Understanding uncertainty in climate models

you this title I was actually surprised when I saw saw the announcement because I it's not the title I sent in because the title I sent in is actually this one but I think Brian and his wisdom realized this might reduce the the size of the audience so he and I think he's probably it was probably a wise choice so I will talk about understanding uncertain inclined models a little bit in general but I'm then gonna focus on this specific topic which is the circulation response so I'm gonna first just give an overview the first half of the talk which will be fairly general and then the second half will get a little a little more technical so a couple of points just to make at the beginning and then I'll ease with some slides in the next ten or fifteen minutes so the first thing is if you consider global aspects of climate change these are it's like surface temperature and so on these are robust both in the observations a robust meaning that it doesn't really matter which dataset you use it doesn't really matter how you analyze that you're always gonna get the same outcome both in the observations and in climate models and I will make an emphasis on what I call physically-based climate models of course there might be all sorts of different models one might use but when we talk about climate models usually mean models based on the laws of physics Newton's law enforcement dynamics and so on these are the foundation of our understanding and the man uncertainties there are how much the rate of forcing this is basically the the the the forcing of the climate system will increase in the future so for example that involves questions like what we do on the human side in terms of mitigation ops in terms of the emissions of the greenhouse gases and so on perhaps yo engineering and on the other hand it depends on what the what the biosphere does because a certain amount of the carbon that we emit gets taken up by the ocean in the land and so that that's another important part of the uncertainty so there's an uncertainty in terms of how much what we call the rate of forcing which is really the greenhouse gas loading will increase so that's what's called a scenario usually in the models you just say we're gonna assume a certain of the scenario and then the other big uncertainty is how much warming results from a given of radiative forcing and that's the possible question of climate to assets but then when you come down to read and those are not these are not small uncertainties I should say but they are but but but we're working with fairly robust sort of tools and observations on the other hand when we climate impacts who will be felt in the regional scale because that's really important you know it's where we live and for us so agriculture is a regional phenomenon flooding is a regional Omron even sea level rise is actually a regional phenomena and if you look consider the regional aspects of climate change which is what we're really dealing what now there are generally not robust either in the observations or in the models and what one of the reasons for that is that that they're determined by atmospheric circulation patterns so just in some sense the global patterns are determined thermodynamically if you like and the circulation patterns are determined to dynamically meaning flow dynamics and these circulation patterns just like with the butterfly effect and whether they are chaotic you have this chaotic variability that makes them heart heart predict and there's very long time skills they actually vary not just on year to year but over a decade alors multi educable time skills and they're strongly affected by model biases so if you're concerned about the the amount of rain and how that major it might change over Britain that depends on work where the jet stream is if the jet stream is in the wrong place the storm track the storm is gonna be in the wrong place and you don't have a meaningful a prediction of how that's gonna change so the valta bias problem becomes very important when you get to the regional scale so just to illustrate this so these are probably the three most famous global indicators of climate change this is the is the surface air temperature it's over land which is why the increase is so large you see it's been steadily going up this is the global mean sea level you see it's being mystic steadily going up the other indicator is the Israel to see ice and so this so this is not a lot of time series obviously sea ice has a seasonal cycle it reaches the minimum in September and then it reaches the maximum about six months later and the minimum over this is the median of the minimum over 79 to 2000 is given by this pink curve here but this was the minimum that was reached just last month which was a new record and so sea ice is really disappearing fast in in the Arctic so as I say there is uncertainty in the future evolution of the rate of forcing and you usually see that in these kind of plots this is from the last IPCC report this is global mean this is not the rate of forcing it's the global mean temperature but still it basically is responding to the forcing of course and this is the past and then this is the future and in each of these different curves corresponds to a different scenario in other words it's a different set of assumptions about how how the rate of forcing might change in the future and you see that there's a big range I think we can all assume that we're not at 2000 emissions so basically of this range here there's a factor of 2 in the response but actually the the that's what what looks like uncertainties around these are actually so these are what this is generated from is a large collection of models maybe 20 or 30 models in the what's called the steamed up 3 ensemble Andy for the different scenarios and what looks like an uncertainty is not really I'd say it's it's a it's a standard deviation but there's a lot of a question about what the spread of these models means I mean there's a sense that the models are I think there's a sense the martyr's never really liked being outliers so there's a sense that they want to kind of get themselves back in the middle of the pack and so if you do what IPCC did then was do a best estimate used on expert in the judgment of what they were card as as the as the likely range of warming and this is only greater than two two-thirds and you see that for example just to take the a 1 B here the the rate the the uncertainty range is a missive is about a factor of 2 so it's much much bigger than this uncertainty so this is the climate this is the rate of forcing uncertainty this is the climate sensitivity uncertainty so why is there climate sensitivity well the atmospheric energy balance is very complex solar you've probably all seen figures like the solar radiation comes in some of it gets reflected back by aerosol and end by clouds especially the cloud feedback is an important uncertainty some some gets absorbed by the up of it by the clear atmosphere a lot gets absorb of the surface and then comes back into the atmosphere through convection and in latent heat and foxes and so on and then there's also a rated of part involving the mreb of which again clouds play a very important part so it's it's really the clouds are the main reason for this climate the the uncertainty in the climate sensitivity but this energy balance also varies spatially and so the these spatial variations they lead to temperature contrasts that lead to pressure contrast as the drive winds and it's interesting I I was you know it's not easy it's not hard finding a figure like this every IPCC report has this figure you go in the way of everyone at this figure much harder to find a figure of the circulation because it's just it's just not in people's heads as much but once we think about regional patterns and you you may not like this one and I thought I should pick one over Europe at least but I couldn't find a nice one at least on Google even though all I hear all I heard about for the whole summer was where the jet stream was and it was in the wrong place you're very conscious of the jet stream here but anyway this is a North American version of course in the old days people were in in the days of sailing ships where all people were very conscious of winds now we're conscious of a jet stream because we travel by air anyway they so this so we have these circulation patterns and the jet streams are very important part of it and because it's it's fluid dynamics and if you know anything about fluid dynamics you know that there's turbulence and uncertainty we have chaos and there's variability so these circulation patterns exhibit internal variability and so this is you've probably heard of the North Atlantic Oscillation perhaps climatologist tend to classify the the actually the variability through these empirical indices the North Atlantic Oscillation is defined by the contrast between the surface pressure and Iceland and the azores the Icelandic low and the Azores high and when this when there's a weak phase of this of this dipole basically the the jet stream gets pushed north and you get cold and dry over Britain and and the warm the the wet weather is pushed to the south of Europe but when when it Jets when this is a strong phase that the jet stream comes right into Britain and you get lots of warm and wet weather here so this is a well known pattern now it's called the Oh is for oscillation but it's not really an oscillation people don't really understand the mechanisms for this but it's as an act as a topic of very active research well what I see Michael Gill was here maybe maybe he explained it but anyway this is a what we call a mode of variability it's a chaotic variability in the system but in this obviously affects weather in the sense that it will set up the jet stream and then you get weather advanced controlled by that so there's modes of variability but they also vary as I said chaotically so this is this is this North Atlantic Oscillation index over about a hundred and fifty years and you can see it's oscillating up and down it's got annual variations but it's also got I just put these in by eye these are not these arrows but you see it's also got kind of ups and downs over very long timescales over many decades and say it's no surprise really it's a chaotic system but we don't really understand all the timescales and what this means is that over over limit of timescales like like our lifetime you can get these apparent trends and that this will this will this will also affect impacts such as droughts there's a lot of research going into predicting these variations but but how predictable they are is is still unclear and certainly in terms of inferring a long-term trend in the record is it's it's very problematical now but these are very important this is work from colleagues at reading Ron Sutton and Bo and dong and and they they as of course I got here in May and it's being wet and people said oh it's this is so unusual those must be people that have a long memory because the in fact the weather has been pretty wet in summer for a while now what they noticed is that there's a relationship between the sea surface temperatures of the North Atlantic and the weather over Europe so this is another index called the amo the amo is the annalen Atlantic multidecadal oscillation it's it's one of the challenges is we don't know all the relationships between all these different modes of variability but this is defined by the ocean temperatures so it's really reflecting the ocean circulation but of course it affects the atmosphere of them too and and they point out that the that the Atlantic temperatures had been warm over the last 15 years but over the previous 30 years had been cold so this for example shows the difference between the last 15 years in the pre and the thirty before that for the spring the summer and the fall you see it's wet here this is precipitation I should say but then it turned out that the the temperatures were actually warm back in the 30s 40s and 50s so if you look at the difference but going backwards so you this is past minus future this is future - past you change the change of sign to get the right clarity you can see that the changes in that period were very similar these these are very similar patterns so there's quite a predictable part of the atmosphere response to these ocean temperatures which it really is important rainfall now so this is a nossa this is going back and forth ups and downs the kinds of swings up and down that we see here this is just a different index of course we we expect that this sum that things should get wetter over Britain in the summer in the future people expect this amo to to keep trending up but it's it's hard to tell from these ups and downs but the real question is can we afford to wait until we really see a clear trend emerge and you know these are pictures from this summer and all the concern of course about rising food prices nothing so this is really one of the challenges is that things may be changing but it's hard to see in the observations because of all this very low-frequency chaos and it's important that because climate impacts of climate change will occur through extreme events this is an example from Canada it's totally outside my area but I find it very interesting this is so I already showed that arctic sea ice has been melting in the summertime so that means that the Arctic coast is it is exposed to more open water the Canadian government loves ACK because they can drill for more oil but of course it's got a lot of impact on on ecosystems and communities and so this is the Mackenzie Delta in the northwest part of Canada in the Arctic sea and so the they had a big storm surge in 1999 and this is a record from lakebed the sediments going back a thousand years and these are different biological species I can't tell you what they are but the ones in green are our freshwater species and the ones in red are the brackish or saltwater species this is not a linear scale but then zoomed up for the last ten years here and you can see that this change has been unmatched in over a thousand years and it seems to have led to an irreversible change in the ecosystems from that one event this is I think how climate change will will have will happen in many cases so we have to worry about this but because of this variability makes it very difficult to discern long-term trends and observe record so even if you have a shifting pattern you may not be able to really see it in the in these time series if you're just looking for trends in a very simple-minded way so this suggest you need a probabilistic approach to climate prediction but how do you get get the probabilities we don't have enough observations generally decimate that so we're gonna have to estimate it from models so this is an example from from the NCAR model in the US what they're showing here is probability density functions or likelihoods if you like this or likelihood this is a relative likelihood of a certain outcome so the peak here is the most likely outcome of wintertime December January for every trends over the 55 years 2005 to 2060 so these are trends in the Eurasian North Atlantic sector now if you look at the at the surface temperature and the gray is a control run without any any climate change forced in the model the red is the model for us with increasing greenhouse gases so of course the mean of the gray must be zero because it's because it's a control but you see there's a big spread here and then what we see is a warming in the surface temperature now for temperature which is thermodynamic in control there's a pretty clear offset and this says that basically over 50 if you can afford to wait 55 years you'll have a pretty clear distinction between a trend in the case of the change in climate and a trend in the unchanging climate but when you go to this side of the thing this is sea level pressure here so that's a dynamically control feature now we see that these things are are offset a little bit but not not a whole lot and these are 55 year trends and it's saying that for example you have about a 30 percent even in the warming world you have about a 30 percent chance of having a trend which is opposite to to the mean trend or predicted then when you look at the precipitation it's controlled both by the by the weather patterns and and the circulation and by the temperature because as as the atmosphere gets warmer it holds more moisture so you expect more precipitation all else being if this if the circulation doesn't change then warmer temperatures mean more precipitation so this so the precipitation is actually somewhere in between you can see that there there's a little more separation perhaps than then here but not a whole long so this means that you know we really have to have a probabilistic approach and it could be that we don't see that the change we observed could be opposite to to the average change even over a fifty five year period so we need to work on developing if we're going to use the models we really need to develop reliable models it's absolutely crucial but the problem is the models have issues this is also work from writing colleagues to Tim Willingham like Blackburn here so this is circulation patterns this is to show that there's a bias problem so bias of course is systematic error and that and that affects the predicted changes these are the lower chopped aspheric they 850 activist kal is just above the boundary layer and the lower troposphere the zonal wind speed these are four of the leading climate models this is the GFDL and the anchor model from the US the French model and the Met Office model and the the contouring here shows the zonal wind speed in the climatology the average dawn wind speed you can see this quite they all have a they all have a wind speed they all have a maximum wind which is more or less in the right place but the magnitude you see is very different in the four models now the contours show that the predicted changes over a hundred years so this is not subject to variability now this is going to be a deterministic change but you see that there's almost no correlation between these four models at all so we don't have any idea how the circulation is going to change if you just look at these four models and this is a bias problem which really has a huge impact on on on prediction so to recap this first part of the other of the talk both chaotic variability and model bias present many challenges on the one hand we may get serious regional impacts of climate change even before the observed trends are statistically significant at least using using the usual tools that climate scientists use so there's a there's a general need to move from a confidence if you'd like to a risk framework so just to make this a little more concrete though I won't say anymore about it but it's it's worth pointing out I think the the scientific community and it's a it's exemplified by the journals that we've published in that's exemplified by IPCC working group one we want to we want to make statements with high confidence so you know highly likely virtually certain all these things have to be 95% confident 99% calm and so on if you ask whether hurricanes and in in the Gulf of Mexico are going to increase we're not haughty we're not confident that they will but if you are in charge of the the levees in New Orleans and you have to worry about whether you might have to build them higher after Katrina because climate scientists tell you that maybe sea level will rise and maybe storms will be human requires you certainly have to compete concern there might be a 20% chance let's say of of the storms getting stronger so you don't have too high have high confidence to have a very significant risk and so there's an important I think it's important it has a community that we think about how to map the traditional statements about confidence into statements so I'm interested in trying to pursue that and actually talking with statistics colleagues here about that sort of thing the other aspect is because the bias problem we somehow need to get useful information from flawed models we can't wait of course for the models to get better they will did that may take a long time so people have to make decisions now and so there's a quite a quite a community working with trying to deal with these flawed models in a sort of calibrate the projections based on on biases and so on but it's difficult to do because it's not really clear how you map the the biases onto the predictions anyway this is another large area in is in you could say calibration of predictions but ultimately I think we these models are all based on on the laws of physics which we which we believe and trust and ultimately we just need better models and so we have to get rid of some of these biases or or reduce them anyway so that's what I'm really interested in and I'll be talking well I'm interested in many things but that's one of the things and that's what I'll talk about in the second half of the talk oh one of the challenges here is that it you think of the weather the these circulation patterns are forced by large squat are called cross P waves a planetary scale which are forced by mountains and by land sea temperature contrasts or also we have weather systems scales of thousand kilometers or so in it I think there was a belief that once that the models got the high enough spatial resolution by which I mean that the think of a model as representing the climate or the weather of the atmosphere with with a grid with a certain resolution so let's say our typical weather forecast model might have a grid of ten or twenty kilometres a climate model maybe 300 kilometres generally so that's you you you represent let's say Britain by one grid box with that resolution so I think it used to be people believe that the the dynamical part was kind of a solved problem in fact when I was getting my PhD it seemed that you know so dynamics was somehow like learning Latin you know something that you had to do you have to know what the Rossby wave was but all the uncertainty was about aerosols and radiative forcing and so on but in fact higher resolution does help but it's not robust in the sense that it's not clear that you don't always get a better result and maybe for that reason centers don't do that because it's of course very expensive computationally and so there it but I think the fact that models don't robust they get better is telling us something it means that actually the scale of the weather systems isn't the only scale that we have to worry about there's also smaller scales so if I say that the model has a resolution of 300 kilometers that means that there's we say but there's a wind and the temperature and the pressure and water vapor every 300 kilometres but there's all obviously the variations have smaller scales so these are not resolved or they're unresolved or they're sub-grid scale although the good scale so these effects are parametrized their representative in a parametric way that's linked to the larger scales this is much more complicated I think we know how to handle the resolved scales in principle but we don't know how to handle these numbers all scales in a very straightforward way and so the fact that there's this challenge that these stubborn biases which persist and have persisted for a long time links that there suggest that they're linked to something unresolved the problem with this is that if you have errors in these parametrized processes they immediately will lead to errors in your winds and that'll lead to errors in the in the in the weather systems and those feedbacks so you got all these feedbacks strong dynamical feedbacks who within the climbing system the really obscure the course of the cause of the error so challenge is to somehow break that because it is a very nonlinear system so the challenge if you want to diagnose the cause the errors is to break that feedback loop somehow so that's what I'm going to talk about a little bit in the second half so what is a parameterised process well as I said it's going to be a process that's below the grid scale and the one I'll actually talk about exclusively here will be Mountain waves there's also other very important process processes or boundary layers convection clouds but mountain waves are an important process so here here's a mountain wave well here's a cartoon of a mountain wave you have flow over of course it induces these ups and downs just as it would in water I think you're all familiar with waves on water those are also those those are gravity waves there's also gravity waves not gravitational waves by the way if you're from physics there's also graduates internal to the atmosphere and the ocean that have to do with us with the stratification so it's not quite as intuitive as the wave on the surface of water but it's exactly the same process it's just the differences those waves can propagate in the vertical so we see mountain ways every time that we see this kind of cloud pattern these waves probably get up and just like the corrugations our road would slow down the traffic these these waves provide a frictional effect on the waves on the winds aloft so we call that gravity weight drag it's a frictional in effect and this is as I say sub-grid scale in the sense that these processes are happening at grid scales below the model and it's not well constrained observational ofcourse we know where the mountains are but we don't know but there's a lot of details in how the waves get generated and how much propagates up and it's a very intermittent process very small scale we don't have a lot of constraints so when people represent these in the model they they don't have a lot of guidance from observations okay so well well welcome back to Mountain waves a couple times so I mentioned the NAO well in the in the winter season it turns out the stratosphere is very active and the stratosphere as Brian said it is above about ten ten kilometres at least of these and latitudes so in the stratosphere there's a vortex big vertical flow around the cold stratosphere because you have that absence of sunlight in the wintertime and you get this strong circulation around there called a vortex and it turns out that this vortex couples with the surface and a couples through through the effect of weather systems so this is now called so the Ennio is is defined by the surface pressure but this some it's called the northern annular mode it used to be called the Arctic Oscillation the reason it's called that is because it's it's got this angular type structure around the pole and it turns out is it it Co varies with the surface and so when you have a positive phase of the oven so I'll use this term a lot this NAMM when you have a positive phase of now mitt means they've got a very strong flow up in the stratosphere and that and that means that the winds over say the latitudes were relevant to Britain are very strong and you got the storms tracking in here whereas if you have a less cold stratospheric and a weaker vortex then you get this phase where the where the where the main storms are pushed into southern Europe I it really amuses me to see this little Sun over Britain I just find that hard hard to believe so so there's this coupled aspect to the whole variability and it turns out that variability in the stratosphere seems to influence the surface circulation this is a plot assault observations of these northern annular mode indices so michiga you plot just like when I showed the NAO this the the index goes up and down it's you're in the positive or negative phase and you can see that if you just composite so what's done here is your T is you take all the observations and when you get a very weak vortex a very strong phase negative phase of this name then then you mark it and then you and then you make a composite and in the historical record there's been 18 of these very weak events and you can see that if there's a weak vortex in in the stratosphere it lasts about a month in the middle stratosphere and it lasts about two months in the lower stratosphere and it seems to influence the the troposphere as well on their hand if you have a strong vortex it's the same thing in Reverse so what this shows is that is that there's actually some influence here and that you actually have a control on the weather systems over a couple of months of course the variability is coming from below anyway the the vortex is disturbed by weather systems and waves probably getting up from below but the the disturbances in the troposphere get quite strongly damped by friction and by heat exchange with the ocean and so on in the stratosphere it's quite quiet and and the radiative time skills are very long so there's actually a lot of a lot of potential memory sub seasonal memory in the stratosphere and that's what gives the persistence of this anomaly and you can see it really affects the surface so it's been known for a long time from these sort of plots that this is possible and actually models have begun to get that right and this is a result from our Canadian models see the Canadian middle our model is called from a golden segment at Toronto this what Sean shown here is observations of mean sea level pressure that's the contours and precipitation which is the is is a color shading following the most dramatic events which are called stratospheric sudden warming's which is about twenty in the observational record here this is the this is the the service conditions between 16 and 60 days after a sudden warming happens so it be like this this period through here and you and this is the model if you take the model and you make a forecast after the sudden warming you get a lot of this the these are normally as I should say this is not not that this is not surface conditions this is the departure from the climatology that you would expect and you can see that it's not perfect but there's a lot of skill in this and so if you thought models that can actually do this it means it opens the door using these forecasts as a test for the models the problem is we can't test the climate prediction because we don't we can't wait for that to happen but you can test for example perhaps some shorter of lead time forecast like like this so this provides this sort of thing provides an excellent test of climate models and people are getting interested in doing this but this effect of the stratosphere on the troposphere extends to not just sub seasonal but extends to any inter annual variability now this is also winter time and what's plotted here our time series from 1971 I guess to to 2002 this again this northern angular mode index and AM index but at the surface now and the black curve shows the observation so you can see that just just like with the NAO the black curve is going up and going down and going up and down so it's oscillating around what's shown here in the red is is ensemble point of climate models with some spread about it where you just imposed the observed sea surface temperatures and I should point out the sudden and Doong paper was about the sprit what was not for the time it's for the seasons apart from winter where the sea surface temperatures seem to exert some control in in the winter time they seem to exert no control because because of the influence I think of this of this now and you can see that the model the model ensemble in red just doesn't really follow the the observations at all there's a correlation appointment 0.35 here but if you constrain this is a historical if you constrain the model to follow the observations in the stratosphere poleward of $25.00 this is so that it's not an effect of the quasi manual oscillation which is a technical point so it's constrained by the sea surface temperatures and by the stratosphere then you see you get almost a perfect reproducibility so that's saying that's telling you that if you can get the stratosphere right then go get the surface right that's because the the stratosphere is the upper boundary conditions so mathematicians don't know the boundary conditions are very important and if you don't get the boundary conditions right you won't get you won't solve the problem I think so that's important for things like like these decay --dl variability but it's also appears to have to to be relevant on longer timescales so this is for climate change this is again this is again from Michael Sigma's work actually with the CMM what this is this is the mean sea level with your response to double co2 this is over the Arctic here and you can see that the results are very different well what's different with these two runs they have different settings of a parameter in the mountain wave or the orographic graduate jag scheme both of which are plausible there within a bloom plausible range and you can see that the answer changes completely depending on this orographic graduate drag and they track this down and it turns out to be the through the control of the lower stratosphere winds which control the the weather systems so there's a lot of sensitivity so gravity wave drag comes in this is example the systematic the sort of structural uncertainty because of the systematic errors and models and even if you look at the bottles with so you might say okay well we know that that the stratosphere is important so let's look at the models that have a good stratosphere so these are the the spark models or the cesium valve models and this is predictions in in the past and the future so from the 1962 21 hundred in the wintertime December January February in the spring March April May but you can see it's a bit of a mess and some models show a big trend in the winter some don't so there's no consistency here and if you look at individual runs you see there's a lot of ups and downs within each simulation so there really is an issue here it's gonna be very hard to say what's gonna happen on the basis of these models unless we can somehow deal with these errors and I'm convinced that they've got a lot to do with the graduated drag okay so the second topic that was on the northern hemisphere I just want to touch on the southern hemisphere as another example so you may or may not know that actually the climate change and the in the in the southern hemisphere has really mainly been detected in the summertime and it's actually got nothing to do with greenhouse gases at all to do with the ozone hole the ozone hole has been a huge perturbation to the stratosphere when you when you lose ozone you yuku cool the stratosphere that extends the winter season essentially in that and that has through the same downward coupling mechanism I talked about it affects the surface so it's like this like the northern annular mode but there's a southern anima mode and if you perturb it in the stratosphere you get a response in the troposphere this is these are observations this is a linear transform I think 1970 up to 2000 of mid tropospheric geopotential height but this is maybe the surface temperature and wind so you can see the main thing is you get a speed-up of the vortex around it Antarctica this has implications by the way for ocean circulation heat uptake carbon uptake possibly this is a model if you force it with an ozone hole and you get very similar pattern so if the ozone hole has led to big changes in the past then we expect the ozone hole would go away over the next 50 years because of the control and see it CFCs it'll take a long time though it's about 50 years then we expect that that the sign will turn around and that means that ozone recovery would weaken or possibly we reverse the Sam trends I should say that climate model predictions are that increasing co2 will also strengthen the surface winds so in the past they were acting in the same direction in the future they'll be acting in opposite directions so this is some from the IPCC models at last around which didn't impose an ozone recovery and these are projections of the winds wind transit in the in the southern summer over 50 years you see a great big increase of the surface winds at 60 South but if you have a nose on whole in there then the recovery period is basically meant nothing happening so it's an important effect this is work from our our Canadian model led by Charles McMahon dress the effects are actually not just on the on the surface winds they're actually on that on the Hadley cell boundaries this is the edge of the tropics essentially the 30 degrees or so and it turns out that the ozone hole has an effect on the southern edge of the Hadley cell which has implications of course for the scent of dry air and so on in these are little complicated plots but what's shown on the on the left is is trends per decade degrees per decade so it's latitude will shift over the past when the ozone hole was developing and over the future where the ozone hole is recovering the blue dots show various kinds of models without and those are all in them and you can see that the trends are the same past and future these these other dots and things show models with with an ozone hole in them and you see that there's a greater trend in the past and there's a reversal in the future same thing if you look at precipitation you get the same kind of a pattern here that there's a different the difference between this bubble and in this bubble is the effective deals on whole and this different bubble in this bubble is the effect of recovery so you have to worry about this if you're gonna if you're gonna get things right and this is a set it's important for Southern Ocean eat up taking carbon uptake potentially also people have claimed for ice sheet stability but the models have biases and gravity wave comes in again this is result from former student Tiffany Shaw who this is so this is the surface again the mean sea level response to to the ozone hole this is with a with a lid that's at about 95 kilometres so we think that all the physics is right in them or at least if the models right we're doing this right and but then this is this is a bit technical but if you don't conserve momentum flux in the graduate drag scheme which is typically actually the case with a lid that's smart in that in the middle stratosphere you get a you you lose the significance which is a shading you really lose a lot of the effect of the surface of this on the surface of the ozone hole so how you treat the the graduated dragging scheme is important you can see the bots there's the probably the major bias is that the models said that NASA as a Sun comes up in the in the summertime the the polar regions he he heat up and the winds changed from westerly to easterly which is to say eastward to westward so the transition to easterlies is the beginning of the summer if you like and in the in the observations it's this black curve that that comes down in time over the course of about a month and you can see the models are almost all late so but this will affect the the surface response and so how much does this model bias in fact all these projections of the surface responds well and so I won't answer that question but what's what's the reason for this for this for this bias and it occurred to me that all the models must be doing something wrong for the same reason and the thing that all these models have in common is they don't have any mountain wave drag at about 60 South which is the maximum of the jet this is a version of model with um data assimilation which is where you bring in observations and make a short-term forecast for six hours incorrectly make a short-term forecast and from the increments from the data estimation you can infer basically what the missing force is that your model doesn't have you see there's a negative force here in the upper stratosphere that comes down as the as the winds as the winds change and so this looks like a mountain wave drag because it's it's it's at least a very slow phase speed wave if you look at the mountain wave drag in the model this is big zero right at 60 self that's because in the model it's ocean at 60 South but actually there are mountainous Islands that provide a lot of gravity waves in observations they're just not represented in the climate models and also graduates propagate latitudinal II not just vertically so those reasons to believe that there should be graduated drag in this gap here when we put something in we actually get the right kind of response so I think you can look at these short term forecasts and try to constrain things and again I think this worth thinking about the orographic graduate drag I'm afraid I'm gonna have to skip the last uh I worked so long in this on this some figure so as a former imperial student ila Simpson who was it was a postdoc but with me she's now after Lamont she was looking at timescales of these variability but it's a complicated topic and so I thought I would straighten with it with a spring and the idea that if you have a damn spring and a forcing that the the displacement of the spring is purport it gets greater if you have longer time skills there's been a lot of work on timescales but I think I just don't have time to talk with that but she's been looking at trying to use constraints on the model to really see where this long timescales come from because if the timescales are too long then that means there's gonna be bias so it's a bias in the variability rather than in the mean and and she's being again using this but I'm afraid I can't I just think right at a time there so finally let me say that um scientific interest is turning really from the detection of climate change which is really being the the big problem for the last ten or fifteen years the prediction of its regional impacts and these impacts are controlled by certain circulation patterns these circulation patterns are flow dynamic well we have turbines and fluids and and they will have very large variability on these very long timescales multi the kaleb timescales chaotic so you can't just fit some sort of simple sinusoid to it as a result the observed trends are generally not robust statistically but this doesn't mean that climate change is not real it doesn't mean that the risks aren't there climate change will be felt through these extreme states so we somehow have to deal with this fact that the observational record does not provide a lot of basis for for for determining what's gonna happen in the future so in the end physically based climate models based on the laws of physics will have to be the basis for prediction but these models exhibits significant biases in circulation patterns and the variability and if you look at the predicted changes over long term say in the future there's that they accept large spread and these are almost certainly related so we need to remove model bias to somehow it's as I said it's likely to involve these promise sub-grid scale processes and how those feedback on the circulation and as i've indicated with a couple of examples / progress can be made by trying to break the feedback loops which confound the model errors and i think that's going to be a fruitful way way forward thank you very much to talk about your first slide in your first graph which was the rise of temperature as you know that as you know the the there is still that I was speaking of parties a political person the temperature over the land and the sea and has been showing this flat for the last 12 years which causes politicians in various countries to say therefore we can ignore climate change but but the the point is the temperature over the land areas now I wasn't quite sure where that was the land area because that shows the flattening well its things been rising and my my you know the Met Office still this the only place in the world we totally puts on its front page of its website in fact the temperature of the land areas is rising and it is you know the IPCC executive summary always puts the land and the sea which of course causes great consternation to Paula Jones and this is I mean that is that is still the situation so I just you know this that's the PRS pay what I just want to say about your orographic which is where I'm a serious scientists I work on that and and and as you know the British Antarctic Survey picking very very detailed studies and you have what we call a low-level explanation of why you get this the highest temperature rise in the world three degrees in 30 years just because you divert these flows due to the Coriolis effects over the over that mountain which of course is not real represented in the current model but you should talk to the British Antarctic Survey while studying this in huge detail thank you very much hi thanks for a really interesting talk the elephant in the room for me are the things that you didn't talk about in that are not in the models and I guess the main one is the biases the the biospheres response to warming and so you haven't talked about the fact that action carbon uptake might become carbon release right so you didn't talk about the changes to ocean circulations okay what factors like ocean of course has I guess two aspects one is that it will it has a mean aspect in the sense of some of the warming will go into the ocean aster in the ocean and actually one of the potential explanations of this of this flattening is that the heat has been taken up by the ocean rather than the atmosphere over the last ten years but of course there's also very important for the variability part because the ocean has its own has its own aspect of variability the the kind of classic picture I guess like I could be corrected by you some like Ted Johnson here but I think the cockpit shows the ocean is basically it's a stable medium because it's it it's you know it's heated from above not below so the only way that the oceans gonna evolve is if you force it somehow and that's gonna be with the wind and some from the atmosphere so the atmosphere is the unstable part of the system the ocean is a stable part but it's still that doesn't mean it doesn't have very long timescales and you can and you can and those long timescales may not be totally predictable just just because something is that stable doesn't mean it's necessary yes right I mean it will be unstable on short timescales in the sense that they the Gulf Stream was unstable and the sort of thing but I guess I just mean that it's well I shouldn't it it was certainly a couple with Yama stirred it doesn't it doesn't necessarily make the whole system more it doesn't necessarily make the system more unstable but I guess it would make the system certainly have more variability than that's restoring so a lot and we don't really you know the anything on a decadal timescale is almost certain to be the unless it's forced by by solar variability or something that's gonna work both with volcanoes is gonna have to be the ocean it's the only thing in the system for that kind of memory we don't really know that all the time scales in the ocean I mean there's some very longtime skills in the ocean so that is an important part and actually I guess another part of some of this spread may come from drift in the ocean part of the model too that's sort of I'd like to ask a question about the last sentence on your last slide and in particular if you tell us a little more about that you know can you give us a bit more information about what you mean or how an example of what breaking the feedback loops that confound modeling or an obscure modeler and then having done that if you would also say just given that these model letters do couple back you sort of think eventually that the models could have zeroth order errors in the even the global properties did you have an idea for the time scales for that to happen Mikey you know how long how long but but I'm actually more interested in I mean I think the two questions are actually related okay so I mean I think just to take this case you have your you're looking so I probably dashed over this too quickly what you're looking at here is you you take a cotton model and you basically constrain it to the observed state of the atmosphere through this data summation process and then you make a forecast and then after six hours you look at the new state of the atmosphere you look at the mismatch between the model and the forecasting and you adjust the model to two sorry the observations are forecasting and you adjust a model to to be more accurate and then you make another forecast that's the basis of weather prediction now if the model was a perfect model the errors would just be from the butterfly effect and the average error would be zero it would just be what sometimes this way sometimes that way if you look at the average adjustment here it's it's nonzero so that means there's some kind of bias and because you've constrained the model to to the observed state of the atmosphere it means that the winds that the way that those large-scale winds are right and the in these Rossby waves the large-scale patterns are right as well so there's no error and those and those in those in that part of the system and the only error that can grow in a six-hour time scale is gonna be this parametrized process which because that's a very fast process so you're breaking the feedback loop because you're not allowing the error in the in the in in the in the gravity waves to affect the wind which then feeds back and everything else so the feedbacks are important because this example for the Arctic was totally oh sorry this might hear the effect of the on the surface isn't coming directly from the graduate drag is coming from the effect of the graduate drag on the winds and then the feedback from that so I'm not saying the feedbacks aren't important but to figure out that it would here we know it's the graduate drag cuz that's what was changing the model but to infer an error is very hard but when you when you constrain the the model by by the observations you're sort of breaking that feedback because your it can't it can't occur in six hours it's just the timescales of the resolved motions or more like days so that's that's the sort of example now in terms of how long it'll take us to get better models I'm sorry I don't think I should guess on that you were talking a lot about the gravity wave like the parameterization of the gravity wave track which is something which is not resolved by the model because the spatial scale of the model is so big so that's why you got a parameter i's it so don't you run to the risk of just using the gravity wave track as another queueing parameter because such as use clouds as a tuning parameter for instance because you can't resolve them as well how much do you see the risk in trying to get too many subscript right processes with too many parameter is ations that in the end you just try to tune against each other yeah well yeah that's a very good question I mean there's to some extent tuning you know tuning a piano is a good thing so something tuning isn't necessarily a wrong thing to do and if you have a set of a parameter range that you that you know from basic physical principles or from idealized experiments or fluid or laboratory experiments or something and then you want to constrain that from observations there's nothing that's totally wrong with that the problem is you you want to make sure it's for the right reason and you certainly don't want to tune to fit something and then say that the ability to fit that thing is a test the model it's certainly not but that's why this kind of thing is good because we're not trying to tune the orographic graduate drag to correct surface bias were actually trying to infer it from these short-term forecasts that's an independent piece of information and and there's actually more sophisticated way than if you if you if you find that there's a drag you can try to infer what the what the errors in the in the parameters are so I think the idea of using weather forecasts or perhaps the seasonal forecasts but anyway some kind of short-term forecast to determine these parameters is a very promising area that's totally in depending but I think it's a tuning the right sort attorney yes I'm an engineer will change Ania I'm not really into climate science but um the reason I'm here I was more interested in the mathematical part of how I deal with the uncertainty in the models because you know as water engineers we get in output form climate models to to use in our hydrological models and you mentioned something that kind of gave me out quality scare but the variability in in the climate models is not statistically significant am i right without oh I'm just saying that if you look at time series over a certain period of time like 50 years or something I know you mentioned a lot of things that are changing but you say it's something that the variability is somewhere is not statistically significant as well what a man I think was that a trend in in a time series one realization over a particular length of time I'd say 50 years may not be significant so if you I'm saying if you take a model as your truth which it's always good to test something on on the model first because if it doesn't work for the model when you know the answer then it won't work for the real Imus for the comparison to realign the sphere and so you can certainly get cases where you like the case I showed with these PDFs where there's really a shift going on you can see that but in any significance test of the whole PDF will show you that if you just take one realization because the climate system will only have one with one realization you would never be able to infer it from that so that's what I meant okay okay okay okay what my question was really the job I thought of getting the impression is that that you you're saying that the statistics is not good enough to show you the is not good in describing the uncertainty in you yeah what I mean I guess and I should say that the statistics that are used in they're often used in climate science I think there's there's definitely scope to improve the kinds of to the scope of methods that are used people in climate science generally I don't have a lot of physical training like myself income from physics or even math math backgrounds McMath and stats are different fields aren't they and and so we tend to use I think butter regarded as pretty old-fashioned methods and that's sort of the what's that's what you need to use if you want to be if you want to get published because that's what the reviewers are comfortable with you know it's different so I think we there is room to there must be room to do better and we somehow have to get more information out of this time series but I guess I'm just saying that the the typical kind of you know let's see if we can see something really emerge above the noise and time series that may not but by the time that happens it may be much too late okay then then I'm secondly or thirdly have you investigated how to to code you measured something you have to move away or combine confidence Vasil risk in a new analysis so I'm wondering now if you want to transfer Oh to understand this uncertainty in the model direct to the product or the model output like temperature like sea level rise and you want to to describe the relationship between the uncertainty and the variation in the output you are you with me there so I'm just wondering since you you you kind of don't trust the statistics so much it's not that I don't it's just all I mean is that I think this example is in it so in this case what we have is we have a whole ensemble of all's you know I don't know how many different copies of the model that are run at 50 or something and you can get a distribution okay we we don't know if this is accurate in terms of the real atmosphere but at least you you can do experiment can do tests in the modeling world and so these are even you know these are all very clear shifts right if you have if you have all the data the problem is that in the in the observations will only have one of these one one one example so we don't have all all the information that we need and I guess you know for for a statistician they would say that really it's a problem of probability but we only have single realizations and the models only produce realizations really and sometimes generate these things but the models are not realistic oh I wouldn't want to say that the statistics are inadequate I'm just saying that there is a challenge that you know we're dealing with with time series that go up and down and yet these so it's gonna be very hard to if you just treat this as a tie as as a time series about any other information it seems to me it's gonna be very hard for you to to necessarily conclude that it wouldn't just go down again and yet if there is something happening we're gonna wind up with with with with flooding so we have to somehow deal with that aspect that that it's the tail of the distribution here that might be really important like oh sorry like let's say over over here or something but but you don't but you're not actually gonna see that because a trend determined it depends on the order of the things happening maybe that's the simplest way to say maybe you're getting more events that are intense but if they happen in the 2020s and not in the 24 to use the trends gonna go the wrong way so trends are very noisy kinds of things like can I add on to that you said about getting useful information from flawed models oh yes so this is saying we only have one realization pair of this but how about the useful information from Florida models ah well um so I guess I mean I need to look into this more but I think the Met Office deserves their their climate assessment that was using some sort of Bayesian approach to try to deal with those flaws and I can't comment on that for them I'm curious about what was done but I can imagine and I am interested I know that in some cases there there are some examples where there's a clear link between the a bias in the model and and the spread in the predictions in other words you you you know a model has a certain bias you take all these 20 different models 30 different models you look at the bias in the in the observable part of the system saying in the client in the seasonal cycle or something like that and and the prediction of how much things are gonna change the future and you find that they lie on a straight line and they and you understand why that's the case because so some basic physical mechanism involved makes sense then you would have a reason then you can calibrate that prediction there's a couple of examples like that but there's not many and I don't know you know if how many of those we can find but if you could find those sort of things of course then you do have a basis to try to use the flawed model but otherwise I don't know how you do it thank you you've given a very nice some talk on uncertainties in models focusing on sub-grid scale processes a lovely example of how you can improve gravity wave drag and thereby improve the representation of circulations in the atmosphere so do you think that in the end all we need is higher in higher resolution models and then we'll be able to resolve all the processes and then that's the answer well in terms of certain so it depends on the kind of uncertainty of course there's always the uncertainties that we won't know about like the future for things end and the biosphere will be a more difficult nut to crack but I think if we believe in the laws of physics we do have to believe that as we get in principle as you and as you have a better representation of things you will do better on the other hand it's not clear that there's really a convergence you know that that's not obvious things do get more energetic at the small scales right and so there's you know mathematicians could think about about this actually it could be that the optimal because the there is a transition from the planet they kind of let's say the weather scales and then at about what is it about a hundred kilometres or so actually if you resolve the thing then you get into a meso scale the spectrum which is very energetic minus 5/3 spectrum actually but at large scales it's not obviously that's a good place to be to be truncating and maybe the better thing is to really resolve these weather systems and then only parametrize and not try to get into this other range but of course the weather models have been getting into that range so we do have some experience from them you know I I'm sorry is I the answer might be yes in principle but you know how far do we have to go the people that work on an aerosol process I mean I have some some Graham Feingold very famous cloud guy he said you know how what resolution do we need before you leave before you think we've models and he said without batting an eyelid 10 centimeters and he was absolutely serious so you know I if the question is if we're talking about a practical answer to the question I think I think we can't wait for that and we have to work with thee and there could be good theoretical reasons actually to try to parameterize the really stochastic parts of the process as you say we can't wait for all these things we sort it but Ted I want to thank you on behalf of everyone for taking us into this round where fluid mechanics starts become dominant again the circulation that actually gives us the climate variability and change at a point where we actually live and thank you very much for such an interesting talk

Channel: Imperial College London

Views: 8,126

Rating: 4.4626865 out of 5

Keywords: imperial, college, london, science, university, UK, climate change, atmospheric physics, lecture, physics, United Kingdom (Country), energy, Power, Global, Change, Warming

Id: 4AjCeXl5tE0

Length: 61min 14sec (3674 seconds)

Published: Tue Dec 11 2012