Global models predict that the Amazon rainforest will act as a carbon sink due to rising atmospheric carbon dioxide concentration into the future. At the same time, soil phosphorus impoverishment in most of the Amazon basin is assumed to limit productivity and growth. Despite that, phosphorus feedbacks are not accounted for by global ecosystem model ensembles to date, e.g. during the Fifth Assessment Report of the United Nations Intergovernmental Panel on Climate Change. To address this, we simulated the planned free air carbon dioxide enrichment experiment in the Amazon (AmazonFACE) with an ecosystem model ensemble (+200ppm; n=14). We show that phosphorus limitation on productivity reduces the carbon dioxide induced biomass carbon sink by 50-58% over 15 years compared to carbon-only and nitrogen-enabled models.

Phosphorus feedbacks prevented the strong increase in biomass carbon sink capacity, but the phosphorus-enabled models simulated considerable variation, ranging from <0.01 to >0.10 kg C m^-2 yr^-1. The model predictions diverge due to different assumptions in regards to flexibility in plant phosphorus use and acquisition strategies that allow carbon dioxide induced biomass growth despite low phosphorus availability to plants. We provide a mechanistic view on the underlying reasons for diverse tropical plant growth responses to phosphorus availability. Our model-based hypotheses enforce the need to alleviate the uncertainties around phosphorus feedback, during the AmazonFACE experiment and beyond.