Friday, August 26, 2011

Reliability of the past

We wrote a sister to the  previous papers on the assessment of climate models using present day climate data. In this latest work we look at the past, the Last Glacial Maximum, when the earth was a much icier and drier place. We all love the Last Glacial Maximum because it was the most recent time in the past when the carbon dioxide level was very different to the modern climate. Thus we hope that studying it closely will reveal new insights about climate change. For this reason, a lot of modelling and data collecting effort has focused on it. In our paper we conclude that the data from the oceans and the models are surprisingly consistent with each other. It is surprising because when you look at a global map showing the data, and another map showing any one climate model output, you would tend to think that there must be some big problems. The models and data seem quite different in some parts of the globe, and this worries everyone. The model-people worry that they are missing important processes in their models, and the data-people similarly worry that they have overlooked something in the conversion of sediment core to temperature. But just looking at a spatial plot of the data next to one from your favourite climate model doesn't give the whole picture, because there is a lot of variation between different climate models and great uncertainty in the data. When comparing the models and data it is important to take the uncertainties into account, and when you do so, you find that the model ensemble as a whole is consistent with the data. So, climate models that we previously found to be reliable for the present day,  now seem to also be reliable for the past. This can only increase our confidence that they may also be performing well for the future!

But it's not all good news. The data we compared the models to are very uncertain, and the models span quite a wide range of results. Thus the concern perhaps now ought to be not that the models and data are inconsistent, but that the data may not be accurate enough to actually add anything new to what we already know about that climate system as represented by the models. What people want to do is use the information we have about the past to reduce uncertainty about the future, but if the data from the past are just a vague smudge roughly consistent with the models then, while this makes us happy that the models are doing OK, it doesn't actually help us improve them any further.  One strategy would be to focus effort on those regions of the globe and times in the past where the whole ensemble is in robust disagreement with the data, although before doing so, it would also be wise to consider whether the causes of these disagreements may be orthogonal to the causes of anthropogenic climate change. 

The paper is Open Access, so you can find all the typos online at Climate of the Past. 

Of course there is more to be done on this topic. We only looked at the ocean surface temperatures for one snap-shot in the past, and the ways the uncertainties in the data are estimated could (I think) be improved. And of course, now that our beloved Last Glacial Maximum has hit the big time by being included in the official runs for the next proper climate ensemble project (CMIP5 - due to star in the next IPCC report) there should soon be many more models to play with, which will enable a more robust result to be obtained. 



8 comments:

C W Magee said...

I thought that one of the main reason that people focused on the LGM was that, at least for the terrestrial record, finding pre-LGM data in areas that got bulldozed by ice is problematic. Got a reliable 50ka ice margin?

David B. Benson said...

Maybe try to encorporate vegetation type?

LGM vegetation map:
http://www.ncdc.noaa.gov/paleo/pubs/ray2001/ray_adams_2001.pdf

James Annan said...

The ocean data is easiest to interpret for starters, though land data are also under examination.

EliRabett said...

There was this great talk Drew Shindell was giving about a decade ago, something along the lines of pre-1900, what should we trust the data or the models. The answer was neither, but both together were a bit better

Steve Bloom said...

I assume this is tied in with the CMIP5 effort Gavin just blogged on.

Have any of the modeling groups failed to participate?

James Annan said...

This is all CMIP3 stuff, there isn't any CMIP5 output yet! But we plan to repeat the exercise as and when...

Steve Bloom said...

This seems as good a place as any to note this amazing exchange between RP Sr. and n-g. In a couple of particulars, the former seems to have gone a good bit farther around the bend than previously, although maybe it's old news and I just haven't been paying enough attention.

James Annan said...

Thanks, someone else linked to that on another post...