By David Holland
Global climate models are used in the Independent Panel on Climate Change (IPCC) Assessment Report Four (AR4) and Assessment Report Five (AR5) to predict future climates.
How have the modellers resolved the uncertainties of climate change predictions?
This article is based on study related to Masters of Environmental Management (Natural Resources) undertaken by David Holland 2016
When entering the world of prediction we are looking into a crystal ball with many possibilities. With Climate change predictions, we may know the past, howbeit in less detail than would be desired, but the future is simply a guessing game. Satellite technologies have produced data since the year 2000 with increased accuracy which has increased the hindsight data available to both AR4 and AR5. Increasingly data is becoming more refined and reflective of what is actually happening on the ground.
Wigley and Raper (2001) as cited in Meehl Gerald A. (USA), Stocker Thomas F. (Switzerland), (2007) as part of IPCC AR4, states the main uncertainties are uncertainties in emissions, the climate sensitivity, the carbon cycle, ocean mixing and aerosol forcing.
But uncertainty in the future is about the best guess based on past experience. We do not know how much meetings like the Paris accord will change the governments of the world to react to the climate issues or how quickly they will react and as a result we simply do not know the volume of future GHG emission into the future.
Volcanologists can predict certain volcanic events but the prediction of where, and how big a volcanism events may be, and then how long an aerosol event may last is less certain. All we can say is that there is a likelihood of future volcanism.
We understand that the sun has a11-year cycles between sun spot activity by looking into the past but in the future things may change. As unlikely as it seems solar forcing could change.
But the most sensitive and possibly most uncertain is radiative forcing changes. This relates to the potential for changes in the concentrations of GHG’s in the atmosphere and the resultant heat retained in the atmosphere from solar radiation. There is a range of variables associated with this process. The feed back loop related to CO2 atmospheric /oceanic flux, the albedo effect reduction as ice caps melt and more ocean is exposed and how the ocean and atmospheric circulations will be affected by all this.
The first coupled models started their life in 1995 by the Climate Variability and Predictability Numerical Experimentation Group, which came out of the reconstituted World Climate Research Programme. They were call “Coupled Model Intercomparison Projects (CMIP)”. (Gerald A. Meehi, Curt Covey, Bryant McAvaney, Mojib Latif, and Ronald J. Stouffer, (Jan 2005) )
Coupled models are more advanced models, which incorporate complex software interactions of data relationships to produce output that mimics a natural system.
They are defined as a complex interaction of the various software components in the model. This interaction produces results that could be skewed by the addition of a spurious variables or a factor in the maths that may be erroneous. So inherently within the model there are at least two uncertainties, the weighting of the variables and the models complexity not fully understood as it attempts to mimic real natural systems.
As time went on several versions of this model emerged and with a variety of data sets being used to run on these models. CMIP3 was one of the better early models but it, as all the models had inaccuracies.
The various data sets from recorded data would produce a range of results from the coupled models and as a result any output from the model would have uncertainty as to which results could be considered correct if at all any were correct.
Land use changes also have an impact of the future accuracy of a model. Land use change can change the dynamics of the complex interactions of GHGs, flux, radiative forcing within a system. If the model does not have this information then the change will not be reflected in the model output.
The CMIP5 model was able to used much less grainy data sets, which enabled CMIP5 to produce regional climate models (RCMs). But as these were on smaller scales some anomalies were observed on the edges of the regions that did not seem to match an adjacent regions boundary. As a result questions were raised as to what uncertainties needed to be addressed to correct these aberrations.
Clearly climate modelling is peppered with uncertainties, but the argument is that with better and more extensive data sets and the ground truthing of existing models, better models will be made in an attempt to reduce internal anomalies. But the fact remains that modelling is still attempting to predict an uncertain future.
Unfortunately there are a variety of data sets available to feed into the models and a range of models.
The next generation of models were used in the IPCC’ assessment report 4. These multi-model means were starting to be used because the various coupled models seemed to give both accurate and inaccurate correlations to the real natural system as recorded in the past. So if the model produced an accurate simulation on past data then it was reasoned likely that future predictions on simple climate model trends data would produce an accurate coupled data result for the future.
The fact was that coupled data results varied considerably using differed data sets and climate model versions. So it seemed to be logical that if the result were averaged, the results of the 5% to 95% results, (which gets rid of the eccentric data results), we will get from a lot of uncertain results a more certain result. This is an understanding of what a multi-model mean is. A mean of many results of a range of coupled models produced from a range of data sets and a range of assumptions of the future.
The interesting thing about this method is that each model has been set up differently with a range of parameters, some with higher GHG emissions for a future scenario, and some with lower emissions. The end result would be that if the majority of the uncertain future predictions now placed in the models were inaccurate, then the averaging out of the results of all the models would predict a wrong future for the earth.
The method starts with uncertainty as if it was a sows’ ear and suggests that it can make a silk purse by averaging the sows’ ears.
Maybe the analogy is too harsh. It is about the opinions of the model managers who input into the model their best guess of the future. If the manager feels that there will be a reduction of GHGs by a certain date and the majority of model managers believe that this will be the case then the mean of the models will trend that way.
So where does this leave us in predicting the future climate? It leaves us with a best guess solution based on the past’s data collection.
The way the IPCC have handled the uncertainty is by creating several scenarios of the future. These scenarios are based again on varies social and environmental predictions.
However in reality, the prediction that recent data has followed is in fact the highest or least safe prediction for the potential to return the climate to a normal state.
References:
MEEHL Gerald A. , COVEY Curt , MCAVANEY Bryant , LATIF Mojib , AND STOUFFER Ronald J. ,(Jan. 2005) Overview Of The Coupled Model Intercomparison Project, American meteorological society, meeting summaries, https://www.gfdl.noaa.gov/bibliography/related_files/gam0501.pdf, cited September 2016.
Meehl Gerald A. (USA), Stocker Thomas F. (Switzerland), (2007), Global Climate Projections Coordinating Lead, IPCC assessment report 4, http://www.ipcc.ch/pdf/assessment-report/ar4/wg1/ar4-wg1-chapter10.pdf, accessed September 2016.
REICHLER THOMAS , KIM JUNSU, (March 2008)How Well Do Coupled Models Simulate Today’s Climate?, In Box – Insights and Innovations, , AMERICAN METEOROLOGICAL SOCIETY, Publ. NOAA, http://www.nssl.noaa.gov/users/brooks/public_html/feda/papers/ReichlerKimBAMS08.pdf
