REE5: Uncertainty analysis of regional and global climate forecasts

Project completed July 15, 2012

Background

At the centre of the climate change debate lies the question of whether global warming can be detected and if so, whether it is attributable to humans. A technique called optimal fingerprinting is a powerful method of detection and attribution of climate change. The method relies on the assumption that external forcings due to human activity (for example, changing levels of greenhouse gases and aerosol loading) and due to natural causes (for example, volcanic activity and variations in solar radiation) are superimposed on internal variability due to interaction of the atmosphere, ocean, biosphere and cryosphere. 

The Intergovernmental Panel on Climate Change (IPCC) said “most of the observed increase in global average temperatures since the mid-20th century is very likely due to the observed increase in anthropogenic greenhouse gas concentrations”. Since methods like optimal fingerprinting are at the core of these statements, the methods must be robust. Researchers at the Oxford Centre for Collaborative Applied Mathematics (OCCAM) have been assessing the robustness using the extended historical record, the present knowledge of climate dynamics and its interaction with human activity. 

Techniques and Challenges

The researchers explored the robustness of optimal fingerprinting by analysing the influence of adopting different descriptions of internal variability when applying the detection algorithm. Optimal fingerprinting is a linear regression model, which allows the decomposition of the observed global mean temperature into signals representing the external forcings, plus a noise term corresponding to the internal variability of the system. To test the effect of the representation of the internal variability, it was represented by one of two very different stochastic models: either a short-memory AR(1) process or a long-memory fractional differencing (FD) noise process. An energy balance model was used to simulate the contributions to the global mean temperature from four forcing signals: solar, volcanic, greenhouse gas and sulfates. 

Furthermore, the researchers extended the analysis to determine whether the robustness of the anthropogenic climate change detection statements is altered when internal oscillations of the climate system such as the El Niño Southern oscillation (ENSO) or the Atlantic Multidecadal Oscillation (AMO) are considered as separate forcings, and not simply included in the noise term in the multi-regression analysis. Similarly, they considered natural oscillations in climate with periods of 20 and 60 years.

Results

The researchers found that, independent of the representation chosen for the internal variability, the greenhouse gas signal remains statistically significant under the detection model, and the attribution of greenhouse gas to climate change was found to be insensitive to the choice between the two representations of internal variability. However, the significance of detection results were affected by the choice of a short-memory versus a long-memory process, indicating a need for more sophisticated statistical checks on the spectral properties of internal variability.

It was also shown that key detection and attribution statements remain robust with the inclusion of additional forcing signals, including the 20 and 60 year oscillations. This gives further evidence that the detection of the anthropogenic signal is statistically robust independent of the model used to characterise the internal variability in the case of the global mean temperature. 

The Future

This work on the global mean temperature is currently being extended to the full spatial and temporal resolution of the temperature in the climate system. Also, the researchers are doing a systematic comparison of simulated climate variability and observational data within the formalism of Linear Inverse Modeling (LIM). This method allows for a sensitive comparison of models and data by contrasting their spatial and temporal spectra.

In particular, the aim is to use control runs of CMIP-5 (Coupled Model Intercomparison Project Phase 5) and observational records pre-1970 to assess the adequacy of internal climate variability in the CMIP-5 models. LIM has proven a powerful method for extracting the intrinsic linear dynamics that govern the climatology of a complex system directly from observations of the system or climate models. Thus, applying the LIM to the output of control runs of IPCC coupled models will help to obtain insights into the structural differences between simulated and real climate variability. It will also allow for the opportunity to generate estimations of more accurate internal variability for use in the detection and attribution formalism. 

References

Allen M.R., Stott P.A.: Clim. Dym. 21, 477-491, 2003

Allen M.R., Tett S.F.B.: Clim. Dyn. 15, 419-434, 1999