1. Large-scale studies have demonstrated that nitrogen (N) and water (W) availability greatly affect terrestrial ecosystems worldwide, and this is especially true for the resource-poor semi-arid grasslands. Yet, experimental evidence is lacking for how N and W availability affect soil food webs across historical grazing intensity-altered environments at a local scale.

2. Here, we included N- and W-addition treatments in an 8-year grazing experiment (with four grazing intensities) to determine how the legacy effects of grazing intensity mediate the responses of key components of soil food webs (plants, microorganisms, and nematodes) to resource addition in a semi-arid grassland.

3. After 4 years of N- and W-addition treatments (with no grazing during that 4-year period), we found that a legacy of grazing, even light grazing, had significant negative effects on the components of plant community and soil food webs. Both N and W addition increased above- and below-ground plant biomass, especially under moderate and heavy grazing. N addition had negative effects on the biomass of bacteria under no grazing, while W addition increased the biomass of actinomycetes under light grazing. N addition decreased the abundance of omnivorous + carnivorous nematodes under light and heavy grazing, while W addition increased their abundance under heavy grazing. Overall, the effects of resource addition on soil food webs progressively decreased from the lowest trophic level (primary producers, i.e., plants), to intermediate tropic levels (microorganisms and root-feeding nematodes), to higher trophic levels (microbial-feeding nematodes and omnivorous + carnivorous nematodes).

4. Synthesis and applications. Our results, which are the first data concerning the effects of resource addition on key components of soil food webs across a historical grazing-induced environmental gradient, show that the strong bottom-up controls of resource addition on soil food webs are mediated by the legacy of grazing intensity. These finding should be useful for predicting the responses of grassland ecosystems to future climate change and suggest that the recovery of degraded grasslands will require more than restoration measure of resource inputs alone.