Abstract
While paleoclimate records show that the Earth System is characterized by several different tipping points, their representation within Earth System models (ESMs) remains poorly constrained. This is because historical observations do not encompass variations large enough to provoke such regime changes, and paleoclimate conditions are rarely used to help develop and tune ESMs, which potentially ignores a rich source of information on abrupt climate change. A critical example is the early to mid-Holocene "greening" and subsequent rapid desertification of the Sahara, which most ESMs fail to reproduce, casting doubt on the representation of land-atmosphere coupling and monsoon dynamics. Here, we show that this greening and abrupt termination can be successfully simulated with one ESM after optimizing uncertain model components using both present-day observations and crucially mid-Holocene (6,000 y before present) reconstructions. The optimized model displays abrupt threshold behavior, which shows excellent agreement with long paleoclimate records that were not used in the original optimization. These results suggest that in order to realistically capture climate-system thresholds, ESMs first need to be conditioned with appropriate paleoclimate information.