 which will also be an introduction really to what we'll do tomorrow afternoon in the practical session then then we'll have a second lecture by David on the MJO and impacts on the middle attitudes and then more on the MJOs as MJO day after the break influence of middle attitude disturbances on the MJO so we'll be looking at those extra tropical impacts on the MJO also so Hylin and Nick Hall will be talking about that and then in the afternoon that'll be the session introduction to speedy so you'll be working on the ICTP laptops they have everything installed so Fred was saying that you won't have to install anything on your laptops and that'll be straightforward so just before we begin I just wanted to come back to that that exercise yesterday where we were doing the MJO phase composite We did pre-SIP and then I suggested oh you could try some reanalysis data like like stream function for example, but it didn't work and the reason it didn't work was I e-mailed to Mike Bell From our data library group. It's because the two data sets So the reanalysis and the indices don't have the same units for time So we needed to convert the The reanalysis which has units of days since a certain day to Julian days to match up with the Bureau of metrology's RMM indices, so if you do that then it works and So that's often if something doesn't work. It could be could be due to something like that So if you look here, this is the the RMM indices. You see Here's the time grid. It's in Julian days And if you look at the reanalysis Here's the time grid It's in days since 1st of January 1948 So that's why if those things don't match up then often they'll just be a cryptic message And it won't work. So it's the kind of thing kind of thing to check So I put that anyway, I put that example onto onto the wiki page so you can you can find that as well Okay, so for today's lecture What I'm going to do is To introduce the the S2S project and say something about Forecasting sub seasonal sub seasonal seasonal forecasting forecasting on different time ranges Then the project itself and then I'll talk about these databases of the database We're going to be using to tomorrow afternoon and there's actually another Subseasonal database that just came online and that's sort of the sub-ex or I'll talk a bit about that and then something about the skill that we have in This this range of two weeks to a season So this is all about forecasting, which is really one of the end goals of the the S2S project And teleconnections is one one mechanism of predictability that that's very important to Give us that give us that give us that skill So weather and climate forecast time scales. We all know the the weather forecasts Now I guess what do you have it? Is there a physical pointer? Oh, yeah forecast for for the next the next few days so Really from the the evolution of the synoptic weather situation and and this this dates back to around 1910, so we've had over a century of of work on on Dynamical forecasting so based on in the middle attitudes Middle attitude barotonic instability or there may be Instabilities in I know easterly waves in the tropics could give you something something similar And then on the seasonal timescale, so this is about atmospheric initial conditions And they're being able to simulate the evolution of barotonic waves for example in the middle attitude On the seasonal timescale then we have these these outlooks Essentially based on the the impacts of lower boundary conditions on the atmosphere so slow Sea surface temperatures or maybe soil moisture Maybe some also from the from the top some some stratospheric influence and so we can say we get maps like this which are In terms of probabilistic forecasts of what's the probability of it being above normal or below normal? On average during the season So we've had that the dynamic forecasts of of these Enso and seasonal forecast they route back to the mid 1980s But but what about in-between range so where you're going where these leave off after ten days or so and And these ones start which is more like on a monthly timescale, so from the next month Can we say anything about? Between this weather and climate between two weeks take two Two months ahead and that's really the intracesional timescale that we're talking about here, so the Topic of the workshop is is the it's the dynamics in that intermediate range What we call intracesional people these days are often calling sub-seasonal In in dynamical meteorology. It's often being called this low frequency variability LFV from ten to a hundred days. What about that? Can we can we make forecasts in that range where the initial conditions of the atmosphere? That the noise is that that information is lost right you hit this this Lorentz predictability limit Around about around about ten days But you haven't had enough time for lower boundary conditions to see surface temperatures to to Impact on on the on the statistics because you're looking at shorter ranges It can we say anything about that and for a long time the the quote the answer was no and so it was it was called a It was called a predictability desert, so there was there was there was no There was no source of predictability in that range that could be harnessed so the models couldn't Couldn't if they're if there were worse sources that the models couldn't predict them sufficiently well So in the weather forecast the main source At least in the Middle Attitudes is baroclinic wave evolution and on the seasonal timescale the main source is the impact of lower boundary conditions on on the atmosphere but in What about this intermediate range or what are what are the sources of predictability there? So this is a little schematic of sort of a some the same thing that I was just talking about where this is the schematic of forecast skill against against forecast range and so you see here in the weather timescale We have very good skill in the first few days and then drops off rapidly as that information in the in the initial conditions is It's lost the skill associated with that and then in the forecast range when we're here in the monthly to seasonal wet We're averaging the idea here is you need to average over some time period in order to average out that noise That atmosphere noise so you'd be left left with a signal So that's why a seasonal forecast It's not for a particular day, but you it's for That's three month average is a free on average is it likely to be above the below normal And then this is showing so this predictability here coming from sea surface temperature Enso is the main main driver there main phenomenon And then what this one in between it's shown here as being less Poor to zero So this this actually comes from an older slide it was based on a slide before You know the s2s days where this is the predictability desert where there isn't anything But what we think now is well actually this could be at least up here so that there is something in this range and it's a mix of The the influences of initial conditions of the atmosphere so it has some some aspects of weather But it also has some aspects of climate So as you get into this where we're looking really at weekly averages here in the 10 to 30 days or or so Where you it some aspects of climate because we're not looking at a single day We're looking at an average over time over over a week or two weeks So there's some aspect of there's some element of it's like a climate forecast It just where we're averaging over a shorter period at the same time. There's some predictability of initial atmospheric conditions and this is a primarily the type of atmospheric initial conditions there are of the Of the the madden julian oscillation Which is the one of the main sources of predictability in that range that we know of So far, it's also thought that the stratosphere is going to be an important source of predictability on this on this Timescale, but that that hasn't yet been that there's much more Work being done on on the mjo then then on on the stratosphere in this range so this this idea in this in the Subseasonal range people often talk about seamless forecast that by filling this gap Between the the weather forecast and the seasonal forecast as to as we call it sometimes bridging the gap between weather and climate So it's toward a lot of people are talking about well. Can you really make a forecast that goes seamlessly from? You know weather timescales up to up to climate timescales, and if you do that How how should that be issued so you could think of it like this as you go out to? One week ahead two week ahead three week ahead that maybe what you should do is this averaging window that you need to average over Should also depend on the lead time so out here Beyond in week three and week four you might have a Two week average So for a seasonal forecast where we're three months ahead We maybe have a three month average and just make this depend on the lead time so and after in week two we might have a one week window and and Four days ahead a four-day window two days ahead two-day window like like that and and so you sort of You go down to weather forecast for an individual day But then as you look further in the future you make it less specific so that you're you're not Forecasting for any particular day, but for some little window of time that's getting that's getting longer and That may also be relevant to applications because as you go further out people may not need to know well Well, you know what what the conditions on a particular day are but rather what the conditions are likely to be You know next week or on the second fortnight of a month so this S2s project Comes from it's sort of driven by in a way It's driven by application or desires to to forecast information that can really inform decisions And so many decisions in in agriculture water water management disaster risk reduction health fall into that time range and The goal of this is this new project, which is it's a joint project So more or less the first time that the world weather and the world climate research programs have come together On a project to collaborate on a project to improve forecast and understanding on the scale and then to promote the uptake By operational centers in the applications community So that's more or less said here. So the idea it's saying well We're gonna we're gonna bring the expertise of these weather and climate communities together to to work on on these things and then toward really toward In the global front framework helping the global framework for climate services so you can produce information that can help Societies manage Climate-related risk through better early warning in that time scale and then it can help adapting adapting to climate So we have two co-chairs Frederick Vita and myself on on this project and the time range here is between two weeks in the season. That's the Target target range where where the project is is is Proposing that we'll we'll we're going to improve understanding and and forecast skill so there's a website you can go there s2s prediction net and Every all the the project activities are listed there and the way that the the research has been Organized here is in terms of some some specific topics have been highlighted and there are actually sub projects You can find those on the web page On those and you can see one of them here is Interactions and teleconnections between middle attitudes and tropics and so high and Christina are leading this this sub project With this one on them the MGO There's one one on monsoons We think that in particularly for in monsoon climates there's there's a lot that can be said about this intermediate range because of the phenomena of active active and break phases monsoon also this issue on set date of the monsoon can we there's something that's very important in Tropical countries and so that one of the targets is well, can we can we predict the onset? In this range more climate service oriented one in Africa Extremes, of course, this is a major focus of the project is a focus on extremes So can we can we improve early warning of extremes in this time range and then forecast verification? So those those are the bars running horizontally and then this cross-cutting issues are in the vertical columns here So the research issues Teleconnections predictability what kind of predictability is in the range What's the role of ocean atmosphere coupling? So in weather forecast most weather forecast models of tended to just Fix the SST boundary conditions and just persist persist SST Normal ease but as in seasonal forecasting ocean atmosphere coupling is obviously a key with ENSO What about in the in the key in this intermediate range? How is ocean that must be is it important as a source of predictability in these ranges? scale interactions physical processes in general Then a lot of modeling issues because this project is really It's it's about it's about improving forecasts so how should models be initialized in in this range in order to to Get the most skill skillful Forecasts get the right spread of the ensemble. How and how should those ensembles be be generated? Should you use some lagged ensemble or all the forecast members launched on the same day? What what kind of resolution do you need you need very high resolution again ocean atmosphere coupling? Multimodal combination, how should the forecast be post-processed? Calibration is another one there and Then the needs and applications This has been done in liaison with a with a working group from the WWW The societal and economic research and applications and then underlying all this is the s2s database That's a what we'll be talking about more in a minute So here's an idea of applications in these different different ranges. So Going up here. We're going across time scales from short-term warnings for short-term forecasts then in the sub-seasonal range Threat assessments and guidance and then seasonal outlook as you go beyond it So if you if we look in this this blue range here in emergency management management, it could be sort of pre-stage emergency supplies In agriculture the scheduling of planting and applying fertilizers and so on so There's many as you go across different time scales There's there's many different types of decision that that that forecast can inform and there's a nice Report that has very recently been written about what about a year ago by the the US National Academy Called next-generation earth system prediction and they talk a lot about the application Hi, were you involved in that report? Yeah, so hi. I was actually one of the authors Of this report. I put the link at the bottom. So This this slide is maybe slightly getting ahead But I think I'm gonna come back to that one. So the s2s database This this This project is involving the global producing centers of Sub-seasonal and seasonal forecasts and they've all been all of these models. There's 11 of them Here they are from all these centers like UK MO and said environment Canada They they all make operationally forecast in this range out to a month or two in the sub-seasonal range and What the project has done is to take their forecast as well as their we forecast I'm cast made for past years and archive them all together In a in a in a database, which is officially archived an ECMWF and the Chinese Met agency and we now have a copy in IRI data library so the I Think you could say first and foremost this project is about a coordination between all of these operational centers to make their make their data available their forecast available and This is what it looks like Those are the centers again, and this is the time range So it's this these are sub-seasonal forecasts some of them are shorter like like the environment Canada that goes out to 232 32 days and some of them are longer like pure meteorology out to 60 days but it's it's in the range of of a month or out after two months various resolutions of models so this is a It's a database of opportunity whatever people are using they have put in this database There hasn't been there the models haven't been reconfigured for for s2s So there's quite a range of Resolution the ECMWF one is is the highest resolution It goes Quarter degree out to day 10 and then after day 10 they reduce it to half a degree Where so the Bureau of Meteorology one is on a on a two degree grid Also, the ensemble sizes are different Sometimes you see large ensemble sizes like 51 for ECMWF, but they're all on the same day Whereas if you look at say N set it's a 16 But this is this is the need to see this in conjunction with the frequency so they make those every day So you can actually take several days lag together to build a larger a larger ensemble size Whereas ECMWF is every Monday and every Thursday twice twice weekly and One issue we have is that these could be issued on different these weekly ones before Centers used to make those on issued on different days, but through this project This has been unified so that now all for all centers that issue on a weekly basis Are issuing on Thursday, so at least we do have some some Some some unification of when the forecast start which is useful if you want to make a multi-model combination Have them all at least starting on the same day So you notice that this is this is much more often than a seasonal forecasters typically Initialize which is month once a month, so it's less often than a weather forecast every day More like on a weekly timescales Then the other important part of this is the is the reforcasts in seasonal forecasting. We call those hindcasts Here they call reforcasts. So the model is run starting For for past year, so for example Bureau of Meteorology has a long Reforecasts period and it's fixed where ZCMWF has a shorter one the past 20 years It's called on the fly so that on the fly means every time they update their model. They run a new set of hindcasts Reforecasts and how often is how often is this done? This is also different The ZCMWF model they they make it on the same days the same dates these twice per week that they make their their forecast whereas some of them such as Korean Meta-agency are just four times a month and then typically these reforcasts sizes are much smaller than the real-time months most of the models are coupled ocean atmosphere models and Many of them also have active sea ice. So this is the sea ice is thought to be an important source of predictability That this database is now in the IRI data library and when that's what we're looking at tomorrow afternoon and I wanted just to introduce you to this this second Subseasonal project called sub-axe. This is a project of of know us Where these are North American centers in the environment Canada is there, but we have other ones are like yeah, we have PRS ESRL from Boulder, we have NASA and US Navy and this is the CCS and force and North American models This is something akin to the the NMME project. He might have heard of for seasonal forecast As anyone does anyone familiar with NMME The one person so I mean this is Since since you guys work are not so much into forecasting You may you may not be familiar with that but for seasonal forecast Forecasting this North American multimodal ensemble is a kind of standard Set of seasonal forecast and here what they've done is essentially they've extended this to the sub-seasonal range So it's just like s2s except that what's nice about this is that the real-time Forecasts are also available in real-time Whereas in s2s they actually delay them three weeks Behind the real-time so if you want to get the forecast made now today You will have to wait for three weeks before you could access it through the s2s database And that's because the the operational centers often have commercial interests in the real-time forecast and The the s2s project is it's meant to be a research project purely so it's we one can look at Forecasts, it's not necessary for research to have the forecast So you know in your in your exercises in the in the next two weeks if you're you're interested in In looking in that in real-time forecast for forecast made today you you you can get access to this one this this database as well and That's also in IRI data library. So what we've been learning yesterday carries for carries for that one as well so so this here's an example of of Forecasts some forecast from from the sub-ex that they have a little viewer here where you can you can actually look at the Real-time forecast so he is up initial condition October 7th to 10th Okay, how about the skill then in this range? this is a Picture from a publication of ours a couple of years ago. It's actually good for s2s But what showing you is the anomaly correlation skill and so red is sort of point five And this is looking in the first week. So that's weather forecast Ranges, but it's been averaged over a week. We're just looking at the skill of weekly average Weekly average preset for the boreal summer season and we just correlate that against CMAP observations over 1992 2008 period and so you can see immediately in week one. There's lots of red So weather forecasts have some skill and have lots of skill generally except in these dry regions But as soon as you go beyond the weather range You know, so as soon as you're beyond day seven, there's a huge drop Especially in the middle latitude you see this this huge drop So you don't there's not much information From the atmospheric initial conditions through these barotonic waves, etc. Once you get beyond about day seven But you can see there are some features as you go through Week two three week four weeks of advance. There's some features that just don't you know, they don't They just persist through the range and so you can imagine that well, why is that you know, why First of all, we thought oh, maybe not just be some bug there You know because this looks a bit odd, but then when we start to think about it more then we realize that This is a source of predictability that that persists. So it's it's it's equally seen Four is in week one and what's that? Well, it's in so that's the there is some signature even of the El Nino in these on these weekly ranges on on these short-time ranges and then if you look here over over the Maritime continent region you can see that there's a there's a broader or Indian Ocean You can see broader skill here that's persisting right through to the end of the month And we looked at that in more detail and and what's contributing to that There's some contribute we found there's some contribution from Enso But there's an important contribution from the from the MJO coming in coming in there Yeah, huh? Yeah, we don't we don't have that we maybe we should make those There'll be a good baseline, right? That's often what people do in in forecasting is to to compare against a simple simple persistent forecast You might see some of this down here would be in the persistent forecast, too I don't think much of this would be so we've looked at some other things and Maybe this one's particularly relevant here. We looked at the the NAO and the PA P&A Indices and this is averaging over weeks three and week four as we take the period the second half of the month from a forecast from day 15 to day 28 we average those two week period and We compare the models forecast with the with reanalysis what what happened in the in reanalysis and this was done over done over I think a Fairly short period it was I think in 1999 to 2010 or something like that and What's shown here is the different seasons and this is the anomaly correlation skill up the side here so he has point five four and You can immediately see that for both in both cases You really see the skills much higher in winter right when the those those Teleconnections are most most active and What's shown here? There's two different models are shown in the different colors So the the blue is is for ECMWF and the red and Genter are for for the end set model And so you can see actually that they have quite quite similar skill for for these for these teleconnection indices And it's it's actually pretty high. It's a over point five which is In week three and week four much higher than we were seeing here in we've seen basically here This is a by the way, this is this is just I just for one model the ECMWF model I'm really not seeing anything repressive But here we are seeing in both the NAO and the PNA Pretty pretty high skill. So this suggests that there's there's some There's some skill in the larger scale atmospheric patterns Teleconnections So what's also shown here, right? So when it says total that that that's just just taking the Those weekly averages straight from straight from the model If it was a sub seasonal what we did was we we subtracted in each year. We subtracted the seasonal mean So that just leaves the amount of this skill is coming from a sub seasonal range because we saw that you could actually in the Seasonal seasonal seasonal phenomena like and so can contribute to these and we found that if you compare So if we just look here at the blue here There's a massive drop in the NAO as you as if you subtract the the seasonal components So it means that seasonal phenomena your ENSO and maybe involving the strategy to make a big contribution To the NAO skill in this range But it's much less the case for the PNA pattern. So it looks like the PNA pattern is as as Not not much of a Not much of an interim inter-annual component Switching to the MGO. This is something that you can do with this S2S data is to look at all the different models And so that's what Frederick Vita did here He looked at well, what is the lead time in the future that you can have an anomaly correlation of 0.6 in the orange or 0.5 in in yellow And then this is shown for all the models here. And so of course EC and WF is the best and It it now has if you think of the 0.5 correlation that we saw in the PNA and NAO It now has that By very correlation out to 34 days I mean, this is this is increased greatly over over past years and this came out I think yesterday and Christina's talk as well That they have they have a I'm not sure if you showed that diagram But they have a diagram showing the the evolution of these scores and they have really increased a lot in the last Since they've been doing this in the last 10 years if you looked 10 20 years ago, this would have been way down way down here 14 days If you look at the other models, they're more or less. They're more or less similar It's it's for the 0.5 correlation out to about up to about three weeks typically in many of the models So the other models also have quite quite a bit of quite a bit of skill if not as good as you seem to be worth So that's for the the MJO index itself So for those RMM indices that we're looking at yesterday, but what about in what about the teleconnections of in the teleconnection patterns in the models so what Frederick is plotted here is the Z 500 anomalies 500 millibar 500 millibar potential height 10 days after an MJO in phase 3 So phase 3 is more or less when the MJO gets to the the convection gets to the the maritime continent So you can see in analysis Greenwiches this Meridian is down here. So this is the European sector. You can see this positive NAO structure and of course in Europe, they're very interested in this teleconnection between between the MJO and the European sector and you can see in the analysis this very strong positive NAO pattern If you look into the models, you can see that it tends to be there Mostly they get the same sign of this But it tends to be much weaker than in the in in the observation So that's a that's a topic of topic of research often. They tend to get the the Pacific Limb of this this P&A type happen too strongly in the models see the insect model there So this is something that we can also look at Starting this afternoon in with when sorry tomorrow afternoon Because the these RMM indices have been computed for all the models and they're stored in the database And this is how they do that. So the computation of the MJO index that follows the the methodology Of Karchuk et al 2010 They do it You did well on that one Some other things you can do with the database and we've done this in in previous trainings actually when we had An s2s workshop here at ICT A couple of years back we did this kind of thing where you can identify some kind of Some extreme event Or high-impact weather event, let's say in in observations and just look to see well How well did the model capture that? And so this is an example of some observed data from this chirps data set We looked at yesterday on a daily basis through the monsoon of 2015 Over over northern India here in the state of Bihar and you can see there was a heavy Heavy rainfall event here wet spell first sort of active phase of the monsoon In 2015 near the onset July 6 to 12. So then you can you can look and see well What did that look like in a map? So this is a weekly average precipitation anomalies and you could see that so it'd be ours in here But this was part actually part of a large-scale structure where you have this dipolar pattern of below normal rainfall over peninsula India and then above average rainfall going up over the The foothills of the In the layers here So that's that's the observations and then we can look. Well, what did the ECMWF model? Forecast and this is its week one where we start at the beginning of the week of the event itself And you can see that it did a pretty good job of getting this this dipolar structure, which could be associated with a Northward propagating intra-seasonal oscillation pulse. It looks like that. Do we have this dipolar pattern? And then we can say well, what about if we look at the forecast that was started a week before that on On June 29. So now this is the week two forecast of this and you can see that's still doing a fairly good job What about if we look a week before that and so this is the week three range You can see that. Well, if you were interested in Bihar now, you really wouldn't see anything at the local level But is this capturing this still capturing this this large scale Dipolar pattern and even a week before that there's still some some evidence of that even at the even if at the local level There wouldn't be much information so one of the the tasks in People working on these forecast is if you have some there's some information in the large scale But at the local scale there isn't any information in the forecast So is there a way to downscale from the large-scale patterns if you have a good prediction of these PNA and AO or all this inter-seasonal oscillation. Is there a way that you could you could Statistically downscale a forecast based on those Okay, so that's about what I wanted to show you this morning So in summary we looked at the weather and climate forecast time scales and We're saying that S2S is this gray area between as some aspect of weather forecasting But also some of some of climate climate forecasting Then this new project it was started the end of 2013 actually as a five-year project So it'll be it'll be completing its its five years Next year and we're actually planning to planning proposing to have a second five-year phase starting at the end of 2018 It aims to improve the understanding of skill and is this what was called before a predictability desert between two weeks in the season range and Teleconnections as a sub project on teleconnections are important research focus The and we introduced these these databases the S2S, but also the new sub-ex database 11 models in S2S and seven models in sub-ex and Real-time forecast of three weeks behind real-time in S2S, but they're in real-time for sub-ex and then for the skill there's We saw that these indices all exhibit skill In these week three week four range Days 15 to 28, but we saw there's some serious biases in those MJO Teleconnection patterns in the extra tropical teleconnections. I'll be happy to answer any questions you might have Thanks