Detection and Attribution on Climate Change

Detection and attribution of anthropogenic climate change is a statistical signal-in-noise problem which is produced by the presence of our natural climate variability. Detection is a word commonly used as a reference to identifying significant changes in climate such as an upward trend in global-mean temperature. The changes must be significantly different and that can be explained by natural internal variability. However, detection does not necessarily give a definite explanation for the cause of change, we must also include the attribution. For example, when looking at enhanced greenhouse effect – after statistical methods suggesting that a change has occurred – we must attribute at least part of such a change to the enhanced greenhouse effect.

There are two components for natural climate variability: internal and external. Internal components interact within the coupled atmosphere-ocean-ice-land-biosphere system (IPCC, 1995). Whereas, external components are triggered by natural changes in the Sun’s output or in the volcanic aerosol loading of the atmosphere. This tells us that due to the natural internal and external processes, the climate is always changing even without human interaction. Only when the changes are relatively unusual to the predicted results from natural variability, do we state those changes in climate are significant.

Detecting a significant change is considered as a statistical problem in terms of nature. By using the standard approach, scientists are able to negate that the observed change in climate can be explained by natural variability, this more commonly referred to as the statistical “null” hypothesis. If the null hypothesis is rejected, it implies that a detection of a climate change at a specifically significant level has occurred. However, this does not identify the cause of the change, therefore, we must apply attribution. To understand the “cause and effect”, an investigation must be conducted which involves a series of experiments. The investigation also involves using the real climate system which systematically studies different causes. The experimentation process comprises of the following:

• No systematic method

• Varying numerous causes simultaneously e.g. changing land surface properties, concentrations of atmospheric greenhouse gases and anthropogenic aerosols

The purpose of using the different methods mentioned above instead of changing an individual cause is that it’s more efficient, by eliminating the need to observe climate responses, and then vary the next cause.

In addition, the use of numerical models is needed for the experiment. To determine the signals for different hypothesized causes, a comparison between model simulations and the observed changes is required. The detection of a significant climate change is referred to as a “unique attribution” which involves human activities and both the consideration and elimination of plausible non-anthropogenic mechanisms. Though, in a statistical sense, this does not mean results will be certain. The reason for this uncertainty is that it’s a difficult task to define all probable “natural” climate change signals. Numerical models cannot be ruled out completely, even though it can only state whether available observations are consistent or inconsistent. This is due to its claim on statistical detection of an anthropogenic signal. Furthermore, according to the Intergovernmental Panel on Climate Change, 1995 there is a distinction between accomplishing “practically meaningful” and “statistically unambiguous” attribution. This distinction is based on different perceptions of risk between policymakers and scientists.

To summarise, in terms of statistics, detection of change is the process of demonstrating that an observed change in climate is highly unusual without providing a reason for the change. On the other hand, attribution processes can establish the cause and effect which provide us with reasons for the observed change in climate. Both cannot give a simple yes or no answer. Also, the results produced from successfully conducting detection processes will not be a discrete value, it will be given in a range instead.