Climate projections from phase 5 of the Coupled Model Intercomparison Project (CMIP5) ensemble show a decrease in interannual surface temperature variability over high latitudes with a large intermodel spread, in particular over the areas of sea ice retreat. Here relationships are found between the models’ present-day performance in sea ice–related metrics and future changes in temperature variability. These relations, so-called emergent constraints, can produce ensembles of models calibrated with present-day observations with a narrower spread across their members than across the full ensemble. The underlying assumption is that models in better agreement with observations or reanalyses in a carefully selected metric probably have a more realistic representation of local processes, and therefore are more reliable for projections. Thus, the reliability of this method depends on the availability of high-quality observations or reanalyses. This work represents a step toward formalization of the emergent constraints framework, as so far there is no consensus on how the constraints should be best implemented. The authors quantify the reduction in spread from emerging constraints for various metrics and their combinations, different emission scenarios, and seasons. Some of the general features of emerging constraints are discussed, and how to effectively aggregate information across metrics and seasons to achieve the largest reduction in model spread. It is demonstrated, based on the case of temperature variability, that a robust constraint can be obtained by combining relevant metrics across all seasons. Such a constraint results in a strongly reduced spread across model projections, which is consistent with a process understanding of variability changes due to sea ice retreat.

Model projections of regional changes in heavy rainfall are uncertain. On timescales of few decades, internal variability plays an important role and therefore poses a challenge to detect robust model response in heavy rainfall to rising temperatures. We use spatial aggregation to reduce the major role of internal variability and evaluate the heavy rainfall response to warming temperatures with observations. We show that in the regions with high rainfall intensity and for which gridded observations exist, most of the models underestimate the historical scaling of heavy rainfall and the land fraction with significant positive heavy rainfall scalings during the historical period. The historical behavior is correlated with the projected heavy rainfall intensification across models allowing to apply an observational constraint, i.e., to calibrate multimodel ensembles with observations in order to narrow the range of projections. The constraint suggests a substantially stronger intensification of future heavy rainfall than the multimodel mean.

Projected changes in temperature extremes, such as regional changes in the intensity and frequency of hot extremes, differ strongly across climate models. This study shows that this disagreement can be partly explained by discrepancies in the representation of the present-day temperature distribution, motivating the evaluation of models with observations. By evaluating climate models on carefully selected metrics, the models that are more likely to be reliable for long-term projections of temperature extremes are identified. The study found that frequencies of hot extremes are likely to increase at a higher rate than the multimodel mean estimate over large parts of the Northern Hemisphere and Australia. This implies that a higher degree of adaptation is required for a given global temperature target. It also found that projected changes in the intensity of hot extremes can be constrained in several regions, including Australia, central North America, and north Asia. In many other regions, large internal variability can often hamper model evaluation. For both aspects—the intensity and the frequency of hot extremes—the total area over which the constraints can be implemented is limited by the quality and completeness of observations. Thereby, this study highlights the importance of long-term, high-quality, and easily accessible observational records for model evaluation, which are vital to ultimately reduce uncertainties in projections of temperature extremes.

No abstract