environmentalresearchweb blog

« Commercial viability of CO2 capture and storage – of course not (under the historical paradigm) | Main | Renewable-energy skills, education and training »

Greenland: snowier than we thought

More often than not, reports about new measurements in remote parts of the cryosphere show that things are either worse than we thought or about as bad as we thought. A while ago, I was describing the out-of-the-way glaciers in the remote southern Indian Ocean. Things down there, it turns out, are about as bad as they are everywhere else. Sometimes, though, we learn that things are a bit better than we thought. A paper by Evan Burgess and colleagues, to appear soon in Journal of Geophysical Research – Earth Surface, shows that more snow has been accumulating in southeast Greenland than conventional estimates suggest.

The Greenland Ice Sheet loses mass through by melting and the calving of icebergs. There has been more of both of these in recent years. The accelerated loss is cause for concern, but we mustn’t forget that the ice sheet also gains mass by snowfall. Working out that gain is difficult, because the measurements are so sparse. The in-situ facts about snowfall on the ice sheet come from a few hundred snow pits and ice cores, in which investigators have counted and weighed annual layers. Over large tracts, no measurements have ever been made, and until recently we had to rely on spatial interpolation to make educated guesses about these data voids. What is worse, we want to know the total accumulation, and we can’t measure that around the edge of the ice sheet, where much of the snow melts and runs off.

Unfortunately, spatial interpolation won’t manufacture surprises for you. (And, if it did, you wouldn’t know whether to believe in them or not.) The contribution of Burgess and co-authors is to bring the skill of a climate model, called Polar MM5, to bear on this problem of missing information. The model stays close to the behaviour of the actual atmosphere by restarting itself from an observed state once a month, and its estimates of precipitation are calibrated by comparison with the sparse observations of accumulated snow. This yields much more spatial detail than could ever be got by trudging around on the surface of the ice sheet and digging holes.

The result is a net accumulation rate of 337 mm/yr of water equivalent, roughly the same as the annual precipitation in Wyoming or Andalusia. Other recent estimates for Greenland have ranged between 290 and 310 mm/yr so, accepting the Burgess number as more reliable, the older estimates were biased low by 10–18%. One of these estimates was by me. I reckoned the accumulation rate as 299±23 m/yr. The Burgess rate carries an uncertainty of ±48 mm/yr, so if you wanted to be picky you could claim that it does not represent a “significant” increase.

But we are looking at two different kinds of error. That plus-or-minus symbol, “±”, stands for an uncertainty due to random factors that cannot be controlled for – a give-or-take kind of error. What Burgess and colleagues have done is to correct a systematic error, or bias. The older estimates were off mainly because in southeast Greenland, where there are hardly any measurements, they were based on guesswork.

The authors managed to obtain a single field measurement from this data void. Where I estimated 950 mm/yr and another worker had 880 mm/yr, their model gives 2,780 mm/yr, but the measured accumulation over two years was 3,760 mm/yr. So it is an open question whether all of the bias has yet been eliminated.

The increased gain is not nearly enough to offset the losses the ice sheet has been suffering of late. It is still shedding mass and contributing to sea-level rise. But we always want the best estimate we can get.

One of the nice things about this work is that it shows that we are making progress. In the first place, we now have estimates of the uncertainty in the numbers, which is a quite new development. And in the second place, the uncertainties, give-or-take plus bias, are not really all that large. When it comes to uncertainty, less is better, but random errors of the order of 15% are bearable, and less than they were a few years ago.

You have to be a subscriber to read the Burgess paper while it is in press. But if you want to keep an eye out for it, its document-object identifier is doi:10.1029/2009JF001293.

Posted by Graham Cogley on November 16, 2009 3:00 PM | Permalink