 Ah, who is the architect of the S2S prediction at ECMWF. And I put some names here. There are also many more people that I need to thank. And some of them I mentioned directly in the slides. So I hope I didn't forget anybody. The outline of the talk. I'll give you a general background on aerosol and their role in prediction in general. Then I will talk about how atmospheric constituents impact NWP. Here I'm focusing on the weather aspects. Of course, aerosols are very important for climate, as most of you imagine already. Then I will talk about the ECMWF experience about the sub-seasonal to seasonal prediction and aerosols. And then I will mention some coordinated studies that are going on now under the WMO umbrella. And I think they're quite important because they show the interest at operational centers worldwide in the topic. And then there is a summary and some open questions. First of all, I need to say that this is a complex system. So this is not only aerosol. It's also what we know as greenhouse gases and reactive gases. But atmospheric composition is a very complex topic. You have emissions from anthropogenic sources like cities and industries and transport, of course. You have natural emissions of gases and aerosols from biomass burning, from desert dust, and the oceans as well, agriculture emissions, and all that into the atmosphere and interacts with the radiation and with clouds and with the atmosphere in general. And that all affects climate and weather. And as I said, the processes at play are very many and very complex. So I focus today on a subset of them. So here is aerosols and weather. So you have, as I mentioned, several production sources, emission sources of aerosols, natural and anthropogenic. And you have also different aerosols that are released into the atmosphere like sea salt, dust, like carbon organic matter. So fate aerosols and ashes from volcanic eruptions, for example, and others from anthropogenic emissions. You have the interaction with the radiation, which is mostly the aerosols have two effects. They can reflect solar radiation in the solar spectrum. And that's their main action. But also they can serve as CCN, which means cloud condensation nuclei. So they can modify, they can serve as agents to form clouds and then modify the way clouds interact with the radiations. But for example, modifying the number of particles contained in the cloud at the same liquid water content. And that changes the way that cloud interacts with radiation. Sorry. That's what there are also other aspects that are not covered in the stock. But as I said, I will focus on this. And there is a little bit of a dilemma, what I call a dilemma when you come to aerosols, because the impact of the aerosols depends strongly on the type of aerosols that are being emitted. Most of them are aerosols, in most cases, are scatterers of solar radiation, which means that the more aerosols you have, the more radiation is returned back to the sun. And then you have a cooling effect. So that's exactly the opposite of greenhouse's effect, which is of warming the surface. And that's more an infrared type of effect. But there are some species, for example, black carbon, which are absorbing species, and they have the opposite response. And they tend to warm the atmosphere. And so they go in the same direction as greenhouse gases. And so they contribute to the global warming, rather than cooling, as I mentioned. It is the case for the other aerosol species. So you have this almost contradiction that if you have a removal of aerosols in a stronger quality policy scenario with reduced emissions, which is good for the elf and for the air quality, as I said, but they can actually have a large impact on climate. As I mentioned, the aerosols can mitigate the greenhouse gases effect. So obviously it's not that we would advocate to pollute more, but it's something that needs to be kept in mind and studied. Also, atmospheric constituents, and here I include also ozone and reactive greenhouse gases, they have different types of... They affect the numerical weather prediction, weather in several ways and across various scales. So you see the scales go from analysis time, which is now. The medium range is the prediction at, say, 10, 15 days. The subsesional range is usually considered between one and three months. And then you have the seasonal range, which goes up to six months or a year. And the impact happens at different scales and through different mechanisms. And it could be dynamics or thermodynamics type of mechanism. The main one being the interaction with the radiation, either the radiation emitted by the sun and the solar spectrum or in infrared or in the UV. And you have also indirect mechanism that are happening, as I mentioned earlier, through clouds. So the way aerosols interact with cloud and precipitation and then they can modify the behavior of the clouds, radiatively speaking. They can impact through what is called the four divart tracer mechanism. This is a bit specific to the assimilation, but the fact that the atmospheric constituents are transported by winds can have indirectly an impact on the wind themselves through the assimilation. So that actually has been studied, but it's still something quite open, this type of interaction within the NWP system. And then I mentioned already the radiative transfer, for example, in the assimilation of radiances for temperature and water vapor assimilation. They can impact the water vapor directly to oxidation. That's the methane in the stratosphere and also have a very important role in the Lansi-Harmus V interface exchanges, particularly the CO2. So as you can see, there is a variety of mechanisms. So it's quite important to look at these aspects. I will start with the ECMWF experience and then I'll talk more in general, but here's what ECMWF has said over the years. First of all, the development of the atmospheric composition in the model used by ECMWF, which is the integrated forecast system. It started actually in the late 1990s with the inclusion of stratostatic zone and then progressively through a series of projects that are listed there, Gems, MAC, and now CAMHS, you can see the acronyms below. Progressively, the system became more and more complicated and couple chemistry and aerosol and greenhouse gases and then now integrated chemistry are now included in the IFS. So it has become a very complex. Initially, ozone was included with a simplified model, but now it's fully in the CAMHS system, it's a fully integrated chemical system. And there's been quite a bit of upgrades to climatologists that are then used in the MWP configuration, both for reactive gases and greenhouse gases and also aerosols. And at the moment, in the MWP configurations, a climatology of aerosols and ozone is used, but in the CAMHS configuration, which runs at a lower resolution and includes all the species that I mentioned, they run with interactive prognostic aerosols and ozone. This is run by the Copernicus Ammosphere Monitoring Service and all products given out are free, and you can have a look at the website and go to the database, because this is funded by the European Commission and it's operational services. So CAMHS deals mainly with pollution on a global scale. They use the IFS model and integrate mostly satellite observations to produce the forecast of aerosols, reactive gases and greenhouse gases. And they use a lot of ground-based observations to verify the model prediction. And then they use anthropogenic emissions and fire emissions to prescribe the fields. And it is quite a complex model. It has a full integration with the meteorology. So it's part of the same MWP model that ECMWF uses for weather forecasts. It uses the 4D VAR system as well, and it has integrated chemistry and aerosol representation and an integrated natural biosphere model. So that's the CAMHS system. The quality of this atmospheric composition forecasts has improved over the years. And you can see, for example, the top panel, you can see that from when this started in, it was 2008, when it was pre-operational, the skill score has measured against independent observations from Ironet. Ironet has increased and you see the black line with the positive upward trend as far as the increase of skill score. And for the ozone as well, you see quite a nice improvement from when the first forecasts went operational. And you see, in comparison with the analysis, now the quality of the real-time forecast is just as good as the analysis, which is quite a success. And you have, for example, here an independent verification using carbon monoxide sentinel 5P observations. And you have to trust me on this one that when I say that the top is observations and the CAMHS model is at the bottom, but it is definitely quite a good agreement with this field, between these fields. So the quality of the forecast of atmospheric composition has improved over the years. And it's quite good. This is an example from the more recent past. There has been a huge event in 2020. And actually this year is quite promising because there have been already a couple of big events. But this one was last year showing the forecasts of this huge dust bloom that affected the air quality in Puerto Rico. You see from the tweets there, you have a picture before and after the arrival of the dust bloom at two sites in Puerto Rico. And you see the forecast already, you had a very good signal already at day five of the arrival of the prune in the right location. So that was quite a good forecast. And this is some verification showing Aronet time series. So you have the time, so that's June. And you see the peak when there is that increase, the blue dots are the independent observation and the solid line is the model prediction. And you see a very good timing of the arrival of the plume. Also quite a good agreement in the amount of aerosoloptical depth that was predicted by the model and then observed as well. And you also get quite a decent agreement in the vertical location. So those are the lower two panels are to Calypso orbit. So that's a LiDAR from space looking at aerosols. And you see like the location of the aerosols in the model, the bottom panels is quite in good agreement with the one shown by the observations from the satellite. Also, aerosolinformation is used to build better climatologies for applications in NWP. I mentioned that briefly. And for example, this is work done by Alessio Botso in using the aerosol from the CAHMS Interim Reanalysis and be able to include them in the NWP. And it's used now operationally since 2016. And you can see that the climatology with respect to the old one, you see the plot in comparison with aeronet observations. You have the old climatology in green was quite showing quite different amount and variability with respect to the current climatology, which is in red. And then you see in black, the prognostic arrow. So you see that overall the average, the acclimatology captures the average behavior very well, including some temporal variability because it's not a fixed climatology gets interpolated. And then, but it misses the big events, which is normal. You don't expect, obviously, climatology to do the same job as the prognostic fields in every day, every single circumstance. And actually this new climatology reduced perform quite well in NWP and reduced bias in the five days. So now in forecast and 925 at Topascal, which was always something that problematic. And I'll show you also what happened in the monthly, in a minute. And there is higher consistency between the climatology and the prognostic arrow source with this new climatology. So yes, going to the experience regarding the S2S scales. So we have a look at the study that I performed together with Frederic Viter. The results are published and they're summarized in a paper in the monthly weather review. So at the time we were running a model version, which is currently quite old actually. We in ECMWF it's measured by cycles. And this was cycle 43 R1 and currently we are in cycle 47 R1. So there's several years have passed and in between several changes have happened to the model. But at the time we had that system and we looked at what happened when we include the fully prognostic arrow so in the radiation scheme. So fully prognostic interacting with the radiation only directed effects were included. No effects on clouds, so no indirect effect. And we used free running arrows with observed emissions however for biomass burning and I'll come back to that, that was quite important. And we chose an ensemble of 11 members with five different start dates. So for a total of 55 cases for more robustness and then we ran a forecast period from 2003 to 2015 and six months of simulation. So we had two control runs and two prognostic runs. So the control runs were one with the old climatology. You remember the green curve that I showed you? So that was by from Teghen in 1997. And the new, the control two is the new climatology that Alessio Botso developed based on the KAMS interim analysis and then two prognostic one initialized from the KAMS interim analysis and one initialized from a free running arrow source simulation. So more initialized from a climatology. And so apologies, the control here in this slide is the control one. So the one with the Teghen climatology, the old climatology which was operational at the time we did, it was being changed at that time but it was still used. And I show here the aerosol impacts on the monthly forecast. So this is the temperature bias at week four. So four weeks into the forecast in the two cases you can see just like having a look at the impact somehow with the color, you see a reduction in temperature bias in most areas in particular Mediterranean basin, the Asian dust belt and also the North Atlantic dust base belt. You have to maybe focus specifically in these areas and the impact can be between 0.5 and two degrees which is actually not negligible. So we were quite happy to see that. We also look at precipitation bias. So this is again week four, again control one and the two prognostic runs. And you can see that, and these are yes, precipitation anomalies. So this has been the bias using the forecasts. And you can see again, well, maybe you have to picture a little bit more on this plot, you know but the precipitation bias are slightly reduced in certain areas, particularly over East Asia and in an amount to, you know, not large amounts but 0.5 to one millimeter a day. And again, you have to remember that only interactive direct interaction with radiation is allowed in this run. So there is no indirect impact on cloud or precipitation. So it's only by modifying the, basically the heating profile that these effects are, you know, achieved. And looking at run probability skills course, this is sorry, it's very busy, like, you know it's a scorecard. So it's not very easy to understand if you're not familiar with them but just of it, you have a series of, you know, meteorological variables in that column for example, two meter temperature surface, sea surface temperature, mean sea level pressure, temperature at various level 50 at topascal, 200 at topascal, et cetera. So you can just, you know, just have a look at that, you know, and you don't need to take it all in. You can find more information in the paper but this is just to show visually, you know in one card, you know, what happens when you change something in the model. So it's quite useful. And you can see one is for the prognostic run one which was initialized with the climatology and the other is prognostic run two, sorry prognostic one was initialized with the camps in terrestrial analysis whereas prognostic run two use the climatology for the initialization. And that's compared to the control with the tag and climatology. And you see a lot of blue. A lot of blue, dark blue is positive. And the interesting thing is you see an impact also a week four, particularly at upper levels. So it's, and sometimes in variables, you know like winds and temperature at 500 in one case or winds at upper levels 200 at topascal. So it is definitely quite an interesting result which we didn't quite expect. So what could be the reason for that? Why does, why do the interactive prognostic arrow so have this type of effect? Well, first of all, we went and looked whether there was some explanation for this impact at the sub-seasonal scale. And we found out that actually the sub-seasonal areas of variability explains a quarter of the total AOD variance. The intracisional variance of AOD. And you can see that there is quite a bit of, first of all, seasonality in the aerosols, which is, we knew that, but, you know that can lock in with certain phases of, for example the MGO. And so that's where we think that the predictability comes into play. And we had a look from our runs exactly at that how the aerosols were formulated by the MGO. And in particular, we looked at the dust optical depth anomalies. And we looked also at the biomass learning. So let me start with the dust. So these are dust optical depth anomalies. And in one column, you have the fields from the prognostic run in the different phases of the MGO, phase two, three, four, five, six, seven, eight, one. Sorry, I'm not gonna be able to go into like, you know, details of what these phases are. But, you know, they are just different phases of the MGO with convection and happening in different areas of the world, across the world. And so you have that panel from the prognostic run and a panel from the counseling theory analysis, which we use as a verification. It's not perfect verification, but it shows the same pattern, meaning that, you know, in the prognostic run in the forecast, we had the very similar patterns as sort of quote, quote, observed because there comes interim re-analysis constrained by aerosoloptical depth observations from the modus. So you could see that there was, you know, an element of observations in there. And the patterns are quite similar actually. Although for the comes re-analysis, you can see bigger, you know, bigger features, maybe bigger anomalies, but they're still like, you know, very much present in the prognostic run. And in particular, opposite phases of the MGO have opposite impacts on the aerosol variability, which suggests that the aerosol modulation by the MGO is a real signal. And you can see that as well in the carbonaceous aerosol. So these are aerosol produced by biomass burning or by industrial emissions. And you can see that the patterns again are not identical to the ones shown by the re-analysis, but they are quite similar. And somehow there is quite a good correspondence. And as well as in an opposite phase, I think in the case of the dust, it's more visible because the dust is mainly, the main source of dust is the Sahara Desert and that's in the tropics. So you actually could see more there, but you can still see quite a bit of an impact as well in the biomass burning in the carbonaceous aerosol anomalies. And of course, you know, the other thing that you have as a byproduct of this is also the prediction of dust aerosols, for example, a manta head, which is not something that has been done before. And it's sort of a byproduct of this investigation, but it could be quite interesting for many applications. People interested in dust forecasts. There are several countries that are very much affected by dust. And so we actually here compare the skill of our runs with respect to persistence that is just considering the previous week and just continuing with the same type of forecast. And we noticed that in particular week one, but also even a week four, the skill of the monthly forecast of dust was lower higher than persistence, particularly when initialized with the KAMS interim analysis, which is the, you know, orange bars. So definitely an interesting thing to explore further. Then, you know, the other thing that is quite relevant is looking at what happens in the case of extreme events. For example, this is the Indonesian fires of 2015. That was a record breaking year for Indonesia. This was a humanitarian crisis because an excess of 400,000 people died as a consequence of these intense fires. And the whole region burned from August to October. And it was possibly connected with a very dry season connected to El Nino. And there was a huge amount of, also greenhouse gases emitted. This described as the equivalent traffic of the entire annual output of Germany, so you can imagine. Here I put some like some plots provided by Mark Parrington for that. And also we had a look at the biomass burning anomalies. And that's reporting in a short article in State of Climate, in Bulleted of the American Theological Society. And you could see that the anomaly were just off the chart in that region. So what happened in the runs that we had? Well, actually, because we used observed emission, as I mentioned, that's very important. So it wasn't really like a forecast. It was more like a forecast exercise. So we knew what emissions we had. So those were prescribed. And you have a panel on the left showing the fire radiative power average over August and October, 2015. And that is from where the emissions are derived and they are fed into the model. And then what happens three months, actually, because we had those emissions three months ahead, we could predict the cooling due to the smoke aerosols. So that is quite an earth off. But, and you can see that the pattern of cooling is actually mirrors the areas, you know, you have like a perfect pattern of the, the, and the areas where the highest fire radiative power, highest like, you know, fires were observed. And you could also see that, you know, six months ahead. So this is started, the forecast started on first of May and then you had the temperature anomalies for October. So six months ahead, the signal was there. That's just because we used observe emissions. Of course, that wouldn't be the case if you didn't have observe emissions. But that actually shows how important this can be, this type of very extreme events. And also it's, you know, incidentally, it also shows that there is a need for predictive fire dynamical model for this type of cases in which you can possibly have like some kind of a predictability for example, for particularly for events that are connecting to El Nino, like this was. This is an example from Tim Stockdale showing the importance of stratospheric sulfate for in this case, seasonal prediction. And you can see that if you have like, you take era interim as sort of like the baseline and you look at a run of the seasonal system at the time was season four, now it's five, but with an incorrect vertical distribution of stratospheric volcanic sulfates, you have a completely wrong temperature response in the seasonal forecast. So you really particularly in case of major volcanic eruptions. So you can definitely see how important to have like the correct aerosols is for certain applications. And recognizing that is the NWF is has led the consortium for a European Union funded project called Confess with the objectives to understand, in fact, the importance of aerosols on the seasonal prediction. And there's a component in this project that looks specifically at volcanic aerosol and biomass burning as well. So very much inspired by the research that I showed you. So we are currently in the first year, it kicked off in November, Project Confess, and it will run until 2023. So we hope to make some more step forwards in the treatment of aerosols and also vegetation and land cover in the C3S seasonal systems to be able to improve the seasonal prediction. And in general, the NWF is very much interested in exploiting the atmospheric composition developments for NWP. And this is being summarized, the requirements and recommendations have been summarized in a tech memo by Rossana Dragani. And the aim is to understand through a thorough and coordinated testing what level complexity or coupling it is needed in order to get the most of NWP, for the NWP aspects. And the focus is on ozone aerosol and CO2. Also at the level of WMO, the War Meteorological Organization there's been an initiative on understanding the impact of aerosols on numerical medium range and sub-seasonal prediction. And this is a project sponsored by the Working Group on Numerical Experimentation, the S2S project and the Global Atmosphere Watch Project. So those are all WMO initiatives. And here I put the names of the colleagues that are involved. And we created, well, the goals of the project are to understand how important the aerosols are at the short range, medium range and S2S time scales. And also what it is, how, whether there is skills too for the prediction of aerosols at all these scales. Well, we know that in the short range, the systems are rather skillful, as I showed you earlier, but at the S2S scales, that's less clear, but there is definitely some good signal there. And also, so it's both like an investigation of the impacts of the aerosols on the NWP so on the meteorological variables, but also to understand how good the predictions of the aerosols in say is for various applications. And we look at, we set up a protocol for these experiments. The forecast will have to cover 2003 to 2018, the focus will be on monthly runs. So it's not gonna be a seasonal impact on the seasonal, but it's definitely the sub-seasonal scale. And we focus on two start dates, May, June, July or August, September, October. The first set of dates is to look more at the impact of Saharan dust, because those are the months in which you have the biggest activity in dust, Saharan dust. And then you have August, September, October is the biomass burning season. So that's where you focus more on the impact of the biomass burning aerosols. And there are several variables to be analyzed, to be like two meter temperatures, surface winds, precipitation, aerosoloptical depth in say, et cetera. And this is all run by CP Tech, Ariane Frassoni, is the architect behind it. And they will also run verification. And for now, the protocol has been distributed and we have a timeline for the completion of the experiment which is two years from the start of the exercise which started last year. So sorry, two years ago now, time flies. So now the deadline for the runs is September, 2021. And we have quite a bit of interest at several centers. And in some cases we already have the simulations I show you a couple of examples right now from various centers, CMA, from China, KMA, Korea, NASA and NOAA from the US, JMA, Japan and ECMWF. So those are our centers that have S2S systems with aerosol capabilities. So that's the requirement. So this is like a repeat of the runs that we did with Frederick with a more recent cycle. As I mentioned, this is a model version, right? And now the current one, it's a cycle 47R1. So many changes have happened, including now the aerosol climatology which is officially the one defined by Botzo and based on the comes in the analysis. And when we look at, this is just the measure of the skill, I'm not gonna go into the details but just look at where the biggest impact is in areas with aerosols. So particularly the desert and the central Africa biomass burning area. And unfortunately the red means degradation. So in this case, the run with the prognostic aerosols contrary to what we found out before is less skillful than the run with the climatology. But so it's not going in the right direction we were hoping for, but the important thing is that it shows a very high sensitivity to the aerosols. So, and that's very important, I think, because it does show again, regardless of the model and the direction of the impact, let's say, it shows that it's very important to get the aerosol correctly in the model. And that's actually a work that Adrian started when he was still at ECMWF together with another colleague, Mark Rodwell, showing exactly that how important the correct definition of the aerosols is. And, but yes, I wanted to have a look at what happened to the Indonesian fires of 2015 in this new model version. And once again, you see the same strong regional signal connected to wildfires and as in previous experiments. So that was good to see, meaning that maybe globally we didn't get the hoped for impact, positive impact of the prognostic aerosols. But when you are looking at extreme events, obviously you still need prognostic aerosols and not the climatology. And this is work from the Chinese colleagues, Yun Chen Zhao Yao, Yun Chen, sorry, I'm not pronouncing it right, but Jiang Chen Yao and Tong Wenwu from the China Meteorological Administration. And they also ran the same setup with their model. And this is just to show a comparison of the dust fields that they obtained. And you have, sorry, the fonts got a little messed up, but when it says dust in parentheses comes, that's a verification using an independent data set that comes to the analysis. And then you have dust direct only, which is the prediction. So the label's got a bit messed up. And you see that the prediction is actually quite remarkably similar to the re-analysis independent data set. So very nice skill in the monthly prediction of dust. And you have like the panel at the bottom on the right that shows higher amounts of dust, particularly in the non-analysis here. This one is from a fixed run. So like having like a fixed climatology, so no prognostic aerosol. And you can see that actually it is the one with the prognostic interactive aerosols that has the best resemblance with an independent data set, the independent re-analysis data set. So it's very interesting that they get similar, good skills from the one month prediction of aerosols like we got in our runs. And we need to also look at that in the current runs for this WM exercise. So okay, I'm down to just two slides. So I'm about to conclude first a summary. Well, I don't think I need to convince you that aerosols are an important part of the earth system. And in general, atmospheric composition plays a huge role in climate and weather. If you have an accurate numerical weather prediction model with physical and chemical processes, this is in general a good start for to model aerosol and atmospheric composition. So like I think the cons forecasts are quite good because they start from a quite good numerical weather prediction model. In return, some elements of the composition can help improve the weather forecasts, or at least we hope for that, at various temporal scales, including the S2S scales. And of course, it's a trade-off maybe between the degree of complexity and the benefits that this offers, particularly when you're talking about specific applications for NWP. And also there is, I think the added benefit that the sub-seasonal to seasonal prediction of the aerosol fields themselves could be of use, particularly as I mentioned for some aerosols like biomass burning and dust. There are some other questions. I mentioned the complexity versus benefits. And it's very difficult to find one size that fits all in atmospheric composition modeling because you would need very high resolution for certain applications and you would need maybe like ensemble modeling for others. So it's very difficult because also it's computationally very demanding. But definitely we need more scientific investigation. There is limited experimentation that's been performed so far. That's why the WGNE, OS2S experiments, WMO experiments are very important for that because it offers a chance to have more models look into this problem and come up with solutions. And we know that climatologists are extremely useful. In fact, right now at the moment, in the WF France, we're not able to beat the current aerosol climatology with the prognostic aerosols but they are not so good for extreme cases. And that's something that I think we need to also look into because extreme cases may be rare but their impact is very important. So that's something that S2S system needs to address. Then there is of course, as I mentioned, the cost of additional model complexity and there are some, for example, single precision which still hasn't been explored. And then there's a lot of maybe optimizing rewriting called rewriting that could be introduced. But these two points are mainly to convince my upper management to look into this problem rather than, you know, so they're just open questions for me and for us, not necessarily scientific but they could definitely be quite important for the future operational configuration of, for example, the Asian level of model or as we have seen the CMA model for that matter. So I guess that's all I have and I leave it here. Thank you. Thank you so much, Angela. Okay, so... Adrienne, shall I stop sharing? No, no, you can keep it there in case you have a question about a specific slide that comes up. Okay. So we have actually had a few questions come through which you weren't able to see, Angela, because they just come through to me, I realized. Yes, I am not able to see the chat. Yeah, I thought they would actually come through to all of our co-hosts, but anyway. So the first question was from Kirsten Tempest. So hopefully Kirsten is now unmuted. Kirsten? Hi, yes. Thanks, Angela, for the nice talk. I was just wondering about the actual model that you're using and it was related to a talk of hers yesterday by Eugenia Kalney talking about the importance of a two-way feedback between human-produced aerosols. So it could be from fires and the Earth system models, so like the normal NWP. And so I was just wondering if this aerosol implementation is that ingrained in the actual NWP or is it like an external implementation? And so is there like a two-way feedback between the model and the aerosol implementation or is it just data of the aerosol on the atmosphere is fed into the model? I hope that was clear, my question. You can let me know. Absolutely, thank you, Kirsten. Yes, I mean, you're absolutely right. It's very important. There are a lot of feedback mechanisms at play. In this case, the aerosols are completely integrated in the NWP model. They are run with a slightly different configuration at lower resolution because of the cost, but they are completely integrated, which means that if you modify temperature or winds, that also in turns modifies aerosols. For example, the transport, but also the emissions because aerosols like Sarandas, they are, the emissions are parameterized as a function of surface winds and other parameters, but mostly surface winds. So if you modify the winds because of the presence of aerosols, that will in turn modify the emissions of certain aerosol species. So, yes, absolutely, you are right. It's a complex system. In this case, it's fully integrated. So, yeah. And I think it's quite important to have it so, to be able to understand the various feedbacks. So, yes. Okay, thank you very much. Welcome. Can I just ask Angela? You say it's fully integrated, but I presume that doesn't include yet micro-physical processes in clouds. Or is it actually accounted for also in the, should we say the warm rain and ice micro-physical or physical processes now? No, you're right. The micro-physis comes into play in the, only as a removal mechanism for the aerosols. So, you have parameterization for wet sedimentation and removal of aerosols, but then you don't have a mechanism, say to generate more clouds because you have more aerosols acting as CCN or isonuclei, for example, dust is known, you know, good isonuclei. But no, we don't have that to waive feedback. Yes. Okay. Yet, yeah. So, the next question is from Sebastian Müller. Thanks for the nice presentation. I have a question out of curiosity. With the present event of Saharan dust over mid-Europe, do you have an educated guess or an estimate or a rule of thumb, how much temperature reduction this brings like for daily maximum temperature? Yeah. Yes, thank you, Sebastian. I mean, they are looking, my colleagues in camps are looking exactly at the case that you mentioned because it was quite extraordinary. And I think you might have seen on the internet a lot of plots of photos of pink snow over the Pyrenees, for example. So, to be honest, given that it's winter and you have quite a fast removal because, you know, it goes snow down quite quickly, I don't think you would see a big impact on the temperatures. I may be wrong, but it's more like a gut feeling and it's actually being studied as we speak. I wish I had some plots to show you about that, but it's still ongoing, you know, that analysis. However, you know, it's unfortunate that the model doesn't have indirect effects because I believe that the fact that it's been, you know, cleared out so quickly, it must have had a huge impact and having a lot of available moisture, you know, to be precipitating the form of snow. I think that might have been the biggest impact than the direct on temperature. But as I said, it's more like a gut feeling that I have than concrete. But if you think that, you know, if the same type of infection had happened, say in the summer, you know, and over, say over ocean, you would have had probably a bit of a, more of an impact on temperatures. Usually you have one degree per one AUD measure. So this defense can have very large AUD. By the time it gets to Europe, not that large, you know, but still, so you would have, you would definitely see the impact. We see that very well the impact of, for example, forest fires in California from last year, from September, even several degrees impact on surface temperatures. So specifically this episode, I don't think so much, but in general, yes, roughly one degree per one unit of AUD. Okay. Thanks very much. Sorry. Thank you, Sebastian. The third question was from Karam Mansour, and he asked me to read it out. So he was saying, when we look at aerosols, we often focus on the anthropogenic or high impact extreme cases. But he was wondering, what about basically marine environments and marine aerosols, such as sea salt? He was saying, how much effort has there been to validate the accuracy of those in these, should we say, reanalysis and forecast products? Do we have an idea of how accurate the marine environment assessment of the aerosols is? That's a very good question because marine aerosols are the most elusive, I would say, although they're probably the most abundant. Because when you think about the ocean surface, you know, particularly if you consider background aerosol as you mentioned. So, yes, I mean, the satellite data are usually actually probably more accurate over ocean than land. So you'd have kind of a good, you know, like handle on the, say, on the AOD. But it is also difficult because it's low values, so you can lose sensitivity and accuracy. And you don't have many stations from the ground, so the aeronauts stations that I mentioned in my talk, they're mostly on land. So it's true. I guess they are the most elusive aerosols and difficult to know. The cost model looked at, and there was a case that when they were overestimated and that has been, you know, the modelers incomes have addressed that, trying to like change the parameterization, but even that is very difficult because there are a few observations directly of marine aerosols, you know, and you may have some ship observations, but, you know, to develop global parameterizations that are valid all over the world, you know, for every condition, wind condition, et cetera, et cetera, it's difficult. So, yes, it's probably, I would say they're probably the weak point, the weak link of the... So do you have a good, I mean, with aerosols, one of the tricky things is you can have sources of errors, first of all, of the assimilation in terms of the atmospheric composition, but then also with the sources, and then you have errors associated with the dynamics, which of course will grow over time with lead time, and then the way you represent sinks, because you mentioned wet deposition. So is there a kind of a feeling, for example, for example, at day five or S2S time scales, which of those dominates? Is it, for example, the meteorological errors in winds that dominate at like week three? And is it like our sources, for example, important for like dust aerosols, for example? It's a very good question, Ideal. And to be honest, not one that we have addressed fully. I can tell you my gut feeling. For certain aerosols, it's the transport, but for others, it's definitely the emissions. Particularly, I would say biomass burdening, you know, because if you just don't have the emissions, there's nothing to transport. So definitely, but for dust, because dust is parameterized using winds, then it's more like, it could be a combination as well, not having the right amounts and the right size, because it's not only, you know, the total amount of aerosol, but also it's like how big they are and how fast they get deposited or how fast they can, how far they can travel because they are maybe finer particles. So yes, it's a very multifaceted problem, but I'd say, you know, it really depends on what aerosols. So probably transport more like, you know, wind emitted aerosols, so natural aerosols and biomass burning, they are natural aerosols as well, but, you know, I mean, like, ones that are more connected to episodic emissions. Okay, so that leads on to a question actually that's related on observations from Bruce. Hello. Hi, Angela. Hi, Bruce. What are the most important observations of aerosol and composition now and possibly in the future? Well, thank you for asking, Bruce. So we, I didn't mention that, didn't stress that, but to be honest, you know, the aerosol problem is highly underdetermined. So we have like too few observations because right now we rely almost entirely on aerosol optical depth, which is an integrated quantity and it sums up the contribution from all species. But for example, at the beginning, if you remember, I mentioned that depending on what species it is, it could have like a different impact on weather and climate, say warming or cooling. So I'd say that if I could have my wish list, speculated information would be, you know, on the top. We have now LiDARs. So we have at least the vertical distribution. They're not used like in the analysis at the moment, but we are working towards that. So, you know, I'd say that there is still room for a lot of improvement. And there is like a very nice ground-based system for that of stations that provide very detailed information like scattering or number concentrations. I am like more thinking for the assimilation aspects, you know, like we're still are somehow under constrained, particularly satellite of global observations, you know, because even the networks obviously do not cover areas. For example, you have nothing in the Sahara Desert, which is the biggest, you know, source of aerosol. So, you know, you have like several layers of problems, you know, is somehow we there are very good observations, but we are still needing more, you know, particularly for assimilation applications. OK, thank you very much, Bruce, for the question. We have two more. I think we have time to squeeze two more in, Angelo, if you're OK for me. Yeah, I'm up now. So the next one is from Kristofer Sanno and. Hi, thank you, Angela, for the great talk. I just have some question about how much time does it take for one week or one month prediction during an extreme case? Like, for example, during the volcanic eruptions or during the Indonesia fires? That's a good question, Kristofer. Sorry. I'm sorry, please continue. Yes, I to be honest, do you mean like computing time or because there are, you know, I guess there are this different level of, you know, how long it takes. The problem right now is that for extreme events, we don't have a mechanism to activate the emissions, you know, so there we would need a mechanism to be able to say, you know, prescribe the emissions or at least, you know, tell the system that it is, you know, there is a volcano. She's roughing and we find that, you know, at the moment, it's it's not quite there. And as far as how long it takes, you know, if you're talking about computational cost, to be honest, at a lower resolution, the aerosols do not add. Well, I say that they add approximately 40 percent to the computational time, so takes 40 percent longer to do that, to do a forecast. But it seems it's a lot, of course, particularly if you're in a time critical type of path. So you have to issue the forecast, you know, within a few hours. At the same time, as I said, we haven't tried any optimization, any specific, you know, like things to make it faster. So it is difficult to to to answer your question, basically, you know, I I don't know, but I don't know if you wanted to ask something else as well, because I came in. So if you if you have more or if this didn't answer, please let me know. OK, thank you very much. So the last question is on the Madden-Julian oscillation. So I'm going to pass the floor to Sergio. Hello, Angela, great talk. Many thanks. And I was wondering if you could explain a bit more about the relationships between the M.J.O. and the oscillations, please. Yes, so to I need to first do a premise. And thank you for your very interesting question. This has not been really studied very much. So we look very much forward to looking at the other S2S runs, you know, to see if they see similar patterns that we do. So what we see is not only specific of the ECMWF model, but it's a robust, you know, physical signal. However, we do think that there must be a connection between M.J.O. and aerosols, because M.J.O. is the main mechanism of like that drives predictability at the monthly scale. And you have a patterns of convection and convection and dryness. And when you have convection, you have aerosol removal. When you have dry, you know, like pattern of dry periods, you may have more emissions, for example, of desert dust. Or this is a bit, you know, maybe stretching it. But say if you are in a very dry spell and it's fire season, because most of the fires, particularly in Indonesia, they are also induced by because of agricultural practices. So they're not only naturally occurring. But say if you, like, have had a particularly dry spell, you know, there might be even more, you know, more fires than you would have usually. So definitely this pattern of precipitation in dry spells. So, as I said, I cannot prove it 100 percent. We can at this point is more speculation, but I don't see why not. You know, how is that possible that the M.J.O., which has such an impact on precipitation patterns in the tropics, as how come, you know, how is it possible that it would have no impact at all on aerosols? You know, it's it's sort of like I answered with the question. But I don't know if I convinced you, but I think, you know, there is there is some like, you know, logical connection there. We just need to explore it more and maybe try to look at observations, you know, see what like observations tell us. Brilliant. OK. Well, thank you very much, Angela. I think that concludes all the questions. It was really a nice way to kick off a series of lectures. So I just want to thank everybody for tuning in. We had almost 100 participants listening to today, which is a fantastic number to have here. So thank you, everybody, for tuning in just to remind everybody, if you think that somebody might be interested in these talks, please pass on the details of the registration page and the description. Next week, the series passes over to Trento to do the hosting. And their first invited speaker, which is the second speaker of the series, is Massimo Volocina from the University of Edinburgh. And that's actually going to be continuing on the aerosol theme. So I'm not sure if that was by chance or by design, that we have the two aerosol topics right at the beginning of the series. But I think it's quite nice, because it gives some continuity to the theme. He's going to be talking about regional impacts of aerosols on climate. So I hope to see you all next week at three o'clock here in the second talk. And in the meantime, again, once again, thank you very much and have a very good week. Thank you again, Angela, and see you all soon. Thanks, everybody. Thank you, Angela. Bye. Bye. Thank you.