Optimal Determination of Time-Varying Climate Change Signals

Abstract

Abstract This paper proposes an optimal method for estimating time-dependent climate change signals from general circulation models. The basic idea is to identify vectors that maximize the mean-square detection statistic derived from optimal fingerprinting techniques. The method also provides an objective and systematic procedure for identifying the limit to which a signal can be restricted in space and time without losing detectability. As an illustration, the method is applied to the Coupled Model Intercomparison Project, phase 3 multimodel dataset to determine the continental seasonal-mean anomaly in surface air temperature and precipitation that is most detectable, on average, in these models. Anomalies in seasonal-mean surface air temperature are detectable in all seasons by almost all models on all continents but Europe; seasonal-mean anomalies over Europe are undetectable for some models, though this does not preclude other expressions of the signal, such as those that include longer time averages or time-lag information, from being detectable. Detectability in seasonal-mean temperature is found not only for multidecadal warming trends but also for cooling after major volcanic eruptions. In contrast, seasonal-mean precipitation anomalies are detectable in only a few models for averages over 5 yr or more, suggesting that the signal should include more spatiotemporal detail to be detectable across more models. Nevertheless, of the precipitation anomalies that are detectable, the signal appears to be of two characters: a systematic trend and enhanced frequency of extreme values. These results derived from twentieth-century simulations appear to be consistent with previous studies based on twenty-first-century simulations with larger signal-to-noise ratios.