Constraints on future multidecadal temperature variability from climate models, reconstructions and observations of the past two million years

The magnitude and spatial patterns of future changes in temperature variability remain debated. Supplementing direct observations with reconstructions of past climate has shown that CMIP-style simulations lack regional variability on decadal and longer timescales, a shortcoming that likely includes future projections. Here, we assess the range of future climate variability based on the differences between reconstructions and simulations of temperature variability during the Quaternary. The assessment uses a multi-proxy database of surface temperatures as well as long-term transient simulations of the past and possible future climates with an Earth System Model. Comparing simulations with reconstructions, we establish a relationship between warming level and local to global temperature variability for annual to millennial timescales. The identified model-reconstruction mismatch provides the basis for rescaling simulations and thus constraining future climate variability. For this, we decompose variability into its long- and short-term components. We then artificially enhance the long-term variability that is underestimated in simulations to reconstruct a possible, more realistic corresponding temperature field. Taking the uncertainty in reconstructions into account results in a wider range of possible scenarios for future climate variability given this past evidence. Our results have implications for climate indices and temperature extremes on short timescales in future scenarios, informing mitigation and adaptation efforts.