Ecosystems are experiencing changing global patterns of mean annual precipitation (MAP) and enrichment with multiple nutrients that potentially co-limit plant biomass production. In grasslands, mean aboveground plant biomass is closely related to MAP, but how this relationship changes after enrichment with multiple nutrients remains unclear. We hypothesized the global biomass-MAP relationship becomes steeper with an increasing number of added nutrients, with increases in steepness corresponding to the form of interaction among added nutrients and with increased mediation by changes in plant community diversity. We measured aboveground plant biomass production and species diversity in 71 grasslands on six continents representing the global span of grassland MAP, diversity, management, and soils. We fertilized all sites with nitrogen, phosphorus, and potassium with micronutrients in all combinations to identify which nutrients limited biomass at each site. As hypothesized, fertilizing with one, two, or three nutrients progressively steepened the global biomass-MAP relationship. The magnitude of the increase in steepness corresponded to whether sites were not limited by nitrogen or phosphorus, were limited by either one, or were co-limited by both in additive, or synergistic forms. Unexpectedly, we found little evidence for mediation of biomass–MAP relationships by plant community diversity because relationships of species richness, evenness and beta diversity to MAP and to biomass were weak or opposing. Site-level properties including baseline biomass production, soils, and management explained little variation in biomass-MAP relationships. These findings reveal multiple nutrient co-limitation as a defining feature of the global grassland biomass-MAP relationship.

Data were collected from 71 sites distributed across six continents.  Data were collected by individual site investigators using the standardized Nutrient Network Protocol. Investigators submit site data to the network manager who performs internal processing and quality control to build datasets.  For this dataset, processing and analysis were performed in SAS and a code package has been included. In summary, data were error-checked and gap-filled as described in the publication.