1.        Understanding the biodiversity–productivity relationship and underlying mechanisms in natural ecosystems under realistic diversity loss scenarios remains a major challenge for ecologists despite its importance for predicting impacts of rapid loss of biodiversity worldwide. Here we report the results of a plant functional group (PFG) removal experiment conducted on the Mongolian Plateau, the largest remaining natural grassland in the world.

2.        Our results demonstrated that the biodiversity–productivity relationship varied among positive linear, neutral and unimodal forms under different PFG loss patterns. Moreover, the form of this relationship with the same PFG loss pattern sometimes changed through time.

3.        The abundance of the remaining PFG(s) before removal and their compensation following the loss of other PFGs were two major mechanisms affecting the biodiversity–productivity relationship under diversity loss scenarios. The abundance effect promoted positive responses of productivity to biodiversity, but the compensation effect caused several biodiversity–productivity relationships, hinging on its direction (positive or negative) and strength. As indicated by the values of the compensation index, negative, zero and partial compensations contributed to the positive relationships, while full compensation resulted in a neutral relationship. Overcompensation at intermediate PFG richness levels created a unimodal curve in our system, but it could also lead to a negative linear relationship.

4.        Synthesis. Our experiment provides a vivid picture of how the form of the biodiversity–productivity relationship varies among different diversity loss patterns in a natural ecosystem. We argue that compensation by the remaining species, which is not revealed by synthesized biodiversity experiments, plays a critical role in shaping the form of this relationship when diversity is lost from existing systems. The direction and strength of compensation are highly dependent on extirpation scenarios. Thus, impacts of biodiversity loss on natural ecosystems are likely more complex than predicted by the canonical positive saturating curve obtained from the synthesized biodiversity experiments. We suggest that models forecasting the consequences of biodiversity declines on natural ecosystems should take into account diversity loss patterns and the ensuing compensation.