 Okay, so roi gydag fawr yn gwybod i fyfnol o'i gweld chi ddechrau eu gweld o'i hawdd yma i'w Llywodraeth Magdyn Juliol Llywodraeth i'w gydag a'u gydag yma ar gyfer'r Cymru, a'r Cymru yma i'w Llywodraeth Ym Mhwy. Fydiwch yn gweithio yn ymdwyll ar gyfer Mjaw. I'm not really going to go back into what the MJO is or anything like that, so if you want a refresher on that now, stick your hand up. So the MJO, Eastwood Propagating Organised Convection in the Tropics, has a period somewhere between 30 and 60 days. The convection is most active over the Indian Ocean and West Pacific Warm Pool, but it has sort of influences on the convection over South America and Africa as well. The heating anomaly is associated with the MJO, excite rosby waves which propagate into the extra tropics and have influences on, particularly strong influences on the weather over North America, but also affect the North Atlantic and the Southern Annular mode, so variability over New Zealand or South America. And it's very closely associated with active break phases of the Indian monsoon, the Australian monsoon, for example, and it's a major mode of variability on sub-seasonal timescales in the tropics, and it's probably our largest source of predictability on sub-seasonal timescales in the tropics. So, given that, it would be really nice if our GCMs that we use for weather prediction on these timescales were capable of simulating the MJO well. So, that's what I'm going to discuss today. I'll say a little bit about the history of modelling the MJO, and then I'll talk about the challenges that we face and why it's a difficult thing to do. And as part of that, I'll give two examples of process studies based on work that we've done at Reading, but there are things that are done widely in the community. I mean many people will have done studies similar to this in one way or another. I'll talk a little bit about high resolution modelling of the MJO, and you might have heard something about that last week. I think you had a talk on high resolution modelling last week, but I didn't see what was in it, so I don't know. And then I'll talk a little bit about the MJO in current sub-seasonal prediction system. Okay, so we don't have a very good track record on modelling the MJO in our climate models. And that's been going on for many, many years. So the first sort of really comprehensive assessment of the MJO in climate models was paper by Julia Slingo. Oh, the wrong buttons. A paper by Julia Slingo. She analysed 15 GCMs that participated in the AMIP project, the first model into comparison project. So for those of you that are not familiar with model AMIP and CMIP activities, they have a very strange numbering. I'd missed out four altogether, and AMIP 1 and 2 became CMIP 3, so it's all a bit confusing. But they found that most of the models weren't able to capture some of the essential features of the MJO. It's period, it's amplitude, or even any evidence of eastward propagation. What they did note in their paper was that the best models tended to have convection schemes closed on buoyancy, rather than moisture convergence, and perhaps maybe I'll come back to that in a little bit. So how many people know how a convection scheme works? Adrian does. Yes, I'm glad he does. OK, so perhaps I need to say something about convection schemes then. I didn't really think about that beforehand. So basically most convection schemes these days, anyway, work on the basis of, you have some model of a convective plume, or an ensemble of convective plumes, that take water vapour and temperature out of the boundary layer and transport it vertically through the model. And you're kind of modelling it a bit like at the plume of an individual cloud, but you're trying to represent an ensemble of clouds. And the mass that goes up is compensated for by subsidence in the environment, and the combination of the subsidence in the environment and mixing from the cloud to the environment gives you the tendency on temperature and moisture on the large-scale environment that's designed to represent the effects of convective clouds. And so the core of a convection scheme has two things. One is how do you model the plume going up, and then how do you decide how much mass is going up in the plume. And the closure is the bit that decides how much mass is going up in the plume. So there are ways of closing it. So often these days it's closed to try and remove the cape in the profile over a given time scale. But other closures include closing it on the amount of moisture convergence in the boundary layer. Or sometimes, especially in the old days, it was closed on the amount of buoyancy that a parcel has lifted over a small part of the plume, over the lower part of the plume. So the closure scheme and changing the rate at which you remove the cape or the amount of mass that goes up in the plume is one way you can control the strength of convection in climate models. It's not a simple way to do it because you twist the lever to make it remove the mass and the whole system adjusts. And so you don't get what you expect generally. So anyway, Julia and others found that schemes closed on buoyancy rather than moisture convergence did better. And remember that. They also found that models with a better representation of the mean state precipitation tended to simulate a better MJO. And this was in the early days when climate models were still actually quite poor at representing many features of the mean state in the tropics. They're still quite poor, but perhaps not as poor as they were 20 years ago. So 10 years later, Lyn et al did virtually the same kind of study on 14 models in the CMIP3 database. They found that most models had less than half the observed variance in the MJO part of the spectrum. They were unable to capture the eastern propagation of the MJO, so we haven't really improved since 1996. However, they note that the best two performing models have closure or trigger functions that depend on moisture convergence. Schemes closed on buoyancy rather than moisture convergence did best. And that's quite a nice summary of how much understanding we have of what you do to a convection scheme to make your MJO better, I think, in the 10 years I believe about what worked best has changed. So later, Lyn et al 2013 compared CMIP5 models, so this is actually the next one from CMIP3 because there was no CMIP4. And they actually found that there was a general improvement in the simulation of the MJO. There was more variance in the models. The relationship between the eastward and westward moving power had improved, so more models simulated some form of eastward propagation. But they still argued that only one in 20 models was really able to simulate a realistic propagation of the MJO. So there's still some way to go in climate models. So why is it so difficult? Well, the MJO relies very strongly on the interaction between the convection and the planetary scale dynamics. The MJO is a planetary scale oscillation, 10,000 kilometer length scale oscillation. Tropical convection happens on virtually every scale from individual clouds all the way up through mesoscale systems, but it's organized on the scale of the planetary scale MJO as well. And we rely on the convective and other parameterizations, the boundary layer, the cloud scheme, the radiation scheme, et cetera, to represent the smaller scales and essentially trust that they will interact with the large scale dynamics in the right way. So this idea about the convection organized across many scales may be one of the reasons why we struggle to do this. So this is a picture, a satellite picture from during the Toga Corps period, which is an observational field campaign in 92-93. And this is a large-scale satellite IR image of the MJO. And you can see the organized convection associated with the active phase of the MJO and then a region of suppressed convection behind it. If we zoom in on this box, this is a sort of false-color IR image, and you can see within this large organized system there are still many different scales of convective organization. So we're now sort of 2,000 kilometer scale. If we zoom in on this box here and look at just what looks like one feature at this resolution in this image, we can see that even this feature has got lots of scales of organization in it. So this is about the size of a climate model grid box. There's many NWP model grid boxes. So this is kind of the thing that we're trying to parameterize in a climate model. But it's got to interact with the systems that are going on here and on these large scales in a good way. This is just over the ocean. This is pretty much over the I-Met Boy region, I think. I don't know, maybe not because this is the equator here. So the I-Met Boy would have been down here just north of the Boy. So this interaction between convection and the large scale dynamics is why this is a difficult problem to do. So what aspects of convection might be important for the MGO? Well, we don't actually know, I don't think. There are a number of properties of the convection which we think might be important and for various different reasons. So there have been a series of papers that argue that the stratiform heating of the part of the convection. So this is convection in the upper atmosphere that tends to get stronger heating in large organized MESA scale systems in that region. It might be important because the vertical profile of the heating essentially determines the characteristics of the waves, the planetary scale waves that the convection triggers. So this might be important in determining the phase speed of the MGO and providing energy for growth. A number of papers have suggested that the shallow heating that you get ahead of the convection, the active phase of the convection, might be important because this shallow heating drives quite strong moisture convergence, more moisture convergence than is required to maintain the precipitation associated with that heating. So that can moisten the atmosphere, so that might be important. It might be that the shallow convection actually moysins the environment directly and there's some evidence that you take moisture out of the boundary layer, it goes up in the shallow clouds, some of it rains out, but most of it is just evaporated again into the environment and that moysins the lower environment and that might be important for the heating convection. And actually just how sensitive the convection is to the environmental humidity might be important. So if you imagine you've got this plume going up and if it's going up through a very dry environment it trains lots of dry air into the plume and that reduces the buoyancy of the cloud and so the convection won't penetrate as deeply maybe, but if it goes up through a moist environment it's in training lots of moist air and can remain buoyant for longer. So that might be an important aspect of the convection. It is entirely likely that no single one of these will determine the quality of your MJO simulation in a model because even if you do something to modify this you're almost certainly modifying this and this and this at the same time. And so it's very difficult to actually really pin down why we've improved our simulation if we do improve our simulation. OK, so what's the current status of MJO modelling? So this is about comparable to the CEMIP5 study. It's a little bit later than that. And this is some results from a big international into comparison project to not only assess the quality of MJO simulations in our models but try and understand why some models do better than others. So this is 27 simulations from 24 different models and the way I've plotted this here is I've plotted a regression of the inter-seasonal precipitation so 20 to 100 day band pass precipitation against a base point in the Indian Ocean. And so this is what that looks like in observation. So this is the base point and you can see nice eastward propagating signal in the enhanced precipitation and the suppressed precipitation following behind it. No, this is suppressed precipitation ahead of it. This is time going up on this so you've got a suppressed phase and an active phase and then the suppressed phase follows. And this is how all the models did. So you can see most of them still suffer from this problem of no real coherent signal on these timescales. Some have a westward propagating signal. Some look fairly stationary. And then there are a few that have eastward propagating signals. About a quarter of them have something that looks a little bit like an MJO in one way or another. So two of those models are the super parameterised version of the community atmosphere model. So this replaces all its sub-grid parameterisation schemes by running a cloud resolving model in every grid box to produce its tendencies. So this is the most expensive convective parameterisation in the world. It's about a thousand times more expensive than running just an ordinary convection parameterisation scheme. The model is about, well these days it's only about ten times slower than a climate model because we can paralyse it quite nicely. But it's a lot more expensive in the way of a convective parameterisation scheme. And this is a version of the same model where they've imposed a vertical structure of the heating from observations based on the diagnosed phase of the MJO in the model. Well, yes, they've cheated in a way. They've actually, in a way, what they've done is they've said we're going to replace one aspect of our convective parameterisation. That is the vertical profile of the heating by an empirical relationship with the phase of the MJO. It's not the observed phase of the MJO. It's the model simulated phase of the MJO. The other interesting aspect is the sensitivity to AC interaction. So this is the atmosphere-only version of SP-CAM. And this is the coupled version of SP-CAM. And that's a much better MJO. So this is the atmosphere-only version of the CNRM model. This is a coupled version of that model. But this coupled version has very big changes to the basic state SSTs in the model. So we asked them to re-run it with using the SSTs from the coupled model in an atmosphere-only configuration, and this is what you get. So this is really quite a nice demonstration that it's the role of the AC interaction that has improved the MJO, not changing the basic state. So in terms of trying to understand why, so Jang et al found no systematic relationship between the MJO fidelity and the basic state. So that's different from that slingo et al paper. And I think that's largely because overall the basic state has improved. And so what you were finding very early on was that if you had a really poor basic state probably meant you had a bad model generally. So you didn't do a very good job of the MJO either. Whereas now the basic state in most models is much better and so you're not sampling really bad basic state models in quite the same way as you were. They found no systematic relationship between the partitioning of the convection between the large-scale and convective rain. And this has been sort of related sometimes to this relationship between convective and stratiform precipitation. Be very, very aware that that is not the case. It's just the way your model partitions its precipitation. But it might have an impact on why you simulate an MJO better or not. So Adrian, some while ago, had a look at the role of this partitioning on equatorial waves in the ECMWF model and found that because the large-scale rain is ultimately tied to the vertical velocity whereas the convective rain isn't necessarily tied to that then the partitioning between these two things can change where the heating occurs in the phase of the wave and that can affect the wave growth. But it is not the partitioning between convective and stratiform part of rain in convection. And no systematic relationship with the vertical profile of the diabetic heating. So that was a bit disappointing. They did find relationships between MJO for the entity and a couple of sort of growth measures of climate models. So one was the relationship between precipitation and relative humidity. So you may have seen observational studies that show this nice exponential growth of precipitation with sort of column relative humidity. And they looked at something that was sort of similar to that and found that those that basically had well, they argue and I'm an author on this paper but I don't necessarily agree with the argument made in the paper overall is that this measure is sort of representing the sensitivity of the convection to environmental humidity. I think it's actually indicating whether the models are able to make a better dynamic range of humidity across their simulations. And then something called the growth moistability which some of you may be familiar with but it's basically a measure where the coefficient convection is at removing and the resulting circulation is at removing moist static energy from the column. So this is about how much moisture is exported from or imported into the column when you have a convective convection going on compared to how much dry static energy is exported or cooling of the column goes on associated with that vertical motion. So that's entirely appropriate in the boundary layer or something like that maybe but outside of the boundary layer the tropical atmosphere doesn't have a constant relative humidity. It can vary between 50% and 100% in the sort of tropical atmosphere on interseasonal timescale. It may or may not be more appropriate approximation on a if you're thinking about a climate change timescale, it's going to be roughly the same 70% on climate timescale maybe. So a number of studies looking at the relationship between MJA's fatality and aspects of the representation of the convection either in these large model into comparison projects which generally tend to find nothing in terms of a systematic relationship and if they do it doesn't necessarily agree with the systematic relationship that was found in other studies or by perturbing some aspects of the model physics rather than comparing different models. So very often these studies don't draw the same conclusions and very often the relationships that they find are between sort of growth properties of the model i.e. this normalised growth moistability or the relative humidity diagnostic and they're very often not directly relatable to the formulation of the physical parameterisation. So although they tell you something about if I have a model that looks like this in this particular way or this particular way then I might have a good MJO it doesn't tell you what you have to do to your parameterisation scheme to make it behave more like that. So I'm going to change pack here and talk about two sensitivity experiments that we've done at the University of Reading. One is the sensitivity to convective entrainment. So this is how much does my plume mix with its environment as it goes up. There are a number of studies that have done this in various ways as early as 1988 for example. This Bechtold and the Hirons paper are changes that were made at ECMWF during well possibly just after Adrian was there but maybe about the same time as Adrian was there that had a very big impact on the MJO simulation in their monthly forecasting system and this is the one that I'm going to talk about. So we perform some experiments in the Met Office Unified Model because before we started this is what the Met Office Unified Model of MJO looked like, climate model looked like compared to observations. So these are composites in the phase space of the Wheeler-Hendon diagram which I think Hylin would have introduced you to last week. So the MJO on average will propagate around in this direction and slowly decay into the middle and this is just for composites starting in each of the eight phases of the MJO. So this is what observations looks like in that diagram and this is what the Met Office Model looked like in that diagram. You can see that very quickly the MJO decays and it doesn't really propagate very well. So we tried a number of experiments in some hindcasts actually to explore the sensitivity to various parameters and convective entrainment is the one I'll focus on because it is the one that had the biggest systematic impact. So this is for an MJO event during the Year of Tropical Convection. You can see the eastward propagating precipitation signal and eastward propagating signals in the winds and this is how the Met Office Model simulated that event. Very little eastward propagation in the precipitation or in the winds. Not a very good simulation. This event in wheel of hand and face so we start here and it goes around like that and we're initialising it somewhere here actually I think. So if we change a number of parameters which is what we did one of them was the convective entrainment parameter and this is now the simulation of that event and you can see a much much better simulation of the MJO in this sort of hofmola diagram and I'm sorry there's too many lines on here because I only really want you to look at two. One is the red line which is this control simulation with the low entrainment and one is the blue line which is this high entrainment simulation. And you can see that in this sort of face-based diagram there's a much better simulation of the MJO. The other two were where we changed the convective or switched the convective momentum transport off and that actually improved the simulation in both of those cases for this MJO event but when we tried it over many MJO events it didn't really have a systematic effect one way or another. So we did that for many hindcasts and these are the composites for those hindcasts and you can see that increasing entrainment has quite a substantial impact on MJO simulation in this model. So what happens if we go back and put that in the climate model? So this is what happens if we put it in the climate model and you can see that although we're still not capturing fully the length of these the amplitude of these composites all the way through we have a much better simulation of the MJO in our climate model now than we had originally and we can see that generally it's still outside the unit circle five days into the simulation and the propagation is much more reasonable around the diagram. OK, so that's some sensitivity to convective entrainment and this isn't the first study that has shown that this is the case in models generally. So perhaps before we go on one aspect I will say though is that very often these models and many other modeling studies where you improve the representation of the MJO by perturbing the way the convection scheme is either sensitive to the environmental humidity or inhibition at low levels improve the MJO but makes for your climate model basic state mean state slightly worse and if you think about the entrainment change if you make the parcel entrain more as it goes up what it tends to do is make the convection not quite go so deep in your model for example and that means that you need to have a slightly colder upper troposphere to get your convection to go deep enough to be able to balance the radiative cooling in the upper troposphere so you tend to get cold biases at the the tropopause level and if you're a national met agency trying to forecast upper tropospheric winds that can have quite a bad for civil aviation and things like that that can have quite a bad impact on that. Okay so we'll move off from convective parameterisation so the role of AC interaction so again 1988 Krishnamurti identified a relationship between atmospheric intracesimal variability and oceanic variability on intracesimal timescales over the Indian Ocean in a specific region. Not a lot was done with that apart from maybe it was part of the motivation for the Togacore field campaign which was this large field campaign in the winter of 1993 which really stimulated renewed interest in the role of AC interaction in the MJO and there's a paper by Wello and Anderson which has a figure similar to this but without some of the observations on it and Heddon and Glick looking at the role of that and there are a number of other studies related to this field campaign and the AC interaction and the MJO. So what I'm showing here are the surface fluxes so if you just look at the black line this is the surface fluxes going into the ocean during this field campaign and you can see regions where there's net heating of the ocean and quite strong net heating of the ocean and then sort of little or no actually cooling of the ocean by the surface fluxes and these periods of strong cooling of the ocean are associated with the active phase of the MJO and this net heating is associated with the suppressed phase of the MJO which is the rainfall and this is the zone of wind stress and so the suppressed phase of the MJO is characterised by very light surface winds and so that reduces the latent heat flux out of the ocean and reduces the mixing in the ocean so you tend to get a sholing of the mixed layer and that means all the sun that comes, solar heating that goes in warms the ocean quite effectively which is what you can see happening here and you'll also notice that there's a very strong diurnal cycle during these suppressed phases and then the active phase you have very strong winds so there's lots of latent heat flux out of the ocean you have cloudy skies reduced shortwave radiation which is these blue ones and there's very strong mixing in the ocean so you tend to take all that heat that's accumulated in the operation during the suppressed phase and mix it down deep into the ocean and so that has a cooling effect on the ocean and mix cold water from below up so motivated by this there have been a number of modelling studies to explore the impact of coupling on the MJO I suspect some of the motivation was also that it was much easier to run a coupled model in 1995 6, 7, 8 than it was in 1989 so this schematic summarises how the AC interaction might work in the MJO so here's the active phase of the MJO you have easterly flow to the east of the active convection into the convection and westerly flow to the west of the convection and up a level divergent over the Indian Ocean and west Pacific at the time the MJO is active you have mean state westerly winds and so the combination of this easterly anomaly and mean westerly wind to the east of the convection means that you have very low winds and reduced cooling and you also have nice clear sky so you have lots of shortwave heating of the ocean so the ocean warms ahead of the convection behind the convection you have strong westerly winds on a mean westerly state so you've increased the wind stress you've increased the latent heat release or evaporation from the ocean so that cools the ocean and you also tend to have more clouds behind the MJO active phase and you do ahead of the MJO active phase and clearly there's reduced shortwave actually underneath the active phase so this tends to cool the ocean behind the convection and warm the ocean ahead of the convection and this warming ahead of the convection if you just make an argument about convection likes to sit over warmer SSTs would make sort of tend to favour the progression of convection over this region I don't like that argument because convection feels that SST through the surface fluxes and you've reduced the surface fluxes here so it's not quite as straightforward as that argument but there are good reasons to think that warming here might for example a warm surface here might generate a lower pressure here and that would increase convergence which could bring additional moisture into the environment so in a composite sense it's about a quarter of a degree but well no a quarter of a degree positive and quarter of a degree negative sort of thing in an individual if you look at an individual event when you composite you tend to average over the spatial and temporal variations in it you can see anomalies of the order of the maybe 0.7, 0.8 ahead of the convection no and I'm not sure there is such a thing as too cold for convection that's again a nice thing if you average but it doesn't take account of the fact that the reason it doesn't convect over water that's 27 degrees C is because somewhere there is water that is 28 or 29 degrees C if everywhere was 27 degrees C you'd still have to have some convection somewhere so that's a bit of a okay so loads of modelling studies I haven't listed the modelling studies that show that coupling improves the MJO because virtually 80 or 90% of them probably do but there are a number of studies that find results contradictory results they tended to be longer ago but there are a number of plausible explanations why AC why these couple modelling studies that find no impact might be right or why they might get this result and be wrong so it might be that AC interaction isn't really very important for the MJO Wojtek Grabowski made an argument actually that all AC interaction is doing when you couple of your model up for bad representation of the variability in the atmosphere you don't have very good representation of the variability in the atmosphere you stick some variability in the ocean and the model responds to that I don't buy that argument because it's quite a lucky coincidence that the way it responds to it is to produce something that looks like an MJO it might be that when you introduce coupling the errors in the basic state actually degrade the MJOs or the errors that you introduce in the basic state of SST might degrade the MJO simulation more than the positive impact that you're getting from coupling it might be that you just don't represent the AC interaction processes in your model very well either because you're not properly representing some of the important processes in the model so it might be the mixing in the upper ocean it might be the sensitivity of your surface fluxes to changes in wind because you've got too strong a drag coefficient in your model or something like that or it might be that errors in the basic state i.e. surface wind or EG surface wind errors lead to errors in the nature of a coupled feedback how many flies do I have to go back so if you haven't got a mean westerly wind in your basic state but you've got an mean easterly wind in your basic state then that will tend to enhance the latent heat fluxes ahead of the convection and you'd get cooling ahead of the convection and warming behind the convection due to changes in the evaporation and if that's the case then your coupling mechanism is going to break down and the other reason is if you've got an atmosphere only model that's got poor intracesional variability in the first place then you haven't got anything on intracesional time scales to organise the surface fluxes that will generate some kind of response in the ocean and then you've got no way for that to feed back on the atmosphere so if you haven't got anything to start with then you might not expect to get anything back overall the consensus now is that air-sea interaction does modify the properties of the MJO but it is not critical to the existence of the MJO the MJO is probably inherently an atmosphere a convection dynamics coupled phenomena in the atmosphere doesn't rely on air-sea interaction for its existence but the air-sea interaction does modify the MJO so I've mentioned this already when I was showing you these diagrams during the suppressed phase of the MJO light wings and clear sky conditions can lead to a strong diurnal cycle in sea surface temperature and so if you're trying to model those diurnal cycle in SST what I should say is this diurnal cycle actually has quite a strong rectification onto the means on inter-seasonal timescale so I'm going to do this from close up so if you can imagine without the diurnal cycle in SST the SST during this period would follow essentially the bottom of that red curve because the diurnal cycle is really just enhancing the SST in the top metre or two of the ocean so if you put the diurnal cycle on that tends to, the daily average SST will increase to somewhere halfway up these spikes and the impact of that on the diurnal on the daily mean SST during the suppressed phase is actually it elevates the daily mean SST by something like a quarter to a half of the degree so that's quite a big impact compared to the underlying inter-seasonal variability it explains about 30% of the underlying inter-seasonal variability in SST so obviously if you want to capture that diurnal cycle well you need to couple your atmosphere an ocean on a timescale that is less than one day and in fact because of the way that the coupling is done in models you really need to do it at least three hourly otherwise you get some aliasing of the diurnal cycle depending on where you are in longitude you need upro ocean resolution of the order of a meter or so because the diurnal warming only really happens in the top meter or two of the ocean and so you need to be able to resolve that otherwise the diurnal heating is distributed through five meters the top five meters of the ocean and you don't get a very big diurnal cycle and there we go this is about the rectification of this diurnal variation on to the inter-seasonal variability until very recently coupled GCMs typically had upro ocean resolution of the order of ten meters and coupled once a day so that's not going to give you a very good diurnal cycle of SST so in 2007 I did some experiments at ECMWF to have a look at the impact of this poor representation of the diurnal cycle in a set of hindcasts for the toga core in the ECMWF monthly system and these figures show the impact of that diurnal cycle coupling or diurnal coupling on the forecast skill of the MJO model for that those hindcasts and so here's a thing that you should think about when you're plotting your forecast skill you should have some reference is persistence of the MJO and so these are correlations for the two PC's that make up the wheeler-hendon diagram so if you just persist those PC's you don't get a very good MJO forecast this is essentially a forecast in an atmosphere only version of the model where we persist the SSTs and you can see even then there was quite good skill to pass out to sort of 12-13 days for this event a black line shows what happens if you use their monthly forecasting system of the time which was coupled to a dynamical ocean with 10 meter resolution in the upper layer they did couple every hour even then so they've got the diurnal forcing of the ocean but they don't get the diurnal response and then the blue line shows the skill if you couple to just a thermodynamic mix layer ocean but that had very high resolution so 1 meter resolution or 0.8 meter resolution was the top layer and it had something like 8 layers in the top 20 meters of the ocean and so you can see for this component of the wheeler-hendon index there's a marginal improvement as we introduce coupling and then introduce the diurnal coupling or the diurnal cycle of SST into that and this is the phase of the MJO that essentially, or this is the PC that essentially describes the variations between phase 4 and 5 over the maritime continent and phase 8 and 1 over the um this is the one that describes um the phase which is active over the Indian Ocean or active over the Pacific Ocean and you can see that for this phase in particular we get some quite extensive improvement in forecast skill up to sort of 5 or 6 days increase in time for which we have anomaly correlations greater than 0.6 many studies in forecast models which have introduced SE interaction have found that the coupling really improved the forecast skill for those phases which are active convection over the Indian Ocean were specific rather than for the Marathon continent active phase so here's just a quick role of basic state errors so this is a paper by Pete Innes um so this is again a regression type look at the MJO this is observations in OLR um and this is when Pete did some simulations in the a very old version now of the Hadley Centre couple model you can see that the we have quite good representation of the MJO over the um over the Indian Ocean but not very good representation of the MJO over the West Pacific and that's because at the time the MJO the basic state of the couple model over the West Pacific this is actually a degradation compared to the atmosphere only version of the model in this region had a mean East Belize in the West Pacific region so if he does an experiment where he does a flux correction of the model um then you can actually recover some of the propagation over the West Pacific because you've improved the basic state of the model so this is sort of one example of how coupling might make your basic state worse and then you lose some of the impact of coupling so back to a Klingerman on Walno this is a different Klingerman on Walno we've looked at the impact of AC interaction in the MJO using a the same high vertical resolution mix layer model of the MJO we've done it in this model because it allows us to fully capture the dinal cycle of SST whereas the standard configurations of the Met Office model at that time didn't really allow us to look at that and using the mix layer model it's quite easy to prescribe some seasonally and depth varying heat and salinity tendencies to the ocean model which allow you to maintain the basic state of your ocean model very very close to observations and that means that you can you shouldn't be changing the basic state of your model and so you can really see the impact of AC interaction and we've done this in both the high and low entrainment frameworks that I talked about before in general low entrainment simulations the coupling improves the amplitude but it doesn't really improve the propagation in the model and in high entrainment simulations coupling improves the propagation but it doesn't actually change the amplitude very much so here's some examples of that so this is for an MJO or composite based on essentially phase 2 of the MJO this is where it was in phase 2 of the MJO we're confused now looking at here it's OLR okay that's good so this is the active phase of the MJO this is observations this is for phase 2 and this is for phase 6 so again I'm showing the phases where the convection is active over the Indian Ocean or the West Pacific this is for the control entrainment experiment and you can see that this is the atmosphere only and this is the coupled version and you can see we've increased the amplitude particularly in this phase 2 but not really done anything in phase 6 but we haven't really improved the propagation signal at all this is time going upwards this is longitude and the zero is here they're grabbed from a paper so the axes are designed for papers not PowerPoint presentations unfortunately if we go to the high entrainment simulation this coupled low entrainment and atmosphere only high entrainment actually look quite similar in a way if we couple the high entrainment simulation we actually see that if anything we slightly reduce the amplitude of the signal and hopefully improve the propagation of the signal both in phase 2 and phase 6 ok so the impact of SE interaction depends on what sort of MJO you've got in the first place in your model so I've kind of summarised that in these diagrams again here but in the interest of time perhaps I won't dwell on this too much you can see that in the control entrainment we tend to increase the amplitude but we don't really do much to the propagation in here if you look in the right places you can see improvements in the propagation and so where is the right place so this one for example we tended to get slightly more propagation into the circle compared to this one and this one is a big improvement in the sense of you really extended that propagation you also seem to improve the lifetime of the MJO event ok I've forgotten about this high resolution modelling of the MJO so all the things I've been talking about have been in GCMs where we rely on our parameterisation schemes to represent convection now recently there have been some advances in computing that allow us to do some simulations at high resolution where we can allow the explicit model dynamics to deal with the convection and there have been a couple of examples of that one globally with the Nikon model in Japan where they have now run simulations down to 350 metres globally in their model they don't run them for very long or originally where you force a large domain with the analysis of the boundaries and allow the MJO in the model to develop and so there's an example of that by Chris Holloway and me right so notes here the resolution apart from when you go down to 350 metres in Nikon is still not cloud resolving it's cloud permitting but we're really not actually resolving individual clouds and so that means that actually we're still not really representing the details of the convection correctly but it is a very different way of representing the convection compared to a standard convective parameterisation scheme these models generally show improved simulation of the MJO compared to their parameterised counterparts even allowing for not changing the resolution it's still not clear whether the improvement relies on an explicit representation of the convection and the ability to capture the full range of scales of interaction so we're resolving from four kilometres up to four hundred kilometres to a thousand kilometres and so the convection can interact with the dynamics on all scales and as opposed to a parameterisation scheme where we don't allow the interaction of the convection with the dynamics on any scales less than the resolution of your climate model or whether it arises just because we've improved some representation of essentially parameterisable processes e.g. the sensitivity to environmental humidity or the vertical profile of the heating so this is just a simulation from the Holloway paper so this is the same event that I was looking at in Nick Clingman's paper of the MJO during the OTSI so this is just the first ten days of that you can see the eastward propagation of the convection these are two simulations performed at a twelve kilometre resolution with parameterised convection and there's no propagation really we've also done a simulation where we've changed the entrainment rate a la Clingman et al by increasing it and actually you do see some improvement in the propagation in that simulation and then these are four simulations where we've turned the convective parameterisation off essentially and generally they show some improvement in the propagation of convection they're very very noisy and this is because we're not really resolving convection properly and so what you tend to get is because you can't resolve the small weak convection because you need you've got to be convecting on a four kilometre scale what tends to happen is instability builds up in the boundary layer and then it goes bang and fires off and comes down again but there's a general improvement and if you look at this in these sort of wheeler-hendon phase plots these three simulations are these three lines here and these are these three simulations we do understand a little bit why this one doesn't show the same kind of improvement as these three okay so despite everything I've said about how bad climate models are at simulating the MGO we do have some skill in sub-seasonal prediction systems so I'll talk a little bit about two studies so this is a study by Frederick Vitard about the MGO in the ECMWF model over time so this is the correlation bivariate correlation between the forecast ensemble mean forecast RMM indices of the wheeler-hendon index and the observed and so this is the lead time at which the bivariate correlation falls below 0.5, 0.6 and 0.8 and so if you're sort of you know 0.6 if your measure of skill then there's nowadays there's skill out to 24, 25 days for the MGO in their model and that's pretty good there's been slightly less improvement in the 0.8 bivariate correlation but 0.8 is quite a tough test you know for a measure of skill and you can see that you know there's been improvement over the years it's not been it's not necessarily been sort of systematic depending on where you look at in the threshold so this is a measure of the correlation but it tells us nothing about the amplitude of the MGO signal in the model this is the amplitude error in the model and you can see that actually up to about 2005 2006 there's very poor MGO amplitude and between all of the 2006 operational system and 2008 operational system the amplitude of the MGO in the model really improved so this is an amplitude error of 0 here and so Adrian was partially responsible for this improvement over this period and then he left I did notice it's flat after you left Adrian yeah so many studies have shown that the MGO skill depends on the initial phase in the forecast so this is just the variation in that forecast day for the 0.6 correlation depending on which phase you've initialised if you've got a strong MGO in your initial conditions and which phase it was in the initial conditions so those that are initialised in phase 2 or 3 or phase 6 or 7 which is MGO active over the Indian Ocean or West Pacific tend to have slightly more skill than those that are initialised through the convection active over the maritime continent or sort of the sort of western hemisphere phases so these errors in amplitude and possibly errors in the propagation speed do have an impact on the teleconnection from the MGO so this is still from Frederick's paper this is the composite of the North Atlantic South webinar 500 hectopascale geobotential in the northern hemisphere 10 days after an active MGO in phase 3 this is what that looks like in the 2011 version of the seasonal forecasting model and you can see that actually over the North Atlantic region and over the pole it really is capturing the signal quite well if you look over the Pacific the signal is actually a little bit too strong if you compare that to the 2002 version of the model where you had very low MGO amplitude there's also a very weak teleconnection response and that's essentially because having that teleconnection relies on a rosby wave that is generated by the heating if you've got weak heating you'll generate a weak rosby wave response and your teleconnection response will be weaker and that actually has quite a significant impact on the no prediction in the ECMWF model so this is the correlation skill for the NAO prediction over ECMWF for cases where there's a strong MGO in the initial condition which is a solid line and no MGO in the initial condition and despite the fact that we know that the MGO is a source of predictability for the NAO in the versions of the model where there was weak amplitude it was actually worth predicting the MGO than at the NAO than it was when there was no MGO in the initial condition and that is essentially because the model was failing to capture the driver of the NAO variability in these forecasts and so despite the fact that it should be doing better because there is a source of predictability from the MGO it was actually doing worse than if there was no MGO in the initial condition as the amplitude improved this teleconnection improved and now a strong MGO in the initial conditions provides additional skill over the NAO so Nino et al examined this MGO prediction skill and potential predictability the paper is mainly about potential predictability unfortunately rather than the skill in a comprehensive set of hindcasts from the ISVHE project intracesional variability hindcast experiment they base their predictive skill on the time at which the mean square error has the same amplitude as the signal that you are trying to predict so this is quite a good measure of predictive skill because it is actually taking account of the amplitude because if you have a weak signal then your error will be larger so this captures both amplitude and phase propagation the black line is the single member skill so this is essentially a deterministic predictive skill of various systems and you can see that ranges between something like let's ignore CFS1 but between 10 and 15 days maybe for a single member skill on average the hash bars show the skill for the ensemble mean in a deterministic sense and there you can get skill for ECMWF up to 27 days for example but more generally around 20 days for most models these bars up here show their estimate of the potential predictability in a perfect model scenario and you can see it so the tan lines should be compared to the black and these should be compared to the hash you can see that in a perfect model potential predictability sense which is perhaps an overestimate of the potential predictability there's still quite some room for improvement maybe and they find additionally they find predictability limits are lower if you have weak MJO in the initial conditions so that's perhaps not unsurprising these are all estimates for strong initial MJO conditions and they find some phase dependence of predictive skill and predictability in some models but not all and the phase dependence is not the same for all models so that figure I showed for Frederick's paper is not necessarily the case for all models I don't know why I've got that figure twice so summary, I've finally got there so simulating the MJO in climate models is still a really big problem and the MJO relies on an interaction between well, because the MJO relies on an interaction between planetary scale dynamics and atmospheric convection and is highly sensitive to the representation of convection in GCMs I won't go into all the details there AC interaction is important but the existence of the MJO doesn't depend on it but despite weaknesses in model MJO simulations operational prediction systems still do exhibit useful skill for MJO prediction