 And here we can get inspiration for our lab, ideas for our lab sessions. Thanks Fred. Good morning everyone. When we were preparing for the Open IFS I asked Glenn whether he wanted me to give this presentation or the one I gave yesterday and he said both. And luckily this was actually ready, this was a work that we presented with Laura Feranti last year at a meeting on sub-seasonal variability in Washington DC. And I thought that some aspects might be relevant to what you will be doing and the way you may want to verify your experiments here. So you all know that 2015-16 was one of the three biggest in linears we had, probably according to some measures the strongest one in the record. But what I would like to point out is that there was a lot of intracisional variability during this winter and therefore what you will verify, so what when you verify your experiments against the observation the result will depend on exactly what period you are going to look at. So this is the outline of the talk. I will show you something about your sub-seasonal variability both in the U-Atlantic and the Indo-Pacific circulation during this winter. We'll talk a bit about the connection between the rainfall in the Indo-Pacific domain and the non-harmic circulation from analysis and from our current operational seasonal forecast system for winter. And now I will show you how the sub-seasonal variability is represented in the ensemble of our operational seasonal forecast. We do 51 member ensemble so the sub-seasonal variability can be quite different among different members and I will show you how this will affect the prediction. And I will talk a bit about the predictability on a shorter time scale than one season. So here is what happened in reality in 2015, winter 1516. These are maps of 500 leptopascal height taken from the NOAA climate diagnostic bulletin. We hope that it will still be there for the next El Nino. So what you see here is that the maps have some common features but also quite some remarkable differences from one month to the other. In December, it is well known that December is not the month where you see a very strong signal in the Pacific. Everybody is thinking about the connection from the El Nino to the North Pacific but all the seasonal forecasters in the U.S. say that they usually look at January, February, March as the peak season for the response on the Pacific North American part. And in fact, if you look at the map for December here, you don't see a lot of signal in the North Pacific. And instead you see this very strong signal here with a low just south of Iceland and a big ridge over western Europe. So this particular anomaly obviously projects strongly on the positive phase of the North Atlantic oscillation. When you move to January, then the pattern is quite different, sort of wave number two pattern. Now you see this traditional low over the North Pacific which is typical of the El Nino response. And you get again the sort of canonical negative articulation signal in high latitudes. The low over the North Atlantic as weak and has actually moved to the south to the point that actually if you project this anomaly on a standard North Atlantic oscillation pattern, you actually get a negative projection. Then we come to February and in February the low over the Central and Pacific is maintained. But somehow the flow over Europe has come back to this tripolar structure that was very evident in December. And so again we come back to a positive phase of the North Atlantic oscillation. So you will get monthly means to look from your open IFS output. So if you average the three months you will get something quite standard over the North Pacific. But for Europe you will not get the sort of canonical answer response with a negative articulation. Because in this particular winter that actually only happened during January. And instead you will look more like a wave number two pattern with lows over the North Atlantic and the North Pacific that resemble Wallows cold ocean warm land pattern. Now what I will show you now I mentioned that this map project in a different way on to the North Atlantic oscillation. So I will show you this through one of the products that we issue for seasonal forecast. Which is the projection of the geopotential height anomaly on to this first empirical octagonal function for this particular domain. And what you find on our website are these so-called climagrams. And these climagrams are basically plot where we basically compare the distribution of this particular index in the analysis. And that gives this is done on a monthly basis. It's actually the only product of our seasonal forecast where you see monthly information. So the yellow and orange bar give you basically the spread of the distribution of this index in the period of the reforcast for system 4. So in this case it's 1981 to 2010. The red line is the median. The orange bar gives the central tercyl interval. And the yellow gives the distance between the fifth and the 95th percentile. So this is the distribution. And here you actually have the projection over DOF which is there in the corner. And here at the bottom you have the two meter temperature average over northern Europe. Well known then these are well correlated. So positive and AO brings warm anomalies over northern Europe and vice versa. The gray bars actually give again the median tercyl and fifth to 95th percentile distribution from the model climatology during the same period. And the purple bar gives the distribution from the actual seasonal forecast. In this case the seasonal forecast started on the first of November. So if the purple bars are shifted with respect to the gray ones it means that the forecast for this particular season is very different from its own climatology. So you have a significant anomaly. And in this case you see that if you look at the first month November, December, January at least from the forecast from November. If you look for example at the medium or the box indicating the tercyl that is shifted upwards with respect to the gray bar. So it means that the forecast is predicting a substantial shift in the distribution of this index compared to its own climatology. And the same is true for the two meter temperature. So the model was predicting positive and AO basically throughout the winter and warm anomalies over northern Europe throughout the winter. This may be most likely what you will get from your experiments. However if you look to what happened in January as I pointed out January was quite different. So in January the sign of the NaO changed from positive to negative. And we actually had some very cold spells in January. So the temperature anomaly in January 16 was actually cold. The red dots are the verification. So of course when you look at the operational forecast you don't see the red dots. You see everything except the red dots. But then aposteriori as a verification these plots are modified so that you can actually see how they observed a value fitted with the predicted distribution. So again if you average December, January and February and you look over Europe in the Atlantic you will see probably a positive NaO signal. But yeah take into account that if you do a verification, if you are looking on a month basis there was actually this significant variability between different months. Now why is that? Well of course one may speculate but the strongest source of intracisional variability in the tropics and in some parts of the extra tropics as well is the mother and Julian oscillation. And probably many of you are already familiar with it is basically a propagation of large scale organized areas of convection that usually starts in the western Indian Ocean and propagates to the east. And Wheeler and Handon in 2004 proposed a classification of this MJO cycle in eight different phases and proposed one index which is now widely used in the verification of weather prediction and climate models. So these are the canonical eight phases of the MJO according to Wheeler and Handon and these phase two and phase three are the phases where the convection is in the western and central Indian Ocean. And a number of studies for example one from Kassou in 2008 on Nature actually pointed out that about ten days after these phases occur ten to fifteen days you tend to have a positive NaO signal in the North Atlantic and following typically phase six after ten days you tend to have a negative NaO. So did the MJO episode occur in 2015-16? The answer is yes. You can actually see again some data from the NOAA bulletin. So these are of molar diagrams with time going from top to bottom. So starts in June 15 and goes up to April 16. On the left hand side you see anomalies of outgoing long wave radiation. So low anomalies are indicative of increased convection. So you see that overall throughout the season you tend to have negative anomalies typically in the NINU-4 region. However there's quite a strong modulation even more so when you actually look to the compensating sinking motion over the maritime continent. And if you actually look farther west in the area of the western Indian Ocean what you see is that you see negative anomalies in December and negative anomalies in February but actually positive anomalies during January. So although the signal in the main NINU region in terms of OLR is quite consistent throughout the season when you look at the maritime continent and the western Indian Ocean you see the effect of this MJO activity. You can see clearly the easter propagation of the signal in this other of molar diagrams that shows the velocity potential at 200 hectopascal. And again you see here for example that this if you look at December you actually have an anomaly of negative sign again indicative of convection over the Indian Ocean. November and early December then you actually have positive anomalies here at the beginning of January which actually vanish again in February. So quite a lot of intracesional variability in the tropics. And now so it's interesting to try to work out how the convection in different parts of this domain actually connects with the extra tropics. And well it's now three years ago with colleagues at ECMWF we looked at the teleconnections in our seasonal forecast system system 4 and particularly we look at the connection of rainfall in three regions of the Indian Ocean and up to the Central Pacific domain with both rainfall in other parts of the world and geopotential height. We focused on three regions one on the western and central Indian Ocean that you see at the top the maritime continent and then in your four region. And what you see is a comparison on these maps from the GPCP 2.2 data set on the left hand side and system 4 on the right hand side. And you can see a sort of a common structure in the these are maps of covariance of rainfall everywhere with rainfall in the boxes. And you can see that there is this tripolar structure which is basically a consequence of the dominant variability in the water circulation. So usually when you have a scent due to increased convection for example in the Indian 4 region and you have the scent of the maritime continent but then again increased rainfall over the western part of the Indian Ocean. The Indian Ocean so these two regions are regions where there is a positive correlation between SST and rainfall. Very strong here about something like 50 to 60 percent here. So the precipitation in the western Indian Ocean is positively correlated with the SST but to a less strong constraint than you have in the Indian 4 region. So there's more internal variability from the Indian Ocean that you have in the Central Pacific. Actually this is an area where there is hardly any correlation between SST and rainfall because actually the amount of rainfall is controlled not so much from the SST but just from the dynamical response to the other two areas of ascent. So you can see that the system 4 does a pretty good job in simulating these teleconnections as we have diagnosed them from observed data, the GPCP data. You can actually, one difference that's quite interesting is that if you look at this particular map you see that the signal associated with rainfall over the western Indian Ocean so the covariance between the rainfall over the western Indian Ocean and the Central Pacific is actually stronger in the model than it is in the observation. So somehow the model tends to like this tripolar structure more consistently than in reality. Now if you now look at the geopotential height anomalies that cover with the precipitation in these regions let's focus on the two, the one on the left and the one on the right. The one on the left is from the western Indian Ocean. The top are the teleconnection maps from Iran Interim and here are from system 4. So from the Indian Ocean you tend to get the wave number 2 structure with a positive NEO signal actually quite stronger in the observation than in the model but still the model captures that. While if you look at the teleconnection from the NEO 4 region if you look at the observed map you see actually a negative NEO pattern. You don't see this very much in the model and the reason we think is because of the stronger link between the Indian Ocean and the Central Pacific in our model. So since the Indian Ocean tends to give a positive NEO if you actually always associate a linear with a strong convection in the Indian Ocean you will tend to get a positive NEO as well. However one has to say that if you look actually to 2015-16 the observed average for the season looks more like this map than the top map. So because as I said it was the main response in this year was not the sort of canonical response to answer that you see in this particular diagram. Now of course you can do things in reverse and say instead of starting from the tropics and seeing what happens in the extra tropics you can start from the NEO signal and it's interesting to do this now to compute these covariances the other way around and to do them month by month. So if we now take, we compute the NEO index as the projection onto the first Atlantic UF and you compute the covariance of rainfall everywhere and you do it on a monthly basis. Well first of all of course you see the traditional shift of the precipitation to northern Europe which is characteristic of positive NEO and then you see in December again this characteristic tripolar structure of precipitation in the tropics. When you move to January the connection to the central Pacific actually tends to vanish but you still see this tripolar structure between the Indian Ocean and the maritime continent. In February most of the signal is actually gone. So when you actually average the three months together you have a rather weak signal. So this is a problem that has been noted by many authors that when we try to connect the response of ENSO to the extra tropics and especially in Europe often doing the traditional seasonal average then to wipe out the signal because the signal is actually different in the early part of winter and the late part of winter. This is something that Fred mentioned yesterday and again so this is some aspect that you may want to see in your model whether you get a different response in the early part of winter when this tripolar structure seems to be stronger. So in fact we found out also from looking at the observation the best index for this wave number two pattern is actually a combination of ascending motion over the western and central Indian Ocean and sinking motion over the maritime continent. So if you construct an index basically as a dipoly index between these two regions same as you can do for example for the NAO between Iceland and the Azores and then you teleconnect this index just in December and January then you get a very strong extra tropical response with this characteristic wave number two pattern. So February doesn't contribute very much to this particular correlation but it's actually stronger in the early part of the winter. Now I mentioned that different ensemble members can represent this in transition of variability in different ways and so I already showed you that if we put all our ensemble members together the model tended to stay in a positive NAO state throughout the season. However some ensemble members did show the inter-seasonal variability so what Laura did was to pick up the best five members in system four in the sense that those that showed the largest change in the NAO index between December and January. You remember we had a very positive NAO in December it went to negative in January. So if we pick up this member then the geopotential height anomaly would look like this in December and the temperature, two meter temperature over Europe will look like this so very strong energy signal and positive anomaly and then in January you move to the structure with actually positive anomaly over the Arctic so a negative Arctic oscillation and that is actually associated with cold temperatures. So some members actually managed to do the same as the real atmosphere but only a minority of them. So we try to understand why it happened and try to relate it to the convection in these regions that I showed you before and these are maps of 200 hectopascal velocity potential anomaly for the five best members on the left and the ensemble mean of all the 51 members on the right and this is what, this is the anomaly in December and this is the anomaly in January and both these five members and the ensemble mean show that this very strong signal that came from the western Indian Ocean so negative velocity potential anomalies are indicative of enhanced vertical motion and convection so all members were probably quite consistent in getting the decrease in the activity over the western Indian Ocean however if you actually look at the maritime continent region there was quite a difference so the best member actually showed an intensification of the sinking motion over the maritime continent and the ascending motion over the Indian Ocean region so somehow the strength of the working circulation while in December was mainly between the maritime continents and the Indian Ocean actually shifts in January to a more traditional El Nino pattern that has in fact the canonical reconnection as a negative articulation side. If you look at the ensemble mean the change over the Indian Ocean is there but you actually, if you actually look at this feature there's actually a weakening of the two centers over the maritime continent and the Nino four regions so somehow what it seems is that the anomaly in the working circulation shifted from being mainly over the west Pacific and the Indian Ocean in December to the central Pacific in January and this basically changed the response from a positive NEO to a more canonical negative articulation scene. So if you run your 10 member ensemble you may get some of them First of all, we have to say that sub-seasonal variability is hard to predict beyond the first, let's say, month or two months in the best cases. So it's not very easy to predict what happens in generally starting from the first of November so don't be disappointed somehow. Usually sub-seasonal variability operationally is limited to 30 or 45 days. So probably some of your members will show the change as it's depicted in these panels some of them will look more like the ensemble mean. So this is just to warn you about the possible differences that you may find in your ensembles. Now in this prediction of course there's another factor apart from the inherent predictability and that is the fact that all models have systematic errors so in some sense it's usually easier to predict some transitions in a model if the model tends to like the state towards which the atmosphere is actually moving. And actually, so if you look at the same climagrams for the North Atlantic oscillation here and the two-meter temperature anomaly from actually operational seasonal forecast starting not on the first of November but first of December and the first of January, you actually see that even from the first of January the model actually did not do this transition from positive to negative NAO even if you don't, one more month. But actually, if we started from the first of January then it actually got the return back to a positive NAO signal correct. So the transition from positive to negative was not well forecasted on a range of two months but the transition from negative to positive was. And we think that the reason is because somehow this in February there was a return to a situation similar to what we had in December which is a situation with which the model tends to reproduce more easily. I showed you that the model likes this tripolar structure and tends to actually enforce a tripolar structure on the rainfall anomaly. So we think that the difference in the predictability on the two-month time scale was somehow related to the fact that the model went back to a state which is more somehow consistent with its own dynamics. Okay, the conclusion is just a summary of what I said and so we hope that this may guide the way in which you look at your experiments and the different ensembles and I think that Glenn has, so we have included this western Indian Ocean as one of the boxes that can be analyzed with the Metview script so that you can actually look at, for example, time series of evolution of rainfall and the system in this particular region which is quite important for the teleconnexions.