I’m extremely happy to say that on the 1st October I’ll be taking up a lectureship at the Open University!
I’ll be sad to leave Bristol. I’ve spent all my years as a climate scientist there, ever since those running the NERC project PalaeoQUMP took a punt on a particle physicist. I’ve been very supported in my science and my public engagement, particularly by my boss for the last few years, Tony Payne, all those who run the Cabot Institute, and public engagement guru and incredible cheerleader for those who do it, Kathy Sykes. I’ll deeply miss regular coffees and cakes with one of the finest humans I know, Jonty Rougier. I’ll feel unsettled without my amazing network of friends, collaborators, and interesting-idea-swappers across the university – from drug use and mental health, to risk of volcanic eruptions, to philosophy of mathematics and politics. And there are many other very lovely friends: thank you for listening, helping, and being very cheering.
But I’m incredibly excited about the move. It’s not just that it’s a – cough – permanent position, or the relief of “making it” as an academic (for now…). It’s that the Open University makes education available to many of those with obstacles to campus learning, such as families, jobs, ill health and disabilities, or prison sentences. It also has a commitment to the wider public, working with the BBC to make wonderful programmes such as Frozen Planet (Principal Academic Advisor Mark Brandon) and making a lot of material free online (OpenLearn and FutureLearn). Unusually, they also put their money where their mouth is when it comes to multidisciplinarity. I’m excited to join Mark, Neil Edwards (no relation – I’m not an Edwards by blood), Joe Smith and others, continuing my research in model uncertainty for future sea level and past climates.
While trying to finish up in Bristol, I’ve been trying to say no to most new requests, particularly public engagement. But one came along I couldn’t say no to (thanks for recommending me, Jon Buttterworth). I’m very excited to be giving a TEDx CERN talk next month. There’ll be a live audience (tickets all gone) and webstreaming, and the talks will also be put online afterwards. The line up looks fantastic, Brian’s hosting, and I’m pleased to finally visit CERN…
Looking for a job this year has contributed to my long, er, hiatus in blogging (as well as my terrible response time to emails…I’m so sorry). Here’s a post I began writing back in January, about our paper on the Greenland ice sheet. It was published in The Cryosphere, one of the Copernicus Publications stable of open access journals that have transparent peer review. I love this public discussion approach, and usually sign my name to my reviews. Like PLOS, like the Open University, I’m a fan of opening everything up.
This post is about a result we published early this year from the ice2sea project, one of two companion papers asking: how might the interactions between the shape of the Greenland ice sheet and its local climate affect sea level rise?
Greenland is losing ice. But, as I’ve written before, this is mostly because it is always losing ice. At the same time it is also gaining it. Permanent ice sheets and glaciers are not really permanent, because these inputs and outputs change each year and season. So the size of the Greenland ice sheet is a matter of accounting. Its contribution to sea level depends on whether the books are balanced: ice in minus ice out.
On the surface, ice is added when snow compacts, or rain and meltwater freeze, and ice is removed when it melts without refreezing (the water runs off into the ocean), or sublimates. This part of the ice budget is called the surface mass balance. (The other part of the budget is ice being lost “dynamically” at the edge of the ice sheet, where ice bergs calve into the oceans, but I won’t discuss that in this post).
Surface mass balance depends on altitude. If you are high, as you are in the middle of the ice sheet, on average more ice is created than lost (the accumulation zone, where the net “ice income” is positive). If you are below about 1 km elevation, along the edges of the ice sheet, more ice is lost than gained (the ablation zone, where the net is negative). Here is an elevation map of the Greenland ice sheet as represented by a regional climate model called MAR (pixel size 25 km x 25km):What happens if we warm the air above the ice, as we predict will happen in the future? More melting, so ice is lost and the surface becomes lower. Being lower means being warmer, because of the temperature lapse rate (how temperature changes with altitude). Warmer means more melting, which means…more ice loss. It’s a positive feedback loop. Here ‘positive feedback’ doesn’t mean ‘lovely praise from your colleagues’. It means amplifying. The effect of the original climate change on sea level is increased.
But when the ice sheet changes shape it also affects precipitation (snow and rainfall). The precipitation part of the feedback is more complex than the temperature part because it can act in two opposing ways. A lower surface can mean less precipitation, because the air does not rise and cool as much, which is another positive feedback. It can also mean more precipitation, because the air is warmer and so holds more moisture, which can add ice, countering part of the effect of the original climate change on sea level.
So we’re interested in the combined effect of the temperature and precipitation changes. Does this “surface mass balance elevation” feedback amplify the Greenland contribution to sea level rise compared with climate change alone? How confident are we about this?
In other words, how much is the surface mass balance affected by the shape of the ice sheet as it responds to climate change? Can we express this in a simple way, and quantify the change per metre in height? We decided to call this number the SMB lapse rate: SMB for surface mass balance, and lapse rate in analogy with…temperature lapse rate. If it’s a positive number, it’s a positive feedback; the larger the number, the larger the amplification of sea level rise.
In this study, we needed to estimate the SMB lapse rate in the regional climate model MAR, not in the real world. That’s because we wanted to use it as a simple way of connecting MAR (which has a simplified ice sheet) with an ice sheet model. It adds in the feedback loop between surface mass balance, simulated by MAR, and ice sheet shape, simulated by the ice sheet model. So we needed it to represent the modelled feedback, to be self-consistent.
It would also be possible to connect the two models by joining up, or ‘coupling’ their code, but this is quite time-consuming and tricky to do. It also reduces the speed of simulation to the lowest common denominator, which is the (veeery slooooow) regional climate model. That’s very limiting, when the ice sheet model is super fast.
Our method means we can simulate the climate once, and then use the fast coupling to connect this climate simulation with the ice sheet model later. Or…with another ice sheet model. Or a group of five ice sheet models, as we did in the companion paper. In other words, it means we can assess the uncertainty from using different ice sheet models, or different settings of those ice sheet models, without the need to run the very slow climate model again and again.
It’s not the first time someone has quantified the SMB lapse rate for a regional climate model. But it is the first time anyone (a) studied the full probability distribution of the lapse rate (b) used it in ice sheet model projections (companion paper) to test how much it affects future sea level rise.
By ‘full probability distribution’ I mean the following kinds of shapes.These histograms are rough estimates of probability distributions for a MAR SMB lapse rate. (The curves are smoother estimates). I say “a”, not “the”, SMB lapse rate because we estimate different ones for the north and south of Greenland, and for the accumulation and ablation areas.
The light grey histogram is our “first guess”, or prior, distribution for an SMB lapse rate. (It’s the one for the ablation zones along the south Greenland coasts). The units are kilograms of ice per metre cubed per year*. The values come from changing the ice sheet elevation in MAR and seeing the effect on SMB. Each value is from a single MAR pixel, calculated by dividing the change in SMB by the change in elevation. It shows that in some areas the SMB lapse rate is positive – the net ice budget decreases as you go lower, as you’d expect from the temperature part of the feedback – but in others it is negative, showing the complexity of the precipitation part.
The dark grey histogram is our “updated”, or posterior, distribution. Normally in a Bayesian analysis this update means “in the light of new observations”. Here we are trying to estimate the SMB lapse rate in MAR, not the real world, so the “observations” are actually another MAR simulation. We have used each value in the light grey histogram to predict the SMB that MAR would simulate, if we changed the ice sheet height**. And we compare this with the SMB that MAR does simulate, when we do change it! (It is less easy to do this with the real Greenland ice sheet). The better the match, the greater the weight we give to the SMB lapse rate value.
This updating has the effect of ‘squeezing’ the distribution, giving higher probability to the best values (around 2 kg per metre cubed per year for this example). Our results shows that overall, when trying to represent the kind of ice sheet changes MAR predicts in the future, the positive SMB lapse rates are most successful: the feedback amplifies sea level rise. These results apply to MAR, but others have found similar results for other climate models.
The nice thing about this Bayesian approach – trying to estimate probability distributions in this way – is that it gives you a fuller picture. Instead of just one value with an error bar, it gives you a shape. Our distributions happened to be quite symmetric, but if they had been wonky that would have been interesting. Not everyone likes Bayesian statistics, but this post is already too long to go into that (very interesting) topic!
We use these (dark grey) distributions in the companion paper to assess how much the uncertainty in the SMB lapse rate affects the sea level projections. We try out the most probable lapse rate – the peak of the distribution – but also the high and low values from the extremes of the distribution. This way we can try and understand the full picture of the uncertainty. Is the effect ‘wonky': do the high and low values have symmetric effects on sea level? For example, you could imagine that the most probable SMB lapse rate amplifies sea level rise because it is quite large and positive, but maybe the lowest values in the distribution have little amplifying effect or could even, if negative (as they are for some of our other SMB lapse rates), reduce sea level rise compared with climate change alone. And how much of the true uncertainty of the climate and ice sheet models have we assessed here?
Tune in to the next blog post*** to find out…
* which comes from kilograms of ice per metre squared area per year (units of surface mass balance), per metre (units of height).
** More detail: the prior comes (mostly) from simulations where we lower the ice sheet height everywhere by a fixed amount of 50m or 100m. The test simulation has the kind of ice sheet height changes we expect in the future under climate change, which are up to a kilometre lowering at the edges and up to about 50m raising in the middle. This means the SMB lapse rate values that are best at representing this kind of MAR response are given the highest weight. Remember we are trying to represent MAR’s response in projections of the future, not represent the real world…
*** or read the open access companion paper
P.S. A reminder of moderation here: if you don’t see your comment appear for a while, it’s because you haven’t commented before (with those details) and I haven’t yet had time to read it and let it through. If I do want to moderate it, I usually try to do this transparently and tell you why. Comments on old posts tend to be ignored though (sorry again).
Open to positive feedback by All Models Are Wrong, unless otherwise expressly stated, is licensed under a Creative Commons Attribution 4.0 International License.