1. Understanding ecosystem responses to plant species loss is essential for the optimal management of grasslands. Recent studies have examined the effects of simulated random species loss in experimental plant communities but not those of realistic non-random species loss resulting from transient extinction pressures in semi-natural grasslands.
2. To investigate the potential effects of non-random species loss on grassland productivity, we established mesocosms with mixed communities comprising 15 plant species, and exposed them to 2 years of high-intensity management (an extinction phase) followed by 2 years of low-intensity management (a restoration phase) allowing recolonization from differentially managed neighbouring plots. In addition, monocultures of each component species were subject to the same extinction–restoration phases.
3. During the extinction phase, species with high monoculture biomass had lower extinction probabilities in the mixed community than species with low monoculture biomass, but there was also species-specific variation. The species that were most productive or most persistent during the extinction phase were not the same as those performing best in the restoration phase.
4. No consistent effects of spontaneous recolonization from neighbouring communities on species richness or productivity of the focal communities were observed during the restoration phase.
5. We estimated that extinction of all but the species with the lowest extinction risk reduced biomass productivity by 42–49%; loss of all but the four species with the lowest extinction risk reduced it by 2–35%. Identical calculations for a random extinction scenario yielded reductions of 52% and 26–54%, respectively.
6. Synthesis and applications. Prediction of the effects of species loss on plant production and on other aspects of ecosystem functioning in semi-natural grasslands must account for both specific non-random extinction processes and post-extinction conditions. For European mesic grasslands experiencing a shift from high-intensity to low-intensity management, our results suggest that recolonization by 'missing' species must be actively assisted if high production is a management objective.