 I mean, you just all introduce yourself. So we're not going to introduce all the people on the web. But some people just did make the effort to present themselves. So I just take this opportunity just to, if you can see us, just three persons for the moment having responded. So we've got, for example, Samane, sorry for the name, if I just don't say them right, from Ivan. Excuse me, from Ivan. And she tells us I'm, let me check, sorry. She's from Ivan. She's graduated in environmental civil engineering. She's teaching industrial wastewater treatment at university. And she applied for PhD of environment in field of pollution. So we would have, for example, Samane. We've got Noor Abdul. She's from Jordan. She's got a PhD from Jordan. She's an assistant professor in the Department of Public Health at the College of Medicine in Jordan University of Science and Technology. And we've got La Thang. And then I'll leave you, Marie, too. Massimo, sorry. So we've got Therese Salame. Yes, that Eric knows. And Therese, who is working IMT in Lille, north of France, Lille-Douay, and is a postdoctoral fellow in atmospheric chemistry, working on emissions, sources, measurement of pollutants. That's all, just to create a link with internet and you. Sorry. That was nice. Thank you for doing that. And we know both, two persons, Noor Abdul. We know also because we met her at the Cyprus meeting last, I don't remember when was that. OK, so thank you very much for inviting me to talk about this topic. I know you might be a little bit tired. So I will try to keep it as more informal and light as possible. So the topic is our quality measurements. And I also added modeling because, of course, I am just a few words about my background. I am a statistician and I work as an epidemiologist in the Department of Epidemiology in Rome. So I am not an expert in our quality measurements in terms of engineering of the monitoring sites and stuff. I'm not really a modeler in terms of our quality models. So my perspective will be entirely in terms of what shall we do with this data, with this air pollution estimates with the purpose of estimating health effects and with the purpose of implementing and health impact assessment procedures. So I think that was important. I will not speak too much about definitions because I think most of us will know what air pollution is just to agree on some of the aspects. I think also we are a little bit late in the schedule. So I will try to skip some of the slides. But basically, when we speak about air pollution, we speak about the presence of toxic chemicals or compounds, not just anthropogenic, but also natural, biological. In the air at levels which might pose a health risk. So this is our perspective here. There are, of course, many, many different pollutants which we might be interested in. The ones I'm reporting here are those which are included in the EU legislation because there is some evidence. In some cases, it's a very massive and convincing evidence. In other cases, not as much, but there is some evidence of their health effects. They are toxic for human health. So there are gases like SO2 and nitrogen dioxide or nitrogen oxides. We will speak a little bit more about PM, particularly matter because that is the pollutant. Also, Carla was pointing out about PM as being in the top 10 risk factors for human health from the global burden of disease. That's the one where most of the research, but also most of the, let's say, policy agenda is focused on. There are, of course, metals, the results on, et cetera, et cetera. There are many, many different pollutants which we might be interested in focusing. When we speak about pollution, we think about sources and ensure there are people in this room much more experienced than me about identifying sources or maybe apportioning different sources to specific amounts of concentrations. But here when we speak about sources, we don't speak only about anthropogenic sources like usual ones like industry or traffic or waste disposal. We also think about natural events and since we are speaking about the Mediterranean area, but there are also people from Iran or other places which are very much impacted by dust events. That's another source which we might be interested in, so natural events which might pose a real threat to human health. So there is a very extensive list of possible sources which we should take into account. And also when we speak, when we think about pollution, first of all, of course, we think about primary pollutants which are those directly emitted in the atmosphere or near the ground where we go to measure these pollutants and this is a list of primary pollutants but there are also secondary pollutants. So basically those pollutants, we just come after chemical reactions so they are not directly emitted by the sources but they derive from chemical reactions which occur in the low atmosphere. One example is ozone. Ozone is not a primary pollutant. It can be transported through long range transport but or basically most of it comes from reaction so it's a secondary pollutant. We already spoke about concentrations so maybe we don't go too much into these specific definitions just to say that the air pollutants we measure them in terms of concentrations which is basically the ratio between the mass of a pollutant we are accounting and the volume of air we have in mind and generally they are measured in terms of, for example, NO2 in terms of micrograms per cubic meters. So as I was saying, this talk speaks about measurements but I will not speak too much about that. There are, just to say that there are reference methods existing for each pollutant that they are defined by law. In some cases these methods are very much consolidated and automated so no big need of, you know, maintenance or going there for the individual operator and retrieve the measurements and that's the case for most of the gases. For example, there are other cases where in principle the reference method for PM would entail the weighting of a filter which is a lot of work from an individual operator but luckily there are equivalent methods which have been tested and proved to be absolutely reliable which are much more automatic and again I am sure there are people in this room which know about this much more than me. So a few words about the particular matter because I think that this will be the pollutant where most of our attention will be focused during this week. Particular matter is, as the word says, it's not just a single, you know, gas, a single molecule, it's a complex heterogeneous mixture of solid and liquid components and because it is a complex and heterogeneous mixture, its toxicity might very much vary depending on the source profiles, depending on the study area, depending on the reactions with meteorological patterns, et cetera. Sources can be very different from power plants, industries, motor vehicles, in cities of course, in general traffic is one of the most important sources but also domestic heating and but also natural sources. Again, this is very relevant in areas like southern Europe or some large areas in Asia also where there is a lot of impact from arid regions. There are several definitions but basically one of the most distinction between different types of PM is in terms of the size of PM because the finer in principle, this is quite easy to understand, the finer the size, the easier the penetration of the particles into human body and so most of the, let's say, research agenda now is focusing on the latter one, the ultrafine particles, which is particular matter with a diameter smaller than 0.1 microns, which is the very finest one and those which might not just stop in the upper respiratory system but it can go down to the lower tract, it can translocate to the blood circulation, it can impact also peripheral organs so determining a number of different adverse effects. This is where most of the research is going but the most of the available evidence is about fine particles, both because in the US it has been monitored since 20, nearly 30 years now but also because fine particles compared with the quartz fraction, which is this one, they also have been shown to cause a number of different adverse effects, not just respiratory but also cardiovascular effects and now the spectrum of the potential effects of these particles has been, has increased a lot. Now we know that it might impact also neuro-behavioral systems and also the neuro-generative diseases and it's bad for pregnant women with a number of different adverse effects on the new babies, et cetera. So there is a wide spectrum of potential effects of the fine particles but also I think for us it's important also to focus on quartz particles. I think for us thinking about, for example, the Mediterranean area where we know that many particles might come from arid regions and we know that the sides of these particles might be also in the quartz, in the quartz mode. So this is one quite typical distinction of a different PM and here this is a nice picture which is always shown just to give you an idea what do we mean by ultra-fine particles? So we are on the same range as viruses, fine particles, the same sides as bacteria, the quartz fraction is more on the sides of the cell or pollen. So this is just give a comparison scale for different sizes of PM. And the source profile might differ between fine and quartz fraction. Okay, so this has been already touched by Carla when we speak about air pollution. Of course we have in mind emissions because air pollution is caused by emissions but they are not the only cause of higher concentrations because in addition to emissions, we need to have transport and diffusion of the pollutants and so there is an important role which is played also by meteorology, by orography, by a number of different issues that will be touched during this whole week. Okay, low limits, I think it's quite important for our purposes because they very often help investigators to define counterfactual scenarios. What do we mean by air pollution as an impact compared to what? Generally, when we want to define the reference compared to which we do the health impact assessment we need some sort of threshold above which to define the impact. And this is also will be discussed in the next few days for each pollutant. There are different low limits and these low limits might be defined in terms of daily variability of air pollution but also in terms of annual averages depending on whether we are interested in estimating the impact on the short term or on the long term effect of air pollution. This is just an example of the PN10 monitoring network in Italy where you can see for example in the red dots that these are places here, all the red dots where there has been an exceedance of the limit values compared to what the legislation say it is acceptable. So with a detrimental effect for the population living in these areas but also with economic impact because when there are these exceedances there is some infringement procedure from the European Commission. Okay, more or less I've just said this. I think this is quite important to bear in mind when we speak about PN concentrations in Europe you see there are, this is not the latest report. Now the situation has a little bit improved but you see there are large areas with a lot of I mean red dots which means that in those cases the population lives in places where the limit values are exceeded and so there is a real threat to human health but it's also interesting to see that if we consider the EU, the European Union our quality standards only 21% of the population live in places which exceed such limits but if instead we consider the WHO the World Organization our quality guidelines which are I mean lower thresholds than those defined by the European Union legislation this percentage is 81%. I think this point also will be touched in the next few days but this is just to say that this air pollution is a real problem for the population in Europe and this is the same also in other continents because many, many people live in places where the levels of air pollution are not at all safe for their health. So we spoke about measurements but when we do health impact assessment we are not and also Carla showed a slide with different quality levels of air pollution estimates. We are not just okay with the few point of measurements that are available on the territory because they are not able to capture the very small scale spatial variability and so we really need to rely on models which complement the measurements and allow us to assess air pollution exposure to each and every individual level in a study area. There are different ways to do this. I will just mention briefly dispersion models because this is one of the most consolidated one but there will be experts I think in the next few days which will provide more details about how dispersion model works but basically the idea of these models is that they simulate emissions, transport dispersion and deposition of airport pollutants and also their chemical reactions. So these models are based on complex systems of mathematical equations using fluid dynamic laws of course so that having all these ingredients, emissions, transport, meteorological fields, et cetera, they are able to predict in this place as an average this year for example how much NO2 will be present in terms of concentrations. And this is just an example of the complexity of the system in the Lazio region which is the region in central Italy where Rome is located where you can see there are many different pieces which enter the chain and the final output will be something like this having either at the regional domain or for the metropolitan area of Rome having a sort of map where you can see the predicted concentrations of the pollutant and then the next step as Kala was saying before will be to assess and to attribute these levels to the population living in Rome and so to make the connection between how much exposed they were and how much their health was impacted by these levels of exposure. Another possibility which is in a way complementary to the dispersion model is the land use regression models again there will be people talking about this I think tomorrow but the idea here is that these models are aimed to predict these are statistical models not as much mathematical models based on specific physical laws these are statistical models which try to predict pollutant concentrations in different spatial locations by taking advantage of the spatial relationship between observations and land use characteristics. These models can be very easy or very complex depending on how many data sources we want to use to better characterize the distribution of our pollutants over space but they generally need as input observed measurements of the pollutant from one or more monitoring campaigns of course then we need data on land use characteristics for example we need to have the road network if we think that the traffic might be one important component to describe the spatial patterns of our pollutant we need to have a orography we need to have a length cover etc. population density there is a need of some GIS expert here who takes all this data handles the data and attributes to each point a vector of different parameters and then once we have this we need to develop a prediction model for the air pollutant we have in our study so this is just a brief example of the complexity that we have developed for Italy where basically the aim was to predict for each day and for each square kilometer the PM concentrations so we built a grid of one by one kilometer for all our country we collected a number of different many different types of data starting from PM monitors we have nearly 700 monitors in Italy and basically we use satellite data in order to be able not just to capture the spatial variability but also the temporal variability yes? no no it will take one minutes maybe two okay thank you so we collected information about the satellite data so the left side is aerosol optical depth which is a parameter which has to do with the amount of particles which are in the column of air as I mean retrieved from the satellites we're on the right side, sorry on the right side here you have a map of the vegetation from the normalised difference vegetation index which is also a satellite parameter we collected this information and as I was saying before using a lot of GIS techniques we attributed it to each one by one kilometer square a number of parameters like in this example population density or we checked whether there were important industrial emission points which might impact the air pollution in each square we considered the Korean land cover database to characterise each point in Italy in terms of land use characteristics the road network to distinguish between highways, major and minor roads in order also to capture the footprint from traffic other special parameters I'm just going over them quite quickly just to give you an idea of how much work you can implement with the aim of predicting PM concentrations in this case everywhere for every day and these are other, for example Saharan dust that's a topic which can be of relevance in many different areas for example in Africa of course or in the Middle Asia not just in southern Europe we detected these desert dust events and we used also those by using a mixture of models including atmospheric models back trajectories etc okay, meteorological data we already mentioned those and we have implemented a multi-stage approach I will not go into this just to show which was the final let's say output of this effort this is the annual, the average across 2006-2012 of PN10 in Italy but basically behind this map there are daily maps like this which can be very useful to whatever health impact assessment methodology we want to apply which is either in the long-term scale so focusing on the special contrasts or on the short-term scale so focusing on the really scale so focusing on the short-term contrast these are annual maps and this is the temporal day-by-day variability because depending on whether we are focusing on the short-term or long-term effects we might want to use either the special or the temporal variability in our estimates so I've finished and of course there are questions I will keep it, thank you thank you Massimo any question? thank you Massimo is that type of effort done in other countries than Italy? you are speaking about the last part about this special temporal country-wide model this has been done in several other countries in Europe at least and in US for sure it has been done in Switzerland in France we are doing this in Spain in Sweden as well so there is an effort in trying to apply this kind of methodology European-wide in order to have a sort of standardized approach to then run all the impact assessment models we want but yes there is an effort in replicating this in several countries thank you Massimo very nice I would like to take the opportunity to see whether people want to intervene on which kind of data they have because if we want to to do you I think satellite data now exists for the entire world for example freely available since 2000 they will stop probably heard that NASA is cutting money etc for now they are still on yes there will be new satellites also being launched very soon at least for Europe but I mean these methods have been used in China I've been used in India I've been used in many other countries not just in Europe so at least you know that for your country you will have that this is a good point of departure plus monitoring if you have stations now it's very up to date to have this remote sensor with participatory networking people carrying that giving you in different places but this is an important step of your I mean of your health impact assessment you really need the assessment of pollution to be sure to make reasonable no I think this is very relevant point I mean I was listening before some of you mentioned that I think it was someone there about having these filters with this XRF and this kind of daily measurements so I think the distinction we might try to understand is whether you have some sort of very refined measurements like the ones you were mentioning before which maybe do not cover the entire space but it's only a few monitoring sites but they will be perfect to capture the day to day variability and so that's one thing because then that's the case maybe you might be more into the short term part understanding whether this really is more related with daily mortality or hospitalizations versus on the other opposite is there are if there are models which are able to describe or to predict the air pollution on the special scale then that's another story because in that case we are more into the long term effect and the more conventional let's say health impact assessment approach so this is a first distinction we might try to make to understand who has what in terms of data Massimo would have a question from the University of Columbia asking are precise the model with respect to actual measurements and for how long in the future can this model predict okay so I guess the question was about the special temporal model which I presented in the last slides so that is a statistical approach it's not a forecasting model so we don't have that model for the future we have it based on historical data so until now we have done it that slide was about 2006-2012 we have it now updated until 2016 and in principle it can be done year by year but not in as a forecasting model so that's I think it's important to understand this second in terms of how precise that is it was very good I mean we did all the efforts to cross validate the model against the measurements so the idea was that we have our monitoring sites we remove some of the sites we predict the model on the others actually we feed the model on the others and we predict on the left out and we repeat this approach over and over again so we are making sure that there is no overfitting let's call it in the model itself and we were able to predict more than 70% of the observed concentrations from the left out monitors which for us being this a nationwide effort was a good achievement thank you for the presentation I have just a curiosity you say that in other countries of Europe this satellite based model is being implemented right now and it could be a reference model for health impact assessment my question is how it relate for example we know that in front Schimmer is the the model for prediction Italy we have but every sorry at European level there is this good great effort to put or to compare all the model based on emission sources my question is how this model can coexist I mean the satellite based model can coexist to the more classical let's say particulate matter or air pollution models thank you I will just give a brief and Isabella you can speak about the French situation they can coexist very well simply because they are completely different I mean they are very much complementary because as I was saying in the slides they come from different perspectives in terms of the agreement with the Italian special temporal model and mini it's quite good it's not perfect I mean it's like 0.5 correlation to me it's good because since they are so much different in terms of you know philosophy they they are very good well correlated but especially if you look at the maps predicted for the two you will see exactly the same hotspots the same pattern special pattern so this is a good thing and the second issue there is now the idea of fusing them I mean the next step will be to kind of come up with a sort of let's say super model where you try to exploit the benefits of both into a single estimate so the idea is that one okay I mean Karaden is completing because we made this work in France because we have satellite also many during the station actually overall I agree with you there is a good agreement but if you really want to assess the uncertainty one model compared to the others you can have some places I remember is that it's not that you have some differences for metropolitan France but so we are going to publish these soon but the trend is really to have one model that include everything and Augustin Collette is going to talk tomorrow or now Wednesday about that that's perfect so we looked at differences between satellite data and we saw that if I remember correctly one was systematically higher I mean slightly not much than the other except in urban areas where the other was systematically higher for each grid point I'll look that up and get back to you if you want for I think it was PM2.5 PM10 yeah so one did I don't want to say better or worse because I don't have an actual but we did also compare with measurements okay we have we can discuss yes sure but we have enough time so we have two, three, four okay maybe Lazis that was telling that but I didn't hear the way the models are going on is to make ensemble modeling systems so you are comparing several models and you realize that most often the average of all of your models is better than any of the models and then the modeling community now is assimilating satellite observation as well as surface observations and this gives you what they call analyzes which is the most realistic simulation that you can get with a full temporal and spatial coverage and that's probably the way that you are going to have the best results for these impact studies thank you and it was just a curiosity because the colleague talked about Schumer and also Minni that is a model used in Italy and I know also for example Cummings that is another Eulerian model used and from my experience a problem with this model could be also the computational time for example to do a simulation all over Italy I cannot estimate now but the computational time is a lot so I was wondering which are in this sense the benefits of these kind of models that honestly I don't know exactly this is a good point of course computational time in terms of computational time I speak about this model it's not that big because we are able to run one full model for one year in maybe a few hours not weeks of course you need computational power you need a large computer a cluster something like that the nice thing is that now everything is parallelizable let's say the technology helps us in optimizing the process in terms of time it's not a big issue because as I was saying this is a very statistical approach what we have done in the latest part was using machine learning techniques to do the job and it was not so much intensive I would say we can do one year daily for 300,000 cells which is the square kilometer of Italy in a few hours you have used approximately 700 monitor sites for calculation my question is what is the minimum monitor sites can be used for good precise calculation this is a very difficult question I mean yes we had in total 700 but year by year they were not 700 let's say on average every year we had more or less 500 but then some of those in some cases there were some more stations but others were kind of closed so in total we had 700 we started in 2006 not because we didn't have satellite data satellite data go back until the 90s but because we only from 2006 we had let's say a reasonable enough number of monitoring sites to train the model for us this reasonable enough was something like 300 out of the whole Italy I know of other exercises where they are only focusing on New Delhi in India they had like 30 to 35 40 monitors which is perfect if the study area is smaller so it's relative to the size of the study area because in the end what you do is to train a model on those points but then you want to predict every other in all the other places so the more reliable these monitors are to represent the study area then it's okay but to predict a very large area you don't trust the predictions anymore thank you very much so we know sorry though that you have okay we have