Detection is the process of determining whether a climate signal has emerged from the background noise of the data. Typically this "noise" constitutes the natural climate variability of the atmosphere-ocean system., particularly variability on decadal time scales which can often be aliased onto longer-time scales trends such as those associated with global warming, thus making these trends difficult to detect unambiguously. A recent summary of the detection process is provided in Hegerl et al. (2007). Detection is largely a statistical issue and is usually determined by statistical techniques, ranging from simple trend analyses to multi-variate analysis.
For a signal to be detected unambiguously, good data for both the signal and the noise must exist. Like all good climate data, cyclone data must be free of inhomo-geneities caused by changes in observing practices. The data must also be complete in that the data sample being analysed must be consistently collected at similar time intervals over the entire period of record. To estimate the magnitude of the climate noise in a particular climate parameter, it may not be possible to just use the available observational record, as this may be too short to fully characterize the long-term variability due to noise alone. In principle, data records much longer than the duration of the climate change signal are required to estimate the range of long-term variability that may occur due to climate noise alone (Santer et al., 1995). In practice, such lengthy observational records do not exist for any climate variables and so alternative approaches must be used to estimate the long term variability due to noise. Often, long control simulations from coupled ocean-atmosphere climate models performed with no changes in external forcing factors are used to estimate the long-term variability of climate variables due to climate noise alone, assuming that the models provide realistic simulations of the noise in such variables.
It is clear that the mere detection of a signal is not an indication of its cause. Further analysis needs to be undertaken to ascribe causes for any detected signals; this is known as the process of attribution. In the case of anthropogenic climate change, we are interested in whether the detected signal can be attributed to man-made global warming. As defined by the IPCC, in order for high confidence attribution conclusions to be reached, a signal needs to be detected that is not only of the expected pattern of change but also of the correct magnitude expected from the response to anthropogenic climate forcing. This response is usually estimated using simulations with climate models forced by increases in atmospheric concentrations of greenhouse gases and aerosols, although theoretically-based approaches have also been used in some instances. Inherent in this process is the assumption that the simulations of climate models have reasonable skill, an assumption that is not justified at present for some small-scale, complex phenomena such as tropical cyclones (e.g. Randall et al. 2007; Walsh 2008).
Therefore, based on the formal Intergovernmental Panel on Climate Change (IPCC) definition of detection and attribution, the following conditions must be satisfied for successful detection and attribution:
• A signal must be detected;
• The signal must be consistent with the esimated response from modeling or theoretical techniques of a given combination of anthropogenic and natural forcings; and
• The detected signal must be inconsistent with alternative, plausible explanations that exclude important elements of the given combination of proposed forcings.
The last point is particularly important in that it provides a means of eliminating alternative explanations to a signal that might otherwise appear completely consistent with anthropogenic warming.
Attribution can also be achieved, but with considerably less confidence, by using statistical techniques to relate well-attributed variables to other climate phenomena. In this case, the confidence of the attribution would depend upon the plausibility of the hypothesized physical association between the variables. The current controversy regarding the influence to date of anthropogenic warming on tropical cyclones arises from this lower level of confidence.
Confident attribution of an anthropogenic effect depends on the likely magnitude of the anthropogenic effect as well as on the data and model simulations and theoretical understanding available for testing. Thus, if over the next few years a series of strong tropical cyclones were observed in the South Atlantic (a region where such events have been exceedingly rare in the past), we would be justified in concluding, with little formal studies, that this was likely the result of anthropogenic climate change. Similarly, if tropical cyclones off the east coast of Australia regularly started to retain their tropical characteristics as far away from the equator as, say, Sydney, we would again be justified in concluding that this was the result of anthropogenic changes. However, we do not expect such massive changes any time soon. So the question of attribution, given the expected degree of change from anthropogenic causes, is more difficult. This means that it is essential that we contrast the various formal and less-formal approaches to detection and attribution, so that we present balanced expressions of our confidence in any attribution statement. Quite simply, there are approaches than can yield strong statements about attribution, and others that can only yield weaker statements of confidence, given the tools and data available and the expected degree of change due to anthropogenic causes.
Was this article helpful?