 Thank you. So our next speaker will be Amy Butler. She is a research physicist at the NOAA Chemical Sciences Lab here in Boulder. She also kindly gave a talk at the Colloquium last, well, two, four and three weeks ago. And today she will be talking about quantifying stratospheric biases and the role of stratosphere-troposphere coupling in S2S models. Thanks very much, Amy. Over to you. Great. Let me just share my screen. Sorry, I'm having little issues. All right. You can see it now. Okay. Let me hide that thing. All right. Can you see it now? Yes. Okay, great. Thanks so much for that introduction, Judith. Today I'm going to be talking about two international projects to quantify stratospheric biases and the role of stratosphere-troposphere coupling in S2S models. And I want to acknowledge my co-authors on this talk. I am Garfinkel, Zachary Lawrence, and Peter Hitchcock. So as I talked about in the lecture a couple of weeks ago, stratosphere-troposphere coupling processes have been increasingly linked to global extremes. And this is a chart from Jomaiuson and Butler 2020 in which we highlight the ways in which the stratosphere has been found to influence tropospheric extreme events. And this extends beyond just what we typically hear about in the media in terms of the stratospheric polar vortex influencing colder outbreaks. There's a number of processes in the stratosphere such as wave reflection events, the quasi-biannual oscillation in the tropical stratosphere, vortex weakening in the southern hemisphere, and chemistry climate feedbacks that have been found to influence tropospheric extremes. And these have a broad range of impacts on infrastructure, health, resource extraction, shipping, and agriculture. But perhaps more importantly, stratosphere-troposphere coupling processes are important to S2S forecast skill. And this schematic is highlighting some of the processes that we think are relevant on these time scales. And in particular, the stratosphere has in many of its processes longer memory than weather. And so we are able to assume that we can get increased S2S forecast skill at least during windows of opportunity when these processes are occurring. But there remains a number of questions about these. For example, what are the biases that are linked to these processes or perhaps to the poor simulation of these processes in S2S models? And also, how can we isolate the role of the stratosphere on S2S predictive skill? Obviously, there's a lot of other things going on such as the MJO that we heard about this morning that could also be driving some of this skill. So picking apart which component is actually attributed to the stratosphere can be tricky. So to answer some of these questions, we are part of a stratospheric network for the assessment of predictability, which is called SNAP. This is a WCRP spark international activity. And it's also the stratospheric sub-project of the S2S prediction project. And I'm one of the activity leaders, and my colleague is Chaim Garfengel at Hebrew University. And this is a list of our steering committee. And the goal of this group is to assess stratospheric predictability and its tropospheric impact. And to that end, we've been doing a number of studies over the last few years. This is the most recent one that was led by Daniella Domisen at ETH. And this was published in a two-part paper in JGR. And it was looking at the stratospheric predictability in the S2S models. And in this study, we divided the models into high-top models, which we defined as having a model top above 0.1 hectopascal and several model levels above 1 hectopascal. And so in our study, we found five of the models we were using to be considered high-top models and four to be considered low-top models. And we found that the high-top models show higher stratospheric predictive skill and that they better capture stratospheric pathways of teleconnections. But today, I want to focus on two current ongoing collaborative projects that kind of expands on these initial results from Daniella's papers. And so that you all be talking about, the first is one in which we assess stratospheric biases in S2S forecast models. This is being led by Zachary Lawrence at NOAA PSL. And it involves a wide range of researchers from 11 different countries. And I'm going to be showing some of the initial results of that study today. And then the second study I'll briefly touch on, but it's more in the initial stages. And it's called the stratospheric nudging and predictable surface impacts project, or SNAPC. The goal of this project is to use targeted experiments to quantify the contribution of stratospheric circulation to forecast skill. And this is being led by Peter Hitchcock at Cornell. And so far, 11 modeling centers have expressed interest in doing these experiments. Many are from the S2S prediction project, but there's also some sub-X and ME models wanting to participate. All right. So before I talk about that experimental project, we're first going to look at the first project assessing stratospheric biases in the S2S forecast models. And there's a broad range of results. I'm just going to highlight some of the main ones. And so sort of our broad point here is that S2S models do show systematic stratospheric biases, which could impact their potential for providing improved predictive skill. And so in this plot, I'm showing the multi-model wintertime zonal mean temperature biases. And here we're comparing to error interim. And the left plot is showing the high top models. And the right plot is showing the low top models. And the top row is showing day one. And the middle is day 14. And then bottom row is day 28. You can see that the biases stay pretty consistent. It's not dependent on lead time, but they just get amplified over time. So they get larger with lead time. So what we find is that in the wintertime and really in all seasons, we see an overall warm stratospheric bias. And because this is a global bias that doesn't have meridional temperature grading involved and it also is not seasonally dependent, we can assume that these are due to some sort of radiative process. And so likely it has to do with the ozone parameterization or the water vapor in the model. We also can identify some dynamical biases. And that shows up particularly in the low top models where we see very cold pole bias in winter. And this is also true during JJA in the southern hemisphere. And this is associated with a two strong polar vortex in the low top models. And so likely one of the things affecting the polar vortex, either the waves that are driving variability in the vortex or the mean state of the vortex itself is not being simulated correctly in the low top models. And then finally, we have tropical biases which show up particularly in the low top model as two cold temperatures near 10 millibars and two warm temperatures near 100 millibars. And we think this is associated with the QBO. And in these models, I should mention, none of them really get an interactive QBO, but they're all initialized with the QBO wins. The low top models in particular struggle to maintain those wins over time. And more work will have to be done to disentangle exactly what these tropical biases are from because there's potential for both radiative and dynamical sources in the tropics here. But to further assess the biases in the QBO in these models, we've looked at a number of metrics. And here's just one example from one model, the CFS V2. And this is showing histograms of the equatorial zonal means zonal winds at 50 hectopascals. And so the gray shading here is showing the histograms for era five, reanalysis. And so clearly we see the bimodality of the QBO. And so normally the winds are either easterly or westerly in the tropics at 50 hectopascals. And this is a, we then do this for a function of lead time. So week one, two, three, four, five, in CFS V2. And so what we can see is that when the winds are first initialized close to week one, the model really captures the bimodality of that. But as we go further away from those initial conditions, we degrade the bimodal distribution of the tropical winds, starting with basically week three and going well into week five. And you can see that they're both the easterly and westerly phases of the QBO are collapsing towards sort of a weak easterly state. And this has influence on the QBO teleconnections in the S2S models. And this is an updated plot from Garfinkel et al. 2018. And here we're showing three metrics related to QBO and its teleconnections and how they evolve in the weeks after initialization. So it's plotted against the week after initialization here. So the left column is showing the individual S2S models and the right plot is showing the air interim data. And here we're looking at differences in these metrics between the easterly QBO and the westerly QBO at initialization. And so for example, here the top row is showing the 50 hectobascale tropical winds, so just basically the QBO itself. And so by construction in the air interim data, what we see is more easterly winds in the tropics compared to westerly winds. And it persists for all the weeks after initialization. And if we look at that in the S2S models in panel A here, we can see that that degrades pretty quickly in a number of the models as the winds tend toward this week easterly state. So we're rapidly degrading our QBO signal in the tropics. This has some influence on the local temperatures at 100 hectobascale near the tropical triple pause, which is shown in the middle row here. And you can see that normally in the air interim, we see colder temperatures for easterly QBO relative to westerly QBO, which is also a very persistent signal. But in the model, some of them managed to hold on to those cold temperatures fairly well from initialization, but especially the low top models tend to lose those temperatures quite quickly. And this also has implications for teleconnections via the Holton-Tann relationship to the pole. The bottom line shows the polar vortex winds at 60 north and 10 hectobascals. And in the reanalysis data, we see that there's on general more easterly winds of the polar vortex during easterly QBO relative to westerly QBO. And we can see that in the S2S models, they capture this relationship in the first couple of weeks, but by week three, it basically completely degrades very rapidly. And this might explain why we don't see a very good QBO influence on the surface in the S2S models. The final thing I want to talk about is the influence of bias on northern hemisphere sudden stratospheric warming risk. So sudden stratospheric warmings can be important for forecast windows of opportunity and generally, well, and occur when the climatological westerly zonal winds at 10 hectobascals and 60 north become easterly. And this means that these events are based on absolute, not anomalous values of zonal winds. And the threshold value of zero meters per second that we use to define them has really important dynamical implications because planetary scale waves can't travel and easterly flow. And so the next time you get a planetary scale wave packet, it breaks at a lower level. And so it has a big influence on how well the signal propagates down towards the surface. So here, what we're showing is the probability of getting an easterly wind for any given day as a function of each target week. So each column is a target week starting on the left with days one through seven. So week one, two, three, and four over on the right. And then we do this for each northern hemisphere winter month from November in the top row all the way down to March in the bottom row. And each line shown here is one of the S2S models. The probability of those models detecting the sudden warming and each little diamond is the error interim risk value subsampled to the S2S initialization dates. And so what we see is that here we're not doing any bias correction on the zonal winds to detect the sudden warmings. And without bias correction, the S2S models generally tend to under predict sudden warmings in early winter, particularly in December and really in January. And then they tend to over predict the risk in late winter at longer leads. So if we're looking at week four in February and March, it's tending to see more sudden warming center actually occurring. So we might think, well, let's just bias correct those winds and we'll have a more realistic estimate of the sudden warming risk. And so that's what we did in this panel. And indeed, the bias correction does seem to improve sudden warming risk in most months, particularly in January, it gets a lot better, more accurate estimation of that risk. But it does raise the question and Charlotte brought this up regarding the MJO, but the question is, how do we adjust the response following the sudden warming that's detected with a bias correction? For example, if the SSW was only detected after the bias correction, the models did not fully capture the dynamical impact that would have occurred if the winds had truly reversed to below zero meters per second. So it's really unclear how to deal with those dynamical influences down the line. Any one minute or so? Okay, let me just finish up here. I just want to, I've already stated these, so I just want to say this project's in its final analysis phase, and we hope to have a publication on it shortly. And I want to thank the SNAP Stratospheric Biasis team for all their efforts. I did want to touch briefly on the SNAP-Z, because I think it's a really cool experiment. And these are modeling experiments, which will isolate the role of the stratosphere on predictable surface impacts. And to do this, we picked out three events to focus on. And this includes two northern hemisphere major sudden warmings, which are shown on the left plot here. One is the February 2018 event, and one is the January 2019 event. And then we're also looking at the southern hemisphere minor sudden warming in 2019 that was followed by really bad brush fires over Australia. And so all these events were followed by well-known extremes. And the question is, how much of this can really be attributed to the stratospheric event that was ongoing? So to do this, we're targeting these three events, and then we're doing four experiments. The first is just a free-running ensemble forecast, so a normal S2S-style forecast run that's initialized. Then we're doing a nudged experiment, and in both the nudged and controlled experiments, only the zonally symmetric component of the stratosphere is nudged. And the intent is to prescribe the zonally symmetric component of the stratospheric flow without indirectly constraining either the troposphere or the planetary waves that play a central role in the coupling between the two. So we let the waves of all freely, but the zonally symmetric component is nudged. In one case to the observed evolution, and in the controlled case to the climatology. And then finally, we have a full run where the entire stratospheric circulation, not just the zonally symmetric component, is nudged to the observed evolution. And we'll have two initialization dates for each event. The goal here is really to highlight or take advantage of a large number of ensemble members, so we're asking for 50 to 100 members to really be able to detect the signal above the noise. And so we have a later date that will be focused on setting the surface extremes after the event, and then an earlier date to understand the evolution and predictability of the stratosphere. And these details will be available in a publication and prep for GMD. We already have a number of working groups, and I'm not going to go into these too much, but if you're interested in being involved, please contact either me or Peter Hitchcock or I am Garfinkel. We are happy to get people outside of the stratospheric community involved in this, and actually we're searching for people outside of the stratospheric community to be involved in the analysis of these runs, which we think will be useful for a broad range of things, including QBO-MJO relationships. So to conclude, SNAP is an international group of researchers, and we have the goal to investigate the role of the stratosphere and S2S prediction. We have two current projects underway, and we hope these projects will advance the state of knowledge about biases in the S2S models and allow us to better quantify the contribution of the stratosphere to surface predictability. So please feel free to contact me if you're interested in getting involved. Thank you. Thank you very much, Amy. On terms of extremes events and generosity, Amy and I were once on a plane, and we were flying into Japan, into a typhoon, and Amy kindly shared her hotel room with me, so I didn't have to spend the day on the floor of the airport. So thank you so much, Amy. We have time for one or two quick questions. Thanks, Amy, for a thought that talk was really good, and I'm excited to see where these projects go. I wonder if you looked at all that relative humidity or humidity biases in the S2S models. You mentioned, I think, at one point that the stratosphere, maybe there were problems with the water vapor that could contribute to some of these, but did you, or do you plan to look at that in more detail? Actually, we haven't looked at that, although one of our goals of the SNAPC experiments, we are asking for output along those lines to try to better assess that, but we haven't looked at it in the current S2S output, but that's a good idea. Thank you, Anish. Thanks, Stuart. Thanks, Amy, for the really interesting talk. My question was related to the nudging experiments that you're doing. Will you be saving the nudging tendencies as well? Is there an effort in terms of both looking at the nudging tendencies to identify model errors, but also how to improve process representation in the model? That was one question, and then related to that, how good are the observations that go into the reanalysis towards which you're nudging? Is there uncertainty coming in from the reanalysis as well? Yeah, so in terms of the first question, I actually have a slide on this. So we have one of the limitations of the current S2S output is the lack of stratospheric data available, and S2S prediction project has three levels in the stratosphere, which we're basing all our analysis of the biases on, but that's pretty difficult actually to determine all the things going on using three levels, and that's even more than most of the other projects have like SubEx or NNME. And so for these SNAPC experiments, we're obviously requesting more output in the stratosphere and specific fields to help us, and that includes a lot of the tendencies that are going on, and to try to better understand exactly what's occurring when we have one of these events. And so yes, basically it will be available. The other thing is that for this experiment, we're encouraging everyone to use the same initializations based on this error 5 data set that Tim Stockdale has put together for us. And so hopefully there will be no differences in the initial conditions in these experiments, which will help narrow, get rid of that possibility of it causing problems. But yeah, generally I would say what we're nudging with is pretty good for especially the dates that we're looking at. So I don't think that will be a huge source of error in these experiments. Great, thanks. Yeah, thanks so much.