 Okay, I think we can start. Atish? Yeah, okay. Okay, good afternoon everybody. Welcome to this second panel of ICTP discussion in its series, ICTP series of lectures supporting the 2022 United Nations International Year of Basic Sciences for Sustainable Development. IYBSSD 2022 as it is called. My apologies I cannot be present in person, but I look forward to the colloquium remotely. So the today's panel discussion is titled Artificial Intelligence for Detection and Attribution of Climate Extremes, which will feature Professor Kumu, who's a professor of climate extremes and societal risks at the Vijay University in Netherlands, and Robert Wotal, a meteorologist and climatologist at the Institute, P.F. Simone Laplace in France. So let me say a few words about the International Year for Basic Sciences for Sustainable Development. This was approved by the General Conference of the United Nations recently, and as a founding partner, ICTP is very proud to belong to this and support the activities of IYBSSD along the themes that are important for IYBSSD. I think this year rightly puts the focus on the important but largely underappreciated role of basic sciences in nearly every aspect of our lives. And it underscores the notion that basic sciences crucial to the attendant of the 17 sustainable development goals adopted by the United Nations in their agenda 23 plan. The International Year will be for sustainable, basic sciences for sustainable development will be officially inaugurated on 8th of July at UNESCO headquarters in Paris. And events and activities will be organized around the world until June 2023. And we are very happy that the first of this, we sort of kicked off this with the first colloquium recently on this topic, the importance of basic sciences and addressing the global energy crisis. This was on the 18th of May. And today's colloquium is the second in the series on an equally important topic, which has to do with climate extremes which are immediate and direct concern now for many people around the world, and also the role that artificial intelligence can play in this detection and attribution. So, I leave the floor now to Professor Erica Coppola. She, she will moderate this event together with Professor Davide Faranda. So, over to you, Erica. Thank you very much. So welcome again to the events. I'm Erica Coppola, I'm a research scientist here in ICTP in the Earth System Physics section, and this Davide Faranda is. And we will introduce now the, the, the, the, yeah, the panelists that will give the colloquium, and then we'll give the word to them. So the events will be like roughly half an hour of presentation from, from the two speakers. Then we will have plenty of time for question and answer is supposed to be an interactive colloquium. So from the people in, in the room but also from the people online I see there are quite many people online and then we will leave the room for ICTP students so the panelists will stay with them and answer their questions. And then last but not least there will be refreshment and cocktail in the terrace above us at seven o'clock. So you are welcome to attend the people in presence, the online one, we will send you a photo. Okay, let's start so the first speaker of today's dim crew, as the director already say, he's an associate professor of extreme weather and climate change at the University of Amsterdam is also a group leader of the Possum Institute of climate impact research is an expert in advanced statistical analysis climate model and machine learning technique. It was getting a lot of funds both from national but also from international funds agency, and in particular also private fund agency high level one focus to focus is research on the most relevant extreme so the one that are mostly harmful to the society. And today you will speak a little bit about this introduce you to what is research expertise. Okay, and for me it's a great pleasure to introduce Robert Votar, who is the director of the Institute PRC Mola plus and researcher director of research at the National Center for scientific research in France. And Robert is also the coordinator of the project. He will talk about that. Robert is a brilliant career that starts more on the dynamics of the atmosphere and with seminal works understanding the relationship between some recurrent pattern in the weather. And then he moved towards the applications of the extreme events in climate and now also coordinating several projects, mostly about climate services, but still keeping the interest for the atmospheric dynamics. So, we will be happy to hear Robert after them. Let's stop with them please get to the stage. Okay, yeah, thank you very much. Thanks for the kind introduction so great to be here. Yeah, I think maybe many of you know we have this summer school this two weeks at ICTP and it's for Dutch guy like me is just fantastic to be here. So, hopefully, see lovely weather, lovely food. So that's great. And of course also important science. I would like to say a little bit just give you a bit of an idea of the type of science that we are doing. So we are interested in how climate change is affecting extreme weather all around the globe. And this is from heat wave to droughts to tropical cyclones, etc. So understand better how that is creating risks for society. For example, impacts on agriculture, you see here, or flooding events. And I think that's, yeah, so in my talk. I would like to highlight the type of research that we are doing the stuff that you are really confident in. And also, you know, the part of the climate system that we understand less well, and where there's more and more uncertainty. And also, what XIDA is going to look into in the coming years ahead of us. So XIDA is in here. It's a quite large consortium of many climate scientists collaborating on these themes and working for the next three, four years. But let me start with this graph. So this shows global mean temperature of the last 2020 years. So here's basically the year zero right so Jesus was born in a relatively cold year, as you can see, and you see it's relatively flat over the first 1000 years it gets a bit colder. In the 1670 18th century, this is known as the little ice age. So especially Europe was was quite a bit colder. This is due to natural variability in the climate system. And by the late, late 18th century. You see here invention of steam engine. So that really tickstart the industrial revolution, right. And since that moment, we've seen a lot of warming so a bit more than one degree warming in the global mean so average over the full globe. And right now we are up here. And so this is a quote from the latest IPCC reports so the IPCC stands for the international. International panel climate change intergovernmental panel climate change excuse me. So this, this does create reports every five, six, seven years, and really tries to assess the state of the arts of our knowledge on the climate system right. I'm familiar with it. And in this quote it's stating it's, it's an equivocal that human influence has warmed the atmosphere ocean and land so that's a very strong statement right unequivocal. It means that we are really certain that this warming over the last 100 years that is really due to a mission of greenhouse gases. So let's talk about that part right this is really proven science and really want to talk about what it means to be up there in the climate system and what it means for the weather that we are experiencing. And we do see increases of some types of extreme weather events. For example, if you look here at the left. So these are analysis where we look at observations from about 50,000 meteorological stations all over the globe, and we just start looking at extreme events and start counting them right. So we see here for example for heat waves, we see a strong increase since the 1980s, and in this figure above, especially since the 1990s we see an increase in these daily rainfall records. Right, so this is happening all over the globe. This is of course creating risks for society heatwaves are often links to wildfires. This is a picture taking of a recent heatwave in Australia. It can also cause damage to agriculture harvest losses and extreme rainfall events of course can cause floodings. We see on a global scale that heatwaves and extreme rainfall events on the rise. If you look at drought. Then we see more regional changes and this is again a picture, a figure that took from the IPCC report. And what we're seeing basically is that those regions that have already relatively dry climate are becoming even drier. You see this, you know this is, you should visualize this as a global map North America South America, Africa, etc. And basically all these regions that are have this brown color. I've seen an increase in drought since the 1950s. Quite notably the Mediterranean right Mediterranean is really a hotspot of climate change. We see winters getting drier and summers getting a lot hotter and more an increase in heatwaves. There we see a clear signal. Also, Sub-Saharan Africa, large regions there, Western North America notable. And so a region like California has seen quite intense drought in recent years. Good. So, the increases in extremes that you see on these slides, right, that that is the stuff that we understand very well as climate scientists. This is directly related to warming of the earth surface. If we want the earth surface summer temperatures get get warmer, and therefore the probability of heatwaves strongly increases. That's what you see in the lower left panel. Also, the air of the lower atmosphere is warmer. And therefore the air can hold more water vapor and that leads to more extreme rainfall and can also intensify droughts. So this is the stuff we understand very well as climate scientists. But there are also extremes that we do not understand very well, right, and that are really much more intense than we anticipated right and sometimes we refer to such events as black swans. Maybe I should illustrate this a little bit so the phrase black swans, it goes back to actually Roman times and initially it was really used in Europe, a phrase for events that were impossible. All swans were simply white. Black swan was impossible. But then in the late 70s century, there was this Dutch sailor, he went to Australia, and he was actually the first European to observe a black swan. So since then the meaning of this phrase has changed into events that are considered to be impossible but then actually happen. And yeah, in climate science. We use this to illustrate events that are unlike anything we observed over the history of observations. We've seen several of these type of surprises in recent years. So just to give you an idea this is summer of 2021. And we had a very intense heat wave in Western Canada, as you can see here already in June. Now if you look at the warmest day of the year in this region. We've looked away it's about between 28 and 32 degrees since the 1950s. There's a bit of an upward trend also. But then 2021 really totally jumps out of this historic range. Very much a surprise, right. Also, if you look at local temperatures. Right so this is the town of Lytton in Canada. We had a temperature of almost 50 degrees C last summer, that is really unheard of the previous record temperature was 40 degrees C. So, almost 10 degrees C more. That's very striking. And of course it raises a lot of questions, why, why do we see these record shattering events. Another case was the flooding in Western Europe so this happened in July just a few weeks later. A massive rainfall right at the border between Belgium, Germany and the Netherlands. This shows an image of the archiver so massive flooding and it's one of the costliest European weather disasters actually and also more than 200 people died mostly in Germany. So if you look again at the data so this shows the wettest day in the warm season. So it typically fluctuates between about 20 to 50 millimeters in one day in this river. And then 2021 really is jumping out of that with more than 90 millimeters. It's a really important question to understand these type of extremes better. As this figure here shows both events were linked to the jet stream and that's actually what we see quite often, right, if we have extremes in regions like the US or Europe. It's often related to dynamics in the jet stream and that is something we really still need to understand much better. This shows another example so this is a snapshot of the state of the jet stream in 2018 in July. So we had a very strongly meandering jet stream pattern that went really all all around the globe. It was also very persistent. And you see here in this part that that is related to high pressure and low pressure systems, and therefore the jet stream really determines the weather that we experience on the ground. And so we had a heat wave in Western Europe also wildfires in Greece, and at the same time we had heat waves in Japan and California. So if we have these type of large scale waves present, it can create simultaneous extremes around the globe. And this can, for example, create risks for bread baskets. If we have, if you have two heat waves in two important bread baskets, say the US and Europe, then we can hard can have harvest failures in both regions. And that can then drive, for example, food prices. Good. So this is a very important research question. Also, something that I personally focus a lot on. So how is the jet stream changing and what does that for mean for extreme weather. So we have data that shows that since 1979, the jet stream has been weakening, which, which you see here. So, since 1979 we have accurate satellite data so then we have fairly good estimate of the strength of the jet stream. So we do see this weakening happening. There's still a lot of open questions on what does this mean for extreme weather in regions like Europe and the US. And there, I think AI comes in. So, of course, we're living really in an in an age of increasing compute power increasing data. So this shows, for example, that compute power has been doubling every two and a half years also over the last decade. This allows us of course to do to create ever more accurate climate models. And also climate models that reproduce extremely events better. This shows a tropical cyclone here. And we, well, we're generating an enormous amount of data from satellites from radar, and all kinds of observational data, and that we can use. And that is very, I think, very powerful it really creates new opportunities for climate scientists. The question is really how can we, what I would call harvest new knowledge from big data, right, and new knowledge I mean, better understanding of the physical processes, for example behind dynamics in the jet stream. There are methods out there. I'm not going to discuss them. But things like explainable AI and causal discovery can really give us better insights into the underlying physical processes when we apply them to data. And that is really what this exciting project is all about. I leave it here. So, yeah. Thanks very much to the audience and so thank you so much. Erica, do you want to say something. No, thank you. Thank you a lot. I CTP also, I must say, as an introduction. It's not the first time I'm coming here. I mean the first time was in 1991. It's been several times. We have, we've had several collaborations with ICTP and member one of the project was on Poland's actually with Filippo, Georgie. And we have, we've had IPCC also with Erica, you know, three years in a video conference. So it was really a pleasure to be here and to meet again with colleagues. So I'm just stepping back a little bit from what Dim was just saying. So Dim, so I'm the coordinator of Excider, but Dim is vice coordinator. So we are doing the work together. So I just wanted to, to, to, to make a step back on why, what we know today and how, how we arrived there so that the IPCC report just showed that we have already observed climate change actually we are witnessing under our eyes, these are our heat waves on the top. These are the regions where we can say that heat waves are increasing both in amplitude and then frequency. And, and the red, the red regions, and the dots gives the confidence level one dot is low confidence and three dots is high confidence so there's almost all over the globe we have high confidence. So in every single region, actually it's not like the previous IPCC reports where we said it waves are increasing now now we say heat waves are increasing in every single region or almost. And it's due to climate change. So we, we are witnessing it and it's a it's a new fact actually from the last IPCC report. It's a bit less clear for for extreme precipitation on, I should use that. Yeah. Yeah, on, yeah, on this panel and on drought on this panel, the original differences and the IPCC report really discusses these differences. The question is where actually do we get. How did we get there actually. It's, it's, it's a very important result we, we, we have to go back actually to the work of class husband man if we class husband man is not the price I suspect you're aware of that. Actually, at that time attribution was not there really but the question was the climate change so you see these are the stripes. The climate change was not there yet. We couldn't see really a signal so but he was anticipating the signal would come at that time. The question was how what are the concepts what are the methods we need to develop to detect these changes and to make sure that we we are detecting things that are due to human activities. So there has been a lot of developments is since they're in particular a study which is a pioneer study actually to two studies to one from my zone and what from Peter start in 2003 and 2004, which showed which put the concepts actually on on single event attribution so it's not attribution of of changes in global temperature on average, it's single event attribution. There was many questions related can we attribute single event can we say something about the the effect of climate change on a single event like the heat wave we had a few few days ago. And so pioneer studies and then many many studies came in the last decade or so. And more and more we try to say something as you might have fun as you might have seen that there's a lot of media attention on that. So, oops, I guess it's going back. Okay, so I should. Yeah. So, how do we do how do we know that it's climate change so I just take a very simple example, which is not on extremes but which is on the mean warming. The picture, which is from IBCC you have simply simulations from climate, you have the observations the black curve of global temperatures since 19, since 1850. And then you have the simulations of the models, including all factors, which count volcano solar forcing and human activities, and on the bottom. And this is including only natural forcing volcano solar forcing orbital forcing but it doesn't change over such short periods. So, we clearly see the difference between the two sets of simulations are many models many evidence now. And this is why we can already, we can say today that the warming since the, I mean, the, the, the, the last century or the middle of the last century cannot be explained today, without putting human activities in in between, in the factors. So, we now know that 100% of climate change is due to human activities, it's not 50% you know we often hear 50% 30% if you ask some many people it's part of climate change no it's full it's 100%. So, for the extremes, it's about the same. It's about the same method I will not detail the methods of course, but it's about the same way we have today's climate and we have a climate in which we, we, we don't have human activities, and we compare these extremes, the extremes we we are witnessing, probably, I don't know, maybe a few days ago, temperature was going above, let's say 40 degrees. Also, it was probably not the case here but just in case. And so you just count the number of time it occurs in the one set of simulation and the other set and you compare the probabilities is very simple. So, in practice it's a little bit more complicated, of course, because there are many technicalities let me show you just an example. This is the 2017 June July temperature anomalies. And it was a very extreme year, probably in Trieste also it was very extreme, all around the Western Mediterranean. And you can see that in the, the, the, oops, no. Yeah, in this temperature series which shows the average temperature over the summer for each of the years starting in 1900 and ending in 2017. In this summer there was some extreme heat waves in the early August, which were called by the way Lucifer, am I correct David because there's a new naming now David is working on naming heat waves actually. So, I hope it's Lucifer because it has been called Lucifer at that time. So what we found in this by trying to to see what what it's very simple we count in a natural world. How frequent this temperature of 39 degrees on average over the, over the the area over the summer would have been frequent would have been probable in a natural world it's 0% actually it's not exactly 0, 0.0 something. In a current world it's a 10 year event has probably today it's even less it's it's sorry more in a 1.5 degree warming world it's 25% and in a two degree warming, it's about what we expect in the middle of the century. In a two degree warming world it would be just a natural it would be just a normal event. One year would be below one be one year would be above so you can take really 2017 try to remember the drought and and all the effects and see what the future will be. So, we have been trying with a group of scientists to do a rapid attribution why rapid because there's a lot there's a lot of attention so I know there's a debate on on that also from different perspectives but it's been an activity of trying to attribute extreme event in near real time. So, how do we do operational how do we go operational on that. So there is a protocol which has been established in this in this paper I'm not going to go in all the details but it's very important. to to to have a very precise criteria on when on when we call that an event or not, and also to. There is a procedure to follow, including evaluating models because we use models and observations, evaluating observations as well. I just want to go in detail. I just want to say that we are now a network world which is called world was our attribution there is already 30 cases. And we use a very kind of, I would not say old fashioned but compared to AI, what can what can what AI can offer it's quite old fashioned way with statistics to compare events in a current climate and counterfactual climate and and see if we can say something about the event. So, yeah, you can visit the website we have done several cases, a number of cases one one case was the three day rainfall amount, which led to actually flooding in the same river. And we found out that the probability, due to human activities of the three day rainfall amount has probably been multiplied by a factor of two. It's not always, we do not always find. We do not always find a clear effect of climate change, this is the case, a recent case on Madagascar. There was a food crisis food crisis last year in Madagascar in the southern regions of Madagascar, and there was claims that we had climate driven event on food effect on food security. Actually, we have looked at the, the, at least the precipitation series and it's what is on the on the right hand side here, and on the region it's absolutely not obvious at all that we have any trend in the precipitation so we do not. We cannot claim like that, that climate change is affecting food, food crisis. It's not climate change is never actually the the only factor in, in, in when we have impacts. And in this case, obviously the, the, the vulnerability of the economy was, was a driving factor, and other factors like COVID, which increased food prices and food chain supply, food supply chain, where, where it was, you know, in question. So we have many questions as dim was presenting we have freaks events or type of freaks events like the North America heatwave. We have, you know, events that we still do not know really how to tackle because we don't yet have the models at that level or we don't have this the observation, long enough or homogeneous enough to, to be able to say something these are elements of storms like wind gusts thunderstorms lightning hail, etc, etc. So we, we need more data we need more modeling efforts. We also have some in some cases conflicting signals between observations and model trends. This is the case of wind storms in the northern Europe actually in the observation it's going down the models going up a little bit. We have a lot of ground events I don't want to detail we had, I don't want to go into detail but we had a very interesting case of frost in April after heat wave in March last year, and I don't want to detail but we can attribute this kind of things. Yeah. So exciter, as dim was mentioning is European project. And for the moment we have overarching goals which is can AI and advanced statistical method help in understanding these events and especially in the detection characterization from impacts you know impacts are often the result of factors, and it's not always easy to model impacts. Just, there are many works around using AI and causality as well is trying to look at chains of causes for for the events is one thing that AI and can do. Also possibilities so we are trying to improve the methods. So we are starting from basic statistics and trying to put machine learning and a new methods to better give more accurate and more useful results hopefully. Thank you. So now the floor is open for question but in present and online. Please don't be shy. Otherwise we, we push you to make question try to be an interactive audience. Any question one. Oh hello. I'm a seismologist so in seismology we talk about early warning systems I mean, is this something that works in the community about abrupt changes in the climate I mean, are we prepared to monitor I mean is there a system at the international level that could basically monitor these abrupt changes and then lead to a sort of an early warning system. So we can start what, what is usually called early warnings in, you know, in operations is weather forecast. First of all, but if we want to have more, more information from longer for longer lead time, then it's, it's what we call seasonal prediction. It's clearly still a challenge. I mean, I mean we cannot today, say that the seasonal prediction for instance for the Pacific heat wave was not. It was not predicting a Pacific heat wave it was a predicting a warm summer. It was not predicting but not such an event. So we do not yet have things like that. However, there is soil moisture, which is a very important factor for heat waves, and for droughts, of course. Soil moisture is something that that tells a story also for maybe not, maybe not two months ahead or three months ahead, but it's an indicator of risks also for heat waves, especially in summer, because of soil moisture feedback with heat. Yeah, I know fully agree so it may be just to add that. So on the sub seasonal to seasonal timescale so that's from a, from a couple of weeks ahead to say the season ahead. There is indeed still a very difficult timescale so whether models have maybe 10 days forecast kill. But on that S to S timescale there is now indeed a lot of activity going also in terms of applying a I to improve predictability. And I think maybe, maybe five years ago we thought that on the monthly timescale, there was essentially no predictability but now we are finding that there is probably more there. And we are hopeful that these AI methods can help us in finding when we have predictability so it's a complex problem but a good exercise for AI. Can I just add one thing. The two extreme heat waves. This one of the last week, at least in Spain, the Portugal and France, and the one in 2019 were predicted with a very high precision one week before. So, well, it's, it's a question of degrees of course you, but it was predicted. So it's the weather forecast are really improving on that. Yes, there is no system in place in the society to deal for to deal with this. I'm just recent, I think last year there was a WMO reports or the World Meteorological Organization and they documented the last 50 years of extreme weather and impacts. And what you see is that damage to society has increased, but fatalities has reduced. So the increase in damage is linked to, you know, we see more extremes. And of course there's also more buildings there's more to be damaged. But the early warning has been on weather timescales have been very effective in reducing fatalities. Yes, so I have two questions from the chat. First one is from Ricardo Barros Lorenzo. I hope I'm pronouncing it right says great presentation. I would like to know if excited eyes currently developing collaborations between weather climate impacts on land models regarding the interaction of these in earth system models. So which topics are being developed. Thank you. And has the second one afterwards, after your answer. Well, I would, I would say we have a full work package, almost a full work package dedicated to, to do that. There is a lot of, there is a lot of activity in trying to detect a strong anomalies in the, especially in the ecosystems, which is one of one of the work packages of excited. So we have a lot of activities in this direction, especially for detection. And one of our work package leaders is in the room. The second question, actually several questions, but it's from Simon Baylats. Simon Baylats. How do you define an extreme event as black swan. Is it just an event that we never observe, or do you also look at the return period of this event, for example, an extreme event with a return period of more than 5,000 years. That's a good question. I guess we never really define it as a black swan. We just, we see events that are far outside of the historical range that we, of which we have data. Right. And the question is, of course, maybe we just don't have enough data. So we're basically the full variability is under sampled. So that is, that is always bit of a question. Is it, is it really new physics that we are seeing, or is it the same physics, which we just haven't wait long enough to see it. So that is a bit of a challenge. I just want to add that in the, in the previous study that you mentioned, WWA study, we made explicit the assumption that it was not actually a black swan, that it was a normal type, normal event, an event which could have happened but with a very low probability. You need this kind of basic assumptions. If you do, if you do want to do statistics, but we are not sure about that. Of course, we are, it's an assumption. It could be a different types of physics actually turned out that we have now within Exida, new new studies coming out, showing that this could be just what we should have expected. Yeah, but maybe, I mean, kind of, in general, if you have an event that is really so far outside of your anything observed before. So you have nothing, your statistics gets very problematic. Right. I am a development student in the system physics section. I want to ask, what is an indicator or weather situation which leads us to compound events. So what is the weather situation that lead us to compound event. What are the situations that lead to compound events. Well, I mean, there's a really different, different type of compound events and there's all range of definitions. One way to define to look at compound events is what we call spatially compounding, which means that you have extremes happening in different regions at the same time. So I talked a bit about these waves in the jet stream. They can cause that also very important El Nino variability and that that creates that can create drought conditions in different parts of the world simultaneously. So that's that's what we call spatially compounding extremes. We also temporarily compounding extreme so that's in the same location where multiple variables come together and then have a high impact on society. So you can think of very warm temperatures, very dry air, maybe, maybe also wind, and then you get a wildfire. So that's this type of compounds like one of the component event is the rainfall runoff and storm searches at a time. And what do you say about this what is the main basically reason about behind this event. I think you were mentioning that if you have a storm system like a tropical cyclone that hits an island or hits a lands, you can create a storm surge. Sometimes you have a lot of rainfall over land, and that is a can generate flood risks. Right, so we've seen that actually multiple of those events in recent years. I'm thinking of the Bahamas, which were flooded in that way. So I think that is a serious risk. And right now, climate scientists are still. Right now I'm treating these risks separately right so they either looking at storm searches, or they're looking at rainfall extremes, but for flooding, you want to have them really both together and model the whole system. But that that is, I think only in the last couple of years starting that type of research. It's really an emerging topic and emerging science. These compound events there are many types of compound events. And we today have some results in some cases like heat and drought, or especially compounding events or floods with different inputs. The IPCC report is really also looking at that, but we only have a few results at the moment, so it's increasing definitely. Climate scientists as well. And as climate scientists were permanently in our everyday life, asked to perform attribution studies. So my father just asked me recently. This heat wave and this thunderstorm that we've had our these is this climate change. And my response normally goes like this. In some way, yeah, this is what this is in line with what we expect in a warmer climate to happen. So, how do you explain to your family. When a weather event is extreme and how do you do this attribution and which words do you choose. We get the same type of questions for instance last week we in the end of last week we had big thunderstorms and or maybe early this week and people asking, is it normal to have a heat wave and a big thunderstorm like that after I say, well, it's, it's quite, you know, we have thunderstorms at the end of heat wave it's not unusual, but what is unusual is the intensity of heat waves at least thunderstorms we it's a bit more complicated so I try to focus you know on what we know. What we know is really heat, heat and cold is really what heat and cold and we also have physics understanding for heavy precipitation it's quite clear now we have emerging signals in many places. So, I just make a step back and say well yes we have all these events but we have to be careful in what we say because otherwise we it could be tempting to go too far and and and give some discredit to science so we have to be rock solid So what we can say really and say that there are many questions that we do for which we don't have yet the answer I think it's very important to, to say it. I mean, I agree with that we have to be of course quite humble. And there's some extremes we don't really understand yet right and we don't. We cannot find a climate change detectable signal or attributable signal, but then still yeah in my view I think the physics are most important so for example for convective rainfall events. It's very hard to detect the signal it's very local so you need super high resolution observational data to be able to detect that to model that is doing that is super challenging. But that of course that doesn't mean that we have good physical IDs, why convective rainfall events would quite strongly increase with warming. The fact that we don't have kind of, we cannot kind of reject our no hypothesis to call it this way. And that there are no changes that we cannot reject it just because we like the data, but that doesn't mean that we should be wary that those type of events are likely to increase. I think I'm, I guess I'm more like we should be very close to the physics, rather than maybe statistics. You just add an example, we take care, we had long discussions with on heavy precipitation in the Mediterranean region, I think Erica remembers that it's a complex region because in the northwestern part of the Mediterranean sea we clearly see an increase of heavy precipitation intensity number, etc. But for the future, they are conflicting signals, the whole Mediterranean is trying in terms of lack of precipitation lack of total amount of precipitation, and also we have more heavy precipitation. So for the northern parts of the Mediterranean, we have a winning process which is the increase in heavy precipitation, but for the southern parts of Mediterranean, it's not clear. Probably what happens is that when we have a precipitation event, it's going to be more intense, but we will have less, less favorable conditions, more anti cyclonic conditions than what we have. So there are, you know, different causes which can cancel each other. So it's, it's, it's a tricky region, and especially in North and Southern Italy may have different signals in the future. Okay, so we have many questions over zoom. There are two people raised their ends. So maybe we could start with them. I think Attish Daborka can start. Okay, hi. Thank you for a very nice colloquium. Okay, my question is perhaps a little bit more like a citizen, I'm a theoretical physicist, not a climate science question. But you talked about sort of modeling and attribution of the physical causes of extreme weather conditions as being, and there you see a very clear pattern, you can sort of demonstrate that it is caused by human activity. In this report, do they also consider, I mean just to convince the policy makers and the general population of the seriousness of the problem. I have people studied also the consequences like the economic costs, which is not a physics question but other models to show that let's say fire, many fire wildfires or floods have increased over the last 10 years. The cost to national economy is to all these kinds of extreme weather conditions. I don't know this is probably to brought a question for to ask for climate scientists. Well, maybe starting with the flood you have these. For example, Munich re like one of the big re insurance company they, they keep these disaster databases, and then you just do see that flooding and damage from flooding has been quite strongly increasing. And the question of course is, you know, is that purely because due to exposure so we're building more and we're building more in full normal places, right like in the floodplain of rivers. Or is this really a climate change signal so I think that is not really certainly not a settled questions. So maybe with, maybe with wildfires maybe Robert correct me if I'm wrong. I think in some regions there are good studies showing that wildfires have been increasing and that this is can be linked to global warming. So then I'm thinking of Western US, and maybe Australian. There are quite a few regions like that. I can't remember exactly I think West, West US, and the main, and which, yes, well yes in some cases it's not fully, but fires are complex because the occurrence of a fire, or even the occurrence of many fires can be due to not to only climate change, but also to the management of forest and the monitoring system. If you don't have firemen, then fire will propagate or if the surface is too, too big to, to go on the place, then it will be more easily propagate. So fires are complex so what we usually talk about when we talk about climate hazard is the fire weather risk or fire weather, fire weather, we talk about fire weather. And we have indices for that. And we have attribution also studies but it's on fire weather. It's really the, the weather component of the risk. And indeed, as Erica was saying, we do have attribution, a few attribution studies for Australia and also for slightly for all the regions like Western America, but not so much. But however, in the future, it's quite clear. We have a quite clear signal. I can remember our table, you know, fire risk and fire weather risk is in many regions of the world. It's increasing really in many, many regions of the world. It includes the Amazon region includes not not. It includes the West of America, Central America, the Mediterranean, Southern Africa and Northern Mediterranean and Southern Australia if I remember well, no everything everywhere in Australia, if I remember well. So there are many, many regions for which we expect this risk to increase. But maybe just to complement the question about this so that there is also this information in the IPCC in the working group two and three. So there are information on the risk. So how much this is causing risk or increasing the risk of some kind. And also there are also study assessment or study on the, on the, which are the possible solution and how much you save if you mitigate I mean this kind of study is there in the other working group, not in the physical science basing by the other two. So we, I mean there are scientists that also assess this part. Yeah. Thank you. Thank you for the great talks. My name is Khalil and I'm doing the PhD at the University of Leipzig with the project. I have a question about definitions of the extreme events because sometimes or for some types of extremes. There are different definitions. For example, heat waves, it could be two days or more three days or five days or more of normal heat. I don't know your opinion, whether it is desirable to have common definitions, or we could continue like this because depends on the working groups depends on, and these different definitions could also have a big influence on the findings and on the results. Yes, of course, it's always nice to have a common index to see what's going on around the world in every region. However, the problem with heat wave in particular is that the impacts are absolutely not the same in different parts of the world. For some parts of the world, heat is already included in everyday life, and in other parts it's not. So, even if I'm talking on my country, France, we don't even have the same definition in each in the north and in the south and in different regions. There is a broad definition, but if we really want to make it useful and operational because we need operations, you know, when there is a heat wave trigger, there's a lot of things happening, you know, hospitals are mobilized, there are networks for lonely people are mobilized and you know that the heat plants so we have to it's a very operational thing so it has to be really suited to the region. That's also why we see differences. I agree. At some point scientists, you know, not operational people but scientists could, and they do actually do studies with the same index everywhere to see what what the differences are but I'm not sure for the operational consideration that this will be useful to have a uniform one. Oh, yeah, sorry. Hi, this is Anthony from my STP. So, over the recent years, we have certainly witnessed the ability of artificial intelligence to produce very good predictors that from data can produce predictions. But often these are very opaque in their functioning. So it's very difficult to explain why a certain decision or prediction has been made. So they are far from being explainable. So in your presentation you mentioned that you want to refer to methods from explainable AI. So what kind of tools you think from the AI toolbox can be more useful to deliver this promise. Thanks. Yeah, great question. So, I fully agree that I guess still a couple of years ago, at least for the climate community, the AI methods that were out there were mostly black boxes with all the problems that those have so and now you see the first studies that use explainable AI approaches, right. So, I mean, for example, I mean, we've just finished a study where we I mean this is aiming at prediction of warm summer temperatures over over Europe. Right, so this is one month ahead. We want to know the temperature over a region, Germany, France, the Netherlands, roughly this region. And you can train an AI methods to forecast that. But then we want to learn like what's what is really giving this predictability and so the sources of prediction. There are tools out there, for example, Shapley values. I can explain much more about it here is the lead author of the study. But what we're finding is that activity in the Pacific sea surface temperatures over the Pacific are very important. So this way we're learning. Well, we are we're getting more confidence in. Okay, we are. We're getting a forecast we get some skill that is nice, but we also understand the physics better. We know that the Pacific and things like El Nino linear variability are often very important for predictions. For example, I think where an explainable AI methods gives us better in gives us better skill and gives us better insight into the physical processes. So those those type of studies, we want to do more. I'm not a real expert on explainable AI. So there's some people in the room, who can really tell more about all the options that are out there. So there are several questions in the chat but I think someone raised their hand a while ago and the hand disappeared. So if that someone wants to ask question then they should go for it now. Otherwise, there's a question from Ivan Geroto, maybe even wants to ask him it, ask it himself. Climate modeling is probably one of the fields of science which is more affected by the power crisis of standard chips and the consequence software crisis meaning compute power is increasing but models are not able to exploit it efficiently. I am wondering if the introduction of AI would even increase this problem or reduce the problem of current models, and I would like to ask whether the community is planning major investments to reduce this problem in the years to come. I don't understand it. Yeah, no, I'm not sure I understand the precise question. What the problem that it refers to, if the AI is going to help with the crisis of the need of always more and more computer power and chips and more powerful or is going to make it worse. I mean, typically, I think, typically AI or machine learning tools are way more efficient than climate models. Climate models are very intensive to run. Right, so if we can, of part of the climate models can maybe use AI tools to the climate models become better or that they can become more efficient. So I think they're rather helpful rather than creating an additional problem. Yeah, I can maybe add on this I think it's an open question actually because, yes, for sure we maybe parts of models parameterizations and things can be probably modeled in a more efficient way. The architecture also of the codes have to evolve. So there's a mutation of codes at the moment. But the problem with AI is that we don't fully and unless we have explainable or I don't know what type of AI. We are not sure you know if if we are projecting climate change, the conditions for training AI. So if it's a trained AI, well, are not the same in the future so whether this will work or not will will still be a question mark so unless we really understand what we are doing. So, I'm not the best person to answer I must say this question but I know it's an open question because we don't know. We had also a problem with the AI is that we can be tempted to generate lots of new data and lots of new data because we can do it faster and quick so this is also adding up. It's only just about the computation but it's also about the storage of this data that it becomes a critical point also because at the end somewhere as to analyze and then I mean we cannot communicate on all the data that we generate so we end up downgrading the data that we have generated high resolution or focus only on a part of the data sets and everything is going so fast that we end up also erasing data that we have just generated. So all of this is also an impact in terms of CO2 emissions in our community. I want to add one thing so maybe one hope is that at least for very high resolution computational model like the one that we are running as you know even the hope is that we can build some emulator based on AI this is what we are trying within this project. The emulator should save us some computer power and some CPU hours, but I mean we are still not there we have to see if we can build this emulator but the aim of the emulator is to reduce the computer power that is enormous at the moment for this high resolution one. And so another question was, what does attribution results mean for policy making. Yeah, do you understand. Do I want to start. I mean we have we had a bit of discussion about that how you are useful attribution studies are I think ultimately, maybe most stakeholders or maybe decision makers in certain sectors, whether it's when you know it can be policy it can be the energy sector the water sector. What I'm learning when I'm talking to them. They would like to know the chance and the probability of events happening now, and maybe in 10 years from now. And that is, that is not something that attribution per se answers right it's attribution science touches upon that we also dealing with probabilities. And typically we're looking of course attribution means what has historic climate change. How has that changed those probabilities and it's a, I would say a slightly different question. So maybe many decision makers are maybe more interested in slightly different type of stuff. I just want to add on this, of course very important question but it's not extremely easy to answer because if you take just the result of attribution as well this event had more chance to occur than it would have had without climate change or without human activities. It's not an actionable result it's not an operational result per se. But it's if if I always put a second message saying, therefore, we should not use only observation from the past to design our risk plans. Because in many areas in many sectors, we still are using people are still using observations from 100 years ago, or from 50 years ago to calculate risks, and they are not using this information that that models that that climate has changed already. So, per se, the message is not extremely actionable or operational. But in the end, if we can, if this gives the message that we have to that risk risk plans must include now climate change for adaptation. I think it is useful. This is something that people here in the policymaking area. So it's it's it's we are probably going to see changes in the future we will never be able to say this is they have changed their way of process of doing because of attribution of course we cannot attribute the way of the change in the way of doing it would be too complicated. But I think I think there is some impact of attribution which may which may not be fully direct, but also indirect also, you know, doing attribution of something that you see from your eyes that we where you see impacts is more talkative. It's telling it's telling a story. So it's not direct impact I would say but maybe indirect later. I'm also a diploma student here. My question with regards to artificial neural network. So, in some of the other studies, they use multi layer perceptron in their training with a prediction studies that they do. My question is that, how can we know that with the number of that with the number of hidden layers that we are using in that method, we are able to predict what we should predict. She's asking, how can we be sure that the architect that we choose and in particular the number of the hidden layer is the correct one for the problem that we have on neural networks. That's what they understood. We can ask Gustav. My answer would be with a little bit of experience. Usually you have to have a number of parameter, the dimensionality of the parameter of the network has to be comparable to the training data set that you have to avoid overfitting or underfitting. I mean, there are for sure technique for choosing the best architecture, but for sure you have to try and see how the model is learning. Right. On the overfitting problem. I think there is now a standard way to go is to run with three data period one data period that you never look at. And two data period with training period and testing period, for which you can test the ideas, but I don't know much more theoretical background than just doing this. Maybe my colleagues, I don't know if you want to interview Gustav, would you have any thoughts on that? So Gustav comes first from the University of Valencia. I guess that the question was how to select the best architecture for the for our network or whatever arbitrary problem. Is that the question so the magic number. It is five. There is not a number. Actually, most of the most of the times you have to cross validate it with different data set in different portions right for training validation and then the test set where you report the results and you don't you have not said it before. And then you try out different different numbers, not only layers but also neurons and learning rate and many different hyper parameters that you have to choose. So I said that there are also theorems that can tell you more or less what are the optimal numbers in theory, but then the practice is more, you know, the reality is more more difficult. Hi, very good talk. I've got a very stupid perhaps naive question but a simple remark just there. A thousand years ago, we fantasized what would be the weather forecast in 2050 in France so we had the TV presenter, and she was showing that map, saying 42 degrees in Paris 44 degrees in Bordeaux. There was a forecast made for August 2050 last week. And we had that map in France. So I'm making what's coded in tools and metal France. So my question is quite simple given what you talk about the freak events as well. Was the IPCC, or the climate scientist somehow to optimistic in the forecast are things happening faster than expected. Are we far off from those, let's say, very bad it waves happening earlier than we expected. There are different ways of answering this. So you're referring. Yeah, I know. And there was a comparison also with the 2019 heat wave for which we had these four. We observed this 42 degrees in Paris and I don't know in Bordeaux, but anyway, one thing is a regional answer that there is still. There is still something we don't know. We don't understand the evolution of temperature of extreme temperatures in summer in Western Europe and in particular in France but not only in France also in the Netherlands. And in some parts of Germany. It's going much faster actually than what the models predict, they basically two times faster in terms of changing both, I think, magnitude and frequency frequency I would not say factor of two but at least in intensity typically we calculated the intensity that we observe in June for the for the maximum temperature in June each year and it's it's basically four to five degrees since the early 20th century in the observations and in the models is two degrees. So, in July it's a little bit less, but we do see differences that we do not fully understand. There are probably, there are certainly things with clouds and aerosols, but maybe some feedback that are not explained so this is regional however this is we do not find such big discrepancies in other regions or maybe not it's not generalized. We also have these events, these events for which I mean scientists when when when the Pacific North American heat wave came. There were plots all over the place on Twitter, showing that models don't fit like that it's it's just it's way above, way above what statistic would predict in terms of extremes. So, in that case, we need a bit more time to to understand, we have a few studies currently showing that we could have predicted that at least some models have such type of events in the world. And maybe we should focus on on this kind of events, record shattering events, a bit more to understand the physics behind. It's something that was left a little bit behind because I don't know why but it and it's becoming more in focus so probably the P&A heat wave was in a sense predictable. Maybe there was no focus on such type of events so so far. Yeah, maybe just to shortly add. I think it's worrying that that the observations said that the temperatures are rising faster in Western Europe than we would expect from models. So that's still an unresolved question. So clouds were mentioned by Robert that that is, that is always something of course we look into in climate models we know that we do not really resolve clouds explicitly and it's all parameterized. So we knew we know that clouds are a bit of a weakness in the climate models. But yeah, I'm, I'm myself looking a lot into dynamics of the jet stream and these type of where we do see changes and looking in whether the models. See similar things so we do see the type of actually, we do see similar type of weakening that I showed in my presentation in models but only if we add much more see you to the system. So by the end of the century, basically. So the question is them. Well, are we looking at that that weakening of the jet stream is that mostly due to natural variability large decadal cycles, or are maybe the models not sensitive enough, right and that is, that is still ongoing debate. So maybe to add also top of these and answer that physicists will like it's the climate models. We have to somehow had some viscosity additional viscosity to make the model run. This is something to understand of course as physicists but of course, what does it means it is the viscosity introduce a different time scale. And if it's an hybrid is considered is not related to that of the air, it's slower. So it means that probably you are good in capturing slower motion but the, the high fluctuations you have some dumping there that forbid you to observe this, this sharp extreme This is a bit the intuition that we have as a physicist working with turbulent flows but Well, we can verify somehow that the observations say another story but then it's difficult because this model of 100 parameters to really find the right button to press and say, ah, it's this one and we should correct this. And just to add also the IPCC report the last one in last year. We have this kind of low likelihood impact events. So IPCC is still willing to describe potential events that are maybe not predicted by models but that would be physically plus plausible. So it's, it's, I think it's important. Although there is such a wide range of possibilities, it's really hard to say, I would tend to think that precisely the critical years are a bit now because we clearly see the signals now. And we have some unprecedented style type of stuffs and probably more surprises will come. And we have to take lessons from them. For for instance in my personal communication. I would say to policymakers that 50 degrees in France is possible. Yes, it's possible. Because many people asked this question, could that happen. Yes, can happen. We don't know when exactly. So it's not exactly very useful, I'm saying, but it can happen. It can happen in a few years already. Yeah. Sorry, 45. Yeah. Yes. Last question from the online. Yeah. So what would be your definition. It's again about definition of extreme events but what would be your definition of extreme events in the literature extreme events can be very location dependent. For example, extreme rainfall in France is not the same as in Colombia. Should we consider only the rarity of occurrences should we also include the potential damage of these events on our society to define them, which would imply for some places to have much less rare even semi annual extreme events. Maybe I can I can try to take this one first. I think it's a it's a it's a good question is also a semantic question but the approach depends what we want to do actually if we want to do if we want to study a basic science and the relation between I don't know. Processes, of course we might want to to be interested only in rare events and trying to see to explore what but rare doesn't mean that it's extreme. I mean, having the mean climate on one year on one day is extremely rare. I mean, climate over Italy, you can you can see it's it's rare. So rare doesn't mean extreme in terms of impacts and at least for attribution what we have done so far because there are so many extreme events. What we have done is extremes which have impacts extremes are defined as having impacts we are trying to define to find the meteorological situation that drives to high impacts and it's quite clear in many cases. But I mean the question is good. Yeah, I mean, I guess. Often societies are quite adapted to what the type of extremes they've experienced in the past, right. So of course this can be very. When we talk about rainfall this can be much more intense in the tropics as compared to here in Europe. So therefore, you know, a very basic like once in a hundred year time type of event is likely to be quite damaging in many regions so that is, I think still a very useful type of quantification that can be used across regions. I think we have to close it now. So first of all, we thank once again our speaker today. We want to thank the online people and sorry if we couldn't take this so many questions that we got it was very nice to see all the attendees so active also the one in in presence. So we can close it here the webinar. Thanks everybody, whatever you are and have a nice evening morning afternoon and see you for the next events of this series. For the present people don't run away so we will have a group photo now then you are all free to go except the student that can stay just 10 minutes longer if the speaker will resist. I don't want to ask any question you want without any revision of all the people. So now I'll come down and that's the photo and thank you very much right. Sorry. Atish you want to say something. No, no, no, just wanted to say thank you. Okay, thank you for the thank you so much. Thank you. Yeah, so come down for the for the photo first.