interpret the LRR outputs, the following criteria were applied: 0 represents a treatment effect identical to ambient, values greater than zero indicate positive treatment effects, and values less than zero indicate negative effects 54 . To determine the significance of the effects of nutrient addition and drought and their combined effect on total biomass at each aridity level, in each site and in each plant functional group, we considered the confidence intervals of each LRR coefficient. An effect was deemed significant if the confidence intervals did not overlap with zero. Further statistical analysis was conducted evaluate how biotic and abiotic factors influenced the effects of treatments on biomass, using generalized linear mixed-effects models with the function glmmTMB using Template Model Builder 55 . Treatments and biotic and abiotic factors were considered as principal effects, whereas site was included as a random effect with block and plot nested within it. For this final model, explanatory variables were selected from the previously described gathered data (Table 1), after evaluation of their correlations and exclusion of those that showed high multicollinearity on the basis of a variance inflation factor, with a threshold of 3 (Supplementary Tables 3 and 4). The set of variables included in the final model were MAP, richness, CV of MAP, proportion of graminoids at each site, and soil nitrogen percentage, all of which were standardized and scaled using z-scores to ensure that the variables were on the same scale (mean 0 and standard deviation 1) before being entered into the model (Supplementary Table 5). Visual inspection of residuals, along with specific statistical tests (Kolmogorov-Smirnov and dispersion tests) using the DHARMa package 56 confirmed that the model adequately met the assumptions of normality and homoscedasticity. To visualize the model-predicted data, we also used R package ggeffects 57 , which calculates the average marginal effects of predictors from the mixed-effects model, and then we plotted them for the significant relationships obtained for the nutrient addition and drought treatments. All the statistical analyses were carried out using R v.4.4.1.