 So thank you for inviting me. Today I will be talking about some of the work we have been doing in our groups over the past few years. And we generally deploy outdoor and indoor networks of low cost sensors. And we also collect personal exposure measurements for health studies. We develop novel analytical methods for the interpretation of the findings for health and policy purposes. Initially I wanted to present a few more projects, but I decided to focus on two projects that are our most recent work. The first one would look at the impacts of the first lockdown on air pollution levels in London. And the second one will hopefully convince you why personal monitoring is important for health studies. They both involve a large number of partners and they're highly multidisciplinary. And they involve a great range of disciplines starting from atmospheric scientists, computer scientists and all their way to building science. I will start from the fact that air pollution is an important environmental health risk. I think everybody is familiar with that. It is estimated that the cumulative effects of indoor and outdoor exposure result in eight million premature deaths every year around the globe. And this does not even include the years of disability. It's not a new problem. It's been around since the ancient times. But one example I always like to bring up is the London smoke event in the 50s where you can see that an increase in over four days and increase in smoke and gas has polluted resulted in excessive deaths, which did not return back to baseline even after the levels went down. And this is because health pollution can have lagged effects. Another thing that I would like to mention is that the burden of air pollution is not equally distributed among countries with the poorest and most vulnerable populations being the worst affected and basically being exposed to the highest levels of air pollution resulting to the highest number of deaths. There are many pollutants that have been associated with health effects. One of the two of the most serious is the NOx nitric oxides. That's the sum of NO and NO2 and particulate matter. NOx is primarily emitted by cars and trucks because it's created by combustion at high temperatures. There is convincing evidence to associate NO2 with an increase in the prevalence and incidence of asthma attacks in children and also with respiratory and cardiovascular effects in older adults. Particular matter can either be formed directly or indirectly in the atmosphere through complex processes a mixture of different chemicals that can be solid or liquid. I will focus a little bit more on particles. We normally classify them by size. I would be focusing on PM2.5 in this presentation because the size is important in a few different ways. First of all, it determines how long a particle will stay suspended in the air and how long it will travel. It also gives us information on the chemical composition. Generally larger particles come from natural sources, smaller particles come from combustion. Emerging evidence indicates that these combustion particles are actually the more toxic, which is unfortunate because they are also the ones that go deeper into the respiratory system. They can even reach on the alveoli and from there pass into the blood and harm every organ, the liver and even the unborn fetus. Because air pollution is such an important health risk factor, governments around the world try to put limits on the exposure of the global population and they routinely measure levels with reference grade instrumentation. While they are very accurate, the main limitation is that they are very expensive to set up and maintain and they require infrastructure. As a result, this is a map I downloaded from open air quality. You can see the discrepancy between developed and low and middle income countries where low like Western world is overrepresented and has a much better infrastructure. For example, in London, we have the London Air Quality Network that has around 100 of these instruments and it's one of the biggest and well maintained networks around the world. However, even so many sensors like 100 of them are not adequate to capture the high variability of outdoor air pollution or the exposure of the population that constantly changes. And most importantly, because there are no resources in countries with lower income, they don't have the capability to deal with this urgent environmental issue. With that in mind, we've been developing low cost sensor platforms that they are very cost effective to deploy and maintain. And their main advantage is that they can increase the coverage both because we can create much denser outdoor or indoor networks, but we can also capture personal exposure, which wouldn't have been feasible in any other way. Here we have an example of a personal monitor, I will talk more about that, and also they don't require power so they can even be deployed in rural settings of the developed world. I'll start now with the first project, I'll talk about the Breathe London. We deployed these sensors as the one shown in the top right corner, we deployed 100 of them as shown in the map here and we expanded the existing network. And they measure everything that is measured on a normal reference station and I would like to refer to two elements, I won't go into detail about that, I just want you to be aware of this. The first one is that we collect data, we collect fast data, so we have one minute normally a station gives you one hour resolution and that allows us to do scale separation. I also will not talk about that right now about how we do that because it's a bit complex. But this can enable us to do source apportionment. Secondly, we can collect CO2 measurements and while CO2 is not an important proxy, for the outdoors CO2 is not a good proxy for health effects, it can still be very useful in the analysis of the data because we can extract emission indices. Shall I go on, does someone want to ask something? I'll continue. Okay. So there are, I would mostly focus on the second outcome, how we use modeling to do the data simulation. And, but before I jump into that, I need to mention how we do the calibration. I mentioned earlier that there are concerns in the scientific community about the accuracy of such sensors. And many of you know that the way that we do the calibration is quite tedious and labor intensive. The first way is either to do it in the lab conditions or we can deploy them outdoors next to a reference system for a few weeks. So to overcome this we came up with a new methodology that we can do this calibration automatically while we collect data. I won't talk about that and I hope the publication will be out very soon. But this, this is a, I think it's a game changer for these technologies. So moving now on how we do the automatic interpretation. Looking at the top, the red line, this is the two months data time series from last year. And the red line shows where the measures came into force officially when it was announced that we should stay home. And those are the knocks before and after and just just eyeballing it. We don't see a big difference in the period before and the period after. So let's take a closer look and see why that happens. Those are data collected by the way with our network. We employ an ADMS model. So this is the ADMS model is a dispersion model that takes into account emissions and meteorology to output concentrations. And we see a reasonable agreement between the ground observations and the black. And that is the output of the model until the lockdown. So now we're going to let the model run to do business as usual and see what happens. So the model has not been updated. It doesn't know that there's been a lockdown and everything changed. So it predicts what would have happened. And now we see that it doesn't agree anymore. Why does this happen? Firstly, it explains why we didn't see an option and an obvious reduction because the meteorology is really important. And so like the effect of the meteorology masks the reduction that we should have seen. So what we learned from that is that local interventions are effective in reducing NOx. So if we want to reduce this harmful pollutant in London, traffic risk restriction would indeed have an effect. So this is a natural experiment that we wouldn't have been possible to do otherwise. So let's go on the data simulation. And this slide is a bit complex or bear with me. So what we did is we took the model and we started tweaking the emissions. The emissions are in purple. We started tweaking them until the concentrations that the model produces agree with what we measure. We see here that the reduction before and after lockdown is only 15%. But if we now look at the emissions, the reduction in the emission is significant and it's in the range of 85%. And so it also teaches us an important lesson that if we want to evaluate the effectiveness of the interventions, we need to use this kind of smart analysis and we cannot just rely on the measured concentration levels because this is not a good way to quantify the intervention. And finally, a look now in the same way, I will look at the PM 2.5 levels. And this is the levels before the lockdown and the levels after. And what is quite unexpected is that there is an increase in the levels. So we do the same thing as we did before. We run the ADMS model and then we compare the model versus the measurement. And we see that there is actually no visible decrease despite the significant reduction in traffic. And that shows that most of the particles in London, they come from the far field. So they might be coming from mainland Europe and therefore local interventions wouldn't be effective in controlling this harmful pollutants. If we wanted to tackle that, we would need to tackle it in the international level. In sum up, this is the first part of this presentation and I showed how we can collect good quality data from networks by using an automated calibration method. And secondly, I showed how by using a simulation of data, we can extract valuable information from these data sets. And by quantifying the different changes that we see, we can guide police. And now I would move on to personal exposure. I would start from some uncertainties that are still out there and I would also include some concepts so that the results that I present are a bit clearer. I've put some contour maps again from the ADMS model that I showed earlier that we used for the Breathe London. And the scale starts from blue and goes to red, lowest to highest. So you can see that London lights up and there are specific hotspots of exposure both for an auto MPM 2.5 because at this time scales and at these spatial concentrations, these pollutants are highly correlated with each other. On the other hand, ozone, because of the chemical properties, shows an anti-correlation with, so when an auto is high, ozone is low. This has important implications in health studies. So this is a graph. I don't know if you're familiar with it. It's very common in health studies to quantify the effects of air pollution versus a specific pollutant versus a specific outcome. And I will come back to this graph towards the end of the presentation because this has been produced without data. So we use ambient levels versus a specific health effect. And we see that all gaseous pollutants that are produced by the same processes, they show a similar effect because they're correlated. We also see that an auto MPM 2.5 are highly correlated and anything that's above the zero line, it means that it's harmful to health. So this high correlation between PM 2.5 and NO2 has been a problem in health studies. And as the Committee of Medical Effect of Air Pollution has stated, it's impossible to disentangle the effect of each one of these pollutants. On the other hand, ozone is below the zero line. So it shows that it has a beneficial effect for health or in other ways that breathing ozone is good for you. This is highly unlikely and this is something that is produced because of this anti-correlation that I talked about earlier. Someone who's living on the cleaner bit of London in the in the peri-urban area would be exposed to higher ozone concentration and lower pollutants and would therefore appear as if the as if ozone was the reason that they are better. So all these correlations when we use outdoor measurements as proxies of exposures are the biggest limitation of epidemiological research and we cannot distinguish causal links. So with that in mind, we developed the personal air quality monitor. We call it the PAM and it's a it's a very small platform with multiple sensors that they measure everything that the reference station would measure. It's very small. The participant can put it over their shoulder with a strap and walk about and get on with their daily lives. And all they have to do is just charge it at night and it will send data to our server through GPS technology and we then further post-process this data. It also collects activity parameters like GPS accelerometry and noise which help us develop a time activity model which I will expand also later on. Here is an example of a plotting GPS level, NOx levels and as the people go about their daily life. So we have GPS coordinates we know where they've been exposed to what levels and while it's not representative of the wider and exposure, it is still gives us an idea of where maximum exposure happens during daily life, right? We're walking on a busy street with the result in high NOx levels. The time activity model is important because as we walk down the street and we breathe in air, how much air we breathe in would depend on how physically active we are. For example, a person running down a street would inhale more air and therefore would inhale a bigger dose of air pollution than a person walking down the same street that has lower physical activity levels. The performance of the pump has been excessively validated and if you want you can look at the AMT public case in multiple configurations, outdoor, indoor and in transit. So we need to understand what is the association between outdoor measurements, indoor measurements and personal exposure and the best way to do that is by employing a very simple mass balance model. And as an illustrative example I would say I have simulated some data. So this is, we start with the outdoor, we simulate some outdoor levels. In the first case we have an inert pollutant that means it's not chemically reactive so there are no losses. And whatever is outside, whatever is outdoor reaches the indoor environment through ventilation, through intended and unintended openings. And then I simulated some sources that they would, some irregular sources that they would simulate like what we do on our daily lives, smoking, quick and clean, everything that introduces indoor air pollution. So the total exposure can be thought as what's generated outdoors, plus what is generated indoors. Right. And the outdoor, so every source would take some time to decay to the outdoor levels and the rate of this decay depends on the ventilation rates of the specific building so it depends on the building characteristics. So the second case we're looking now is a reactive pollutant so that means once it comes inside there are indoor processes like chemistry or the position on surfaces that would take away some of the pollutant that comes from outside. It would get lost. So this is what we would see, we would see generally lower levels than the one outdoors. And now I introduced the same sources as in this case, but this time the peaks are not as steep because some of it is lost. And also the decays are much faster because it's lost both due to ventilation as in case A, but also due to the indoor sinks. So while it seems like a very simple concept, it is quite hard to model it in real life, because a lot of the building characteristics, the materials, the location, how often the occupants open the windows and what kind of sources they introduce, all of that they introduce great uncertainty. And as an example, I am going to briefly mention some of the work I've done during my PhD on indoor air pollution in London schools. So this is an older building, it's a Victorian school, and it's close to Main Street, and this is a contemporary building farther away in a background area. And what we've seen is we see very different characteristics in the air pollution profiles of the two buildings. This one has higher NO2 levels because they are produced locally and then they infiltrate indoors because the building is very leaky. This one has much lower NO2 levels. And as a result, the asthma prevalence in the school is like 10 times higher than the asthma prevalence in the other school. And while this might cover some other health inequalities between the two groups of children attending the school, I think that school exposure is quite significant in these outcomes. So now moving on to some PAM data, they've been collected over a week. So we have one participant here, the PAM data are in blue. You can see there are seven spikes and they probably reflect cooking events. We collected this data in China and the red line are the outdoor monitoring stations that is very close to the participant's home. So we see the same features that we saw in the simulated data. Now CO and NO, at this time scales can be considered inert. We see a distinctive spike that decays over time and reaches outdoor concentrations. In the absence of sources indoor equals outdoors and both for NO, but then when we look at the NO2 or OSON or PM2.5 they're all reactive, lower indoor levels in the absence of sources and much faster decays once a source has been introduced. So I'm hopefully convinced you that far that personal exposure does not equal outdoor exposure and because we spend a lot of time indoors it might be more correlated with indoor air quality. And the reason for this big discrepancy are both the activity patterns but also the modification effect of the building characteristics. Shall I take these questions now? I see that there are three questions in the chat or shall I wait to take them later? It's entirely up to you. You can see what the questions are and take them now or we can do a longer question session at the end. I don't know, I'll have a look. So I guess there's nothing specific to... Yes, okay, I will take them to the end. Thank you. Great, thank you. So now another concept that I would like to introduce is the importance of the time activity model. So what we do is we have input parameters that we collected from the PAM. Here is the example of the road GPS data collected in the China deployment you see as the participants when here and there. And we use a mixture of artificial intelligence and rule-based models and so on to distinguish the major environments based on time use metrics and space use metrics. So we know when they've been home, other locations are transit. And then we further break down transit into modes of transport because as I've said earlier, it's important for the physical activity levels. And we can in that way characterize exposure and dose in a very high spatial and temporal resolution. And now I'm going to bring an example of deployment with the PAM. It's just a time series. It's a simplest example I could find just to illustrate this exposure and dose issues. It's a colleague of mine. He offered to carry the PAM for one week. So he lives here, he comes to work, he cycles about on the weekend, he takes the bus, he goes to London. At the same time, it's a time series of one of his representative work days. He sleeps, he comes, he cycles, he's at work just before nine. He cycles back home, he does a little bit of shopping and he cooks and he sleeps, right? So what's really important is that his maximum exposure during his daily life, for example, to NO, happens while he cycles at a rush hour. You can see here the accelerometer, you can see clearly when he's out and about. And his maximum exposure to PM 2.5 is when he's cooking. So this kind of short term events that affect personal exposure would not have been captured with an outdoor network. So the only way to capture these things is with a miniature sensor that the person wears on them. It doesn't obstruct them and they get on with the day. And so the inhalation rate, I got this from the exposure handbook, depends on who you are, personal characteristics. Are you a man or a woman? How old are you? What's your weight? But also on the specific activity that you've been doing. So I estimated us now in two ways for this one day. The first way is I take the pump concentrations and then I multiply it with a generic inhalation rate and I get this white bar. And the second way is I calculate the inhalation rate for my friend and then I calculate the pump measurements and I multiply it with his specific inhalation rates for the specific task that he is doing. And now you can see that cycling, although a tiny fraction of his day contributes significantly to his exposure. And the difference between taking into account the specific inhalation rates or using generic can affect the dose by a factor of almost two. So it's a massive difference that you need to take into account. So personal exposure does not equal dose because physical activity levels are important. And now I would jump into the airless project and I'll try to explain to you how we use these previous lessons to address remaining uncertainties. So we recruited 250 participants in an urban site in Bezin and the peri-urban site just outside. We asked them to carry the pump one week in the winter, one week in the summer. And then we asked them to come to the clinic three times to take detailed measurements of biomarkers and other health outcomes. So here, for example, is one participant that has been trained to use a spirometer for daily lung function measurements. And we take these. So now we have an integrated database that we know exactly how much air pollution they've been exposed to. And we also know what are the specific medical outcomes. And then we try to run some statistical analysis to find the links between exposure and health and, more importantly, uncover the underlying mechanism of air pollution on health. The two cohorts were quite different. The urban one in Bezin lives in high-rise building. They have centralized heating, a lot of occupancy density. They are, and while the peri-urban cohort, they are mostly, they mostly occupy themselves with agriculture. So Pingu is very famous for some agricultural products like peach and some flower, chrysanthemum flowers. And the richest of these people, here is a participant from the rural side coming to the clinic, to the rural clinic. And the richer of these people can use coal for heating and cooking and generally for domestic energy use. While the poorer ones, we've noticed that they use corn cob or generally any leftover waste that they can get their hands on and they can burn it and satisfy their domestic needs. And while we were there in winter, and you can see there is a, there's a paper on the characterization of this period. And we had frequent haze episodes. So haze is like increased the levels of some pollutants and particularly PM. And so they, it's a quite visible event. In the summer, we had very high levels of ozone. So we have a very different environment in winters and summers. So we collected the data and and we analyzed them. We cleaned the databases, we did everything we had to do. We applied the model and now we look at their time budgets. They spend a lot of time at home. They spend as much as 90% in the peri-urban side. And as you can see, the people in the peri-urban side do not cover such a large spatial distance as the people in the urban side. They're generally far more mobile. And then we looked at their diurnal pattern. So they, they, the most likely time that they would become new then would be during the day. Now let's look at the doses of these people and please bear with me because this slide is a little bit complicated. We calculated those in two ways. The first way, A, is a white bar, is what a standard epidemiological study would do. It would take ambient concentrations and then generic inhalation rates because I don't know what you're doing, what this person is doing, multiply them, estimate what they've been exposed to. And then method B would be the refined method that we're trying to develop. So we know what you've been exposed to, we know what your activity, let's find out what your dose. And then when we compare it, we see a massive difference between the two estimations. So outdoor exposure does not, outdoor measurements do not capture personal exposure. And what's even harder is that the extent of the misclassification changes between sites and it also changes between pollutants. So it's a very difficult, it's a difficult way to, to model this misclassification if you don't have a personal monitor. And finally let's have a look what's happening in the UK, how this compares to the UK. So here we have exposure of the airless cohort and we compare it against, so the solid is airless and the hatched one is the UK and collected this data during a pilot study with 35 participants in London. And the first thing we see is obviously outdoor levels of an O2 in China were about three times higher than levels in the UK. And this is also reflected in the personal exposure. And the same for PM 2.5, they're about 10 times higher and see the same difference, like proportionally they have the same difference. So personal exposure will to some extent be affected by what's happening outside. But what is a wildly different between the two locations is that the home in China is really a very unclean place in terms of air pollution, while in the UK home is one of the lowest exposures. On the other hand, what I think is another striking result is that PM in the UK are generally very low, apart from the time that people go in the tube. And here we see quite high levels. And this is because these particles are produced locally by the wear and tear of the trains, and they are not so resuspended by passing trains. And it's also a very different chemical composition. It's very metal rich particle composition. So in the UK PM are disproportionately high in the underground. Now, looking at the normalized dose. And so I just normalized it so it's easier to compare because the differences are huge. And you can see that because of the long time that this participant spend at home, and also the high levels at home, and home is really this environment really dominates their dose. And this amounts for more than 80% of the total dose highly the UK, because home is much cleaner, where you can actually see the effects of commuting, particularly active, active commuting such as cycling and walking. And this is or some or also for the other environment. So, so yes, it ends up that commuting is important for the UK. We did see this exposure errors between outdoor and and personal exposures. Why is this important? And I'm going back to the graph I showed earlier on in the presentation. So I selected one health outcome. And it's a it's a C reactive protein. It's just produced in the blood when there is inflammation. So it's we hypothesize that the mechanism of air pollution on health is by produce by producing inflammation on on on the body. And we use the mixed effect linear model. It's, I won't go into detail it's imagine it's like linear regression is just a little bit more complex to account for for the fact that the observations are independent are not independent. So, we run this model. And what we see is that all pollutants seem harmful to her apart from ozone and and we use ambient measurement. Now we do the same analysis to see what we would get with the palm measurements. And now we see that no two is no longer significant. And that is that is exactly the point of this exercise that we're doing now we're able to break the correlation between this two pollutants and therefore improve our exposure estimates and reduce the bias that we get in health studies. Of course, it's concerning that ozone remains to have a beneficial effect on health, but as I've said this is the zero and thought model, and it's a standard way that the standard model that epidemiological study uses but we're able to realize that now we have far more complex data, both in terms of exposure. So we have very detailed exposure data, there are, there is all this project, there is all this evolution in the fields of omics. We can collect very detailed medical biomarker and the fact that that maybe this model is not enough. So my, what I'm doing right now is trying to develop better models with mathematicians and statisticians. So, to sum up about the personal monitoring. What we've seen is that the, this exposure error is not innocent because it can, it can introduce significant error in health models. And this has also significant implications because as we've shown earlier on the presentation, depending on which pollutant is important and we're trying to tackle, then it might mean that we need different types of interventions and different scales of intervention. So we're hoping that this improved exposure methods and more detailed health outcomes would provide adequate guidance for policy. And to sum up what I've talked about today. Firstly, I showed you what were the effects of the first lockdown on air pollution levels in London and we, and how we extracted this information from a low cost network that it's been with appropriate methods for quality and assurance and how we can use modeling to evaluate interventions and that that this kind of information are important to drive policy. And I think that one of the most important outcomes of this project is that we use London as a case study, but we developed a methodology that's transferable in any setting. And for the personal monitoring, I showed how exposure and activity are necessary in health studies and how we can achieve it, even, even when we recruit hundreds of participants and assessing this kind of health effects during daily life is revolutionary in the field of epidemiology. And this improved understanding can really improve the effectiveness of interventions. And I would really like to thank my group, none of that would have been possible without the Rod Jones group. And on the left hand side, I put a picture of the ABH project that was that was the kickoff of this project that was part, it was a part in this very big multidisciplinary project. Thank you. Thank you so much Leah for a fascinating talk covering such a wide range of data collection methods and data analysis that was that was great. For anyone who has to head off now please do join us for our next webinar in early July. Otherwise, Leah if you'd like to take a look at the Q&A box we've got some great questions coming in and I also have a couple from the panelists that I can pass on. So Jill Thompson asks, two questions. First, at what scale was the meteorology and did lockdown affect the micrometeorology? Did this impact sensor emission and particle recording? And also can the PEM measure virus particles such as COVID? Okay, metrology or meteorology is the first comment. It looks like metrology. Yeah, measuring things. Metrology is the way that you measure things. Sorry, I don't understand the first question. Sorry. Okay, Jill, if you could pop a clarification in the chat, that would be great. And in the meantime, Simon Usher from Plymouth asks, if there are any chemical speciation measurements of PM 2.5 to give clues as to what PM 2.5 material is in elevated periods. So what we try to do is we develop a method, a low cost method to characterize the composition of the particles based on their hydroscopicity. But again, this is a very specialized question and it is work in progress that we do. And we can do some form of source apportionment with the methods that we develop. So there's a question about the palm and now the palm cannot measure virus particles and generally one of the biggest limitations of every optical particle counter, like all the OPC is because all these logos technologies work with an OPC. One of the limitations of every OPC, even the most expensive ones is that they cannot go, they cannot really detect ultra fine particles, like anything that it's smaller than 0.3 or 0.1 depends how good the sensor is. And therefore we kind of lose a lot of mass on this range. So what we do is we try to extrapolate this mass that we lose. Thank you. A question from the Digital Environment team. What's the difference in terms of cost for manufacturing and maintenance of street level outdoor devices versus the personal sensors? Sorry, what is the? What's the difference in cost in manufacturing and maintenance between the street level outdoor devices and the personal sensors? Oh, you mean like the, so shall I stop sharing? Sure, sure. Yeah, so normally this kind of reference instruments they cost depends how good it is, right? Obviously it can start for 40, 50,000 pounds depends what it is. The sensors of the palm, everything included is cost like 600 pounds. So the difference is quite significant. So how in terms of maintenance do they each require? The reference instruments, at least they need like, they often need calibrations on cylinders and gases and for example when we were working in Bangladesh where the resources are very limited, the Department of Chemistry called to find the resources to buy this kind of cylinders from Germany, right? While the low cost sensors, the most of the effort that you put, I think the most expensive part of it is the effort to develop this post processing algorithms because an instrument is only as good as its algorithm. That's true for any instrument. So generally I think that depends on the environment. The low cost sensors can live for a few years. We found, for example, that in a very dirty and dusty environment the sensors tend to disintegrate quite fast. For example, the OPC has got a mirror inside so the dust deposits there and it stops being so reflective. So we saw the sensitivity of the instrument, the gain like changing over time. So you need to consider this kind of limitations but still, and there are many different costs in that sensors too, right? We are not really on the low cost sensor. We are on the lower cost. There are even sensors you can find today for particles that they get the cost like five bucks. A clarification from Jill Thompson that she did mean meteorology, not metrology. So Jill says in the first part of the talk the weather was taken into account at what scale. After lockdown the local ambient environment might have changed in addition to the emissions. How did this affect the models? The model takes into account the meteorology. That's what the DMS does, the dispersion. And how was that meteorology obtained? Is it at the same scale as the sensors? What metestations do you use? What's the details of the model? How does it get the weather into the model? I am not sure. The DMS is a commercial software. I could not answer that. Great. Okay, so Jill, yeah. Please do get in touch with Lea if you'd like more. With the CIRC. And so the second part of the question, after lockdown the local ambient environment might have changed as well as the emissions. So not just the emissions themselves but actually how the streetscape has changed. How do you think that may have affected the model? How would the streetscape change? Again, Jill, sorry I'm interpreting your question potentially badly. So the local ambient environment, I'm not quite sure what Jill might mean by that. Okay. I think that, sorry, can you repeat it? Sorry. So after lockdown, the local ambient environment might have changed in addition to the emissions. How did this affect the models? Again, that might be one to talk about later. Sorry, it's not clear to me. No problem. No problem. So yeah, Jill, please do send Lea an email and you guys can arrange to talk offline. Next question from Simon Creer, who enjoys your talk very much. What evidence is there that elevated levels of NO2 in London is impacting human health? Do people die prematurely or have high respiratory disease in London? And how important is bio-aerosol co-exposure in models in relation to respiratory disease? I think that from the work I've done and it agrees with what other people have found is that, especially for children in schools, the only important pollutant was NO2. So the issue now is that I don't know if it's NO2 directly that affects the health outcomes and I don't think it's clear to anyone. NO2 is NO2 because it's primarily emitted by traffic. It might be a proxy for any traffic-related pollutant that can be emitted with that. I have looked a little bit on bio-aerosol, but not in my current job previously. And it looks like that their pollution makes the effect of the aerosol even stronger. So living in a polluted environment makes you even more susceptible to viruses and allergens and all the things that you would be able to deal better in a cleaner environment. But again, all of these are open questions. Still, the evidence of NO2 is not clear. And also in my work in Cambridge, I worked with the COPD cohort in London. So we monitored people with chronic obstructive pulmonary disease for many years. And we found two, three years and then we found that the exacerbations were mostly attributed to NO2. Again, as I've said, the limitation is whether it's a causal link or not. So there's a lot of effort towards this direction, doing more detailed measurements. So one of the things I'm trying to do, and I haven't included that in this talk, is trying to use sample pollutants as proxy for more detailed measurements. And so in collaboration with people that do detailed chemistry, can I guess what other pollutants are co-emitted just by using emission indices in the same way that you would do it for the outdoor? Can you do it for the personal exposure? So this is the next step I'm taking, trying to break down a total exposure into indoor, outdoor, and what are the ratios of different pollutants and what can you derive out of it? Thank you. It's really interesting to hear kind of what the next stage is. And I guess following on from that, another question from the Digital Environment panel. Do you see a role for citizen science within this kind of? Yes, absolutely, absolutely. I think that the biggest thing we can do with that is be involved in the outreach and empower people and empower communities. For example, one thing we could do is deploy this kind of low cost sensors in schools. CO2 in schools is important, PM is important, especially taking into account the current challenges that we're facing. CO2 can be a proxy for your probability of getting an infection. So there is the Welsh Riley equation that it can directly associate CO2 levels, probability of viral infection. So we could do that, like deploy this kind of sensors in schools, empower the people, learn about their exposure, reduce their probability of catching infections. I think that there is a great, great opportunity. And yeah, and also big data is a part of it, how you would serve this feedback at real time, how you would enable it. Yeah, I think that's a whole other subject, you know, I should give a whole talk on how you process your data and I don't know more about that, but I think we'll be coming to the end. There's just one final question from the panel. So we've talked a little bit about the cost of one of your sensors versus one of the standard monitoring sites. What do you think the comparison is in the worth of one super calibrated sensor versus lots of sensors at coarser resolution that maybe aren't quite as accurate. Interesting. That's a very interesting question. Yes, lots of lots of less high quality data versus smaller numbers of very high quality data. That's, yeah, that's a that's a brilliant question. I think that both approaches are valid depends what you want to get out of it. Yeah, it might be, for example, for citizen science, you don't need the super calibrated down to the PPP levels as we do. So now with them, with the calibrations we do, we can get down to the two, three PPP level. Maybe you don't need that. 10 PPP might be fine. Yeah, and that that brings us on very nicely to some of the the other questions that will will be addressing in the webinar series you know what what is the resolution that what is good enough. Yes, that's good enough. Yeah. Okay, well, with that, I think we should draw the webinar to a close. Thank you so much Leah for your contribution that was a fascinating talk. Thank you everyone for your questions. And please do join us again at the beginning of July for the next talk by Professor Jane Hart.