 So I have the unenviable just for lunchtime slot, trying to keep it to time. So I'm going to talk about some modeling experiments, very much in the, you know, continuation of what Hylen has just been presenting. And I'm going to talk about the influence of the extra tropics on, generally on propagating signals within the tropics. So this is work I've done with Severin, Thiebaud and Patrick Marcicello. And so I was going to start with a kind of just a list of references. So, but we've had so much kind of quality review material just up to now. I don't really need to say very much, especially on the models here. Hylen has just spoken about some of these modeling studies where they've taken the model of the tropical band and tried to look at how the boundary conditions can influence the simulation that you get in those models. And a lot of that work has been done, originally I think the first paper, which is the kind of progenitor of what I'm going to show is from Gustavs and Weir where they took MN5 in a regional domain around the maritime continent and found that the boundary conditions were crucial for propagating the MJO. That was kind of short case studies and then Palavre and Chidong Zhang did a series of papers on isolating different aspects of the boundary conditions and how important they are, not just the lateral boundaries but also the communication around the tropical band and they found that both of those things are important, but particularly the lateral boundaries, the boundaries with the extra tropics and they found that it was necessary to have those right in order to have a simulation which was capable of propagating an MJO. So what do I contribute to this? A lot of this stuff is kind of short runs case studies and it's almost like anecdotal if you're looking at a single case and so my contribution was to do a longer run 20 years and have a bit more of a thorough examination to try and bring out the systematic behavior, the generic systematic behavior of these models looking at a more statistical sample of boundary conditions. So what is our hypothesis? Which of these pictures do you think best represents the influence of the extra tropics on the tropics? Top left is a forced response. So is it a forced, do the extra tropics basically control what happens in the tropics? Or is it more like this one is stochastically excited internal mode, so these bells, they ring in a certain way, it doesn't really matter how you shake them, they're going to ring the same way. Or is it the girl on the swing, that's a resonant response, she needs to be pushed in a certain way and if you don't push her right she's not going to swing and she's not going to be happy. Or is it Jimmy? Jimmy, he can hold that perfect note, he can sustain it by having the right feedback with his amp and it'll just ring on. So is it an unstable internal mode of the tropics? So we can fiddle with the boundary conditions and try and at least eliminate some of these hypotheses and that's what we're going to try and do with a modeling study. So first the sort of caveat on these types of modeling studies though. This is reality, okay, this is the extra tropics and this is the tropics and we arbitrarily almost draw some lines which is the limits of our domain and we'd say this is our domain and this is external to our domain. And so from that conception we will interpret reality like this, we'll say in our domain we have the tropical solution and external to our domain we have the mid-latitudes which will supply boundary conditions for our experiment. So from that premise we will then perform an experiment and the experiment in blue now means it's a model, okay? So we have these boundary conditions from observed extra tropics and we have the model solution. So I don't know if you can see where I'm going with this but this is philosophically not robust because what if there's a problem with the experiment? There's no reason to suppose that the conditions on these boundaries is entirely the result of what's happening in the extra tropics. It could also be the result of the observed solution within the tropics. So maybe these boundaries should be red and not green, okay? So are we feeding our model domain with external influence or are we just echoing back the observed realization within the tropics back into our model for some combination of the two? We're not sure about that, okay? And none of these studies really solve this problem. So I'm going to call that the elephant in the room, okay? So I don't know if you know what that expression means. The elephant in the room is everybody agrees not to talk about this very obvious big thing which is there, we just talk about something else, right? So I'm not going to talk about it either, not to start with, but I will come back to it towards the end. This is our experimental procedure. We impose boundary conditions from observations in a tropical channel run. So where do we put these boundaries? I'm going to show you a short movie. It's short. It's just a case study. It's anecdotal. But this is the observed 850-minibar wind, zone of wind. This is a warf simulation, model simulation. I'll get to what warf means in a minute. With the boundaries at 20 north and 20 south, and then this is with the boundaries at 30 north and 30 south, and we're just to see if we can see an event propagating through with those two different choices for the positions of the boundaries and is it any different between these two runs? And it's very difficult to tell just from the case study, but here it goes. You can see something's going to propagate through here. And it's difficult to look at all three of them at the same time, but it'll cycle around. But there it goes, going through. And so it seems to happen OK in both simulations. There's a slight difference between the second one and third one, but I think we can be reassured that we can confidently try to simulate this kind of propagating system with the boundaries at 30 north and 30 south, which I think is the safer choice, given the caveat that I showed you earlier. I don't know if you noticed the subliminal message in this video. That was for a different presentation. I was trying to persuade people back home, the French, they loved their acronyms, and I was trying to persuade them to form a group on ocean atmosphere and teleconnections. All right, so what is our experimental procedure? Because we want to isolate the influence of the boundary on the tropical solution. So we'll do this with the WRF, Weather Research and Forecasting Model, with the boundaries at 30 degrees north and 30 degrees south. On those boundaries we'll impose the NCEP2 reanalysis. There's a specification in the model, it's a one degree resolution, and we'll do the following experiments. We'll do 20 year runs from 1993 to 2012, and for each run we do it twice. We'll do two experiments. I'll get back to that in a second. So what do we do with these boundary conditions? We have a reference run where the boundary is everything and it excludes nothing. So we just put the NCEP2 reanalysis on the boundaries. Then there's what we call a notch run. I borrowed that terminology from Goofstabson and Weir, who did a similar experiment. And on the boundary we put the dyno cycle, the synoptic scales, the annual cycle, the inter-annual variability, everything except this band of frequencies, the MGO band, which we've removed, so that's why it's a notch. And then we have a climatology run where we take away everything except dyno cycle and the annual... a repeated annual cycle, which is the same every year. So there's no inter-annual variability, no low frequency, or even synoptic variability, just this. Then there's a couple of other runs where we... these things are done to the boundaries and also to the C surface temperature so to isolate the separate influence of those two we do a couple of cross experiments. Ref star, which is like ref, but with SSTs from notch, and notch star, which is like notch, but with SSTs from ref, okay? Just to check that the SSTs aren't too important in our conclusions. And for each of these runs, we do it twice. So 20-year runs, one is with the initial conditions from the beginning of 1993 and the other, we just bang in the 1st of January 1994 as the initial condition, nothing else changes. And it'll be a shock, but it'll be a shock which the model adjusts to after a month or two. And after that, what we have is two identical runs with identical boundary conditions and the difference between two of them we treat as its internal variability in the model. So that's another way of looking at the importance of the boundary. The difference between those two twin runs is independent of the boundary conditions. And yet it's consistent with having realistic boundary conditions. So here's the kind of validation. This is the winter state of the simulation. So here are average fields for shading is precipitation and contours is low levels on the wind. This is the, here's the observations. Here's the reference run. So the wind is the right sign, well placed, a little bit strong. The rainfall is a little bit strong as well, but it's very well positioned. The notch run is very similar. And the climatology run is rather poor. And that is because it's missing those crucial synoptic timescale transient momentum fluxes which screws up the climatology in the tropics. And so we get a poor simulation because we're missing important fluxes on the boundary. So how about the variance? So here I'm showing you the reference runs and I'm showing you the two twin reference runs. And we're looking at the variance of 850 millivar wind again. And we see that there are places in the maritime continent where it is maximum. It's a bit strong compared to the observations. It's extremely similar between these two runs as you'd expect. This is a 20-year average variance. And this is the variance of the difference between these two runs. It's not the difference between these two pictures. It's the variance of these two different time series. If you take the time series of these two time series subtract one from the other, you'll get a difference time series. It's the variance of that. And so what do we expect? What do we expect it to look like? The variance of the difference between two things is the variance of the one plus the variance of the other minus two times the covariance. So imagine that these two things are completely independent. Then they'll have zero covariance. And this should be twice as big as this. Or imagine that they are perfectly correlated. Then this will cancel with these two and there'll be nothing here. Somewhere in between the two, what we have is a measure of how independent these two runs are. And the only thing that makes these two runs not independent the fact that they're not independent is because they have the same boundary conditions. So the degree to which they're not independent is also the degree to which the boundary conditions are important. So what do we see? We see that the difference here is not twice as big as this is not zero. It's about the same, right? Which makes us think that, well, the boundary conditions are important in determining the variance here. They account for about half of it. And the other half is internal variance. So that's one way of looking at it. Here's another way. Let's look at the notch run. Now the notch run has weaker variance than the reference run. It implies what I just said that some of that variance is coming from the boundary conditions and it's coming from the boundary conditions on those MJO time scales because that's what's been removed here, remember, for these notch runs. These runs are the same as the reference runs except we've removed MJO time scales from the boundary conditions. The difference between the two of them is closer to twice the variance here, okay? Which again, which implies that this variability that we're seeing here is basically internal variability. We've removed the boundary influence. Now let's just go back, okay? That's with full boundary conditions. This is... So the structure is quite similar, okay? But the magnitude has gone down by about half, okay? It's all very consistent. If we look at the climatology run, same story, the difference is twice as big and this is all internal, but the structure is totally different because the climatology is very poor. We've ruined the basic state by removing too much high-frequency information from the boundaries. So this is not a useful experiment. It's useful as a benchmark, but it's not a good experiment because it messes with the time mean and Ray and Zhang had this same problem with their work. So here's a summary slide. So run reference, notch, climb, and this is the variance and this is the variance of the difference. You see, this gets weaker. This gets stronger. Right, let's move on. Oh, yeah, a quick talk, a quick word about the mixed boundary conditions. So this is the reference run again. Remember, this is the ref star run. So this is the run in which I have the boundary conditions the same, but the SST is from this notch run. And you can see it looks very much like ref, which kind of implies that the SSTs are not the determining factor, it's more to do with the lateral boundaries. The notch star run is the opposite experiment. It's like the notch run except the lateral boundaries... sorry, except the SST is from the ref run and it looks like the notch run, again implying that it's the lateral boundaries which are determining the solution, not the SSTs. So how does it propagate? Is this model able to propagate the signal? This is a hovmula of correlation. One point correlation at this point here with the OLR. So what we see shaded in colors is the OLR and the contours are the zone of winds at low level. This is a lag from minus 20 to 20 days function of longitude. So we see, obviously we have a correlation of one here, the observations you see this OLR signal propagating eastwards with time and the wind structure, well basically you have low level convergence where you have enhanced convection and it's propagating at MJO speeds and it speeds up over the Pacific. This is a signal, the well-known MJO signal which we're trying to simulate with the model. Here's the reference. It has some aspects of the convection signal propagating. It does do that to some extent but not as well as you might hope. It's basically a large part of it is a standing oscillation but the winds do propagate and they do it in quite a realistic way. So we're fairly happy with the simulation, the dynamical simulation here. It changes its phase speed and it goes into the Pacific just like in the observations and we have these precursors here for the next event in the mid-Pacific. So that's the reference run. Let's look at all the runs. So again, here's the same one that we saw before. This is the... So in the next column is the twin run. So run one, run two and the difference between the two. See that the reference run, the twin runs look the same. The difference between the two does not have that nice propagating signal. It kind of dies over the Pacific. It's not able to keep it going. Then the notch run, it's not quite as bad but it doesn't have those strong precursors and it looks rather weak when you get to the mid-Atlantic and this difference field is not so different to these ones as that is to these ones. So we are missing important information from the boundaries and it's affecting the propagation. The climatology run is rather poor. So another way to look at that is to look at one point correlation maps and here we have many of them. So this is the simultaneous correlation of the low level wind with OLR in this box. So this is the Indian Ocean and you can see that you have eastlies here, westlies here. Because this is keyed on positive OLR which is suppressed convection so that's low level divergence there. But it doesn't really matter. It's a linear analysis. So then you see the propagation downstream after that and you see these precursors upstream of the correlation point. And you can see a kind of rospigia structure here in the response as well. It's like a response to deep convection. And the reference run gets a lot of that structure. The magnitude is slightly weaker than in the observations but it's all there. Then if you look at the notch run the simultaneous response looks similar but the upstream precursors are definitely weak. And the downstream propagation you don't have that same sense of propagation. So there's a difference here and it's consistent with the Hofwellers I just showed. So we can draw some conclusions now. So we don't believe it's just a forced response because we think those variance patterns in the notch run look very similar to the variance patterns in the reference run. So the structure looks as if it's internal to the tropics. It's not imposed by the extra tropics. And yet it is amplified by the signal coming in from the extra tropics, especially on MGO time scales. It may be that it's so the fact that I haven't ruled out the idea that it's some sort of internal mode that's electrically forced. It may well also be something that needs to be hit at the right frequency. That seems to be one of the conclusions from this work. And I don't believe it's just an unstable internal mode although Jimmy might make a comeback if we talk about global instability. That's something I haven't addressed with this experiment. So what are the next thing which is all right, that's all just to do with variance. I've lost my... I've lost my little squeezy thing here. It doesn't seem to matter. So that's all very well in terms of average variance. Does that actually translate into anything useful for prediction? So here is... We're going to start looking at different phases of the MGO now. And this is the sort of RMM diagram. We have two EOFs of OLR. This is from the observations. And projections onto those EOFs will follow around. This is the A phase, the B phase, the C phase and the D phase. And there's a kind of N zone here where the signal is weak, where the projections are weak. And what I'm going to do is copy some work by Adrian Matthews where he looked at the progression between these four phases of the MGO that he defined. And in particular he looked at initial events and successive events. So an initial event is an event which seems to just grow, just appear and then propagate. And a successive event is one that appears to be the continuation of a previous event. So the definition of an initial event which I've just taken from Adrian is that it starts with an N moving onto an A. So there's nothing beforehand and then there's an A phase. And then you have to have a complete sequence A, B, C, B. And that's called an initial event. Now a successive event starts with an A phase which comes from a D. So there was something there before and the sequence A, B, C, D again. So just remember that. How are we going to diagnose that from the model? Well, it's a little bit tricky because the model didn't have such a great propagation for the OLR. These things are normally defined in terms of OLR. So we're going to define it in terms of the wind because the model did fairly well for wind. So the first set of pictures here are for A, B, C and D phases and these are the composites of the wind for the four phases of those EOS which are based on OLR and the observations. So this is the observed wind associated with each of these four EOS. Well, the two EOS but four phases. We don't want to look at OLR in the model. So what we did was to recompute composites where each of these phases we take these patterns which we've deduced from these composites and we project onto those patterns rather than onto the original OLR EOS and we get some new composites and these are the composites from those new... those are the new composites which are now based on just wind and of course they're not the same it's not exactly the same every day but unsurprisingly it's extremely similar. This is still just the observations but now we have a basis based only on the wind it's not so we've left the OLR behind and then we can then project the model onto these composites, onto these same patterns and look at the composites from the model run and they also look quite similar to the extent that the model run is realistic but they're not the same, there are some differences. So now we can compare sequences time sequences of phase occupation based on this wind metric between the observations and the model and we'll see if you have this sequence of phases in the observations, does the model produce the same sequence of phases? How does it work? How does it do for the reference run? How does it do with the notch run and that information from the boundaries? This is the most beautiful figure I've ever produced I think I mean I think I just stopped talking for 20 minutes and let you look at it everything is here so this is all 20 years 1993 each line is a year to 2012 and the top stripe is the observations based on the OLR second stripe is the observations based on the wind and the third stripe is the model Now, lightening shades of grey denote ABCD phases if it's white it's just an end phase there's nothing happening and the coloured bands are when we have either an initial event or a successive event initial events are warm colours and successive events are cold colours so how well does the model correspond to the observations? This is a good example of when it works here's one of where it misses one you see so and here it hits quite well so we can start to gather some kind of traditional forecast statistics on these type of things and well I don't want you to look at this table I've got a much nicer way of presenting it but here is the data here's where we got we collected all the information and decided to assess the result according to three questions that we asked the first question is how does the model do in just getting the phase right including the end phase where there's nothing happening the second question is well regardless of phase does the model manage to get if there's an event or if there isn't an event so that's just n or not n and then the third is given that there's an event how well do we get the phase and so for each run first of all there's just the occupation the phase occupation which is pretty much even we tune the thresholds in order to make that sound then there's the score for a BGC n so this line gets 100% in the scores so it's based on that this line tells you how well the O&R metric corresponds to these are how well the model is doing and so in brackets you have a Monte Carlo simulation where the years have been scrambled and it's been worked out thousands of times and you expect to be somewhere near 20% for here somewhere near 65% this is what you get from random charts and so you can see if the model is doing better than charts and it is for the reference runs that's all for the other runs it rarely does any better than charts and not significantly so okay the first one is right to virtually the same does it mean initial conditions do not take a role at all? absolutely I don't think the the initial condition is just a kick tropical initial conditions do not take a role at all? no because this is a 20 year run so after a month or two you don't care the only reason to have that change is to produce two different runs anyway that's that's not easy on the brain so this is a nice way to look at it this is the percentage of phase occupation for observation, reference the difference between the reference runs the notch run and the climb run and here are my three questions how well does it get ABCDN compared to chance and the answer is pretty well for the reference run but really not very well for the others okay how well does it get in event regardless of phase slightly better than chance for the reference run no better than chance for the others and given that you have an event how well do the phases progress one after the other how well do they line up with observations and the answer is quite well for the reference run a little bit better than chance for all the others as well and I think that's just because it is an internal mode of variability so if you have an A phase slightly to be followed by a B phase you've only got to get one of them right and the others will fall into place so it's just luck that the event is in phase but given that the event is in phase the rest of it should follow so I think that explains these results but basically the reference run rules and that is another confirmation that you need that MJO timescale on the boundaries to get any kind of correspondence between the run and the between the simulation and the observations so here's the event skill so I talked about hits and misses there's this vocabulary for forecasting hits and false alarms so the number of primary events we have here the number of times the model produces a primary event a number of primary event I think I called it an initial event it's when one grows from nothing and the model produces this number of which this number corresponds to what happened in reality these are false alarms and well it's not very impressive is it for these initial events it's barely any better than the other runs for successive events that's where you have a previous event coming through when the model produces one and follows it it is obviously better than the other runs so this kind of fits in I didn't talk about this but the expectation is what Adrian said in his paper is that initial events tend to start their life in the tropics whereas successive events have some input from the aftertropics and that I think goes back to what Eileen was just talking about this extra tropical route of influence and so you'd expect an initial event to be more influenced by the boundary conditions and initial event the boundary conditions don't seem to be helping very much so I think this sort of still you don't have much skill here but the little extra skill you have compared to these runs is consistent with that idea now if we turn it around these are observed primary events of which the model gets most of the time but it gets hardly any better than the other ones whereas for the successive events there's clearly a difference between the model the reference run and the others so conclusion how much longer have I got 15 minutes the Worf model produces propagating tropical signals that are weakly coupled to convection twin experiments can preserve the model climatology and produce clean assessments of the relative strength of internal versus boundary force variability for a variety of boundary forcing frequencies now this is I quite like this approach because you don't suffer from the difficulty of having the wrong boundary conditions and so the climatology of the two runs is realistic and yet you still have a measure of the boundary condition of the independent part of the solution our experiments point to an important role for MJO band extra tropical disturbances in triggering propagating tropical disturbances now we weren't sure what to expect here I mean there is some there's some work that shows that stochastic forcing is also important so synoptic scale influence adds some skill but that's not what we found boundary influence appears to provide an organized upstream precursors that's from those Hoffmullers particularly for convectively coupled signals now I didn't show this but everything I was showing in those correlations was keyed onto OLR you can also just look at correlations with the wind and you don't get such a clear answer it just looks like Kelvin wave is going past so the fact that we're keying onto convection here is an important aspect of this conclusion hindsight skill is poor but clearly influenced by the boundary conditions especially for successive events not really for the initial events right so was all this just a badly posed experiment are we just looking at an echo chamber where we're providing influence from the tropics to simulate the tropics well I would say no for a couple of reasons let's go back to this picture so these precursors are clearly upstream and it's hard to imagine a physical mechanism whereby the tropical solution will influence the boundaries upstream at a previous time so that's in my defence further to that well I think this is I thought I'd show you this because I think this is really neat this is a paper by Adamis Itzel and I think it's a way of tackling this issue because what I've done is just to put the reanalysis on the boundary so we put the winds on the boundary the fluxes on the boundaries this is another way of cutting up the solution in the tropics into different components and what they've done is they think about vorticity and divergence as if they're sources of the flow so the vorticity is like the charge for an electric field which is a stream function and you look at the vorticity everywhere you can invert it to find the flow where you need two things you need the vorticity and the divergence to get the rotational and divergent flow but you can look at that vorticity in different regions and if you look at the vorticity for example just in the extra tropics and invert it, it will give you the flow globally including in the tropics so you'll have a tropical flow which is independent of the tropical vorticity and divergence so this is what they did they just looked at the extra tropical vorticity and divergence and inverted to find the flow in the tropics and they call that the background flow and it's quite weak it's totally different to the full tropical flow but you could attribute it the extra tropics objectively in a certain sense and this is their MJO when it's over the maritime I can't remember which phase it is and this is just the background flow which is consistent with the full tropical flow and also they have all the teleconnections in this diagram but that's simultaneously with presence of convection over the maritime continent they also showed the usual 8 phase diagram so what we're looking at here is just the background flow they also did it for divergence in the tropics and they separated those two components as well it's a really neat piece of work but we're just looking at the background flow because it represents the flow in the tropics which is attributable to the extra tropics and well these are the 8 phases of the MJO so if you look at this one for example where it's over the Pacific there's a very strong extra tropical load associated with that background flow and so that's the influence of the extra tropics in that particular phase of the MJO which you can then imagine propagates and here's something I did earlier just while Andy was giving his talk same MJO phase so it's phase 8 and this is this low so it's also present in the IRA data set so you don't need to read that baby to do it yourselves so this is a fairly strong westerly in the tropics just associated with these stream function anomalies and we can attribute it to the extra tropics right another thing and this is really going back to what Highland was talking about earlier in which I will talk about in great detail tomorrow morning okay so too much about this slide you're going to see this lots of slides like this tomorrow morning but we have a model we have a simple GCM which is basically the GCM that I'm in was showing I've added a seasonal and annual cycle to it and so if you can just think of this as the observations phi is the observations psi is the model and the forcing on the right hand side of these equations is defined by using the observations as a sequence of initial conditions for the model itself now I don't expect you to understand that this quickly because it takes a while to grasp this but I'll go into this tomorrow so what we have is some long runs which include an annual cycle and so here's the NCEP2 data which we used to calculate the forcing and model so you can see the model has a reasonable climatology in winter and in summer and so let's look at the tropical variability in that long run so what we're interested in is the influence of the extra tropics on that tropical variability so rather than just imposing the boundary conditions we can nudge the extra tropical band the extra tropical zones towards the observations okay I guess there's only like three people in the room at most to understand this cultural difference here let's we're going to do a nudging experiment anyway and so here's the NCEP2 reanalysis so this is a really cool spectrum of 850 millibar wind and so you see Kelvin waves you see Rossby waves you can see this big potato here which is the MJO this is a free run of the model with an annual cycle now there's not much of an MJO there there's something and I'm not Highland was showing you something earlier on and he filtered out all these very fast Kelvin waves that the model has and just concentrated on what was left here and it has remarkably a similar structure to the observed MJO but it's not it doesn't hold a candle to the strength of the real MJO at least not with this measure the 850 millibar wind okay so you don't have much of an MJO in a free running model so there's no nudging here this is the model's own solution because so then let's nudge the extra tropics this those are nice external Rossby waves their Rossby Harwitz waves yeah there any extra tropics sorry something in the sponge layer so there's an MJO there there's an MJO there in the observations I'm saying there's no MJO there with apologies to Highland saying there's no MJO there not to the same kind of entity in the model in the free model what happens if we nudge the extra tropics so we nudge them outside 30 degrees north and south and cute things on this again no no difference so we've got observations outside 30 degrees north and south not much happening in this dry primitive equation model so let's bring it in a bit bring it a bit closer outside 25 degrees north and south and look what we've got see now where's that coming from is it the model's dynamics producing an MJO I don't think so I think it's just that we've got a bit too close with the observations and that's what the spectrum is picking up even though the spectrum is calculated within 20 north and south so I'm not just there's no observations in this spectrum here but it's influenced so that's a kind of rough idea of don't go too close with your boundary conditions but close enough and of course I think that would be model dependent resolution dependent for all that sure I'm not saying there's nothing from the extra tropics here I'm saying that there's something from the tropics you see there's a risk here that we're polluting our solution with observed tropical okay I get what you mean it could be extra tropical but it's observed extra tropical yeah sure but it's a difficult one I mean where do you draw the line that's my title here and I quite like this approach where they they talk in terms of the vorticity and divergence which is effectively what we're doing here because this model solves the equations for vorticity and divergence I'm nearly done alright so this is off topic for this talk but it's not off topic for this meeting so I'll just finish with this you can do it the other way round rather than nudging the extra tropics you can nudge the tropics the model solution in the extra tropics and so here's just a few quick experiments where this is the DJF what is it it's the kind of faux geopotential on the 20 to 90 day frequency band so it's low frequency variability of the geopotential high field at 250 and this is for DJF in the observations this is a free run of the model so no nudging of the tropics the model has its own tropical solution and then if you nudge the tropics this is the variance that you see in the extra tropics and you can see it's quite a bit stronger much closer to the strength in the observations so it seems that having a more realistic solution in the tropics has an impact on the extra tropical solution at least in winter not so much in spring although there is some influence or in summer although it seems to be a recurring theme in these things that you have this strong variance here which you didn't have so much here and that seemed I think there was you David showing that earlier that's one of the best responses that the models have to an MGL is to this variance in the north pacific and I've reproduced that with the simple model as well okay anyway that was just a kind of show and tell which has nothing to do with the rest of the talk that's where I'm going to stop I think