Detecting Greenhouse-Gas-Induced Climate Change with an Optimal Fingerprint Method

A strategy using statistically optimal fingerprints to detect anthropogenic climate change is outlined and applied to near-surface temperature trends. The components of this strategy include observations, information about natural climate variability, and a “guess pattern” representing the expected time–space pattern of anthropogenic climate change. The expected anthropogenic climate change is identified through projection of the observations onto an appropriate optimal fingerprint, yielding a scalar-detection variable. The statistically optimal fingerprint is obtained by weighting the components of the guess pattern (truncated to some small-dimensional space) toward low-noise directions. The null hypothesis that the observed climate change is part of natural climate variability is then tested. This strategy is applied to detecting a greenhouse-gas-induced climate change in the spatial pattern of near-surface temperature trends defined for time intervals of 15–30 years. The expected pattern of climate change is derived from a transient simulation with a coupled ocean-atmosphere general circulation model. Global gridded near-surface temperature observations are used to represent the observed climate change. Information on the natural variability needed to establish the statistics of the detection variable is extracted from long control simulations of coupled ocean-atmosphere models and, additionally, from the observations themselves (from which an estimated greenhouse warming signal has been removed). While the model control simulations contain only variability caused by the internal dynamics of the atmosphere-ocean system, the observations additionally contain the response to various external forcings (e.g., volcanic eruptions, changes in solar radiation, and residual anthropogenic forcing). The resulting estimate of climate noise has large uncertainties but is qualitatively the best the authors can presently offer. The null hypothesis that the latest observed 20-yr and 30-yr trend of near-surface temperature (ending in 1994) is part of natural variability is rejected with a risk of less than 2.5% to 5% (the 5% level is derived from the variability of one model control simulation dominated by a questionable extreme event). In other words, the probability that the warming is due to our estimated natural variability is less than 2.5% to 5%. The increase in the signal-to-noise ratio by optimization of the fingerprint is of the order of 10%–30% in most cases. The predicted signals are dominated by the global mean component; the pattern correlation excluding the global mean is positive but not very high. Both the evolution of the detection variable and also the pattern correlation results are consistent with the model prediction for greenhouse-gas-induced climate change. However, in order to attribute the observed warming uniquely to anthropogenic greenhouse gas forcing, more information on the climate's response to other forcing mechanisms (e.g., changes in solar radiation, volcanic, or anthropogenic sulfate aerosols) and their interaction is needed. It is concluded that a statistically significant externally induced warming has been observed, but our caveat that the estimate of the internal climate variability is still uncertain is emphasized.