 I forgot to ask you that. OK, well, I'd just like to welcome everybody back this week to our next installment of the climate series of seminars that we've been organizing together, ICTP and the University of Trento. And this week, we're actually returning now to the impacts theme we've had in the series of talks so far, one or two talks that have touched on climate impacts or machine learning for climate impacts, which is the talk we had last week. So this follows on very nicely as a topic. So I am very pleased to give a warm welcome to the series to Veronica Huber. She's a climate impact scientist and she's had a special interest for a long time in human health and aquatic systems. And she's employed various modeling techniques to analyze empirical data for climate impacts. Currently, she's a senior researcher, a UPO in Sevilla in Spain. And she was a sectorial coordinator. She's a sectorial coordinator. I should use the right tense for health in the intersectorial impact model into comparison project. Another of these MIPS that we see in climate. And this is for sectorial impacts. And this is actually where Veronica and I got to know each other when in the first round of the projects called us the acronym EZMIP. We were actually working together on some of the aspects of health where we were contributing various different aspects to that health impacts of the modeling. And I'm super pleased to hear that she's just had a proposal accepted. So she's now going to become a Marie Curie fellow moving to LMU in Germany and continue on a good work on climate impacts. So with no further ado, I'd just like to thank you for taking the time to give this talk to us and I will hand over to you. And I will mute myself and clear the way. And thank you. Thank you very much, Adrian. Yeah, I'm most happy to speak in this series. And thanks a lot for inviting me. It's great to see that you have so many great female scientists who presented in this series. So even more happy to be in this queue of female scientists. So I will share my screen. OK, so can you see this? Yes, absolutely. Great. Yeah, so welcome to everyone. And as Adrian said, so this talk will be on the health impacts of climate change and specifically on heat and cold related excess mortality. And yeah, just the outline of the talk. I will first talk a bit of my personal motivation and why am I studying this and give you some introduction. And then I will present some models and of the type how we measure actually cold and heat related mortality with an example from Germany. I will present something on future projections. Also some work in progress on accounting for adaptation in future projections. And then at the very end, a global perspective on limits to heat adaptation. OK, so let's start. How did I get into this research field? Because I actually did my PhD in immunology, so working on climate impacts on freshwater systems. But now since a couple of years, I'm working on temperature related mortality and health impacts. And so how I got motivated is actually by being exposed to climate skeptics arguments. This is really some years ago where I helped battle some of these arguments. And so this one argument that we came across a lot was that, OK, with climate change, you would expect increases in heat related mortality. But at the same time with global warming, you would also expect decreases in cold related mortality. And actually the net effect would be an overall decrease in mortality rates. And so this is just one book that also includes this claim. And so at the time, I was just, OK, brought a little bit interested, but I didn't work on this. And then my first entry point in all of this was working on economic assessment models. And I think you also had a talk in this series by Tamar Kalten, who was really an expert on all of this. But this was just a study looking at one of these economic assessment models. And where the assumption was, and also the results were that really with climate warming, you would expect a significant decrease in overall mortality related with temperature. And so in this study, I won't go into details, but we just identified a couple of flaws. And by correcting for these flaws, you would actually got the opposite results so that climate warming, you would expect that you would get an overall increase in temperature related mortality. OK, so this is really my entry point. But now, let's really look into mortality data. And so here is a plot of daily death counts in Sevilla, where I'm based. And Overline is just a spline function. And you can see that there is the seasonal cycle. And what you can also see if you look at when are the peaks, so the peaks are in the winter month. And so this is generally the case that you have about 10% to 30% more mortality winter than in the rest of the month. And OK, so this hints a bit at temperature, but of course, there's also other seasonal factors that might underline this. Then of course, there is this public awareness of heat waves and their effect on mortality, most striking in Europe, the 2003 summer heat wave and here is a photo of Paris. And this is including an information to the citizens that it's like if you're looking for a victim of the heat wave, then there's a hotline that you can call. Because actually, there were so many people vying during this heat wave that they had to set up refrigerated tents outside of Paris because the morgues were overcrowded. So we do have these dark impacts of heat waves that we are aware of. And here actually, this is also a nice comparison because it shows the mortality impact of the 2003 heat wave and it compares it to the first wave of COVID-19 last year. And here what you can see in this figure is in light gray, it's just the daily death counts in the years 2000 to 2019 and then emphasized the death in 2003 and then the COVID. And you can see this extremely strong peak in during the heat wave in 2003 with actually on its most acute actually, even then the COVID wave, of course, the integral of the COVID wave is much bigger. So more people died in this first COVID wave, but really the daily death in France during this heat wave during these hottest days in August 2003, they were extremely high, extremely strong excess death. OK, then something else to keep in mind when we talk about these heat-related excess mortality. So most of these deaths are actually due to cardiovascular or respiratory causes. So very little are really heat strokes. And also important to know is that respiratory, cardiovascular causes are important, but there are very, very many causes that show a relationship that show that there are excess death due to heat. This is actually why in the following in the studies that I will show you, we normally work with all cause mortality. So just using the best registry of the statistical offices without looking further into cost specific mortality. OK, so this was an introduction. Let's now look into the methods of how do we actually measure cold and heat-related mortality. And so this example comes from a paper I published last year with data from Germany, where we had 23 years of data. And we did have all-cause mortality time series, daily all-cause mortality time series. And we also had climate data from local weather stations from these cities. And OK, so now let's look into the methodology of all of this. So the typical way of analyzing this data is you use quasi-poisson time series regression models. As the outcome, you would have the logarithm of the daily death counts. And then you have a function that describes your association with temperature. And then of course, you also try to control for confounders. So what is typically done is that you have introduced some baseline trends to control for long-term and seasonal trends. But you can also include some confounders varying on a daily basis, which could be like holidays or day of the week, but which could also be other data such as influencer or other possible confounders that you have data on. OK, and so applying these type of models, the relationships you usually find and look like that. These are just now two examples from my German cities that I worked on. And so looking at Berlin, for example, what you normally find that there is a temperature of minimum mortality, which would be around 19 degrees in Berlin. And below and above this temperature, you find that the relative risk of mortality, so this RR relative risk increases. What is also shown on this figure is below it's a histogram of the daily temperature data. So you get an idea of the distribution of the temperature that a currently sees. And actually, the type of models that I've been using and that I usually use in the field now, what they do is they also account for a complex leg structure. So what I'm showing here is actually a cumulative risk across time legs. And behind all of this is this type of leg structure. So you have flexible splines that allow you to also look at nonlinear effects here. And so you have this three dimensional relationship of the relative risk and in the function of temperature and also the leg. And if you, for example, look at this three dimensional figure and you take a slice at a certain temperature. So you, for example, in the first case, you would take a slice at temperature minus five degrees and you would get the results shown here. So it shows that actually the cold effect on the effect of a day of mean daily temperature of minus five degree. So it would only occur a few days after the exposure, but it would also be long lasting and lasting around two to three weeks. And on the other side, for heat. So taking again a slice through this three dimensional function at a temperature, mean daily temperature of 25 degree, you would find what I show on the right side. So the relative risk is actually high on the same day on the day the exposure happens, but it then falls pretty quickly. So heat, the heat effect is optimal for you. And what is also shown that is interesting is that at some point after exposure, the risk falls below one. So this is something that is referred to as short-term harvesting. And this is the idea that some people may die from heat, but they would have anyways died a few days after that. So it's kind of, heat is limiting the pool of the individuals that would die anyways in a few days. So all of this is taking into account in this type of models that we use. Okay, so the type of models used normally also what is done that there's also a second stage where you do a meta regression because in all of these multi-city studies, what you do is you use the information from all of the cities to improve your estimates. So for example, for cities where you would have less samples or less certain estimates, you can improve this by doing meta regressions of the model parameters. Okay, so once you have your exposure response functions estimated and set up, then the next step is really attributing mortality to non-optimal temperatures to cold and heat. And so what you're interested in is in the attributable fraction. So you look at the percentage of death attributable to temperature out of total death. And to do that, what you normally, you define your heat as all of the temperatures above this minimum mortality temperature and you call this all of the temperatures below the minimum mortality. And then just the sum of these cold and heat attributable mortality is the total temperature-related mortality. And for example, for my German data, what you find, and this is very typical for many locations around the world, I will show you this a bit later, is that there's around, yeah. For example, you see all of my cities and on the left side, there's the city average. So in our case, we found like a little bit of about 6% of total mortality attributable to non-optimal temperatures. And then there would be like around five attributable to cold and around 1% attributable to heat. And to make this a bit more accessible, and of course, you can translate this also in mortality rates. And so for example, for heat, this would translate into eight heat-related animal excess death per 100,000 population, which for Germany would be like something like twice the mortality rate due to traffic accidents. And important here is to say that this is of course, average, no? Do you would see much higher numbers if you looked at certain heat-wave events, of course, no? So this is really average to cause all of these 23 years. And so how does this compare to studies that have a more global perspective? And so here's an example from a multi-city, multi-country study that looks at hundreds of locations around the world. And again, so here you see the heat and the cold-related mortality, but this time it's also split up into extreme and moderate cold and extreme and moderate heat, no? And what you find in general that the cold fraction is much higher than the heat fraction, but of course, if you look at the extremes, the extreme cold and extreme heat is much more comparable to the burden. And maybe one more thing to point out in this figure is that this is really across different climatites, no? So most studies are done for temperature regions, but if you look here, there's also traffic countries and subtractive climates. And even in these countries, you find this very general pattern of cold and heat-related mortality. Okay, so let's turn now to the climate impact to future projections. Okay, so again, this is from the study we did for Germany. And so just that you understand the results that I show now is in this case, we were interested at looking at mortality estimates at different levels of global warming. So it took a particular approach, so where you have your scenario data from your climate models, and then based on the global mean temperature data for these models and scenarios, you identify time windows that correspond to certain levels of global warming. Now, so here we looked at one to five degrees of global warming. And so once you have identified these time windows, you can then get the local temperature time series to put into your model for the different scenarios and for the different climate models. And in this case, it's also important to point out that when we do these studies at the local scale, there's an additional step of doing calibration of your global climate model, your DCM data, to make sure that you calibrate the data with local, with weather station data to account for the bias, the model bias or to remove the model bias to the extent possible. Okay, so these are the results here. So again, I show on the left side, it's the city average for all the cities. And then there's also examples for individual cities. And you can see here now on the x-axis, the different levels of global warming. And then on the y-axis, it's the axis mortality, it's this attributable fraction. And the squares are the observed. So what I showed you before, and the dots are now the projected. And you find pretty much what you would expect. So you would, you find an increase, a projected increase of heat-related mortality as global warming intensifies and a decrease in cold-related and mortality. And the total, yeah, there's a slight increase, a special for the very, very high levels of global warming. And maybe one interesting point to mention here is only that for some of the cities, for example, for Leipzig on the upper right, you see that actually heat-related mortality exceeds cold-related mortality above three degrees of global warming. So this suggests like a reversal of this pattern that is seen under current day conditions. Okay, but let's not turn to my motivation. So what did this data on Germany tell me and tell us about this question of this climate skeptics claim. So does global warming actually lead to a net increase or decrease of temperature-related mortality? And so here you see again the different levels of global warming. But this time it's just the difference against the condition of one degree global warming. So against present day pretty much. And so you see on a city average that our projections suggest that net increase actually above two degrees global warming with a very few exceptions. So for example, Hamburg down there and the bottom right, you don't find this clear net increase in mortality. But of course, this is just one sample for 12 human cities. And there's more studies that look at this at a more global scale here. This is another example from the same project that unites many, many mortality data sets and the climate data sets. And so here now for 23 countries doing the same type of projections, actually to point out Germany is missing still in this map but now it's there. We contributed our data also to this collaborative effort. Yeah, so this is the result from this paper and I hope you can see it. It's pretty small but importantly so it's kind of the same figure that I showed you before for Germany where you see the change in heat-related excess mortality, the change in cold-related excess mortality and then the black squares is the net change. And this time here it's done for different RCPs for different climate change scenarios and it's done for different decades in the future. And also this is change against the current decade. And you can see pretty quickly that in most of the regions you would find the same net increase in mortality especially for the high emission scenarios. And then there's also some exceptions where rather modeling suggests a decrease in net and temperature-related mortality. So for example, for Northern Europe, East Asia or Australia, this modeling suggests a decrease. But I think what is most important here is to find out that the strongest net increases projected and occur here in the less developed in the most populated countries, not for example, or regions. For example, Southeast Asia or South America. So it's really the strongest impact is projected for regions with least adaptive mortality. Okay, so this is now this climate skeptics claim revisited and showing that really this climate skeptics term at least in these very general terms cannot be held. But of course, one big caveat in all of what I've shown you before now is that in these studies they don't account for adaptation. And they also don't account for shifts in age distributions. They don't account for demographic shift. And so now I would like to present you a bit of working progress and where we try to account for adaptation in these projections. And yeah, there's of course very different ways to go about this, but one approach that we have taken and in the study that we are working on right now is that what you would generally find is that this minimum mortality temperature. So this optimum temperature and also the temperature when the temperature related mortality risk is lowest, it really depends on your local climate. So here's an example from Spain with many cities and they are ordered here in terms of endoming temperature and you can see that in general the higher the endoming temperature, the higher is this minimum mortality. So there seems to be some acclimatization to heat. And yeah, then the same for example has also been done at a global scale. This is really from many, many, it's like a kind of a meta-analysis study including many studies from around the world for different climatic zones. And here relating the minimum mortality temperature again relating it with indicators of local climate. So most frequent temperature, a certain percentage and again the endoming temperature and you find again that there is the optimal temperature. So the minimum mortality temperature tends to be higher, the higher the warming. So strong indication again about acclimatization. And then there's also fewer studies but also some evidence of shifting up this optimal temperature in time. So here's an example on the left side from Sweden with a very long data series where they find that the minimum mortality temperature has increased quite substantially, quite substantially doing the last 100 years. And another study from Japan where they actually also find that the minimum mortality temperature has increased over the recent decades. Okay, so then our study how and so we, the plan was really to use this type of evidence from the past to include this type of acclimatization in future projections. And so now this time we worked with daily mortality data from Spanish cities. And so we had 40 years of data and what we did this time is that we fitted our temperature mortality association in different time periods. So we divided our data in eight, five year periods and we then with some sophisticated method we obtained these exposure response functions for these different periods. And we also got these estimates of the minimum mortality temperature for these different periods. And so we had these 88 data points on the minimum mortality temperature and we regressed that against different climate variables. And so the climate variable most planetary that turned out was the mean summer temperature. And so we then established a relationship between these minimum mortality temperature and the mean summer temperature. And here what we wanted to do is to decompose this into the temple and the spatial component. So yeah, I think I won't go into too much detail but we looked at the intra-city variability. So really in the variability in time of these mean summer temperature and then also in the spatial variability not in the cross-sectional variability. And so then we fitted these mixed effect metagression model on this and with the plan to then use this coefficient to be able to project the mean minimum mortality temperature based on future projections of the mean summer temperature. And so this is done here now. So for the example of two cities, Barcelona and Madrid, you see in black the observed data. So you have, these are actually five, these five year periods. No, so this is like each point is like, always these five year periods. And so you see the mean summer temperature and the minimum mortality temperature in black. And you see really how the minimum mortality temperature tracks the mean summer temperature. And then based on the relationship I showed you before, we then use the projections from five GCMs of mean summer temperature to project these minimum mortality temperatures out into the future for two different scenarios. And then we use these to define acclimatization scenarios. So we have a default where we just do what we did in my first German study. So we just use the fixed minimum mortality temperature. You don't shift anything, which is the solid lines in these figures. And then we have another scenario, which is the acclimatization scenario with shifting these minimum mortality temperatures according to these projections. And so now this is a bit complicated, but let's look at the highest emission scenarios at the red lines, no? And you find in first, if you look at the solid lines, you find pretty much the same as we found for our German studies. This is actually the city average here. So you have a strong increase in heat and a decrease in cold. And then the total you have like at the end of the century, you would also find a net increase in the mortality. But then if you assume this acclimatization, if you just project these minimum mortality temperatures out into the future, then you see that really, yeah, there's a very strong effect. And this suggests that the time and pace of acclimatization that we see in our data in our past four decades, if you assume that the same pace is maintained into the future, then really the expected increases and changes are much, much, much lower. And so the question really is, okay, what is behind these shifts in the minimum mortality temperature that we see in our data set? What type of mechanisms are behind that? And I think this is really something for future research, but just to mention a few things, of course there is air conditioning and there is unfortunately not a lot of very good air conditioning data, so residential air conditioning data out there, but some studies have now included this. And so one of these studies, and they have never related to the minimum mortality temperature. So this is something to be done, but they have tried to incorporate it in a different way. And this is just an example from a recent study where they had air conditioning data for countries and where here you can again see attributable fractions and they compared, so the beginning of the time series with the end of the time series. So in dark blue, it's the beginning of the time series and in light blue, it's the end. And you see that there is a strong reduction attributable fraction and due to heat. So this is really focusing on our own heat. And then you can do a counter-factual analysis where you just assume that the air conditioning would not have increased. And so you see that this is like this middle blue. You see that you would still find a strong decrease in the heat attributable fraction, but of course not as much as the without accounting for this effect. So that the air conditioning has played a minor role according to the study in changes in the vulnerability towards heat. And then there's this other fact, of course, also studied is the question of green areas in urban spaces. And yeah, there's a lot of evidence on the buffering of green spaces in terms of vulnerability towards heat. So this is just an example from Berlin that showed that in districts where there is less densely built up structure, the effect of heat waves on mortality is less than in this more densely structured urban area. So this is really definitely also something to take into account if you want to build much more concrete and explicit adaptation scenarios. And then another question is, of course these heat health prevention plans or early warning systems, all of these policies that were set up, especially for example, here's another study from Spain, which looked at the effectiveness of heat health prevention plans. And here they split their data into the period before the 2000 heat wave, 2003 heat wave, and after the 2003 heat wave, because this is really after the 2003 heat wave, a lot of countries implemented more of these heat health prevention plans. And so here in this study, they really tried to evaluate whether these heat health prevention plans had any effect. And so, yeah, this figure shows you on the X axis, they're kind of the comprehensiveness of these heat health prevention plans. It's, yeah, the more on the right, the better is the plan. And on the Y axis is just the difference between the second period and the first period in the attributable fraction. So I'm on the heat, on the heat burden. And so this suggests that, yeah, these heat health prevention plans that did have effect and they reduced these heat attributable fractions, at least if they were well set up. And so this is another thing, of course, to look into in more detail if you want to really understand this acclimatization phenomenon better. And we also want to include this in a better way and in adaptation projections in the future. Okay, so then, yeah, let's, I thought, then when I structured my talk, let's also take a bit of the bird's view and zoom out a bit. And so, of course, there's a lot of work on this topic. Yeah, heat and climate changes, of course, a very important topic. And so there's also many, many studies that look into this. I'm not working with mortality data. So it's not more climatology than epidemiology, but I thought I would also present through three studies to end with on this field of research. And so there's one study that I took a really nice approach also to that they did a huge search on reported lethal heat events. And they identified the climatic conditions during these events. And they paired that with also climatic conditions in the same locations who then use machine learning techniques to identify conditions where we are heat fields. And so this is what you can see on the left side. They found that really the main predictors are average daily temperature and average daily relative humidity. And all of these black downs dot correspond or crosses correspond to these lethal events. And then these blue and red line, they really, they define your deadly conditions. And so, and then using them in this case, these red line of saying, okay, beyond that, these are deadly climatic conditions. They looked into the occurrence of these conditions on the present day here above historical and then also for future projections. And here, so in this on the right side is now shown the number of days per year where this threshold is crossed above the deadly threshold. And you can see that in the historical data, there's some locations in the world where, yes, they are crossed for a number of days. But then if you look at the end of the century for these very high emission scenario that you find, yeah, this extends extremely in these lower latitude and you have actually even locations where you would have these deadly conditions year round. So this is, I think, a really interesting study. Of course, it's caveats, but it's a really interesting study to look at how, yeah, how, yeah, this challenge of heat, especially in the lower dilatitudes. And another study coming pretty much to similar conclusions is this one where the authors looked at something what they call the human climate niche. So they used big data sets on the distribution of the population across the globe. And they also looked at what are the mean temperatures that these populations are exposed to. So on the left side there, you see this plot of the population density of the globe in different time periods. So during this light blue, for example, you would also have this green, which would be 500 before present. So they have these polyclamatic data sets as well and estimates of historic and prehistoric population distributions. And you pretty much find that in all of these 6,000 years, population distribution was pretty similar. So you have these two peaks running around 13 degrees and then around 25, which is pretty much the Indian Muzun region. But of course, you can also plot into this future projections. And here again for higher emission scenario, RCP 8.5. And this would be your red line on the left side. And you see how really the climate change, the unmitigated climate change will remove a lot of the population out of the niche that they have experienced for 6,000 years. And then there's another way of showing this is if you, for example, look at the threshold, you look at this limit of the mean and your temperature of 29 degree. And you look at the globe and you see where is this mean and your temperature exceeded in the world. And so today it's these black blocks. So it's only in the soil regions that you have a mean and your temperature of about 29 degrees. But then with this high emission scenario, this area would extend to many, many regions in the low latitudes. And so with population projections, because these are also the regions where population is growing fastest, you would have a great, I think it's 30% of the projected population is exposed to these extremely high mean and your temperature is something that humans haven't experienced in the last 6,000 years. And where they haven't lived in the last 6,000 years. So this is another study pointing to these heat challenges. And maybe last but not least, I think my time also is coming to the end because I think this links also nicely with climatology. There's other very interesting studies that really try to define absolute limits of human adaptability, looking at thermodynamic limits in terms of how can a human body cool itself? And so based on this, you can define the wet bulb temperature, which is like, it's defined by just the temperature. If you cover normal thermometer bulb with a wet cloth and you ventilate it properly, then this is the temperature. So it's really combining humidity and temperature. And yeah, and so on this, based on the thermodynamics thinking, you can define the lethal limit, which is an absolute lethal limit. And so then you also can look at, has this ever been reached in the past? And you find not. And then you can look at, in this case, extremely high risk case scenarios. And you would find that many parts of the globe would actually experience this in peaks. And of course this would be, yeah, in a way non-adaptable because if you stepped outside or your air conditioning failed, this would be lethal. So I think, and there's a lot of, and now also studies looking at this and looking at this in more detail, and also even finding that for very high emission scenarios for, until the end of the century, you would even cross this limit in some individual locations of the world. So yeah, just I think this summing up this enormous challenge that climate change is. And in this time in terms of heat exposure. Okay, to sum up very, very quickly. So yeah, I think I showed you that how cold and heat exposure is related with excess mortality. I showed you that if under assumption of no adaptation, most studies suggest that we should expect the net increase in temperature derivative mortality, which is contrary to other widespread claims of climate skeptics. Then I also showed you the study on accounting for acclimatization and that I think these type of adaptive measures that we know of, they need to be studied further and they need to be included in mortality projections. And then at the very end, yeah, if mitigation efforts fail, large region of lower latitudes will be exposed to extreme heat conditions and they will probably exceed the adaptive capacities. And this is something I haven't said yet, but of course these studies often conclude that the only way out is massive migration. And I think it's just also an important factor to keep in mind when we think much more broadly from a birth view and about climate change. Thank you very much. Thank you very much, Veronica. That was really interesting. And as I suspected there have been some questions coming in on the chat while you've been speaking. So we'll go straight to the questions and I will invite in order to unmute. So hi, thanks for the talk. So my question was, as you focused the first part of the project focused on the mortality in cities where obviously there are more people and more data available. And you talked a bit about the difference between.