Multiple resource limitations explain biomass-precipitation relationships in grasslands

Interannual variability in grassland primary production is strongly driven by precipitation, nutrient availability and herbivory, but there is no general consensus on the mechanisms linking these variables. If grassland biomass is limited by the single most limiting resource at a given time, then we expect that nutrient addition will not affect biomass production at arid sites. We conducted a distributed experiment manipulating nutrients and herbivores at 44 grassland sites in 8 regions around the world, spanning a broad range in aridity. We estimated the effects of 5-11 years of nutrient addition and herbivore exclusion treatments on precipitation sensitivity of biomass (proportional change in biomass relative to proportional change in rainfall among years), and the biomass in the driest year (to measure treatment effects when water was most limiting) at each site. Grazer exclusion did not interact with nutrients to influence driest year biomass or sensitivity. Nutrient addition increased driest year biomass by 74% and sensitivity by 0.12 (proportional units), and that effect did not change across the range of aridity spanned by our sites. Grazer exclusion did not interact with nutrients to influence sensitivity or driest year biomass. At almost half of our sites, the previous year's rainfall explained as much variation in biomass as current year precipitation. Overall, our distributed fertilization experiment detected co-limitation between nutrients and water governing grasslands, with biomass sensitivity to precipitation being limited by nutrient availability irrespective of site aridity and herbivory. Our findings refute the classical ideas that grassland plant performance is limited by the single most limiting resource at a site. This suggests that nutrient eutrophication will destabilize grassland ecosystems through increased sensitivity to precipitation variation.


56
The productivity of grassland ecosystems around the world is strongly driven by precipita-  a. Nutrient addition is expected to increase both biomass measured in the driest year (b d ), and the precipitation sensitivity S (proportional response of biomass to a change in rainfall), estimated from the graph of biomass vs. precipitation across time at a site. Black denotes the relationship in control plots, red in nutrient added plots. b. Across space, precipitation sensitivity is expected to decline from arid to mesic sites. Since sensitivity is already maximum at arid sites, nutrient addition is expected to have no effect on sensitivity at arid sites, but a strong effect at mesic sites.
driest year biomass) are expected to vary among sites, especially with relation to site aridity  However, precipitation-production patterns may depend on nutrient limitation. Most grass-81 lands are not only limited by water, but also by the supply of available soil nutrients; e.g.,

107
We quantified how precipitation sensitivity and driest year biomass were affected by 5-11 108 years of continuous nutrient addition in 44 grassland sites around the world. We also tested 109 whether effects of nutrients were altered by the gradient of aridity among sites and the simul-110 taneous experimental exclusion of vertebrate herbivores (at 36 sites). We specifically examined 111 the following predictions derived from the hypothesis that water is among multiple resources 112 that co-limit grassland productivity -   3. Driest year biomass will increase from arid to mesic sites, as the amount of water received 118 in the driest year is higher at mesic sites as compared to arid sites. 4. Nutrient addition will have no effect on driest year biomass at arid sites, yet will have    Table S2). We defined 162 aridity of a site as the log 2 ratio of mean GSP divided by mean growing season PET at a site.  We fit linear mixed effects models of the following form at each site: Peak biomass at year t (log 2 transformed) was the response variable. Predictors were GSP, 171 fencing, nutrient addition, as well as all interactions between the these three, allowing both slope SPEI is a normalized metric of water availability in a given year relative to the precipitation and 178 temperature history of the site. This metric is positive if the previous year was wetter than the 179 mean, and negative if it was drier than the mean. We also included a random effect for blocks 180 within sites, to correctly account for the design of our experiment. All analyses were performed 181 in R version 4.0.0 (R Core Team, 2020). 182 We measured precipitation sensitivity (S ) as the slope of the relationship between log 2 (Biomass) 183 and log 2 (GSP). A slope value of 1 means that biomass value doubles when precipitation dou-184 bles. S < 1 indicates that a change in precipitation results in a less than proportional change in 185 biomass, and S > 1 indicates a greater than proportional change. Fitted models and parameters 186 are shown in Appendix S1.

187
Driest year biomass (b d ) was directly estimated as the measured biomass in the driest year 188 during our experiment. We calculated the effects of treatments on this biomass as the log 2 ratio 189 of biomass in treatment plots over control plots in each block, during that driest year. 190 We also fit linear relationships to the data of biomass and precipitation (Huxman et al.,    Four regions in our study had more than 5 sites each, and were amenable to examination of  significant increase of precipitation sensitivity (median +0.08, V = 640, p = 0.09, Figure 3a).

233
Since our study is on natural temporal variation in precipitation at each site, we also checked 234 whether the observed effect of nutrient addition on driest year biomass depended on how extreme 235 the driest year was at each site. We found that the nutrient effect on b d was not significantly 236 associated with the SP EI g s value of that driest year (Appendix S1: Figure S4). year biomass at many sites. While the mean effect was not different from zero, there were 243 sites with both strongly positive and strongly negative legacy effects (Figure 3c, Appendix S1: 244 Figure S5). The proportion of total biomass variance explained by legacy effects ranged from 245 0% to 74% (median 14%, IQR = 5% to 33%). In 21 out of 44 sites (48%), previous year's water 246 availability explained more variance in biomass than current year growing season precipitation.

302
We found that sensitivity did not significantly change between arid and mesic sites, matching