The IPCC sea level numbers

The first is the treatment of potential rapid changes in ice flow (item 4 on the list above). The AR4 notes that the ice sheets have been losing mass recently (the analysis period is 1993-2003). Greenland has contributed +0.14 to +0.28 mm/year of sea level rise over this period, while for Antarctica the uncertainty range is -0.14 to +0.55 mm/year. It is noted that the mass loss of Antarctica is mostly or entirely due to recent changes in ice flow. The question then is: how much will this process contribute to future sea level rise? The honest answer is: we don’t know. As the SPM states, by the year 2095 it could be 10 cm. Or 20 cm. Or more. Or less.

The IPCC included one guess into the “model-based range” provided in the table: it took half of the Greenland mass loss and the whole Antarctic mass loss for 1993-2003, and assumed this would remain constant ever after until 2100. This assumption in my view has no scientific basis, as the ice-flow is almost certainly highly variable in time. The report itself states that this ice loss is due to a recent acceleration of flow, and that in 2005 it was already higher, and that in future the numbers could be several times higher – or they could be lower. Adding such an ill-founded number into the “model-based” range degrades the much more reliable estimates for thermal expansion, mountain glaciers and mass balance. Even worse: to numbers with error estimates, it adds a number without proper error estimate (the observational uncertainty for 1993-2003 is included, but who would claim this is an error estimation for future ice flow changes?). And then it presents only the combined error margins – you will notice that no central estimate is provided in the above table. If I had presented this as an error calculation in a first-semester physics assignment, I doubt I would have gotten away with it. The German delegation in Paris (of which I was a member) therefore suggested taking this ice-flow estimate out of the tabulated range. The numbers would have become slightly lower, but this approach would not have mixed up very different levels of uncertainty, and it would have been clear what is included in the table and what is not (namely ice flow changes), rather than attempting to partially include ice flow changes. The ice flow changes could have been discussed in the text – stating there that at the 1993-2003 rate, this term would contribute 3 cm by 2095, but it is bound to change and could turn out to be 10 cm or 20 cm or more. However, we found no support for this proposal, which would not have changed the science in any way but improved the clarity of presentation.

As it is now, because of the complex and opaque way of combining the errors, even I could not tell you by how much the upper limit of 59 cm would be reduced if the questionable ice flow estimate was taken out, and one of the reasons provided by the IPCC authors for not adopting our proposal was that the numbers could not be calculated quickly.

A second problem with the above range is that the models used to derive this projection significantly underestimate past sea level rise. We tried in vain to get this mentioned in the SPM, so you have to go to the main report to find this information. The AR4 states that for the period 1961-2003, the models on average give a rise of 1.2 mm/year, while the data show 1.8 mm/year, i.e. a 50% faster rise. This is despite using observed ice sheet mass loss (0.19 mm/year) in the “modelled” number in this comparison, otherwise the discrepancy would be even larger – the ice sheet models predict that the ice sheets gain mass due to global warming. The comparison looks somewhat better for the period 1993-2003, where the “models” give a rise of 2.6 mm/year while the data give 3.1 mm/year. But again the “models” estimate includes an observed ice sheet mass loss term of 0.41 mm/year whereas ice sheet models give a mass gain of 0.1 mm/year for this period; considering this, observed rise is again 50% faster than the best model estimate for this period. This underestimation carries over from the TAR models (see Rahmstorf et al. 2007 and the Figure below) – this is not surprising, since the new models give essentially the same results as the old models, as discussed above.

Page 4 of 6 | Previous page | Next page