 Zdaj. skala se od vseh, kaj je svojo svojo, in z zelo smo zelo se ospešel na vse, ki se potentivno od Veseljeva, pa se počutila, da je bolj tamo zelo, in vseh, kaj je Andrei Aranzi, z Italy, ne bo. v izgledu svičko izgledaj bolje v oniče, avštakaj ali vsim v Stafođe, imeljamo teh, kaj smo podaljali, od glasbeni demografičnih data, in so n Telefone odbožajo Ноče, tudi nekaj postočajo v početnih, tako, pa si počemo, otvaram z vse, in tukaj pa, različno. Čekaj, kaj mi je, Karla. Vsih je, da mi je vzelo. Vzelo. Vzelo. Vzelo. Vzelo. Vzelo. Vzelo. Vzelo. Vzelo. In invited, and myself and from the Swiss Tropical in Public Health Institute in Basel. And then, later on this afternoon, Andrea will give you an introduction history to use, yes, in this context and also gives you a few examples of how that can be applied. That is sort of the plan, and I just wanted to check that everybody can hear me okay, Andrea? Yes, it is okay! Ja. I can see you, Andrea, so I feel that it's perfect. Also when you sit down. Yes. Okay, so this is about an hour. I think we will try to the first half. I will give you a definition of exposure. We will go through exposure pathways. Talk a little bit about exposure misclassification and then we'll go really to talk about air pollution as an example in the exposure assessment. And the second half there will be some examples of air pollution exposure assessments in studies. We touch on a bit on using satellite data, a very new, exciting data source we will try to use and give you a few flavor of a few other studies. So that's the plan. So the first thing really is to tell you or talk about what exactly exposure is and there are three definitions of exposure. I hear a little bit of background noise all the time, but it's not me. Is that okay? I will just take more. And so maybe it's somewhere somewhere that's better. So this is the exposure definition. So we have three definitions here on the screen and all day. The real, the same thing in these definitions is that we need something, an event consisting of contact at a boundary between a human and the environment, the specific contaminant concentration for specific interval of time. So what it really says here is that you need something which gives you the contact between a receptor, a human being in this case, if we talk about human exposure and a contaminant in the environment. So what this means is that if there is a contaminant somewhere in the environment, but there's no human exposure, human in that environment, there will not be an exposure. So really this is key being that the contaminant need to be in the same micro environment as the person or the human being. So that is exposure and we can see here for instance what pollution can do to human body, give you some of the pathways. So we see here on the left. Actually, is my mouse also showing on the screen? I think so, yes. Yes, yes, yes, perfect. So we have air pollution, water pollution and soil contamination in this example. And you can see from for instance air pollution, we have emissions turning into certain pollutants, certain air pollutants, and these air pollutants can have different effects on the human being. For instance, VOCs, they can cause cancerists or skin irritation, particulate matter, especially as effect on the respiratory illness. So you can see that certain pollutants have certain effects on the human body. The same for water pollution and for soil contamination. Soil contamination, for example, pesticides is a very big, a good example of that, having many different effects on the human body. So you can imagine that that has some implication for if you want to assess exposures and what that can do to the human body. So there are four exposure processes really. And they can be ingestion, inhalation, dermal contact and external exposure. The ingestion is taking it into your body by eating or drinking. So this is all about when ground water is contaminated, surface water, your drinking water, but it also can have, if you have soil contaminants and the fire, the soil, it goes into the food and you eat it. So this is the ingestion process. The other one is inhalation. And of course this is when you have air concentrations, you breathe it in and it gets inhaled in your lung. And this can also go through groundwater, surface water and soil because from there it can get into the air and thus into your lungs again. The third pathway is dermal contact. So here we think about water, soil and air when it gets into contact with the skin and through that skin it can get into your body. So that's the dermal contact. And the last one is external and this could be radiation, so radiation of course is not something you eat or you inhale or dermal. It's something which goes through your whole body. So those are the four exposure processes and just to stress how this sits within the environment and health chain. So this is a nice figure where you can see on the left, it starts with the source and on the right we have an outcome. So the source could be an industrial installation, a power station for instance and here on the right could be an outcome which could be a respiratory illness. So you need from the source you need to have an environmental concentration, so this comes say out of the stack into the environment and then this bit here is where the stressor and the receptor meet. Thinking back to the definition I mentioned earlier, this contact here is what we call exposure and from exposure we can calculate a dose which is something I will explain in a minute and this we come to an outcome. So this is this chain and this is very important in the hill in whole environmental exposure assessment chain. Now there are many ways to assess exposure and this is a diagram which shows you in ways of precision. So we start at the top here with a high precision and at the bottom of a low precision of exposure assessment. The highest is what we call biomonitoring and you will probably hear about that today or this week as well. But biomonitoring is where you take samples of the skin of the hair of the teeth of the blood of the urine and measure certain pollutants in there. And that if you can relate that to a source gives you of course a very accurate measurement of the pollutant in a human body. Very high precision because you know you measure straight at the person at the individual stage. So that we call a exposure assessment with the highest precision. From there onwards we of course start to more and more model this. So here we also have a measurement which is called exposure monitoring. So this is where we would stick a personal monitor on the human body for instance to sample air pollution. Again it's very accurate but it's not as accurate as biomonitoring of course because it's not actually measuring what you inhale it tries to assess that by monitoring just outside your body and you carry this monitor with you so it's pretty good but slightly less accurate than biomonitoring. And then we start to come into modeling because you can imagine that if you have a cohort or a population you cannot do these measurements that well because it's very expensive for starter takes a lot of time and you can only do a few individuals from your larger population. So what you need to do is you need to start modeling and all these modeling techniques here will give you an estimate but there are of course some misclassifications attached to that. I will talk about a few of these modeling techniques and you see here at the bottom which is something I will not go into any further but that's one of the lowest category of precision is questionnaires and diaries that these are questionnaires you give to the population and they will fill it in and give you a sort of a subjective idea about their exposure. So it's one of the lowest but of course also very often used in exposure assessment of populations. So this is this how we can assess exposure and here an example of the different exposure pathways in in a contaminated site. So imagine here on the left we see a source of exposure the source of pollution I must say so here we have some drums some oil drums for instance they've been there sitting lying there for a long time so they start corroding and they start spilling into the soil from there on it gets into the ground water from the ground water it gets into a drinking water borehole for instance and gets into the house and gets ingested by by drinking it other pathways are from the air you have volatile components which get volatilized into the air and then get inhaled by the children by people living nearby. And another pathway is going through the soil and then gets taken into biota by food and or eaten by cows and groups from that way it gets into the human body. So you can see schematically how how that would work. And there's also of course a prevailing wind direction which is important. There will talk about it in a minute so. Metrology plays a huge role as well in in all these expositions another way of looking at the same pictures is this way so this is the same pictures as here but then in a diagram schematic diagram again we have the drums here which had been leaking and sitting there for long time and we see the different environmental medium so we had the air the soil the ground water and the biota and we see then from there we go to exposure points and from there to an exposure route to the potentially exposed population and you see this way you can really illustrate or disentangle how from one source you can have many different potentially exposed populations through many different exposure routes which is what you need to do to describe these exposure pathways. So we can actually calculate exposure doses and I will go quickly through that and this is all I should say also the previous slides are all from the very nice public health assessment guidance manual which is produced by the ATSDR which is the agency for toxic substances and disease registry in the U.S. And it's a very interesting document if you really want to get into exposure assessment and these are examples taken from that and here we see the generic exposure dose equation. So we see D is the exposure dose and there are a number of variables to calculate that and so we have seen the contaminant concentration so that's for instance the concentration in the air. We have an intake rate which is for instance the how fast you breathe or how much air you breathe in, you have a bio availability factor, you have an exposure factor and a body weight. This is the generic exposure dose equation but for every and I don't expect you to see this very clearly on the screen. It just shows you all the different exposure doses available so we have for instance here we have water ingestion, we have inhalation, we have fish ingestion, food ingestion, etc. So all these different routes I was talking about earlier have slightly different dose equations and for instance here we pick one out the inhalation exposure dose equation so very important for air pollution of course and there how we calculate that is we have a D is the exposure dose is a function of contaminate concentration so that's the concentration in the air, the intake rate, the exposure factor and we also need the body weight and of course it's very difficult to assign values for every single person so what they did in this manual they gave you and they give you default air intake rates for instance so we have here different air intake rates for different subgroups of the populations from infants, child, girl and boy, female and middle and this is of course based on American population that it might be different if you do this for European population or for a national population somewhere else but these are the ones the EPA have assigned they also have standard default values for body weight for exposure duration etc. So this is how you can then do these calculations. The next one I will just give you to look into later yourself because I was told yeah I will move on a little bit quicker. This is just the calculation and you can do that yourself how you can actually get through a calculation like this. So let's go to exposure modeling because what I said before it's very difficult to measure these things and to do these assessments at individual level. What we normally have to do is we need to model for a whole population and broadly speaking there are two ways of applying these, this exposure model. So we have an interpolation technique and this really takes advantage of a of measured values in the environment because we measure at certain points but we are interested in the concentrations of the exposures everywhere not only where we measure so we have to do some interpolation so for instance creaking and I will go briefly into that later, creaking or infras distance weighting are interpolation techniques which can be used for instance to produce surfaces of soil pollution which can then be assigned to population. So interpolation, the base of that is measured data. We also have the second is the source receptor modeling and this really is where you know and you simulate erasitips between source and receptor. So for instance this would be dispersion modeling would be an example of this. We know and I will give you more information about that in a minute. So these are the two broadly speaking ways to do exposure modeling interpolation and source receptor modeling and then briefly what makes a good exposure measure? Well, I mean these are quite logic. Of course, I mean they need to be specific. They need to be really on the specific to the pollutant. They need to be accurate. I mean, there's no point in having very inaccurate how I see somebody in the screen. And they need to be robust, flexible representative and practical. I mean, these are all logical things, but these are good to keep in mind when you do an exposure model. And also something we need to keep in mind is exposure misclassification because models really are a simplified representation of reality. How we make generalization every time we apply a model we generalize about processes, the interactions and the feedbacks. But the reality might, of course, be slightly different. So we always have to keep in mind that it's not reality, it's something we represent by assumptions and generalizations. Exposure models also make assumptions about the spatial patterns of the environmental hazard concentrations and also make assumptions about spatial patterns of individual or populations under study. An example of this is a lot of the air pollution studies assume we all live and work at home. We don't really move at all, which is, of course, not the case. We travel to work. We work in a different place than from where we live. And then we go back home again. So we move around in the environment. And so this is something, of course, we have to take into account as a slight misclassification. So there are various aspects of uncertainty associated associated with each method of estimating exposure. So let's now go to the theme of of this also this week, of course, air pollution now. But we have to go into this too deeply, because this is probably already been handled, but just to make sure we have primary pollutants and we have secondary pollutants, primary pollutants are the ones emitted and the secondary pollutants are being created in the air, in the atmosphere by reactions. So very important is where does it come from? And we have, of course, many different air pollution sources. An important one is, of course, natural sources like volcano, wildfires, forests. And then we get to cities and agriculture, which are area sources, which pollute. We have stationary sources like industrial point sources, power stations, waste incinerators, et cetera. And one of the big pollutant sources are, of course, mobile sources, traffic airplanes, et cetera. And so, of course, good to know, to realize when you investigate pollutants to know from where they come from. And here another similar picture shown the same thing. So these anthropogenic sources basically mean human-made sources. So these are human-made pollutants. So we have traffic sources here and I just want to pick out that we don't only have pollution coming out of the exhaust pipe like here, but we also have brake and tire wear, which is also a source and which are still sources when you look in, for instance, electric cars. They still have tire and brake wear. So just to maybe point out that although electric cars are many, many times cleaner than the normal fuel cars, there are still some pollutants coming from there. And also not an important indoor sources in the house. And just to pick out one particular pollutant, PM, so we know PM10 is the largest of the particulate matter, which is all particles smaller than micrometer. PM2.5 is a fifth of that, or fourth of that is everything smaller than 2.5 micrometer and that's something we are studying now more than we did in the past where we just looked at PM10. And more recently, ultrafine particles, which are even tinier, which are only all particles smaller than 0.1 micrometer, which is now the sort of the pollutant of interest in health research. And you can imagine that when you look at these sizes, you can imagine that these smaller particles probably could cause larger health effects because they can penetrate your lung deeper and this cause more damage. Just to say that the diesel particle, for instance, is indeed an ultrafine particle. So when you blow this little one up, you can see here the diesel particle. So hence that's a pollutant of interest. So we can of course measure air pollution. And here are a couple of ways to do that from very expensive routine monitoring stations dotted around all countries to passive sampling techniques like here where you stick a passive sampler just on the lamp post and leave it there for a week or so to sample. So the difference between passive and active sampling is that active sampling, you need a pump to pump the air through the monitoring device, whereas with passive sampling you can just hang up a filter and this filter will then within a week collect pollution on its filter after which it can then be analyzed. So there's many different ways of measuring air pollution and there's also many different health effects as attached to air pollution. And here is a nice graph showing you different health effects at the different stages of life. So we start already at birth weight. There's already studies done showing reduction in birth weight by high levels of air pollution. And you see all sorts of asthma, long function, etc. Health effects which have been associated with air pollution. So it's actually a very large list. So who is affected by air pollution? Well, I mean, obviously mega cities are a huge problem at these days, but not only mega cities, but also cities in Europe. They also still have high pollution levels. I mean, this is just an example of Beijing in 2013 with pollution levels greater than 500 microgram per cube, which is a huge amount. Actually, these days it's China is going down a little bit in levels. It's now cities in India who are top of the list of most polluted cities in the world. But it's not only a problem in those mega cities in the developed, it's also a big problem in middle and low income countries. And that's mainly to do with cooking and the way they cook indoors. So as you can see here, it's using solid fuels, cooking whilst being exposed to the fumes coming off that. And you can see that here as well in this in this map had 3.5 million premature deaths per year attributed to household air pollution from solid fuels, which is a huge problem. So just to had to hold down the message. It's not just the outdoor air pollution, which is an issue indoor air pollution, especially in the middle and lower income countries is huge issue. Of course, Europe has lower levels, but actually we still have a problem there. Also, we still have a problem with health because there's no threshold of toxicity for PM. I mean, there's nowhere level of zero micrograms per cube, so there's always an issue with that. And this is something actually we are investigating currently in a big project we're doing in Europe. Trying to get a grip of these, how bad still is the lower levels of air pollution for health. So this is, of course, where we are exposed and where do we measure? I mean, obviously we spent a majority of time into our houses. We sleep there, we eat there, we spend our evenings often there, and then we do some activities outside. But mainly we spend our time indoors, but often we, the only measurements of air pollution, we have our outside at a routine monitoring station. So in studies what we often do, we measure also inside, of course, to capture that environment. But ideally speaking what we would want to do is attest a monitor to every individual to get an ideal or most accurate assessment of air pollution. But often what we are left with really is the routine monitoring site outside. So with that we need to do some assessments and we, and I will come to that in a minute. So we do monitor air pollution a lot, especially in Europe. This is a map of 2008 with monitoring sites measuring particle to PM10. In 2008 we see the red dots are greater than 40 micrograms per cube. And you can straight away see sort of a pattern of PM10 concentrations across Europe, where especially Italy, Spain, also in the in the Balkan, they are very high concentrations of PM10 and they get lesser in the north of Europe. So this is what is happening in Europe, but this is, if you would make a same picture of a map of Africa, you would hardly see any monitoring sites very much. Yeah, in Europe we monitor a lot and in the US and Canada as well, but other parts of the world we are monitoring less. But why are these measurements important? Well, I mean, one thing, of course, is you can look at worst cases. So here, for instance, we, this is a nice list of monitors in different cities across Europe, where they monitored annual mean and two concentrations. And we can see, actually, Florence in Italy, here on the left, measuring the highest pollution level annual mean. So this is pretty high, 100 micrograms per cube. And the guideline is 40 micrograms, so we're more than two times over that. This is a site, obviously, somewhere on the road site in Florence. And you can see nicely that, you know, picking up all these high monitoring sites, here is one in London, for instance, Marlebone Road, et cetera. So that's a nice way, where you can straight away pick up, you can order these and pick up the worst monitors. You can also, of course, look at it temporarily and look at transient air pollution data. So here, we look at the air quality in London over many years. We see that two levels have steadily dropped over the years, nox values as well, but interestingly ozone levels are going up over time. Stabilized a little bit in the latter years. So very important monitoring is to keep an eye on these trends over time. And then we can, of course, model. And this is an example of a three day simulation over Europe in 2008, generate 2008. These are done with very large models, European models with a lot of data going in. So air pollution monitoring data will go into these models, but also meteorological data, emission data, et cetera. And with that, they can then give us, these are hourly estimates of air pollution levels across Europe. And very nice simulation can clearly see high levels of air pollution taken away by meteorological circumstances. So that was done at European level. We also have maps at world, at global level. This is an example of a map done by colleagues at Dalhousie University in Halifax, Canada. They modeled PM 2.5 across the whole of the world using satellite data, and I will tell you a little bit about that in a minute. So what methods can we then use to do all this modeling for air pollution? So we have, I will show you examples of four methods. One is proximity-based methods. The other is spatial interpolation. Then I will show you example of dispersion modeling and then lastly, land use regression. So the principles of interpolation and energy statistics, as a matter, is the first law of geography from Tobler, and it says that everything is related to everything else, but near things are more related than distant things. So just to make this a little bit more to illustrate this, so if I measure air pollution here at my home and I measure it maybe 100 meter further, they probably be very similar air pollution levels. But if I measure then a kilometer away from my home, or maybe even a different city, they will be different, so things nearby are more related to where you are than distant. So this is the underpinning of all interpolation and statistical analysis. So here are some examples of modeling methods. So for proximity, we have, for instance, foranoid tessellation, which creates areas around each point containing locations nearest to that point. We can also use buffering techniques, which create zones of specific distance around points. Then we have distance functions, and I will give you an example of the inverse distance waiting in a minute. We have global interpolators, like transervice analysis, and local interpolators, like Kriging. And here we see this in action. So at the top we see a actual service. So this is the actual pollutant service we are trying to model. And the only thing we have are these points here, which are measuring points. So we know at these measuring points exactly what concentration is, but we don't know what's happening in between. So this is what we try to model. So for instance, the foranoid tessellation, or also called surface model, that's a disjunct surface, we see what we do, we can model it and say, okay, the nearest point to a measuring point, I take that concentration and apply it. And as soon as I get near to a different point, I take that level and so on. And you can see that you get a surface, but probably not very realistic. Another way of modeling it would be using inverse distance and you see you start to smooth a little bit this surface. You can use a global surface, which puts a global surface through the points, or you can use a locally smooth surface. So you can see you have many different ways to go through these points. And you can judge a little bit, okay, well, probably this one here, locally smooth surface, would probably best reflect the actual surface. But which method is the most appropriate? I think the real issue here is we need validation. So we need some independent points to be able to assess how well the model surface works. So for instance, what you could do is you could take, if you have many points, you could take a subset out of it and don't use that to model and then apply the model to those validation points afterwards to get an idea of how well the method works. So inverse distance waiting. This is the formula, but I want to go straight to the example because that's. Easier to explain. So we have a point here in the middle point one. This is where we want to assign a exposure. We only have measurements here 2, 3, 4, 5. We know the concentrations. So the concentrations at those five points are these, these are the concentrations. So at point two, we have a concentration of ten at three, we have a concentration of five, et cetera. We also need to know the X and Y coordinates of these points because this is the base of it because we need to calculate the distance between each point and the point we want to estimate the concentrations. So we calculate the distance here. Then because we do a one over d square inverse distance waiting, we then calculate the distance square, the one over distance square for all these points, which will give us a weight. And this weight, we can then apply to the concentration measure that those points. And add them together to give us a total inverse distance weighted calculated concentration at point one. So for point one, we estimated concentration of 10.24 micrograms per cube based on the points here and the distance to it. So that's how inverse distance wasting works. So that's the, the end of the interpolation techniques. There's also Kring, of course, which is a very so interesting method, but it's too, too difficult to explain now, but I would Yeah. I mean, you could, you could read up that yourself, but that's, that's the sort of the most complicated one in the statistical methods. So now we go to methods, which actually use propagation models. So they, these are dispersion models and you can see here the plumes you, you see when you go outside and it's a nice day. You can see these plumes coming out of these factories and what we want to do with dispersion modeling is we want to model and want to predict how these plumes go through the environment and how they reach us on the ground where we actually breathe these plumes in. So these atmospheric dispersion models are using mostly Gaussian equations to model the transport through the atmosphere and they were originally developed as a tool for regulatory compliance modeling. And they are traditionally used in environmental impact assessment. They require a lot of detailed input data. So you need to know emissions for, for instance, for industrial source, we need to know the characteristics of the stack. We need to know the height of the stack, the diameter of the stack, the emission rate, temperature of exit gas, et cetera. But we also need to know meteorological parameters. We need to know the wind direction, the wind speed, temperature and we need to know this as a ideally at an hourly temporal level. So how it works? So this is the real picture. We then say, okay, did the plume goes through the air, it rises because it's hot, so it rises in the air, but at some point the wind will take it with it through the atmosphere. There might be some rain, so we might have some wet deposition, but at some point this plume will reach the ground, it will go over altitude levels, which will cause some different flow patterns over complex terrain, et cetera. Now this is quite complicated, but thankfully people already worked on this and they put this into a nice equation, the Gaussian plume equation, where they use a lot of these parameters you see here, but they put it into a mathematical equation. And this is then underlying of a dispersion model you can buy from these companies and what it will produce is something like this, an actual pollution surface around, in this case here, a point source. And this is then the Gaussian plume equation. Again, I won't go into it at the moment because we need to carry on a bit, but you can look into this later, but this is something which is all built into this dispersion model, so it's not. We don't have to thankfully do this ourselves. So what I already said here, for instance, this is in a typical flowchart of a dispersion model. This is the model itself, air mod in this case, we have all sorts of inputs, we need emission data, source data, we need meteorological station data, we need, if it's in a complex terrain, we need elevation data, we need some land cover data. All that goes in. Together with where we know our receptors are and then we can calculate ground level concentrations at the receptors. This is an example of some of these methods and then overlay it with each other just to see what different exposure surface they give. This is, for instance, a situation in London where we have monitoring stations and two monitoring stations across London, and if we use the very first method I mentioned, this foranoid tessellation method where we use the nearest monitor to assign and expose your assessment to population, we get a service like this across London. And you can, you all agree, this doesn't look very realistic, so the next one would be inverse distance rating. If we apply that to these monitoring sites, we get a concentration service like this. Already looking a little bit more realistic. Land use regression techniques, I will explain it in a minute, will give you a surface. Again, already a huge improvement and dispersion modeling would give you a surface like this, so you can see from a very crude method of exposure assignment to the two most accurate ones, you can see the change in exposures. So land use regression modeling then. So this is a method where we use monitored data and information around these monitoring sites to come up with a regression model. So imagine here we have three monitoring points A, B and C. We measure air pollution at these three points, so we measure NO2, these are the three levels we measured. We also have different variables which reflect the sources around these monitoring sites. So we know some information about traffic there, we know some information about housing and altitude. What we can do then is we can do a stepwise regression and predict NO2 using a function of these, in this case, these three variables. In reality, this would be a lot more variables we can choose from. And with this model we can then actually estimate concentrations of energy at unsampled points. So here we have now where our population is living and we have also there the information of traffic housing and altitude because we have that information everywhere and using that model we can then apply that to these points and get a predicted NO2. So this isn't very short. I realize I do that very quickly explaining land use regression, but you can always go back and read up on it a little bit. So land use regression model, the outcome variable is the, is a annual average pollutant concentration. The predict the variables I was talking about, we use land use data, we use road data, traffic data. We can use population data. All this data basically are reflecting pollution sources of the pollutants. Then we do supervised stepwise forward regression. We do some checks in the models of spatial autocorrelation, et cetera, and we do a leave one out cross-validation to see how robust our models are. And this is, for instance, an example of a model development in the UK. We see, for instance, at the first stage one variable, the heavy traffic load within 50 meters comes in as our best predictive variable, giving us already in our square of 0.67, meaning that 67% of the spatial variation is already explained by this variable. So when we do a next iteration, we see that road length within 500 meters comes in and gives us an additional 15% of explained variance. And lastly, we also get a variable, which gives us information about residential area, giving us a final model explaining almost 86% of spatial variation of NO2 in this instance. We also see that our sigmas of the predict the variables are small. They cannot go over 0.1, otherwise we will not accept them, and we also check for variance information factor, which means that if variables are very highly correlated to each other, they would be picked up with this factor. So this is how a land use regression model is developed. We did a big study in the EU, in the Europe, and I will go through this bit quick, because I see that I sort of have about 10 minutes time left. And you can read on this later, what I want to show you is really not this either. The map we produced eventually from it. So here we have a model, and we can apply that model, and this is the nice thing about it, this is the powerful thing of land use regression. We can apply a model to the whole of Europe. So we measured air pollution in many sites across Europe here. We extracted predict the variables. We developed this model, and then we applied it to every 100 meters square in Europe, arriving at this NO2 map. And you can see as well, this graph here is a transakt through Paris. So here I blow it up a little bit. You can see here, so land use regression, the red line is our NO2 estimated concentration. So you can see this is the center of Paris. Levels go up, and these are the outskirts of Paris with lower concentrations, which can also see nicely here the different contributions of the different predict the variables. And so, for instance, road variables, we have a major road variables giving us peaks of air pollution as if you combine all these variables together, we get to these red exposure estimates. Now, satellite data is something, so we are recently including in our models, because it's something since early 2000s, it's a data set we can use now. And you can see, I mean, this is only a very small amount of satellites on this picture, but there's many more going around the Earth, and we're using some of this data now to predict air pollution levels on the ground level. So, for PM, we use this aerosol optical depth measure, and the definition of that really is the, it measures the light extension by aerosol scattering and its absorption in the atmospheric color. So, imagine we have a satellite going across our globe, and we have the sun here. So, we have a direct sunlight going through to the satellite, and the satellite picks this up and measures it. At the same time, the same sun beam goes to the surface of the Earth and come back to the satellite, the satellite picks this up as well. And you can imagine that when this beam goes through the troposphere, it will change because of aerosols. So, because of aerosols, it will scatter, so not all the light going to the surface will come back to the satellite. And the difference between this beam and this one is what we call aerosol optical thickness. So, it's a measure of the amount of aerosols in the atmospheric color. So, that's what we use in our modeling. And recently, there's been a method developed to use this data, and this has been developed by Colin Sovarez, Hittai Klok in Israel, and Joel Swatch in the US, and they've used this method in a lot of regions, especially in the US, now also in Mexico, and Massimo, who will talk after me and myself, we've both been applying this method now in Italy and in Switzerland, quite successfully. And this is how we do that. So, we have to, the key points here are that the AOD, as you remember from the picture I showed you earlier, it will measure light scattering by column of air up to the satellite, so it will measure AOD over the whole column, but we only are interested, of course, in concentrations near the ground. And some days we have more of the particles near the ground than other days, and this probably is related to the mixing height. So, mixing height is this height below which particles mix fairly well. And some days the mixing height is high, which means the particles are mixed up and we get lower concentrations in the ground, and other days this mixing height will be very low, which means that all the particles are trapped near the earth and our concentrations are higher. So, this is why we use this mixing height as a variable, as an interaction term in our modeling. And this is how we do it. This is a flow diagram of our modeling approach. So, we do this at a grid level. So, we get AOD data at one by one kilometer spatial level. For some of these AOD cells we have a PM2.5 monitoring site, those are the black dots. So, if we can then fit daily calibration curves through these cells with PM2.5 monitors and we can fit a model, fix the effect model explaining PM2.5 with the AOD data and this mixing height plus other spatial temporal predictors. So, then we have that model and what we can do then is we can apply this model to cells where we do not have a PM2.5 monitoring site, but we do have AOD data. So, we can then apply that model and estimate PM2.5 grand level estimates. So, then we have all the cells, we deal with all the cells with AOD data, but the cells, some cells left without AOD and these are the cells where on a particular day there's cloud cover and if there's cloud cover, we do not get an AOD measure because it's obstructed. Clouds obstruct the earth from the satellite. So, we do not get a AOD measure, but we then apply some spacial smoothing to still get an estimated PM2.5 using information from the cells nearby. So, then we fill up this hole in our last step, we estimate PM2.5 of all cells and we can also go down in scale and that's we do that by taking the residuals of the measured PM2.5 at the monitoring stations and the estimated PM2.5 at full one by one kilometer and with those residuals we can then apply using support vector machine learning algorithms to actually go down in resolution and then ultimately predict at the 100 by 100 meter resolution. So, for Switzerland, I did this and Massimo did something similar in Italy and here you can see we have PM2.5 stations across Switzerland, 10 of them, which were not really enough to do this, apply this method, but thankfully we also had PM10 monitoring sites and many more, we had about 100 of those, so what we did is we looked at correlation at co-located sites between PM2.5 and PM10 and we found very good correlations between PM10 and PM2.5 at these co-located sites and we also saw that temporal signal was very, followed each other very well, so what we did is we applied these regression equations to PM10 sites without PM2.5 to impute PM2.5, so in the end we had 100 PM2.5 sites instead of 10, which was more enough to apply method. We had lots of predicted data, I won't go into that, and what we were able to do then is to model PM2.5 for all these years, 2003 to 2013, at a 100 meter resolution across the whole of Switzerland, so these are annual means, but we could also go down to daily in this and here we see four consecutive days in July, and we see some very interesting patterns of air pollution, and we see, I don't know whether you know Switzerland, but around the first of August it's Swiss National Day, where they light lots of fireworks, and so this is most likely the effect of fireworks lightning off in Switzerland around those days. So that's the end of my air pollution exposure assessment talk. There are a couple of examples here, which you can just look through and maybe pick up the papers to have a look yourself of exposure assessment in other contexts, but because if it's not really related air pollution, I'll just show them to you at the end, you can then look into them later. Just to say that Andrea later this afternoon will introduce you to geographical information systems, and this really is the base of all the work I've talked about earlier. The GIS is really the way to link all this data together in a really useful system to work. And Andrea will explain that to you this afternoon. So that's, I think you, first we got some questions, but I think then you will have a nice cup of coffee. So this was the end of my lecture. Thank you. Thank you, thank you very much, Kase. I think we have a quarter of 15 minutes for questions coming from the audience. Yeah, I'll bring them, you stay here. Thanks for your good, very good lectures. Just you show on your slide air pollution simulation. My questions, which you can use. I could hear, I could hear very well, but I missed the last bit. Okay. Thanks for your very good lectures. You show on your slide air pollution simulation. Which model you used? So for that, I assume you talk about the dispersion model. Yes, yes. Yes, this was done with ADMS urban. Okay. And ADMS urban is produced in the UK. It's often used model, but I mean, there are many more, obviously. Another model I've used in the past is AirMods. Okay. Which is the US EPA model. And that's actually for free. Okay. Thanks. Yep. Case, this is Karla, because yesterday we were talking about what we can do if you don't have, you know, money enough to buy the model system, and I told the students that you and Andrea in the afternoon maybe can provide some link to find where it's possible to download this model, because I know that someone, some models are available for the general public. Ken, maybe Andrea. Yes, they are available, but they are still not so easy to use, because they are, they're not nicely fit in the user-friendly software environment. So this is the added value of ADMS urban, of course, is that it's all produced in a nice user-friendly software. Or it's the AirMods model from the US EPA. You can buy that as well in the nice software, because companies have used that model and produced user-friendly software around it. But you can also download the source code, use that, but that's a lot more difficult to use. I will come back in the afternoon with some slides on ADMS, and I agree with the case, you will see that the interface of the ADMS is user-friendly instead of AirMods, yes. Thank you. My question is that, you have said that the models are free of charge, but my question is about the input, the data that we use, are also available or not. And then, in my knowledge, I think that we used some experimental data for our start point for the modeling. So how many sites, how many experimental data we have used in the input for such a modeling? Thank you. Okay, so these are two questions, I think. One is about the input for dispersion models, if I correct, they understood. And the other one is about actually other models where we use monitoring data. Experimental data, he says. Yeah, so the first one is about input data for dispersion models. Okay, so that's, of course, some metrological data you should be able to source, and depends a bit on the country where you apply these models. For some countries you get free data for free, hourly metrological data like temperature, wind speed, wind direction, et cetera. You could get for free from the metrological institute in your country. Other countries might, you know, you might have to buy that. The real problem is the emission data, of course, because there is data coming from industry. So this is something which also will differ by country. Some countries, they will have to provide this. Industrial sources have to provide these two registers in their country. So this is how you could get this for free. Okay. It should be public. But, for example, I want to model in regional sites of Megacity, for example, in the north of Morocco. I have only a few sample sites, four or five, that's all. Can I use interpolation, modeling for getting an idea about all the city? Yes, so four or five sites is not really enough for giving an indication of the spatial variation across the city. Minimal would think about something like 20 sites to do as a minimum to start doing these sort of methods, apply these methods. Yeah, four or five is not that much yet. But there are ways you could improve or add to these numbers by doing some monitoring yourself. Would that be possible at all or not? Thank you, Keis. Before the next question, just to let you know that we have people here in the classroom in the classroom who are connected via webinar for, I don't know, 25 countries all around the world. So very different context. Absolutely. And now the next. Good morning. My name is Mutrajev from Kenya. Mine really may not be a question per se. It will be a compliment because finally somebody has answered the question that I've kept posting on research gates without no answer. So I've also interacted with my poster outside, know that I'm linking air pollution to human health. And I'm set to defend my thesis in the next one week or two weeks. But I already know that the major issue raised by the external examiner is the practicality of using satellite column measurements and linking that to the station level human health. So what I learned from you now and I feel so much empowered that I can go home now is the use of planetary boundary layer characteristics, like the boundary layer depth and the vertical mixing coefficient to explain the loading or the concentration of PM. On the days when the mixing length is so low then you can use that as a proxy within the earth surface. So really mine is a compliment and I feel I would want to know whether this can still be used in places like NO2. So NO2 is only applicable to PM. But thanks so much. I feel I may go home now. So I thank you. Don't go home yet. You have another week to go. But no, thank you very much for that. And it's absolutely true. So this mixing height is especially applicable for PM. I am now trying to apply a similar method for NO2, which uses different satellite data. But NO2 is also possible to use the satellite data in this modeling approach. Other pollutants are not yet developed yet. So this is still only possible for PM. Akis, we've got other questions, but could you just put maybe, we've got a coffee cup, which is fantastic in the middle of the screen. Could you, it's quite tempting. Could you try to change and share just your face in the middle for all the people to be able to see you and ask your questions. If you manage. I wouldn't know how to do it though. No, you've got NO2 on sharing the screen. If you don't manage, it's not important. Yeah, it's working, fantastic. And I've got the mic, I've got a question from someone from the webinar. Oh, I did another bool. Thank you very much, closely related to the comments he made. I want to find out what the correlation between AOD and PM is. Because, well, I'm not an expert in AOD retrievals, but I know that AOD is actually a composite. You know, you could have PM, you could have total carbon, carbon, et cetera. So what is the correlation really between AOD and say PM or carbon? How do you, if you have AOD data, how do you derive PM concentrations, for example, from it? Thank you. I mean, that's a very fair comment. Of course, the AOD doesn't distinguish between particulates and other aerosols in the layer. So the raw correlation between PM2.5 and AOD is not high. It's actually quite poor. So this is why we have to do a lot of modeling in addition. I mean, maybe what we could do is we could give some links to papers where we can see how this method is being applied. So maybe we could add that to the list, Andrea, this afternoon, by giving some links to Massimo and my paper from Italy and Switzerland. How we applied this method. Hi, Keith. Frans, what you like speaking, to go on this topic. How can you cope on the transport of high aerosol concentrations in altitude when you are looking to these AODs, for instance, dust or biomass burning? Most of the AOD is not at the surface, but in the free troposphere. Absolutely. So this is, of course, and this is what I was alluding to earlier. So, yeah, the AOD takes care of the whole column. But what we try to do is we try to find a relation between, and we're quite successful in that, in between ground level PM2.5 and AOD using all sorts of different variables. So this mixing high variable is a very important one that will explain a lot of it. But it's really, the crucial point of it is linking it to ground based measurement. So we calibrate this AOD values to ground based levels. And we don't really, we cannot look into this, yeah, this regional transport of biomass burning, et cetera, because that's not, our model doesn't really allow for that too much. Ok. I have another question, please. On one of your examples, you presented an equation, which was a point-to-point dispersion equation. And on your application, I wanted to know, do you apply a point-to-point equation or a line source emission to do the example on traffic? Is it the point-to-point dispersion equation or a line source dispersion equation? So dispersion models, they can take care of point sources, but also of line sources. So in ADMS, for example, you can input roads with traffic volumes. The dispersion model will then simulate the dispersion of these promoters from traffic to the surrounding environment. So dispersion models allow for both point and line source. I hope I answered your question. Can we use Gauss-Bloem model to estimate nuclear radiation concentration of nuclear radiation in case of nuclear accident or no? That's a difficult question. I think that's a completely different area, and I would assume there are models available to do that. I mean, nuclear, these are just, I think what you need for that are these big regional transport models, chemical transport models, because nuclear fallout is, of course, so very large area, whereas dispersion models typically are for small areas, like city-level areas. What is suitable distance from the source of pollutants to receptors to obtain a good result from Gauss-Bloem model? Oh, okay. So, well, typically it depends on the source you're looking at. So if you're looking at a point source with a very high stack, you should look around in 10, 15 kilometers, because this is how far will it travel with wind speed, et cetera. But if you look at lower level sources like traffic, dispersion is, of course, a lot less far away, so then you can, the distance is less. Is that what you were alluding to? The gases or particulate matter? Sorry? Gases? Gases of particulate matter. Were you talking about gases or particulate matter? Both, actually. I mean, the dispersion model will take care of both, and also the dispersion of these will be typically in those sort of distances. Okay. I think we don't have any other question from the classroom. We have some from the people attending by webinar case, so please. So we just got one question from Nur from Jordan. I mean, it's a bit like the question we just had now, but I'll just ask you. She was just wondering how many monitors are needed in an area of known size to get good predictions, and should they be equally spaced, either method to determine the location or the distribution of monitors beforehand? Yeah. Okay, that's a good question. I mean, so I said earlier, I mentioned 20 sites, but that's really a minimum. I would certainly aim higher. I would like typically something more like 40 sites. These 40 sites, they don't have to be equally distance away from each other. What is more important is that they reflect different areas of traffic pollution, levels of traffic pollution. So we would like some sites near roads, we would like some sites away from roads. We want to capture all the different levels in our study area. So that's more important. But I mean, there are some papers written. And again, we could add another paper from Girat Hook on Lenju's regression. So this is the. Thank you, guys. We have the last problem. Yes, okay. Thank you very much for interesting presentation. So you talk about satellite optical techniques that can be used to reflect somehow the stress of the evolution. So I have a question about, can the satellite thermal imaging can reflect the stress of our evolution somehow? Especially that we have been here from the first lecture. There is very correlation between the temperature and the evolution stress. I'm not really interested, sorry. Have you heard correctly or not? I don't completely understand it. Maybe can you come here and I can see you talk. Okay, I ask about the thermal imaging. Can be reflected. Yes, can be reflected somehow the stress of the pollution because that we have been learned that there is some correlation between the high temperature and the air pollution. Okay, okay. Yeah, that's interesting. I mean, of course, we certainly temperature is one of the variables we allow into what we have in our models. So temperature is very much a predictive variable of polluting concentration. So that's correct. We actually haven't used satellite data, temperature satellite data to inform our models. And that could be, of course, also an interesting way to do it. But what we do, though, is we do estimate temperature levels using the thermal imaging. But that's a good point. Okay, thank you very much, Kizir. Speaking, Erik, I have a question about, you said that we'd be best to have monitoring, individual monitoring using some monitors that you could wear, take with you everywhere you go during the day. And I wondered if you have already a set of those monitors that are maybe validated, maybe not cheap and could have a package that we could propose here in this audience and say, if you use this kind of monitors, then you get a good exposure measurement. Yeah, well, I'm afraid I can't recommend anything at the moment. There are good validated methods available, but they are very expensive still and require quite expensive monitors and polluted monitors and they're not heavy, not light, they're quite heavy still and they require a lot of effort. The so-called low cost sensors are still, I would say, unproven in their accuracy. So people are using them now in studies, so they're certainly available and there's a lot of these sensors available on the market. But I wouldn't dare to recommend any of them yet for a study because there's still a lot of inaccuracies attached to them. But I know of a lot of studies trying to use them, but we could certainly give some links to maybe some of these sensors we trust more than others, so maybe that's a better way of phrasing it. Last question. Yes. I have another technical question for the monitoring site. What is the minimum required period for sampling measurements that we can use for modeling calculation? The minimum required period for sampling measurements that we can use for monitoring sites. Ok. There are many ways of doing this, of course, so it depends a bit on the app loot that you're interested in. So if you're interested in NO2, there are passive samplers available, which are very cheap to buy and which you could apply in many locations. You could leave out for, say, a week or two weeks to get a two week or two week average. So that's available, but if you're more interested in particles, you really need to go to active sampling, so you need some more investment to get a pump and filters. So it really depends on the type polluted you're interested in. Also, the type of roll resolution you want, whether you want it by minute, by hour, by day, etc. Ok. Ok. I think we have all thank you very much case to be so kind to sitting, standing in front of your screen for more than two hours. So we really thank you. Excellent. And thank you. Ok. Thank you very much for allowing me to do it remotely. I would have wished to be in Trieste, of course, myself. But maybe next time I would be able to. Ok. Thank you. Ok. If anybody actually has more questions, I don't know whether you, there is an email address that you could maybe provide. We will provide. I would be happy to answer questions also by email of people. Ok. Ok. I want to come into contact. Ok. So thank you very much. For now it's time for coffee. For you too, case go and have a coffee. Thank you again.