D5 Detection and Attribution

Topic

Detection and attribution is the methodology used to determine, first, whether the climate system has undergone a statistically significant change and, second, the causes of that change. Detection compares observed trends with the internal climate variability that would exist in the absence of external forcings. This variability is estimated using long-term control simulations in which no external changes are introduced. When the observed trend clearly exceeds the range attributable to this internal variability, the change is considered detected, and its cause can be investigated through attribution.

Attribution employs the optimal fingerprinting method, which compares observations with spatial patterns characteristic of the climate response to various external forcings, such as greenhouse gases, aerosols, solar activity, and volcanic activity. Observations are represented as a combination of these fingerprints, and the linear combination that best reproduces the recorded change is determined. In this way, the relative contribution of the different forcings to the observed change can be estimated.

The optimal detection method improves this comparison by seeking to maximize the signal-to-noise ratio. Both the observations and the fingerprints associated with the various forcings are projected onto combinations where internal climate variability is lower. This reduces interference from natural climate noise and makes it easier to distinguish the signals corresponding to different external factors.

Once attribution is obtained, a consistency test compares the unexplained portion of the observations with the internal variability estimated from control simulations. If these residuals are consistent with that variability, the attribution is considered statistically consistent. Furthermore, scaling factors allow for the assessment of the intensity of simulated signals relative to observed ones. A factor greater than one indicates that the model generates a weaker signal than observed—thereby underestimating it—while a factor less than one indicates that the simulated signal is too intense and the model overestimates it. This same logic can be applied to specific phenomena through extreme event attribution. Its objective is to quantify the extent to which anthropogenic climate change has altered the probability or intensity of a specific episode, such as a heatwave, flood, or drought. To do this, a factual world—corresponding to the current climate with human influence—is compared with a hypothetical counterfactual world in which such anthropogenic influence does not exist.

One of the metrics used is the attributable risk ratio, obtained by comparing the probability of exceeding the event's intensity in both worlds. It is calculated by subtracting the ratio of the event's probability in the counterfactual world to its probability in the factual world from one. The less likely the event would have been without human influence compared to its current probability, the closer this ratio comes to one, and the greater the fraction of its probability associated with anthropogenic climate change. Another common measure is the return period—the average interval between events of a certain magnitude—which can also be compared across factual and counterfactual scenarios.

Thus, extreme event attribution goes beyond merely stating that climate change can influence a phenomenon; it seeks to quantify how much its probability or magnitude has changed relative to a climate without human influence. Similarly, long-term climate attribution attempts to quantitatively separate the contributions of various external forcings to observed trends.

Together, detection and attribution provide a statistical framework for moving from the observation of climate changes to the quantitative identification of their causes. Detection establishes whether a signal exceeds natural internal variability; optimal fingerprints allow for the separation of contributions from greenhouse gases, aerosols, and solar and volcanic activity; consistency tests verify that residuals are compatible with internal variability; and scaling factors evaluate the correspondence between simulated and observed signals. Applied to extreme events, the same principle makes it possible to estimate how much of their probability or intensity is attributable to anthropogenic influence and how much could have occurred even in its absence.

ID:818