Modelling the interactions of the ocean and atmosphere to predict the future climate of the entire Earth is one of the most breath-taking achievements of modern science. The sheer chutzpah of the endeavour is inspiring.
And much of the debate about the impact of climate change centres on the reliability of these ‘Climate models’. So I read with interest the review of climate models included in Chapter 9 of the 5th Assessment report of the state of the Intergovernmental Panel on Climate Change (IPCC) and enjoyed the presentation at the Royal Meteorological Society meeting last week.
I am not a specialist in this field, but I was impressed by the report, by the talk, and by just how good Climate Models are. The report draws on two ‘Coupled Model Inter-comparison Projects’: CMIP Phase 3 which covers 24 models and CMIP Phase 5 which covers 41 models.
Each model makes predictions for one possible evolution of Earth’s weather and its results are then averaged over time and region to yield Climate estimates.
Each model is fed data on the past state of the climate up until (say) 1900 and then calculations are made in roughly 15-minutes steps to see how the climate evolves as the Earth turns, the Sun shines, Volcanoes erupt, and carbon dioxide levels increase.
We then look back at our actual climate records and see how well each model performed. Of particular interest is the average performance of the models – which represents our collective ‘best estimate’ for what will happen.
What struck me most strongly is that the authors highlight where models get things wrong. This is such an unfashionable writing style one could easily get the sense that none of the climate models are ‘correct’. And of course none of them are perfect. But it is this obsession with error and uncertainty which is a hallmark of a community genuinely concerned with accuracy.
Actually, the models do pretty well. For me the most amazing graph was Figure 9.35 on Page 803. It shows the model’s predictions for the variability of the air temperature above the ocean surface in a particular region of the Pacific Ocean. Most models show a pattern of variability with peaks every 2 to 7 years – similar to the observed variability of El Nino events.
But ‘predicting’ the past is relatively easy because ‘bad’ models can be eliminated.
What about predicting the future? Can we say how reliably the models will predict the future? The authors summarise the state of the art thus (Page 745)
In general, there is no direct means of translating quantitative measures of past performance into confident statements about fidelity of future climate projections.
There has been substantial progress since the AR4 [the 4th Assessment Report in 2007] in the methodology to assess the reliability of a multi-model ensemble, and various approaches to improve the precision of multi-model projections are being explored. However, there is still no universal strategy for weighting the projections from different models based on their historical performance.
The models represent our very best attempt to consider all the physical factors of which we are aware, and to work out what is going to happen. Using multiple models and looking at the extent of agreement and disagreement between them is one way of assessing the likely accuracy of the model predictions.
But the long and the short of this is that ‘we just don’t know’ what will happen in the future.
However this shouldn’t diminish the achievements of understanding that these models embody, even if they prove inaccurate in some predictions. Similarly, it would also be unwise to believe them absolutely, even if they prove accurate.
We are talking about the future, and we need to remind ourselves of this. The results of climate forecasts can guide us, and it would be bonkers to ignore their guidance. But the real challenge is to make policy choices now in the face of the real uncertainty.
Tags: Climate Models