Constraining Uncertainties in CMIP5 Projections of September Arctic Sea Ice Extent with Observations

In agreement with observations, Earth system models participating in phase 5 of the Coupled Model Intercomparison Project (CMIP5) simulate a decline in September Arctic sea ice extent (SIE) over the past decades. However, the spread in their twenty-first-century SIE projections is large and the timing of the first ice-free Arctic summer ranges from 2020 to beyond 2100. The uncertainties arise from three sources (internal variability, model uncertainty, and scenario uncertainty), which are quantified in this study for projections of SIE. The goal is to narrow uncertainties by applying multiple diagnostic ensemble regression (MDER). MDER links future projections of sea ice extent to processes relevant to its simulation under present-day conditions using data covering the past 40 years. With this method, we can reduce model uncertainty in projections of SIE for the period 2020-44 by 30%-50% (0.8-1.3 million km(2)). Compared to the unweighted multimodel mean, the MDER-weighted mean projects an about 20% smaller SIE and an earlier near-disappearance of Arctic sea ice by more than a decade for a high-greenhouse gas scenario. We also show that two different methods estimating internal variability in SIE differ by 1 million km(2). Regardless, the total uncertainties in the SIE projections remain large (up to 3.5 million km(2), with irreducible internal variability contributing 30%) so that a precise time estimate of an ice-free Arctic proves impossible. We conclude that unweighted CMIP5 multimodel-mean projections of Arctic SIE are too optimistic and mitigation strategies to reduce Arctic warming need to be intensified.