 Good morning, good evening, good afternoon from wherever you're joining us from. Welcome to day two of the IBM Research Horizons event focused on the future of climate. I'm George Silevsky, I'm a research scientist, I'm also the manager of the Think Lab here at Yorktown Heights at our research headquarters. We have a great agenda for you with our leading scientists and our climate initiative and we're just so happy and delighted that you're joining us. Before we start I just have one housekeeping item. If you're joining us via the platform you should see a chat to your right of the video. Please use it, our scientists are online, they're eager to interact with you and to answer your questions and even for the pre-recorded sessions we'll have our speakers on live immediately following the session to answer your questions live. So please use the chat. If you found this video just through YouTube you can also join the chat by clicking on the link in the description that will take you to a registration page. You can register for the event and then you can also participate in the chat. So with that out of the way the theme for today is adapting to climate change and the role of research and technology can play in helping businesses and organizations adapt to our changing climate. I can't think of a better way to kick it off than to have Dr. Hendrick Hamann who is our chief scientist in our future of climate strategy and a distinguished research staff member here at the Yorktown Heights Laboratory. Hendrick how are you? Good to see you. I'm very good thanks so much George. So please take it away. All right so yeah also for me good morning good afternoon and good evening wherever you are. My name is Hendrick Hamann. I'm the chief scientist for the future of climate here in IBM Research and yes I have the privilege to kick off our second day of our horizon event of the future of climate. Marcel please push out the presentation if you could. Yeah let me start sharing some interesting but perhaps worrisome data which I think says a lot about the future of climate. Although this is actually data from the past you will see similar data later in the IBM presentations not because we like to repeat but because this data tells a very complex story. What is shown here are the observed temperature distributions during summer for each decade from 1950s to 2010 where the distributions are normalized by the local standard deviations observed way back in the 50s. So you might wonder what the magnitude of a standard deviation is that's around three degrees C but of course it's different in different locations. So what is this data telling us. Well first it tells us the climate between 1950s and 1980s did not change much. The distributions stayed the same. It became neither warmer nor colder it became neither more extreme nor less extreme. It also tells us that starting in the 1980s there are suddenly drastic changes which have accelerated. It tells us that it is progressively getting hotter but not only that. It tells us that the extremes are becoming much more pronounced right. The distributions are widening drastically and that's certainly a big concern. But what does that really mean right. So for example during the 50s the average summer heat in New York City was 24 degrees C with three or four days above 30. Today the average summer heat in New York City is almost one standard deviation higher and we have roughly 20 days of heat above 30 and three to four days above 35 which used to be 30. Now while these trends are striking the data does certainly not tell the whole story what climate change really means right. The impact of climate change will be different in different locations. There will be different at different times. It will depend on how much you prepare how vulnerable you are how much your business is exposed to such risk how much resources you can spend on preparing how much you how much time you have to adapt and many many other other factors. Now yesterday on day one of our horizon event we asked the question how information technology AI machine learning can help with mitigating the impact of climate change slowing or even reversing the trends which we've seen on the previous shot. So whether this is by removing carbon through materials innovation and carbon capture whether this is by improving the carbon performance of business operations such as order fulfillment cloud computing business travel. Today we'll focus on the adoption to climate to the impact of climate change. And what is important is that this is not a simple choice between mitigation or adoption. Clearly we must do both and that is why we do research in both areas in mitigation and adoption. In fact I want to spend a few minutes discussing the fact that the technology opportunities to advance both the mitigation and the adoption are driven fundamentally by the same underlying trend which will provide context for today's discussion but also to yesterday's discussion. Now what is the underlying trend enabling us with mitigation and adoption in technology? Well the trend is that the data and information we have today about our environment our physical world our climate is richer deeper and bigger than ever and that is because of continuous technology innovations which we've all witnessed at a minimum as consumers data from the environment data from the internet of things from sensor networks from satellites from drones from cell phones is growing exponentially by something like 40 exabytes 10 to the 18 bytes per month so vast growth of data and information. Now what does this trend then do and why does it matter to the future of climate? Well this trend digitizes our physical world our environment our climate and that is really the foundation to model analyze apply machine learning and AI to help to make better decisions and improvements right how to sequester carbon in the soil how to prevent methane leaks or how to forecast the next drought and predict its economic impact on the community. It may be also important to remind you that digitization of a sector or industry is nothing new right other sectors in industries have been almost completely digitized think about media industry the finance social networks so why am I mentioning this? Well because history has shown that digitization enables through AI machine learning cloud computing massive efficiency improvements and that is really the opportunity we're talking about in this conference. Now the environment our climate the physical world is arguably one of the most difficult ones to to digitize it actually may be one of the last frontiers of digitization which holds the promise but also the opportunity to discover the required knowledge for all the information which is growing every day so we can make better decisions on how to battle climate change most efficiently. Let's have a look at what we heard yesterday and what we will hear today. If you think about the role of technology there are many however one of the key benefits of cloud computing AI machine learning and some of the techniques we're talking about today is that we can accelerate the discovery of finding the best solutions for battling climate change whether this is adapting or whether this is mitigation. To do this you follow a process and that's very much the same process you use for scientific discovery right you must first learn from all the information which is out there then you ask the questions you hypothesize you model your test and then you go back around so if you think about what you heard yesterday you heard how we can learn faster using AI to find new materials for carbon capture you heard how we can model much faster carbon footprints and carbon performance and finally you heard how we get answers to our questions through testing the solutions in the marketplace through applications so today we follow a very similar pattern we start with core technologies which allows us to accelerate the learning from all the data with our geospatial capabilities with our modeling frameworks for climate risk and impact we then discuss AI machine learning approaches to accelerate climate impact modeling and finally we show how we apply these insights and test these solutions in the real world and then become full circle around now let me stop right here with this great quote from Nelson Mandela which is I think a stark reminder of what we should not forget there's really no better way of saying this please read it and and don't forget and this quote also provides a great transition to our keynote speaker which we're all very excited because it's not only an outstanding individual and leader but for sure he will fire us up with his passion and George will introduce him next and back to you George thanks Hendrik and thanks for a great talk and as we go through the day especially later on in the day we're going to hear more and more about those technologies that you're referring to in that in that kickoff but as Hendrik said we have an outstanding keynote address coming up you know this event is focused of course on technology and innovations and how technology can can you know help us mitigate and adapt to climate change but I think we're going to all agree the real ultimate purpose is to positively impact people's lives especially those that are most vulnerable to climate change and so this larger purpose in mind it's an absolute delight to introduce our keynote speaker Mr Ibrahima Cheek Jong who is currently the UN assistant secretary general and director general of the African risk capacity group this group was formed by the African Union and has delivered over $720 million of insurance coverage to over 72 million people this tremendous tremendous impact on the continent prior to his current role he's had several leadership roles including and consulting at the African consulting and trading group banking at BNP Paribas and in governments where he serves as a special advisor to the president of the Republic of Semicol so we're delighted to have you Mr Jong it's great to see you and we can't wait to hear about all the wonderful work you're doing with the ARC. Thank you very much George for that warm introduction and also a big thank you to Handrick for making my life easier and making the case by the gravity of the situation when it comes to extreme weather let me before I bring you the voice from Africa where I'm based right now in Senegal let me express my gratitude to IBM for helping us to make the case from Africa and bringing an African perspective but more importantly a human perspective of climate change it is extremely uplifting to see that we can act and build public and private sector partnership with IBM so that we can combine our forces to help the communities across the world let me make four important points number one just walk you through some of the problem which I address particularly as far as Africa is concerned when it comes to climate change and then second the mandates have been given to my organization by the African Union to address those concerns but our approach to working with a greater partnership to make sure we can provide some practical solution to Africa and as George said earlier we've been quite impactful but the impact goes beyond just monetary impact I'd like to walk you through also some of the great value addition that we provided to Africa so far and obviously the most important part of the conversation as George said earlier is to see going forward how can we collaborate with organizations like IBM and other technological companies to make sure we innovate the way we act at profile risk on the continent I think on the problem if there was any doubt that we're dealing with extreme weather situation you just have to watch the flooding situation in the US the flooding in Germany and recently also in Africa just to demonstrate not only the earth is warming up but it's creating tremendous challenges to many countries including developed countries I think second what is not obviously hopeful is the fact that I think I lost can you hear me yes yes we can absolutely we can hear you I think what is not absolutely optimistic is the fact that as George as Henry said earlier unfortunately the earth is actually warming up so we're not out of the world yet and the question is how do we anticipate and making sure before this disaster occur we can take the necessary measures as well now if I bring it down to Africa the problem I just talked about are very specific problems that we're dealing with and that is when they come to drought in many parts of Africa particularly in the Sahel region drought causes tremendous problem in terms of food security but also in some cases pushing communities to get rid of their assets so they can survive and take care of their families and be able to keep them in school in some cases also be able to generate some incomes as well we're also dealing with tropical cyclone which is a major problem in Madagascar in Mozambique again was a similar human aspect across Africa and obviously flooding remains a major challenge in Africa because of the climate change situation so because of all of that we've noticed that what doesn't get talked about quite often in this conversation is the human face of climate change even though Africa contribute less to the pollution and the warming up of the earth but yet we happen to be suffering the most when it comes to climate change so the big question is what is Africa doing about what I just talked about in 2012 the African union has decided to lead by creating an institution called the African race capacity entirely dedicated in making Africa more resilient in dealing with the impact of climate change the way we do that is what I call the 2P and the 1R the 1P is about helping countries to actually profile their risk with a hope that visibility allows government to decide whether to return some of the risk or to transfer the risk to the insurance market before it actually occurs because it may be too late when it happens the second P is the preparation of the government particularly different government services that are involved in responding to disasters and oftentimes a lack of any coordination at the level tend to affect the life of many Africans sort of going forward so therefore coordination is key and last but not least is the response and that is that what are some of the financial resources that can be mobilized through insurance by transferring the risk that allows government in addition to the national budget to have access to additional resources in a speedy manner and this is in the context of Africa extremely critical because when there is a drought while you wait for the international community to come to your rescue people are dying lives are affected and livelihoods are also affected as well so therefore it's really important to show that face in fact yesterday I spent the entire day visiting with the communities that have benefited from our support in the past and it was very heartwarming it was inspiring it was also extremely emotional just to see how some of these communities if it did not come with the support either they would not be surviving or the families be affected as well so the question is what has been our impact since we got started because it shows normally the potential in terms of our intervention but also demonstrated we actually the concept has been tested there's a proof of concept and we can scale up across Africa going forward I think number one impact is the value addition that we provide in convincing African governments in the importance of early warning systems disaster does not discriminate and if you wait till it happens it may be too late so I think slowly and surely we're building the culture of disastrous management of contingency planning so government can actually be prepared before they actually it happens I think the second impact I see is the way we are empowering African governments to be on the leadership front in dealing with disasters in other words we can provide all the support necessary but ultimately government has to decide how much of the risk they would like to transfer but also how much they are willing to coordinate themselves so they can get the necessary capacities to react to the needs of the communities I think the third impact that we have and that is our ability to model the risk I mentioned earlier would be known for quite some time in modeling the drought risk in West Africa and other parts of Africa to a model that we designed called Africa risk view which allows us to use some data and anticipate two three years down the road when drought is going to happen based on the rainfall in a particular country and assess the impact it has on the communities and be able to anticipate as of today we are we've launched a tropical cycling product we're in the middle of designing a product in flooding and we have some great good conversations with IBM to see how we can leverage and you know how to be able to actually add as many protocols possible so we can respond to the demand of our member states I think the issue of number four is modeling in today's world any disaster extreme water can be modeled and I think I'm preaching to the to the choir here knowing that the majority of the audience as scientists you believe in the power of modeling I think what is important for us beyond just the academic exercise is the ability of how modeling can help our member state to anticipate transfer the risk and use that information so they can make some sound public policy decision as well and last but not least is the financial payout that we provide as I said earlier when the country decided to transfer the risk through us what they also do is the possibility if that disaster actually happened and triggered they can get a payout in the early stage which allows them within two three weeks of the disaster happening till they get money which they can use to support the communities that are vulnerable across Africa and I can assure you based on the visit that I did yesterday talking to some poor communities in Senegal it saves lives it saves livelihood so this is not a job when it comes to climate change and the African continent in our job is making sure we protect them now I'm mindful that the scientists I'm talking to here attach a great deal of importance to mitigation and adaptation but Africa also is dealt with the issue of resilience and what I have just described so far is to make Africa more resilient while we're looking at investing into adaptation and mitigation as well let me wrap up by saying a couple of areas where I see we can work together with institutions like IBM and others number one you obviously are the leader when it comes to using technology and modeling and I am confident with the discussion that we are having we can certainly leverage a new track record and know how and using technology to actually anticipate and predict some natural hazard that information as I said earlier in the case of Africa is extremely critical in saving lives but also making some sound decisions as well I think second area where I see some challenges as far as concern a modeling has to be accurate it has to be in a way that when we go and talk to the government there is no ambiguity whatsoever of what's going to happen to us from now because the type of disasters we're dealing with they're not as frequent as you will expect them in Africa and that's the reason why we can anticipate and be able to actually model them so we certainly will look forward to looking at your modeling capabilities so we can model the disasters and be able to have a timely response in due course I think it's very important that we use your know how given that in many African countries there's so many competing needs in other sectors whereby to get them at the focus on the impact of climate change it's a bit daunting in some cases using a lot of advocacy to convince them to take more seriously but I believe with what you bring in with technology we could certainly be able to use that use of evidence to demonstrate that if you're dealing with this type of disasters when you wait till it happens maybe too late so convincing by evidence is really something that's important I think to wrap up as going forward we are absolutely committed in addition to our partnership with the public sector to work with the private sector to develop a different type of PPP in this case you being the private sector and the government that we support being the public sector but we also come in as a continental organization so we can use our mandate and you know how to be able to support Africa we're very much committed to exploring that opportunity as well I think second most of our donors that are supporting us provide what we call partial support but what this organization needs to grow is also technical partnership we need to build our capacities of our R&D department to make sure it could model the risk so any obviously offering that you have and making sure that we can build our technical capability it will make us credible it will make us much stronger to support our member states we are certainly uh of takers I think last two points to say is that we are also interested in building our own capabilities in co-developing product so it's not just a matter of transferring technology I'm also interested in the transfer of know-how so if you can actually co-develop some of this product it allows us to actually bring in the African perspective the reality of Africa the ecosystem of the continent so that the product we design can also help us to further respond to the African countries and I think last but not least there's so many developed institutional organizations such as the World Bank and other intellectuals they are doing similar exercise it is about time for us to join our forces so that we can have a menu of solution matched to the demand of Africa so we can be helpful so my takeaway message is the following what I just described if there's one thing to remember is that what we do save lives and what we do save the livelihood of Africans so therefore ultimately unless we can demonstrate that all the work we do can lead to make Africa more resilient or take life we're not being successful in our mission I am committed to doing that I have no more hair because of that and I have no doubt together we can join forces and bring the support that Africa needs so that Africa has to be more resilient so thank you very much and over to George. Thank you Mr. Jong for a wonderful and an inspiring inspiring talk you know if we have time for a couple of questions I listen to your talk that a couple things came to mind one you know you alluded to technological innovation financial innovation and modeling risk and then a sort of political I don't know about innovations but you know empowering local governments to act I'm just curious from your you know leadership position which of those three you know create the largest barrier that you see is the most you know difficult to overcome because they're very different and they're all critical the success of your mission so I'm just kind of curious what keeps you up when it comes to those three there's three areas George the short answer is all of them keep me up now I think that's just they really go together if you're looking for example technology in Africa one of the biggest challenges obviously the fact that the access to the internet and other infrastructure in the IT system makes it quite challenging to come up with something that's like it's suitable to the African model and I think second on the modeling side the using active technology or modeling to make public policy decision it's not something embedded in many African government way of making decision so therefore we have a lot of work to do to advocacy by demonstrating to existing trusted actually concept whereby what we do can save life and use that to convince African governments to believe in the use of technology to model the risk and eventually make the right decisions to go forward but obviously George all of that has to be combined with a strong political commitment so one of my big job is to really convince African heads of states that they become champions about the power of disastrous management yeah yeah yeah I imagine that that all of those quite quite challenging but based on the impact your organization has had already I have no doubt that that coupled with your leadership that will see you know even more impact ahead I was really uh I love the way you described your you know your thoughts on collaboration not just technology transfer but know how transfer and working you know hand in hand co-developing technology so that you not only get a piece of technology but you know the scientists and the researchers in your organization and organizations throughout the continent actually gain the know how as well and so I'm just curious if you have additional thoughts on that model of collaboration also you know for those technological institutions out there IBM and others you know how to get involved and how to how to you know kick off a collaboration like that yeah but I think the number one message is important George to mention and that is that we have the continental African mandates to do what I just described but the solution do not exist all of them act in wooden organizations so we have to partner with others to actively fulfill that mandate so that's the number one statement to make I think second to IBM out of technological companies it's not just matter just using what you offer the question is how do we adapt them to the African reality so we can actually respond to the needs that I described earlier so I think what the discussion we have with IBM right now it's really the right approach we'll be having a discussion with them understanding our ecosystem understanding our demand and our need with the hope from that exercise we can lead to co-development and eventually building our capacity so that there is a legacy that's actually comes out of the conversation we're having right now we're very much open obviously with other technological companies to do a similar exercise but I think co-development is really the key because it build ownership it build legacy it build leadership and that's basically what we say yeah that's wonderful to hear I think quite like-minded in that way I mean we really strive in our partnerships it's not about just building something and throwing it over a fence it's about building it together transferring the the know-how and the technology and then bringing the the local expertise to inform to inform the technology itself well I think we're out of time but thank you so much for your presentation thank you so much for your leadership in this area we're looking forward to working together and also to hearing about the great things to come from ARC in the future thank you very much George I wish you a great conference and look forward to the outcome thank you thank you thank you so much okay uh so we're now going to move on to the technical sessions where we're going to focus on technologies developed by IBM and IBM scientists again focus on climate adaptation the talks are pre-recorded but the scientists are on live to answer your questions in the chat and then following each presentation they'll be on live to do a live Q&A something to keep in mind our first speakers are Ann Jones who's a research scientist in our UK laboratory and Johannes Schmouda who's a research scientist in our Yorktown Heights laboratory and the title of their presentation is geospatial analytics for climate impact so with that we'll let them let them begin hello everyone and thanks for joining us my colleague Johannes Schmouda and I are here to discuss and demonstrate how geospatial big data analytics and AI enhanced impact modeling can address the challenges of climate change adaptation by quantifying climate risk at scale so the impacts of climate change are already being experienced across the globe climate change causes more extreme weather and changing both the severity and the frequency of extreme events as the global climate warms temperatures we consider extreme are occurring more frequently and temperature records are being broken heavy rainfall events are becoming more frequent and more severe rising sea temperatures are increasing intensity of tropical storms rising sea levels are increasing the risk of coastal flooding so the impacts of climate change are the downstream consequences to communities infrastructure agriculture and other systems and climate hazards are the weather and climate events heat waves floods droughts windstorms wildfires which cause these impacts climate risk is the potential for losses due to these hazards so from coastal flooding disabling critical infrastructure to prolonged droughts disrupting global food supply networks this graph shows the frequency and the cost of billion dollar weather events in the united states from 1980 to 2020 and 2020 set the new annual record of 22 events and it was the sixth consecutive year in which 10 or more billion dollar weather and climate events have occurred so these changes are unprecedented in time in location and severity so we can't rely on existing approaches based on past experience to mitigate them but fortunately there's an increase in volume of data climate and geospatial data and sophisticated models of a variety of forms which can be deployed to help us quantify climate risk and build resilience so what does this mean for businesses well climate change adaptation and risk mitigation means building climate intelligence into operational and planning decisions and here are a few examples across industries energy and utilities require both physical infrastructure to be resilient which means climate risk needs to be quantified for that infrastructure and operations to be resilient which means building climate risk into planning and operation decision making in finance the management of financial risk requires a quantification of physical climate risk affecting your portfolio which in itself may be diverse across assets and industries and in consumer goods and retail supply chains are impacted by extreme events disruptions which needs to be prepared for the climate impacts business decisions across timescales from responding to events that are in progress to predicting individual events days and weeks ahead so looking at longer term predictions of the probability of impact for events over a season or years ahead weather and climate data and models are available to address all of these timescales but applying them requires highly specialist technical and scientific knowledge so here at IBM research we've been considering how technology can help streamline this process without losing fidelity or flexibility the further challenge that's relevant to businesses is the spatial scale so weather and climate impacts are experienced locally so changes need to be quantified in a locally precise way but the domain of interest can be at the country or the continent scale so solutions need to be efficient and scalable so how exactly do we quantify climate risk so to quantify climate risk we need to combine hazard exposure and vulnerability but to quantify the climate hazard models of a variety of types so physics based simulation models machine learning models for example are needed to generate and predict climate extremes and quantified hazards so flooding wildfire drought heat waves etc and these hazards that then need to be combined with exposure for a given user application for example the geographic location of physical assets for all communities relative to flood risk areas or the properties of the building which may be impacted by storms and then to calculate the overall risk and damage modern needs to be used to map the hazards and severity to damages or financial costs so here at IBM research we've been developing tools and frameworks which enable us to better quantify climate risk in a flexible scalable way to address some of these challenges that I've mentioned so I'd now like to hand over to my colleague Johannes Mooder to demonstrate how the IBM pairs geospatial and net exchanging can rapidly bring together the diverse geospatial data sets needed by researchers and analysts to quantify climate risk. Thanks Anne. What we're looking at now is something we're all too familiar with this is the general trend in annual temperatures and this particularly case for the contaminants of the United States for the last 40 years. The trend line we added here shows a warming of something like 0.023 degrees Celsius per year. This development turn drives other change such as rainfall so what we're looking at here now is aggregate rainfall across the United States and again you see a change you see a trend which is not quite as pronounced but still there. Now a key question for us is how to make business decisions in this reality. This regards both the question how to contribute to avoid further warming but also how to understand our changing environment and how this affects our assets our supply chains etc. This of us is a spatial problem. You need to be able to process large amounts of geospatial temporal data to understand how different assets in different areas are affected by this which is why our and our clients work here is powered by our geospatial big data platform IBM pairs which I will demo to you in the next couple of minutes. What we're looking at now is the webinar phase to IBM pairs. Pairs hosts some six petabyte of curated and analysis ready data. What we're looking at specifically is the data that was used to generate the temperature time series we looked at two slides ago. Specifically this is data from the Prism initiative which shows daily temperatures and rainfall amounts across the United States at something like four kilometer resolution. Crucially this data obviously depends on time. While we're looking at you as a snapshot actually it's the aggregate for one year. If I click somewhere here I actually sort of make a database request and pairs in real time to go back to the back end and pull out the time series now for the entire 40 years for this particular location. You can see this is already nice. I can basically explore different conditions across this entire 40 year range across the United States. Of course while I'm showcasing the United States for this demo this is an entirely global project. This is an entirely global platform. You could also work outside the United States. Now we were saying temperature is one thing and as this basically is the data we used to generate this time series and we crunched this with pairs for this but of course there's also things in here like precipitation we set here. Now and again now we see rainfall aggregated over 2020 for the United States. At the same time now you have the rainfall time series for these various locations. You can see the differences between you know what happens here in the Midwest versus what happens in the Rockies etc. Now pairs contains many many data layers. The whole point is that we don't have just the data from Prism or data from other data sources but there are satellite images, there is aerial photography, there's land use data, weather forecast etc and all these things are trivial to join. There's no data fusion that's entirely trivial to join any of these things together to generate insights and they're very rapidly accessible. So let's sort of continue with our story then. We were asking this question how can we understand this problem of climate change more locally okay and for this first let us return again to our temperature data and originally we used this for this warming trend okay so we aggregated it spatially we pulled it together to actually on single point looked at the time series but now the question is if you want to say how are different parts of the US you know affected by this in different ways we want to do a more granular more local analysis so what we did for this is we did the same calculation we did earlier with the time series that gave us a sort of you know linear development this linear growth and we did this for each point okay and we crunched this with pairs and what we get now is the temperature change over the last 40 years at each point in the United States okay so now what you see is actually very interesting picture you see the biggest rises temperature are in the west you know you see california you see the rocky is you see you know the rest of the west the east is a bit less pronounced and then you see development actually technically speaking a part of the north has even gotten a tad colder okay and this is something you know you know you see how suddenly different parts of the country are affected in different ways we did the same analysis with precipitation okay so you look at precipitation and you see precipitation change again is regional okay we said there was an overall effect there was a clear point of precipitation growth but now when you do this analysis more regional you see there's again there's a very different you know how this affects pronounced against the east versus the west you know you see increasing rainfall in the east decreasing rainfall in the west and you have to be a carefully of course when you go very local you use less and less data the more local you get obviously then statistics plays role you have to be able to think a bit more about errors but the regional effects are still a lot of hold okay you still see that there's definitely things going on in different parts of the country in different areas of the country okay so this is already progress on what we said before you know instead of having to sort of this you know entire aggregate view of the entire country now we can sort of think about different regions different areas of this affecting there but what we really like to understand is how businesses are affected how business processes are affected how supply chains are affected etc and i said earlier that pairs processes lots of different data types and so far we only looked at data from this presentation source so let's bring in something else specifically let's add to this land use data okay so this is land use classification from Copernicus which is essentially the european union okay what we see here now is for the same area how different parts of the country are used you know this means is it forest area is a crop land is it urban use etc okay and let me add a few comments joining different data types you know it's actually entirely trivial you know because some points like you know there's different data providers different sources but also things like different resolutions i told you earlier that the prison data we used has a resolution of four kilometers what we're looking at right now is a hundred meter resolution you know and if you just want to join these things effortlessly you have to deal with all these things and this platform does it for you automatically now how does it deal with our business decision with our concrete problem to understand better how different things are affected by climate change but what we can now do is we can take this data and pull out for example the urban cover okay so how much of the united states is actually covered by urban areas and we filter this out and we consider the temperature the precipitation change now just for these areas which have urban land use and then you find you know for the urban case all right we said we have something like 0.024 degrees Celsius per year warming in the entire united states you'll find actually for the urban areas it's roughly the same amount okay 0.024 as well however you find for the rainfall it's roughly double what is the average across the united states just by filtering down to these areas which of course relates to the fact that we said you know east versus west is more pronounced you know urban areas are more on the coast and certain other areas so they're not uniformly distributed across united states now the urban areas are one thing what is more relevant than is let's look at um agricultural use crop land okay so this again we took the european union data we filter out areas you know where there is um agricultural use you see very pronounced you know the calm but across united states you see agriculture in california you see you know sort of things along the mississippi and the south and again you can take the um temperature data the history we had the 40 years you can calculate the change across the entire united states as we did you filter then to only select these areas with agricultural use and then again you find a different insight okay you find that yes united states has an annual warming of something like 0.023 degrees Celsius per year but you know for the agricultural areas it's actually only 0.014 degrees Celsius you know so it's not as heavily affected which says how growing conditions will change in the next five years is different to what you would expect from the the rough value similarly you know for rainfall rainfall increases in these areas more heavily which you know should be pretty obvious because we said there was a sort of delineation and rainfall from the east versus the west more heavy in the east less heavy in the west and most of the agriculture is in the calm belt so you see exactly what we saw on these maps earlier so let me sort of sum up a few things give a few more comments okay so in the last couple of minutes we showed the web interfaces geospatial big data platform we were discussing how it pulls in different you know data sources and a reasonably simple example still and how we crunched this data that means we had this land use data we compared it with um historical weather data you know like these temperature time series we said you know you didn't see this life but we said how the platform that is behind this crunches this data down to do these sort of computations to get you these trends to enjoy this with the land use data and give you the idea of what different trends in different areas of the world were depending on land use okay and of course you could do the similar analysis with you know similar things either you place your own assets into this and say how are my assets affected you could say how is for example corn growing areas affected versus areas where other things are growing something like this okay let me finally give you a few more numbers um as I said you know this is geospatial data platform we add lots and lots of data that is relevant to a large audience that means we ingest something like 20 terabytes per day you know there's a six petabyte archive that is ready to use clients analysts can use their own data and add it to it as well for them to join in this analysis and process together and we use it to power our work in this area and so do our clients so now I'll go into some more detail about our environmental and climate impact modeling framework which leverages pairs to quantify climate hazards so we want to build a technology to accelerate the process of responding to climate change by better quantifying climate hazards such as floods, droughts and wildfire and this climate impact modeling framework SYNF is a flexible scalable cloud-enabled modeling framework designed to onboard, execute and enhance hazard models with AI these models translate climate variables such as temperature rainfall into hazards by simulating the dynamics of environmental processes and combining climate and other geospatial data so the framework contains generic components to enable modeling of scales so across different spatial scales across timescales and across different hazard types it supports automated workflows to easily onboard configure and deploy any environmental or spatial temporal model so machine learning models and simulation models hybrid models etc and included in this framework are pre-built models a combination of IBM priority models and open source models for impactful hazards such as floods and wildfire and this is all deployed on the scalable hybrid priorities and open shift so let's look in more detail at the components of the framework and how it can be used users can quantify hazards by considering pre-configured forecast horizons so for example current climates or sub-seasonal forecasts seasonal forecasts or future climate conditions and it can define hazard severity according to configurable metrics for example by selecting specific flood depths and or durations which are impactful to their application and another possibility is running what if scenarios so for example driving a flood model with different extreme rainfall scenarios derived from our rigidity another feature is compounding so this is considering the probability of events occurring simultaneously or in sequence or different hazards occurring in the same time place to support easy model deployment we have tools to help generate the spatial model domain for given target location and automate the choice of models and data and the workflow in terms of running multiple models and sequence we've pre-built hazard models which I'll show in more detail in a moment and with more developments and we have modules which use AI in a variety of ways to improve hazards quantification and I'll show some examples of those as well but finally we're building developer tools to support external clients and collaborators in using all the features of the framework for example by bringing their own models to run the framework so this framework pulls in data from partners and our own research developed weather generator and enhanced climate predictions it can also be connected to external APIs and the hazard maps generated by the models in the framework push back to paths to be made available as parts environmental intelligence suite so let's have a look in more detail at those pre-built climate hazard models so we've illustrated here four different workflows for three different types of flooding and welfare hazard which will available in the framework these use a diverse range of geospatial datasets and models so for example our fluvial flood workflow uses precipitation observations forecasts and scenarios together with geospatial surface features so elevation soil images and this combined to then simulate surface water flooding from heavy rain to quantify coastal flooding we use Nemo a physics based model to simulate storm surge driven by pressure and wind fields along with bathymetry data and then we use LiDOT derived elevation data to map flood implementation for fluvial flooding and river overflow we use an AI model to replace expensive rainfall runoff simulations and predict stream flow and then we combine this with elevation and river network data to predict implementation then finally for wildfire we use an open source wildfire model which combines climate and other geospatial datasets to predict a large number of different wildfire metrics according to standard methodology developed in the US, Canada and Australia and so these raw model outputs are all transformed into configurable metrics so for example the probability of flood above a certain depth the number of days with flooding above a certain depth and the depth simulated for different extreme rainfall events so how can AI machine learning help improve and accelerate climate hazard on this quantification so here is one example one challenge in climate hazard quantification is easily validating models against observations or comparing models so we need to understand how well simulation models we produce past events before we use them for future predictions and this can be challenging if ground truth observations aren't easily available so for example in the case of flooding traditionally flood extent information is difficult to obtain it's inconsistent across geographies is not reliably available but fortunately we can use remote sense data to address this and obtain full extent and we do this using satellite data combined with geospatial analytics and deep learning so we then deploy this in the framework and automate this process across geographies and across different flood types and a further benefit of this ground truth detection is that we can also record historical hazard exposure for retrospective analysis of past events another way AI can help is improving simulation models by tuning their parameters to local conditions so for example in flood models surface properties are set from land cover and soil data sets but these need to be refined to be correct locally so traditionally this is done by manual tuning which is laborious and also doesn't yield a lot to the results so using cutting edge machine learning approaches such as Gaussian processes and Bayesian optimization we can efficiently tune the models to local conditions and do this in an automated scalable way and these parameters then can then be used in deploying models to quantify climate hazards not only that but by using Bayesian methods we can build in uncertainty quantification and predict a range of outcomes according to the uncertainty learned by machine learning model but using the framework we can also combine this uncertainty with climate uncertainty to simulate a whole range of plausible scenarios providing much richer information so I'll now hand over to my colleague Barad-Witt to demonstrate how we use the framework to quantify flood risk in extreme rainfall. Thanks very much Ann so today I'm going to be telling you how easy it can be to use the climate impact modeling framework or synth that we've developed to calculate hazard risk maps so in this case we're going to be looking at a clue of your flood risk and we'll be doing that using a set of historical data using a climatology so that involves running our simulation model to calculate the flood risk for a given month of the year for a set of years in this case from 1999 until 2018. So when we do that we will essentially within the framework in an automated way be running all these simulation workflows in an optimized way for every year in parallel and then combining the outputs. So to start with we input some packages and then the first thing we have to do is define the spatial and temporal domain over which we want to run the models so for a spatial domain you can see here that we have defined it as a bounding box in this case two particular bounding boxes of interest but you could also do this as a polygon to have flexibility over the area that you want to run the models. As it's a climatology and we're interested in a particular month of the year we define that month here so in this case August and this will allow us to calculate the start and end days for each simulation run. We can then see and check the bounding boxes that we've defined and then we can move on to setting up the rest of the simulation workflows. So to do this using SIMF we have a user-friendly API payload that we can set up so you can see here after specifying the type of workflow we're interested in we tell it the spatial domain the start date and end dates the the year range that we're interested in and then we can in a very straightforward way tell it the datasets that we wish to use to drive the modeling. So in this case we tell it the precipitation data, the elevation data, soil and land use that we want to use to drive the models. Then there are some model dependent parameters, optional parameters that we can set here and the available options are documented alongside the framework. So now we've done this we can then simply submit that to the SIMF API and that will spin up a model run with these 20 years running in parallel simulating the different flood scenarios. If we now wanted to do that for a different resolution for example with that smaller bounding box we all we would have to change here is the spatial domain if we wanted to change that the DEM dataset that we're using and that will allow us to run a higher resolution and then if we need to change any additional optional parameters we can do that too. In the same way if we were interested in running with different precipitation data for example all you would have to do would be to change this entry here in the payload and the models would be run with the appropriate precipitation data. Again you would submit down the same way and whenever you submit a workflow request you should get a response giving you an ID related to that tag to that simulation which will allow you to track and monitor the status. Status monitoring can be done as well through the API passing a simple API call and when the models are finished that will also return you URLs to download the outputs from the different model runs. So in this case all the different years that were part of the ensemble. When all the ensemble members are completed the metrics of flood hazard will be calculated using all the the model the different ensemble members as inputs. If we wish to calculate additional metrics after the fact we can do that using a simple API call and here is an example of such a call where in this case we would be measuring the number of days above a particular flood threshold for three different thresholds and the same here with flood probability and again with the same three thresholds and you just pass that to the API to have it calculated. Once the metrics have been calculated they're then uploaded to IBM pairs where we can then go to inspect the data further or integrate it in further use cases. So we'll go to pairs and we can set up a query like this and in this case we're going to set up a query to look at all the input data sets and the output data. So in this case we have the land cover, the precipitation, elevation and soil which are the driving data sets for the model runs alongside the expected flood days and the flood probability for the three different flood depth thresholds that we specified. So now we look at the data so we can see that we have the elevation data set which will determine the direction of flow of the water. We have the soil soil type and the land cover and these will be converted into hydrological parameters within the model workflows and see directly into the model to tell the model again how the water will flow across the surface. We have the precipitation data in this just for as an example this is the total for the month of August for one of the particular years but it's actually a time series of precipitation that is fed into each model run. And then we have the outputs so we have the expected number of flood days that we can see here for 15 centimetre threshold, 30 centimetre threshold or 50 centimetre threshold. So you can see once we have the data in pairs we can query alongside a range of other data sets and overlay that to understand what these impacts mean in the real world context and determine the consequences of in this case flooding. Thanks very much. So finally what are the ways potential external users and clients can interact with that framework? So here are some examples. First an end user such as an asset manager or supply chain manager might want to quantify the risk of flooding and other hazards to their physical assets without doing any complex modeling. The entry point for this would be to use pre-configured models and metrics which could consist of pre-computed hazard maps of different prediction time scales made available via pairs or pre-configured what-if scenarios for extreme events. Secondly a risk analyst may want more control over the hazard models without needing extensive climate or impact modeling expertise, may want to try different data drivers or configure or calibrate models for new locations or do validation or post-hoc analytics. And then finally a data scientist or modeler might be someone who wants to use their own proprietary models with different climate drivers or in a more scalable way or could bring their models to the framework for onboarding configuration and deployment. Due to the flexibility of the framework these models are not even restricted to climate impact models it could in fact be any spatial template model which consumes geospatial data. So these are all just examples models of engagement and interaction which we'd welcome the opportunity to explore with you. So please get in touch if you're interested to follow up. Thank you. Excellent thank you Anne and Johannes and I believe we have them both here live to answer a couple questions. Nice to see both of you. Good morning. Yes yes good morning good morning and good afternoon Anne. So a few questions in the chat. The first comes from Jen and the question is what I guess this is for Johannes. What are the computational capabilities of pairs? Like what's a typical runtime? I know you spent a few minutes talking about you know petabytes and terabytes but maybe expand a little bit on that question. Yeah thanks actually that's a good point because I think it's some nice details about this because obviously what we showed the demo where the results of running queries okay so it's a good question you know how heavy weight are these things how much do you have to weight and so the things we ran um they took sort of in the extreme uh 8 to 10 minutes to run you know and it's good to put this into context you know 8 to 10 minutes you might say hey this is maybe the time it takes to make a cup of coffee but is this like good is it bad you know what does it mean to write this data. So think about it what did we do we had some 40 years of daily data that means we're crunching a time series of something like 15,000 timestamps you know then for each timestamp it's actually a map you know a map covering contaminants United States at four kilometer resolution that means each timestamp is half a million data points okay so roughly the ballpark number is going to keep in mind you know 40 years 15,000 elements of timestamp for this we did basically half a million linear regressions and this is before we started joining in the high resolution radius data okay so you sort of the ballpark numbers to think about to idea computational capabilities thank you yeah yeah wonderful thanks thanks for for that additional additional context I see another another question um and it says uh we need to let people provide their address and determine for example how much insurance goes up how much has been on hazard mitigation growth and property prices etc I guess I mean certainly insurance companies use you know your risk profile is based on where you live but I wonder as the technology evolves as you get even more granular on a geospatial level you know I this this will continue to get more sophisticated this modeling do either of you collaborated or have thoughts on on how the technology you know could be used in the insurance industry and how maybe getting even more granular will impact folks at that in just in that industry yeah I think maybe a bit yeah yeah so I think it's a question for both of us and let me just take a stab at it first and maybe you're at it because I think um obviously both parts of the story you know sort of are very relevant there you know obviously the idea to really you know have access to vast amounts of geospatial data we heard this a lot this morning already you know how crucial data and modeling is to this and on top of this the ability to run complex models and stuff you know this is really true things that sort of allow you to sort of quantify risk and see impact and being able to do this at this very local scale of course you know makes it able to make better decisions you know that's what we want to do you know better decisions business business wise you know how are my assets affected you know one thing is to snow all right there's going to be a change for the entire united states for the entire world but really what we wanted to know and what we're trying to touch on is you know as a decision maker in enterprises you know but also in government you know you said how are certain things affected this is always a local question and this means you have to run these things at high resolution you know simp can run models at and correct me it this is a 30 meter result looking at you know a very high resolution you know you can go much finer you know we have data up to a few meters down so yeah we're going to be able to answer very granular questions that affect businesses yeah yeah and the other point to make of course is that we want to be accurate at that scale as well so we're using AI to improve our modeling to introduce that localization to make sure that we're locally accurate as well and then the beauty of pairs is being able to combine the hazards with all the other geospatial data sets that you need to bring together to quantify the overall risk yeah yeah while I have you and I have one I think questions but listen to your talk specific for for you so when it comes to collaborating and I alluded to collaboration at the end but like you know how would a user bring their model to simp you know and what kind of models can you actually support in that in that framework yeah well we support any type of computational model so you know we we currently have physical simulation models in their AI models hybrid models empirical models based on rules and written in different programming languages Fortran Python and for example the only requirement is that containerized so if a user wanted to come and run their models in their framework it would just be a case of working with us to create the interface that specifies what the data sets are that requires what the parameters are and then also the requirements in terms of the output so how they might want the output summarized or do they want raw output or would the user want this shared via pairs which would then enable them to do further analytics on it that's great that's great and and I'll just add both Larry Light and Allison Davidson can help you connect and collaborate they put their contact information in the chat and I encourage you to reach out to either of them or our speakers if you want to you know try out the technology bring your models you know all the things that Anne Johannes just just mentioned so with that I think we're up on our break thank you again both for a wonderful presentation please stick around if you have some time in the chat there's a few more more questions coming in I think they'd love to hear from you and with that please come back and six minutes five minutes we'll come back and continue our program okay welcome back everyone I know it's a short break but we have again great content ahead here the next sessions are going to build upon that foundational work that Johannes and Anne described our next session is entitled AI for Better Climate Predictions and it's a conversation that is led by Campbell Watson who's a research staff member at the Yorktown Heights Laboratory and he has two guests with him that I'll allow him to introduce when the session starts so with that let's hear from Campbell and the team welcome to the session on AI for Better Climate Predictions I'm Campbell Watson a climate and AI scientist and co-lead of the Climate Informatics Research Program at IBM Research. Climate Informatics is where climate science meets artificial intelligence. Our aim is to develop core technologies that accelerate the prediction generation and understanding of past and future climate events and we aim to provide seamless predictions across time and space of extreme climate conditions going from weather to seasonal forecasting and all the way out to long-term climate information. Our final aim is to deliver this data to platforms like PAIRS to drive risk and impact modeling and allow organizations to take action to enhance resiliency to climate change. I'm thrilled to be joined today by my co-lead Bianca Zadrosny a fellow IBM researcher in the Brazil lab. Bianca, if you'd like to introduce yourself. Hi Campbell, my name is Bianca Zadrosny. I'm a research manager at IBM Research Brazil and I have been working with Campbell co-leading the AI for climate risk and impact thing that we have in IBM research. So my background is in machine learning and over the years I have worked in different projects doing research in applying artificial intelligence and machine learning to different product projects in different industries using geospatial data. The second person joined by is John Williams from the weather company and IBM business and very excited that John can join us today. John? Thank you very much Campbell. Yes, my name is John Williams. I manage the forecasting sciences team at the weather company which is part of IBM. My background is in mathematics, artificial intelligence and weather. My team is responsible for using AI and physical science methods to sift through the deluge of environmental data that's available and create the world's best weather forecasts. So jumping straight into the questions, John you have a very long history in AI and weather. You are the past chair of the American Meteorological Society Committee for AI Applications as well as a project scientist at the National Center for Atmospheric Research and around seven years ago you started work with the weather company. So can you please tell us a little bit about how your group, the forecasting sciences group, is powering weather forecasting with AI? Yeah, first of all I think a lot of people may not know that the weather company is part of IBM and that's really important because weather impacts every business and every individual. So the weather company is probably best known for our weather channel mobile app and our weather underground and weather.com websites which together serve over 400 million users every month. We also provide the weather forecasts for Google, Facebook, St. Samsung and many others and specialized weather data to all over 3,500 businesses around the world including 40 billion forecasts a day that reach two and a half billion devices every day and our MAC system drives the vast majority of TV weather broadcasts in the US as well as many internationally. So IBM is actually the world's leading weather provider. Our forecasts to get back to your question are they're generated with a combination of automated AI methods with oversight from a group of trained meteorologists. And our fundamental approach is to grab observations and forecasts from modeling centers all over the world as well as IBM's own global high resolution atmospheric forecast or graph. And the AI is designed to correct the forecast based on their past performance and then blend them together in a way that optimizes the forecast performance at each location at each lead time and for each variable. Of course I'm skipping over a lot of details here but the upshot is that this approach AI combined with humans who can nudge the forecast like when smoke unexpectedly from the western fire fires blocks the sun and lowers the temperature this combination creates the best forecast in the world and I'm not just saying that the summer there was an analysis by an independent service called forecast watch that compared the weather company and 16 other providers using only over 80 metrics of accuracy and found that we were the most accurate most often about three and a half times more often than our next competitor and that gap grew every year between 2017 and 2020. Wow it's a really interesting combination of automation with things like AI and humans. And I think that gets to the question of dealing with uncertainty so weather as we know weather forecasts climate as well inherently uncertain. So how can you ensure or enable smart decisions when there's uncertainty in a forecast? That's a great question and when you look at your weather app or your weather forecast it's always obvious whether it's a very confident one or whether there's some uncertainty. And of course a lot of us enjoy looking at what the weather is doing. It's a big topic of water cooler conversations but really the main reason that we as IBM are in this business is to help people in businesses make better decisions and uncertainty is a huge part of that. The weather enterprise has long focused on what we call decision support which basically means displaying relevant weather information to users in a digestible form. We're going beyond that with what we call decision services which move farther up the weather value chain to actually make decision recommendations for a user or even take action automatically. And as you said uncertainty is a key part of that. Our decisions framework includes four questions. First what could the weather be? How does the weather translate into impacts that a user cares about? And how would the available actions coupled with those impacts affect the outcomes? And then what's the value of each outcome? So just as an illustration consider a citrus grower faced with a near freezing forecast. How does she decide whether it's worth the expense of deploying misters and blowers to mitigate the risk of crop damage? Well that depends on the cost of the modification the probability that the temperature will go low enough to significantly damage the crop which in turn depends on the uncertainty of the temperature forecast. But it's really more than that. You need to know whether the wind will be caught to know if the blowers will help or if it's dry to know if misters will help so there are actually a lot of different weather variables at play. So what we've done is to combine and calibrate this large set of over 150 different weather model inputs that we collect from all over the world to create a set of weather scenarios or future weather trajectories at each location and over the next 15 days that are equally likely and span the set of possible outcomes. Each of those weather scenarios can then be translated to its impact on the crop depending on what action is taken if it's no mitigation deploying blowers or deploying misters and then those results can be fed into a separate model that assigns a value to each outcome combining the cost of mitigation and the dollar value of crop damage. And by the way it also matters whether this might be a widespread freeze or a very local one since if there's widespread crop damage your harvest will be more valuable. So at the end of the day each action and each scenario gives an economic outcome value and you could take the average to find the expected value and choose the best one that's within your risk tolerance. And I should say that we've developed this probabilistic scenario approach for hourly weather out to 15 days and daily weather out to seven months in the seasonal time frame. Yeah I think from what you're describing the overlap with climate risk is is very clear and as part of the future of climate research that we're doing we're focused on predicting and characterizing specifically extreme conditions at these longer time scales from months like the seasonal forecast that you just referenced to years and even decades ahead. And we at IBM Research have been very lucky to work closely with people like John and others at the weather company to develop new technology for sub-seasonal and seasonal forecasts as well as longer time scales. So Bianca can you please tell us a little bit about the sub-seasonal or seasonal forecasting technology that your group has been developing? Yes so building upon the work of the weather company that John mentioned our IBM research team has been looking specifically at targeting what we call events that matter. So for example if you're looking for the likelihood of flooding rains in southeast USA or a heat wave in southern Europe in the span of the next weeks to few months we are building technology using AI to rapidly create and deploy models for these targeted sub-seasonal to seasonal predictions. And the reason why we think that AI can help in this case is because it even though it's not in these cases it's not possible to model very well in like a small scale like the events are predicted exactly where it's going to rain the models are able to get information to extract correlations from like long distances relationships for example El Nino events and use that in the model that generates the probability of extreme events. So we have been able to show that it's possible to train these targeted models and get better accuracy than just using the physical models or blending of physical models. Is so these models that you're developing for prediction are they targeted to specific features of the climate system? Yes so for example we we say for example we have an extreme precipitation model that is targeted specifically at predicting extreme precipitation we could also have other models for specifically for heat waves. And as I said those models they don't predict they have some uncertainty built on them so when we see predictions for a few days ahead or even a few weeks ahead the as John was saying that you have uncertainty and you have multiple possible scenarios that could happen. And what we are trying to do with AI in this case is actually to use what is called generative models to facilitate the creation of scenarios even in the cases where the signal from the forecast model is very weak. So with that we rely on what we call what if scenarios so from the in the what if scenario a human generates a hypothesis that would be something like what if you have a some a stronger a longer period of rain than expected. So for example if you have a hurricane that lasts one day more than the one that you just had and then you could using generative AI models you can create very realistic patterns that would simulate that kind of weather pattern based on using historical data and also physical information from the dynamical models and then generate those patterns that can be fed to the impact model. So we go many times from low resolution information that comes from the the forecast and very uncertain information so you may say you have a certain likelihood of having some extreme rain in this region to a very detailed map of rain for that region if that condition were to be true. But the thing is that you don't have just one you have multiple scenarios that are compatible with that hypothesis that you are making. So our tools allow you to generate all these possible realistic scenarios that could be realized and then you can estimate the risk of different events. So you'd estimate use that for wildfire modeling for flood modeling for assets damage estimation using those possible realistic scenarios that could happen under climate change for example. And another thing yeah go ahead. I was gonna say these so these what if scenarios sound incredibly powerful and I imagine they could also be used for counterfactual analysis. It's similar to what you described with say a hurricane Harvey what if it lasted for one extra day what if there was 30 percent more rainfall and at the heart of this technology of generating these targeted scenarios is the ability to develop very high resolution weather data that's realistic. So there's a lot of technical challenges in here. Can you just describe briefly some of the ways in which you've been addressing this. Yes so the one of the technical challenges is that we rely on historical data to build the weather generators using statistical techniques or AI generative learning techniques. But the fact that they were trained with historical data actually goes against the idea that you want to generate scenarios of that have not happened yet. So the way that we are addressing that is by combining this with other techniques from extreme value theory and other ways of trying to extrapolate the scenarios using for example physically informed neural networks that could capture use not only the historical data but simulate physically possible scenarios. So these are avenues that we are researching right now of how can you actually create realistic scenarios for certain locations in for cases that have not actually happened in the past before. And this technology will be available through the climate impact modeling framework and also PERS which I think is a really nice way for people to interact and explore possible climate outcomes with this with our technology. So final question for John from your experience of tailoring weather information for decision making. What advice do you have for the audience here with respect to climate risk? That's a really important question and you know I think we all saw the UN intergovernmental panel on climate change report this summer which made clear that we're now in the situation of needing to mitigate the climate change that's already happening. So I think of climate risk as basically future weather risk and so it's very related to the work that we're doing at the weather company and of course we can't accurately forecast what the weather will be months or years in the future but we we can use modeling and AI frameworks to produce a collection of weather scenarios that span the possibilities of what can happen. And then the same decision framework that I outlined earlier still applies you can evaluate each of your possible actions against the set of scenarios to understand the expected value and the risk associated with each which then allows you to choose the best actions. Perfect. So I'm going to end with a call to action for the first time the IPCC said it is unequivocal that humans have worn the skies the waters and the lands but despite their confidence oftentimes the level of precision needed to take action is not readily available we need the right tools the technology and models in place the emerging technology cutting-edge research and a world-class foundation in weather that we bring at IBM puts us in the right place so come work with us to build the models and the data needed to understand your vulnerability and exposure to climate change from rapidly deployable forecast models for extreme events that matter to what if weather generators to help you understand the strengths and weaknesses of your organization we're ready for joint research projects and more. So for more details on how to engage with us please reach out to Alison whose details are in the chat or any of us speakers. John Bianca thank you so much for joining us today signing out. Thank you. Thank you Campbell it's a pleasure. Hey wonderful discussion and I believe we have Campbell Bianca and John on with us here they come one two three nice to see all of you thanks thanks for such a great discussion there are a few questions in the chat they're actually from my point of view quite related you talked a lot in your discussion about uncertainty and confidence because you know taking action has a cost so whether you take that action will depend a lot on how confident you are in the models so let's start with Paul's questions where he asks what is the state of the science with respect to how far out we can reliably predict weather risk to meaningfully inform mitigations of risk. Sure I can take that so I started to type a response to you Paul but I was dancing verbally. Key things here are what type of weather risk what are you trying to what sort of actions are you wanting to take and what's your uncertainty threshold things like will the temperature be getting warmer I think the state of the science is pretty clear yes absolutely but is that useful for the risk that you're trying to mitigate if indeed the question is more like what is the likelihood of you know receiving 100 millimeters of rainfall on a given day in a specific location in this particular decade there's a little more uncertainty with that we can certainly say things about trends and put uncertainty bounds on that but a lot of it has to do with well well defined questions so the way I respond to it is for some things where uncertainty is a little too high we prefer sort of a counterfactual or climate storyline approach for these long-term projections where you'd actually pin the specific event you're interested in to to larger-scale dynamics or circulation so it might be atmospheric rivers hitting the coastline of the west coast us and and and trying to tie let's say relatively local features of the climate to specific events and then searching through say cement six archives data for for these larger-scale events like atmospheric rivers that we know are relatively well resolved in the models so yeah but it's a it's a great question and it's something that I think more of a consultancy role like we have with our gbs group is is something that can really come into play to to help people with that yeah I think there's actually a good follow-up to that we talked about you know working directly with businesses comes from I believe it's Olivier who asks do you have practical examples of correlating weather scenarios with the performances of business processes of an enterprise and of predicting process KPIs according to various weather conditions so very very to the point so I'm curious any any thoughts on this on this question I think that's maybe for John yeah I thought so too yeah so that's a great question and you know we've worked with different industries and each one is different I think one of the really interesting ones was working with a major airline who was interested in making decisions about whether to delay or cancel flights the day before and and so depending on what the weather is at the airport and what the FAA says about how many aircraft you can land you have to decide which of all the many flights you could make you are going to make and if you can proactively delay or cancel your flights the the night before then you can save a lot of money and time if you catch people before they head to the airport you can reschedule them so that was a just a really interesting example of where our scenarios went into determining the airport capacity which then went into the flight planning and the all the airline has you know of course models for for how to do that and then at the end of the day the ability to compare different playbooks so I can't give you you know a precise KPI or maybe even a cookbook for how to do that it's going to depend on the industry but I think there you know we do believe that that this scenario that based framework that can be really powerful that's great I think so thank you for that that that detailed answer I think we're just up on time but please do stick around in the chat and if you can there are a few other questions I think the the attendees would love to hear more from you thank you for the discussion and for your thoughtful discussion and for answering the questions that we got to and and just thanks for participating I'm looking forward to seeing you guys in the near future pleasure thank you thanks George absolutely so we're going to move on to our next speakers we have Jitendra Singh and Schmidt Marvaniya who are both research scientists from our India lab and are focused on the future of climate initiative the title of their talk is climate aware applications demand forecasting so with that let's let them let them begin welcome to our session on climate aware applications my name is Jitendra Singh and I'm a research scientist at IBM research I'm a part of leadership team of the future of climate initiative focusing on climate informed applications in this role I'm passionate about infusing weather and climate insights into enterprise applications to make them resilient against weather and climate risks during this session my colleague is Smith and I will be discussing these applications and demand forecasting we encourage your questions throughout our presentation which you can type into the chat or at the end of the presentation we will have time for a couple of questions as well so let's get started thanks Smith for joining me today for discussion could you briefly introduce yourself hi Jitendra hi everyone I am Smith Marvaniya I'm also a research staff member at IBM research my work focuses on building innovative research applications for many real world insights using NLP ML AI and computer vision in the areas of education agriculture and supply chain over the last year we have been building climate aware capabilities such as multi horizon uncertainty aware demand forecasting and also how to enable climate forecast explainable to enable resiliency in supply chain awesome Smith can you explain at a high level what are climate aware applications what do you mean by that sure the climate our application basically means the enterprise application in which the operational and planning decision are significantly impacted or influenced by weather and climate at multiple timescales there are many applications in supply chain and asset management such as demand and out of stock forecasting inspection and maintenance which are impacted due to climate change traditionally this application did not utilize weather and climate insights fully as we already experiencing significant climate change in terms of changing weather patterns and weather extremes such as heat wave cold wave storm flood climate resiliency has become a significant imperative for businesses and climate our application could help them achieve that okay so to narrow down the topic a little bit let us pick supply chains what do you mean by climate resiliency in the context of supply chains sure let us first look at few events of climate and weather impacts on supply chain which are fairly recent as you can see on the slide on the left hand side energy supply chain last year in september 2020 we saw severe blackout due to record racking electricity demand in taxes due to a heat wave the energy supply chain got disrupted due to significantly increase of usage of cooling system similarly in february 2021 millions of people left their homes and they were without power and water due to deadly winter storm similarly on the right chart as you can see in 2020 fission and upper retail products suffered from low cells and high inventories due to pandemic however in november 2020 daily region significantly experienced the coolest temperature in a decade that resulted in increase of consumer demand for winter clothes this example shows how destructive extreme events can impact supply chain resiliency in a big way due to climate change such events are projected to to be more frequent and intense over research is focusing focusing on building resiliency against such events so that businesses are prepared to address them in a proactive manner rather than the reactive one climate resiliency and adaptation is indeed in the top of the mind for everyone could you talk about how climate resiliency is connected with climate ever demand forecasting sure this is an important question as you know demand forecasting is the key first step in supply chain operations such as demand and inventory planning precision planning climate ever demand forecasting use weather and climate data in an intelligent way along with other informations such as holidays or a product information to forecast the demand thanks to advances in weather and climate science over the last decades weather and climate events can be forecasted more reliably and accurate much ahead of time there are multiple solutions available commercially and also in the open source could you share what are the limitations of those solutions and why can't they be used in the context of climate change sure it's a good question the state of the art forecasting technique do incorporate short term weather attributes along with other parameters but in a deterministic manner however these approaches do not model uncertainties or probabilistic nature of weather and climate which is the most important factor in the context of climate change has the forecasted forecast made by these existing approaches may not be accurate and has reliable enough for informed decision making so one has to come up with an innovative approach that models uncertainties in both space in time while building demand forecasting applications this has been our research focus and we have got very good success in in this by building uncertainty of a neural network models this model basically learns the uncertainty in input space from a large volume of geospatial weather and climate data as a part of model development okay so what are the technical challenges in building such models and how is IBM best position to address these challenges this is a very good question definitely we need an infrastructure to manage large volume of geospatial data for this we have a big geospatial analytics platform called pairs to manage all this data in an efficient manner we also have access to best in class weather data from the weather company which is IBM owned as well as from other forecasting agencies such as ECMW at IBM research we have dedicated research group focusing on further improving weather and climate forecast prediction of extreme events such as flood storm heat wave cold wave beside this we also have deep expertise in artificial intelligence that is required to model this complex data for various downstream applications can you explain how climate ever demand forecasting can benefit users or businesses sure there are many impactful real-world applications that can benefit from climate ever forecasting and specifically demand forecasting here is one such scenario for retail supply chain as you can see on the screen retail manager typically plan for high volume sales event in advance such as black friday or christmas in this chart you see two forecasts generated with and without climate data for the product category of thermal clothing the height of the bars shows the amount of errors for predicted demand the blue bar represents the prediction error for climate ever model whereas orange bars represents the prediction error for non-climate model as you can notice the model with climate insights shows improved forecast as you closer to the black friday event these forecast insights can be useful for retail managers for more informed planning of demand and inventories up to six weeks ahead of time similarly precision planning for seasonal goods require more lead time and accurate demand estimation a few months ahead of accounting for changes in weather patterns such as early winter or late summer climate ever demand forecasting could bring economic benefits for this use cases as well this is interesting what research innovations are needed in building scalable climate ever forecasting this is a very important question there are two core components which need to be considered while designing and developing climate ever forecasting solution first we develop a scalable climate data pipeline which is integrated with pairs which is a geospatial analytics platform the climate data pipeline enables data download data processing extracting customized uncertainty aware derived features support for use case inspired features to give an example we want to derive features related to heat wave for a particular location and and the time frame for this we need the weather data for the period of interest and also the historical climatology data representing what happened over last 20 to 30 years this data pipeline are designed in a way so that all these operations are done in an automated manner second we design a modular neural architectures that include a set of sub neural networks to specifically model the individual weathered or climatic features expected from our climate data pipeline and also simultaneously learn the spatial temporal complex relationships between different types of features and the target variable such as demand thank you for sharing the research innovations and example use cases i can see a lot of benefits and utility in the supply chain management but what could be the other use cases outside of supply chains where this solution can be used excellent question yes in fact there are many use cases of supply many use cases of climate aware forecasting in food supply chain such as production forecasting that we are currently exploring similarly newest cases are emerging in energy supply chain related to increased renewable power integration in the grid given that renewables such as wind and solar are intermittent in nature power and demand forecast at multiple timescale is crucial for load balancing and grid stability our demand forecasting solution leveraging artificial intelligence can be used in this context for enhanced renewable integration and grid operations this is very interesting and i can also see how AI can help in better forecasting however with these sophisticated deep learning models that you refer there is also a concern of explainability and trust of these models are you also looking at these aspects of the AI models also what are your future plans direction on of research on this topic absolutely thanks jitan for asking these as a part of future work we are building tools to enable climate aware forecast explainable so that users or subject matter experts can take decision confidently for this we have already have explored explainability toolkit ax 360 that we plan to use and enhance while learning a reasoning knowledge draft to allow user to do interactive what if queries in an intelligent way further we are also exploring advance AI technique to learn use case inspired future representations that will reduce the dependency on domain specific data furthermore this advanced AI technique do not require hand crafting of futures which is a time consuming activity in building the models we are also planning to expand to other applications in energy supply chain this is exciting exciting future directions and research thank you smith for an interesting discussion and also to our audience for listening today thank you again okay excellent and we should have our speakers on jitendra and schmidt nice to see both of you how are you doing hi george good morning good good good so a couple of questions in the chat um let's start with jen who says uh can you integrate other demographic or social data such as population in region specific buying patterns in climate aware demand forecast forecasting capabilities so can folks bring their data sets i guess is the uh the the main question yeah sure so i think from the geospatial data perspective i guess it's uh we can integrate many forms of geospatial data in a seamless way because of the infrastructure pairs that we have to manage and you know curate such data sets also the underlying models that we are talking i guess they are independent of uh you know any specific data so they can also integrate such data in the feature space and then do the forecasting so the answer is yes for this yeah uh a question from Olivier who says uh is it possible to forecast the demand for electric cars uh for instance depending on medium-term global warming or the evolution of temperatures at a country level any thoughts on on electric car demand is meant you want to respond yeah i think it's it's definitely possible but i guess for this use case there would be many other factors like you know policies and things like that uh but definitely here i see that you know even the temperatures and evolving you know weather patterns could help improve plus the policy signatures included in the model as well so yeah i think it's a nice example to to try it out as well yeah i bet the the mine share also improves the the demand there right i mean people know that we all have to do our part in electrification of our vehicles is one one one thing you do we're gonna say something schmidt i don't know if i cut you off yeah yeah definitely george so this is this is a very good question straight as uh electric car definitely needs it's uh the the for charging right you need the you know the the power to charge your battery and how your input source is dependent on right whether it is really reliable on your renewables or whether it is reliable on on on on the maybe you know some other power sources and we do uh see that efficiently forecasting the energy demand in the domain of you know the electricity or energy which could actually help you to forecast the usage of the electric car as well as like how often they are going to charge and how much is the energy demand which is going to be needed for this specific use case so this is definitely going to be the the future i mean like in in the domain of you know the dynamometer sustainability as well as in the the electric car perspective yeah yeah i haven't thought about the infrastructure piece uh when i saw his question but that's uh that's an excellent excellent point thank you for bringing that up um we have let's get to one last question we're a minute over but i do want to get to rachel's question uh where she asks um you mentioned using demand forecasting for load balancing could that do enough balancing to support an entirely renewable energy grid which is actually related a bit to your to your previous answer schmidt but but any of you want to take that that question uh yeah definitely because as as as we even uh gave a few examples right because of the heat waves as well as the cold waves right the usage of some of the the electrical you know the the impact is like uh increasing or significantly higher right so how can we efficiently estimate the the peaks of the energy demand and the plan much ahead of time so that some of the blackouts kind of you know the events would not happen and how can we have uh sufficient you know the the power in the in the backup so that you know that we can enable the resiliency as well as the stock out kind of you know the scenarios in in real world situation right so definitely uh using the climate data and accurately forecasting this kind of you know applications have a lot of potentials to enable the resiliency even in the domain of energy and utilities and subject matter experts can take appropriate decision making by how much is the additional demand is going to be because of what kind of climatic you know the change or extreme events yeah it's a very powerful tool great great um i think we're a couple minutes over so i'll leave it at that there are a couple additional questions in the chat so if the other if you have a few minutes to go in and interact with our audience that would be wonderful thank you again for a great presentation and for joining us live uh for a great discussion i appreciate it thank you george thanks smith yeah thanks thanks everyone thank you um so we'll go on to our next session uh which is led by my colleague uh leventy climb who's a research scientist here at the york town heights uh research center working in geospatial analytics the title of his presentation is climate aware applications decarbonizing the electricity grid so with that let's let's hear from leventy renewable energy is the dominant energy source that is generated in many parts of the work in europe there are countries where more than 40 or 50 percent of the total energy is generated from renewable energy sources like wind solar hydro or from natural gas in the united states the overall percentage it's around 10 percent of the total energy that it's coming from renewable it is known that traditional energy sources like coal or fossil fuel is generating around one kilograms of carbon dioxide for every kilowatt hour in order to improve the carbon footprint of the electric grid and also the energy generations needs a need for a higher integrations of renewable energy sources in the overall energy generations across the globe with the increased penetrations of the energy from coming from renewable energy it has been observed that the overall cost of the generations is decreasing this trend has been demonstrated across the last decade and indeed nowadays the energy produced from renewable sources are going to be comparable in price with other type of generations in general there are a couple of things that are driving this one overall decrease in the cost of the productions that it's coming from cheaper solar panels and wind turbines that could be easily integrated into the electric grid and many of the technologies that are making the integrations are also decreased in price but renewable energy has its own challenges there are intermittency that is generated in many cases from these energy sources meaning that the power that is generated at a certain moment of the day it's going to exceed what is going to be the demand but some other parts of the day the generations of the renewable energy may be very tiny or inexistent such the case would be for example if dates are cloudy days or dates are wind-free days in various parts of the world also dates a mismatch between what are the ideal locations for generations of energy sources and also where is the highest demand for this type of energy so for example locations in the united states that are ideal for solar radiation and for solar power generations are going to be in Nevada or in Texas but the highest demand for energy is going to be on the coast which are most populated so there are multiple ways in which we could address the price drop of the renewable energy sources and there could be potentially new transmission lines that could be built in order to move the generations of the power to the locations where it's a high demand for it but that tend to be relatively expensive as it requires investment in the infrastructure in the power lines and the transmission infrastructure that is going to carry such power the need dates also need for energy storage mainly coming as batteries as the batteries are integrated into the electric grid they are going to be take up the excess of power that is generated in certain locations and then provide this energy at the moment when the demand is going to be very high both the transmission line and also the integrations of batteries are relatively expensive technologies one way in which we could address the cost drop for renewable energy it's potentially to have better forecasting better forecasting it's really going to provide an understanding about what is the availability of the power at every moment of time and the same time the same forecasting tools also could predict what is going to be the consumption at different moment of the time now in order to generate a reliable energy forecasting multiple data sets needs to be combined and those are coming in the form of for example weather models it's going to be satellite observations it could be internet of seeing type of sensors on the ground like pedestrians combination of this energy these data sources could provide the possibility to increase the granularity of the forecasting extend the time horizon and improve the overall accuracy IBM developed a technology called pairs that it's a big data platform that it's able to take various data sources and then do spatial temporal analytics on some of these data sets combinations of various data sources it enables the pairs technology to improve the forecasting that is generated from renewable energy sources like the solar and wind and at the same time apply advanced machine learning analytics to improve the local forecasting at the specific locations on the globe so some snapshots of the technology it's shown where we could analyze for example that wind forecasting and understand what is the best locations where for example a wind farm needs could be placed and based on historical data and also some forecasting that it's coming from climate models or weather data we are able to identify as the best locations where these farms are going to produce maximum of energy such that now the investment could have a return of investment shorter compared with other locations in order to do this we have to analyze the weather data for example wind and also the variability of these conditions across multiple timescales and also analyze it geographically to identify the locations also we could combine this information with satellite information in order to analyze what is going to be for example cloud cover and then predict what is the best locations for solar generations and then try to match the production with the demand at various places across the continental US as these data sets are combined we have better ways or pools that are using advanced machine learning techniques physics and also statistical methods in order to identify what would be the uncertainty in some of these predictions and as one example you can see here it's going to be the cloudness or how much sunness it's going to be in a certain location versus the time horizon as we are extending the time horizon we are going to observe that the uncertainty it's increasing bringing a larger error and also some of these laws are going to show some exponential dependencies meaning that as we are expanding the time the uncertainty it's increasing also we have power law distributions combination of these techniques into an overall framework like Paris technology enable us to do a quantifiable power forecasting for every locations on the globe and at the same time quantify the uncertainty this type of informations could be used by energy and utility industry to better plan where to generate the power when and what would be the best conditions to meet the generations with the demand that it's coming from the consumers our first example or demo is going to be around solar radiation forecasting using the Paris geoscope tool we are going to visualize based on informations coming from various weather models the distributions of the solar power availability across the continental united states what we are going to see that some areas that are depicted here in red are showing regions where the potential for solar radiations it's increasingly high while areas that are going to be blue are showing regions where the solar radiation is going to be significantly lower the variability it's coming from the cloud cover that it's detected on every day we could analyze the time series for this particular data sets in order to understand what is going to be the variability every single day some days we are going to observe the day it's going to be maximum solar power generations while other days are going to show some decreased generations due to the cloud cover this variability it's going to be geographically dependent region in Nevada it's going to show high productions while areas on the east coast even on the north part are going to show much smaller ready small solar radiations that is generated all these informations it's really related to some fundamental physics coming from the cloud cover at the moment that dates high cloud cover it's going to be much lower solar radiations available for generations while areas where we have much smaller cloud cover it's going to generate more solar radiations this variability is going to be seasonally dependent and it's going to show areas where we have consistently much lower cloudness for example in Nevada or in Texas area while other areas are going to show increased variability or cloud cover that it's going to be in the northwest or it's going to be on the east coast in general the power productions in the ideal places are characterized by regions where the population's density it's going to be relatively small so what we see in this particular graph is going to be the distributions of the people per square kilometers and red regions are going to show areas where the density of populations decreased these are really the areas where the highest demand for generation for energy it's going to be high and accordingly we have to match the demand with the productions what tools like perioscope can do in this particular case is the possibility to go and match the production with the demand by analyzing patterns and historical data about what is the demand for energy in various locations and then placing a distributed solar panel or solar farms in those particular areas we could decrease the overall demand and at the same time match the production with with the demand in doing this type of technologies we could plan accordingly what would be the best area to place some of those solar panels and at the same time enable energy and utility industry to better manage what is the overall productions of the energy what would be the best way of the distributions and what's going to be the best way for the transmissions our second demo it's going to focus on wind forecasting for the state of Texas data from the pairs technology could reveal the spatial patterns of the locations that are ideal for wind power forecasting as you can see in the below map regions that are going to be red or yellow or in colors are region with a high probability of having high wind speeds and the directions of the wind to be consistent other areas that are represented by the blue or the greenish region it's going to show a much higher variability both in the wind and also in the wind directions time series analyses are going to show that there are some seasonal patterns in the spring time and the winter time it's going to be a high variability in the wind speed while in the summer time it's going to be a much lower variability there are also spatial distributions so nearby the Gulf of Mexico we are going to have higher wind while in other regions for example in the southern Texas it's going to be less availability of the wind for generations of the wind power the spatial patterns could be taken into account in order to better design where to place the generations of the wind power and at the same time to optimize such that the constructions of new transmission slides or the distributions could be minimized at the same time now wind at the same time it's also a risk factor many areas are going to have increased fuel availability and the wind can drive in various locations wildfire appearances and then distributions some of these wildfires can interrupt the productions of the wind power or the wind power distributions and a comprehensive view needs to be taken in order to understand what is the best locations for where the wind power could be generated and at the same time minimize various risks that are coming from the wildfire and in this particular image we're going to see some areas that are going to be red in color indicating that those regions are going to have very low risk of the wildfire and they may be also close to the locations where it's a high availability of the wind for the generations of the sources for a utility company both the generations of the power and the possibility to distribute to various customers is going to be highly important so in order to maximize the consumptions of the renewable energy or the wind energy in this particular case some of these built close to the distributions or the distribution the transmissions and the distribution wires such that we could maximize the overall consumptions of the power generated by the demand coming from the customers and in order to do that we could analyze informations related to the distributions of the assets that could be observed for a particular areas and analyze how some of this information in conjunctions with a possibility to generate maximum powers could be matched locally and then use smart grid technology in order to better manage the demand and also the forecasting type of technologies such that we could minimize the overall waste of the power in various locations and overall these new technologies could enable companies in energy and utility industry to better manage their availability of the power overall the renewable energy forecasting it's one of the advanced tool that enable that enable a smart grid operations many of the technologies that are required for smart grid for example variable price pricing of the energy or also demand forecasting really require a better understanding of what is the availability of the power at every single moment of time and combine that one with a demand that is coming from the consumers big data technology like the one provided by pairs provide us the tools to better forecast at the higher granularity both space in time and also reduce the uncertainty or the error in such predictions combinations of various sources that it's enabled by pairs technology it's opening the way to lower the overall cost of renewable energy and at the same time reduce the overall carbon footprint of the electric grid wonderful so i believe we have Leventi on hello sir how are you good to see you good to see you too um so uh a couple of questions in the chat but um one that that i have uh very curious about myself is um you know about network resiliency and i suppose efficiency um you know we could as we do have one giant monolith grid another option could could be to have micro grids that are local um kind of what are the trade-offs between those two from say a resiliency point of view from a you know cost point of view really curious to hear your thoughts on on on that so that's an excellent question george i think that the idea is that if you look we have the electric grid right and it's already there so you can take advantage of it you know in many parts of the of the united states building you know micro grid it's really just an investment that local communities needs to needs to do so they have to really come up with the generations which is mainly going to come from solar and then they're going to need some type of storage where they're going to be able to take this type of energy right and store it for the moment when they not going to be able to generate so from resiliency perspective you can imagine that if you have a micro grid and it's like a huge uh outage for the whole grid you can still maintain the power so you would have continuous power on but that's certainly going to come at the cost if you want to have certainly um you're connected to the higher grid then potentially the cost going to be slow smaller but at the same time you're going to have higher risk of managing you know the power so you're relying on the utility companies to provide this power for you yeah are there any examples of that uh happening in the in the US for example i'm not aware of any personally but are there any there are a couple of examples in brooklyn so uh i've seen examples where small communities they started to form and essentially they made the investment of generations and then also the storage and they are trying to somehow uh fence themselves from from the electric grid and and it's a viable options uh it's more popular in uh europe right now than it is in the US but examples of working and doable it's it's certainly something that it's out there and it's probably going to be the future that's really interesting i wonder how that will be interesting to see how that how that all uh plays out um great and then i guess this is more of a general question um but there's there's a cost associated with decarbonization um and so i guess you know what's your view on that i mean how do we best estimate that how do we prepare for that uh going you know down the road so i mean there are various policies that are out there so mainly coming from the US government that it's proposing that by 2035 we are going to have a carbon-free uh electric grid and electric generations and there are various studies that are coming from government agencies that are looking into how much it would really cost and it turned out that uh with just a few billion dollar investment less than 10 billion we should really transition by 2035 probably to a carbon-free electric grid and that's going to require potentially more solar more wind and eliminating some of the older generations that it's coming from coal and then the fossil fuels and potentially increase also the the overall balance of the nuclear energy such that we are able to compensate for these intermittencies or variability in the generations that it's coming from the solar and wind yeah it sounds like policymakers are going to have to have a big a big a big role a big role in that as well okay we're a couple minutes over uh thanks again for the great presentation and for the time you took to answer questions feel free to hop in the chat there are a couple of their questions there from the audience i'm sure they'd love to to hear from you thanks great thank you um so we are now on to the final uh session of the day uh we have a panel discussion uh this panel is led by Bianca Zidroshny who is a senior manager in spatial temporal modeling research out of our brazil laboratory and she's going to actually introduce our three panelists today they come from various there are various that are industry leaders in their respective industries so let's turn it over to Bianca and the panelists hi everyone thank you for the introduction george um as you said uh we are glad to have with us today three business leaders who share with her with us their perspectives on how their industry are industries are responding to climate risk and how they are adapting to reduce in fact as best as possible so start by introducing uh Hagen Hench who serves as the director of the distribution operating center west at uh one electric delivery so Hagen's organization is responsible for monitoring and operating the western half of of course electricity distribution system which serves approximately 3.5 million customers in this capacity he provides oversight over day-to-day system operations and service pressurations efforts as well as change leadership and process engineering for technology implementations so Hagen to start uh can you please tell us more about on core operations and how they are affected by extreme weather events caused by climate change i would love to thanks for having us on this panel Bianca and just to give the audience a little bit of background on core is um a transmission and distribution utility in the northern part of texas so we are a part of the aircraft grid and we are a platform or delivery provider only we do not generate electricity we do not sell electricity but we are basically the transportation system in our service territory that transports the electric commodity from the generators to the end users as such an open delivery platform we are really technology or generation agnostic and and and what commodities are being used to generate our technologies or end consumption is taking place on the other side of the grid so keep us in mind as i'm sharing our view and perspective how we respond to weather events climate change disasters in order to reliably operate our grid i'm obviously coming from a delivery platform perspective predominantly and as you might imagine this delivery grid that we are talking about the large transmission long distance lines the inner community distribution feeders are highly exposed to their environment to any weather situations just as much as to human activity construction activity drilling driving all those factors not just climate and weather affect the delivery of our services so thanks for having us thank you i will now introduce uh brad gamel co of earth solutions a company that provides a software as a service platform for energy gas and electric utility and municipality telecom and media industries to help ensure the reliability of critical networking infrastructure before joining earth as co brad was global managing director of ibm's energy environment and utility industry and he has also previously served on the boards of the terawatt initiative and the pulse on institutes council on sustainable urbanization he was the founder of the global intelligent utility network coalition and i found a member of the grid wise and alliance brad can you tell us how the industries served by earth are being affected by climate change and what is the role of the technology provided on their adaptation so enjoy being on the paddle of particularly glad to see hoggan it's been a little bit so actually glad to see hoggan we've had the opportunity to collaborate in the past so critical network infrastructure is um is kind of something that people walk over or walk around every day and don't i think have the fullest understanding until there is an outage of some kind or disruption of service um that it's criticality to delivering the energy transition which is you know a decarbonization is heavily going to be driven by automobiles going electric and um power sources that on core delivers um from the wind fields or the solar being delivered reliably across um our software company and many others are focused on how do you um ensure that you understand risk and where we're focused on to that infrastructure and that can come in a number of ways our where we we grew up is ensuring that with someone excavates or does construction that those facilities are not damaged but as we're moving into the future for the industry it's key to understand how climate events can affect um those that infrastructure itself and and damaging it um protecting current infrastructure um just that our gas pipeline systems here in the the u.s with methane released every year equals about 69 million cars on the road just from methane coming out of our long haul gas distribution network so there's the side of identifying things that you can control now that are um basically um failures of the infrastructure as it sits today that can be corrected but also how do you have the right information to make the best use of investment into growing the infrastructure whether it's electric infrastructure or primary the electric and distribute energy resources to lead us to to to lead us to um a carbon neutral future as far as far as the grid goes um understanding how climate will impact how economic growth in particular areas is going to create density of use where electric vehicles will deploy more quickly understanding how that infrastructure needs to be protected to allow that to happen is very important both for external risk and physical risk from those from those particular assets and so that kind of what I would call insights that technology like IBM provides for pairs combined with other data sets starts allowing you to have a much more economical way to start viewing risks and adaptations need to be made to that infrastructure to protect it thank you Brad so now we go to Jared last time who is vice president in Goldman Sachs sustainable finance group where he advances work across divisions in sustainable data and analytics corporate decarbonization climate risk and climate related investments Jared received the master's degrees from Harvard Kennedy school or where he focused on climate policy and finance uh climate policy and finance using a winning a university-wide prize for his working environmental economics on climate change finance and working as a dukakis fellow on environmental markets in the California government's office so Jared could you give us a brief overview about sustainable finance and how it actually facilitates good investment decisions in the phase of climate change great question and uh it's uh it's a pleasure to be here today so um I think broadly what we're seeing is a transition in financial markets that's taking place where there has been a growing awareness given I think increasing secular trends of the importance of incorporating particularly non-financial risk into financial risk assessments when you're making lending decisions when you're making investment decisions and there's no area that's actually more important for that incorporation than on climate risk and I think you know over the long term physical climate risk in particular which has the uh you know potential to really change structurally the nature of how capital markets work how risk in return will be realized what our major growth assumption should look like how much volatility will occur across the marketplace and even things as distant as inflation and how that will impact that impact elements like probabilities of defaults and uh and uh rates across the lending markets so a lot of what the work that I do is to think about how we can take the information from the climate science community and actually incorporate it into a variety of decisions that we're making across our institution and a lot of our peers are on similar journeys and so what I'll describe is really not unique to Goldman it's rather I think a broader you know set of trends that we're seeing realize realized across the financial markets the first is really across our risk division really thinking about how physical risk can be incorporated into lending decisions what level of concentration appetite we actually have for various physical risks across our portfolio and similarly how that interacts with transition risk and what sort of transition risks our lending might be exposed to in the future and how we want to mitigate that on a forward-looking basis we're also thinking about how to apply scenario analysis into broader investment decisions that we're doing on the base on the basis of our client touch and that involves looking at things like how do we actually think about changes to the structural macroeconomic environment on a go forward basis do we think that there are going to be there is going to be substantial as I mentioned long term volatility or depressed growth due to physical risk changes that will impact the broader environment how will that impact larger asset class changes whether that actually impacts the you know breakdown for example of how emerging markets might be able to perform versus develop markets what should that mean for our for our investors and then in the context of specific investing portfolios really thinking through granular investment decisions about is there particular concentration risk associated with municipalities with sovereigns with with corporates that might be actually exposed to physical risk in a particularly asymmetric way could there be clustered events or concentrated events that might actually cause us to realize substantial losses if a particular set of climate scenarios occurs and I point you you know to Hurricane Ida or you know what's happened in the past across the gulf in terms of hurricane impacts as an increasing of frequency and intensity but also the industrial cluster issues that we saw around the oil and gas sector with the recent event that has caused a lot of production to go offline or let's say the texas event that we saw more recently related to you know shock freezing happening across utilities and all of those things are required for us to address from a system resiliency and an evaluation of how that might actually impact financial returns from an institutional perspective. That was very interesting so I'll go back to Hagen now to ask if you can give us a little more detail about the kinds of technologies that you currently use the advanced analytics that you actually are employing now at Encore to predict damage and risk you can go into a little bit more detail that would be helpful. Yeah that's a very good question but I also think it is important when we when we look at what opportunities we have to deploy technology advanced analytics all these things that allow us to make better and more intelligent decisions more proactively anticipate and respond to these circumstances that we're talking about here that it's important to see what type of decision do we try to impact you know just like any infrastructure operational organization you can roughly organize your thoughts in three categories how do we plan an infrastructure system to better withstand environmental factors especially if they occur at a higher frequency at a higher severity those kind of things how do we need to reevaluate our design standards our design approach what kind of technologies and equipment is available that we can incorporate into our grid to make this grid more resilient so those are very basic and very important decision both to Gerritz and Brad's comments that will naturally evolve over time as these as this informed view that we have of the future will now inform investment decisions on our assets and incorporate technologies that may have not been available in the past so the data the insights that we are generating from both our predictions from our forecast what we expect to happen down the road will inform our today's investment decisions how we design structurally topography as well as from a technology and equipment perspective so how can we incorporate more automation the business cases will change some of that technology may have been available in the past but now with a revised view on the frequency and the severity of events you now have a better business case at a larger scale for deployment technologies the cost of technologies the last cycle cost all those kind of things have an effect on how we look at our investment decisions around the grid that's one category the second category is a little bit more focused on specific prevention you know many people talk about grid hardening where do we want to invest in underground facilities where do we still want to be exposed on the overhead side what kind of redundancy can be incorporated and go back into our existing facilities and saying what kind of measures can we take that will help us to better withstand any kind of events you know whatever nature that may be so that's a that these two categories are well-established practices where the new insights and data flow in to to drive and inform these decisions and lastly of course technology and analytics for that matter has a big impact on how we respond to these events so it's not just planning or a specific event prevention but now we're talking about how can we better respond to events how do we gather insights from our environment the situational awareness how we deploy our resources how do we understand what situation we are facing out on our system so that we can have a more effective and a faster restoration response associated with any kind of disaster so so that's high level kind of the categories and then we can dive in in each of those categories and saying okay so what specific technology what specific data points do we have what kind of activities can we identify here and and there's many examples if I go on the disaster response area as well as on the prevention side I want to piggyback a little bit on what Levente shared earlier it's not just the utilities internal data sets that drive that so what we're getting back from our equipment the meters the skater equipment our crews there's a lot of data points that we are collecting that inform our decisions and that we can analyze to drive better in decisions but it's also the remote sensing and environmental data sets that we can receive from the outside and I think a very promising technology there is clearly satellite deployments that really have changed in the past years there are lots of new satellites new data coming back and it is particularly interesting in that field because it is so broad scale so if you think of a large transmission distribution system like on course we're dealing with almost 140 000 miles of lines there's a lot of possibilities when we look at one particular location what we can do and how we analyze it but it becomes very challenging to identify the high risk and the response options when you look at it at scale over 140 000 miles there's limits how many eyes you can put on the system like this at the same time just physical limitations and so through remote sensing satellite imagery that is detailed enough and produces actionable insights for us there's a lot of opportunity there that allows to quickly identify those areas that deserve the human attention deserve resources another example is obviously in the in the planning area where we really try to identify what is the population growth especially here in north texas we are overwhelmed by just the investment volume both on the commercial and residential side we are connecting almost 100 000 meters new every calendar year just keeping up with the construction volume changes the topography of the grid how do we augment the system to address both the accommodate the growth that is taking place as well as the the the operational risk challenges due to weather so so you it's not a siloed only decision that all we are doing is only geared towards accommodating a big hurricane or tornadoes or a bad winter storm our decisions obviously are also influenced by where is where can we accommodate the new growth how we can get a timely construction completed to to address new factories new homes new subdivisions and all these things funneled together and and you can see how data analytics and and and complex modeling really helps us to make better decisions in all those areas thanks just to complementing in this planning decision how far in the future do you do planning for for encore I mean it's like how many years ahead do you start planning a new transmission or can you complement a little bit on that yes I mean as you might expect in any business you have you have different stages in your planning horizons so we would typically go out probably no further than five years that's a realistic planning horizon for us just you know the uncertainty increases dramatically if we go past that horizon so and then you take it back you have your high-level plans you expect where your growth areas are around your metropolitan areas where new subdivisions new growth will take place based on land use you know so you identify your redevelopment areas your your your expansive new development all those things and you get a general idea in what direction the growth in a population center you know which is different in the rural areas versus in the metropolitan areas so you have a general understanding where your growth goes and you try to position your facilities to accommodate that growth without incurring the risk of stranded investments where you're building over capacity where you're burdening your right payers risk cost that we shouldn't so that's kind of the long-term view what I call five years then you bring it down to you revisit that every year and then you basically have an annual review where you're revising your plan and make specific construction scopes that that then will be executed to expand the capacity in areas to improve the reliability due to maintenance or to accommodate specific customer requests that that require construction and so the annual plan then is what we execute and then but it's all based on the five-year plan that goes out thanks so going out to Brad Hagen mentioned that you know the availability of new data sources like more satellites is helping their could help in their case to be able to more practically make predictions also assessments of damage to have a faster response so in your case for earth can you mention a little bit about the kinds of data sources sensors satellite etc that you use and and how they you see changing the future the types of data for monitoring and response yeah i think he's hog is exactly right and it also plays into some comments Jared made is to really take advantage of the period of time we're in we're heavy investments going into infrastructure in north america you know the infrastructure bill has 73 billion dollars in it just for grid infrastructure if that that happens so for a period of time where there's a focus is how do you spend that well to reinforce the current infrastructure against climate impact and also a different growth pattern in our economy right now where that should be placed properly to manage risk i think it's also interesting right now and i'll get to the data that there's also mass support for deployment of 5g across the country and what's interesting is that smart grids and telecom go together at the same time because you have to have the communication infrastructure in order for an intelligent grid to be more responsive and reactive and resilient the way it operates so to this kind of asymmetric point that jared brought up these these linear networks are all linked together you know it's the telecom network it's the electric network itself as we learned in texas how you're how you're protecting the gas infrastructure matters because that's a key fuel you know for balance of the stability of the grid itself when you start looking at that it means you need to look at this linear infrastructure that delivers these services holistically and the only way to do that is look at all the potential risks that that are out there it's the age of the assets that are out there that the utility companies own it's the cloud changing climate conditions it's not just storms if you move out to the west it's temperature of the ground where you have pipelines buried potentially and you have rising ground temperature that can affect the fidelity of pipelines themselves in the delivery of their services so being able to merge together climate information current information of the environment that the physical asset is sitting in understanding the complementary nature of those different services coming together in a region and how they have to work together to be resilient to deliver the full service I call them fuse systems is you is bringing all that data together in a view that you can then query against scenarios of what what would happen today what would happen five years from now and start judging economic impact what what kind of industries are in that area how critical are the industries and the logistics they provide and the goods they provide where I think that's really important to bring that together is then that helps you direct investment it helps the the utilities of the world make better cases to regulators banks put pressure on and financial investors also of where they see the need in order to deploy capital you've got to have insurance that the economic vitality is there you need to understand what social services are in those areas and understand impact like that so I think this whole area of fusing all these different data sources of what's the economy in that area look like what is the environment what's the physical infrastructure what is climate change going to do bringing those together in a way that you can analyze and plan it is very very very important for the way cities plan the way banks plan the way utilities plan and that's something we're very focused on working with company like IBM but how do you bring those data sources together and make it easily queryable so you can start doing the kind of analysis in the planning phase you know if you're trying to basically adapt or the things you want to be doing right now to mitigate like I mentioned with pipelines where you've got you know leaking pipes of methane that are kind of undoing all the good you're doing with wind farms in west texas or when you have a big out power outage in california or in the third world you start turning on diesel generators and mass and there's in in the development where there's about 300 gigawatts of diesel generators there that you know you know again offset all the good work done in other places if you don't have a reliable electric grid so I think data is at the core of you know how we go into this more sustainable future thanks bread that makes a lot of sense I like the example of the pipeline temperature soil temperature so going to back to Jared I think that you can also add some perspective in terms of the kinds of data that you use in your within financial so in the same sense that Brad and Hagen commented can you give us a little bit overview of what kinds of data that do you and different sources and also types of data doing corporate and how you do that for different problems in the financial industry yeah absolutely so I mean when we think about physical risk you know I tend to break down the world in a similar way that the insurance industry has broken down the world historically which is you have hazards which are what you know particular companies might be exposed to how is the climate going to change what is the increased frequency of flooding of extreme heat events both chronic and acute and then you have exposures so are your companies actually located in a particular area where those hazards occur and then finally you have vulnerabilities and so vulnerabilities are let's say that you have an asset that is exposed in a particular area to a particular hazard and that hazard occurs what's the actual damages that are going to take place contingent on that you know chain of logic in that hazard occurrence and it turns out I think that we have a pretty good way as financial institutions and increasingly good technology that's out there on the hazard end there are a lot of different companies that are serving the financial services industry to really think about what how how is the climate going to change how does that actually materialize at a more you know localized geospatial level what does that look like in terms of you know multi hazard events and different frequencies occurring how can we think about that in terms of broad based or broad brushed geographic exposure but one of the things we don't have a very well right now is actually on the front of knowing where the locations of the some of the companies that are invested in are as you can appreciate this is not something that's probably that is widely disclosed across all industries there are sometimes national security implications involved in that public disclosure sometimes there are economic competitiveness implications involved in that disclosure and there isn't really a disclosure standard or method or financial institutions to actually get access to those geospatial locations so there are emerging developing data sets out there that allow us to look at where some types of businesses might be located but even when we can actually locate those businesses and know where the facilities are we often don't know how important those facilities are which is another critical step in the logic chain so if a facility gets hit is that 80 percent of production is that 10 percent percent of production for a particular company how important is that from the perspective of financial disruption the hurricane comes through and takes that production offline that's a question we can easily answer right now and has a lot to do with the lack of facility exposure data and then finally we get to the vulnerability and things which arguably to me from a financial institutional perspective is actually the most important for us to understand but also the hardest piece of information for us to get access to which is what actually makes a company vulnerable you know and Hagen and you know Brett have actually done a really good job of kind of laying out a lot of the interdependencies when it comes to broad-based you know grid resilience and some of the interdependencies on things like fiber optic networks and actually understanding a long value chain in terms of how it's installed what some of the underlying resiliency conditions of a grid might look like and that's something that we may not have the access to from the perspective of a financial institution but it is actually something that we can have influence over so when we talk to companies and when we engage with companies one of the most positive points of engagement we can make as a financial institution is actually to talk to companies about enhancing their resiliency issuing specific financing that will allow them to actually increase the level of resiliency that they've in adaptation that they've built into their their systems whether that's a grid or location of a generation you know location of their generation moving it from a basement to a third floor in a flood from district there are plenty of examples out there that we've kind of explored to actually be able to you know not only think about assessing risk but building resilience and that's an area where I think we do need things like new types of disclosed information disclosure standards potentially to really understand what are the actual adaptation and vulnerability actions that can be taken in partnership with the companies that we lend and invest to in order to build that resilience over the long term one thing I'd quickly add here is that nowhere in my opinion is this more important than when when we think about the concentration of risk in emerging markets when we think about resilience and adaptation and the ability of you know develop the physical risk you know one of the critical dynamics here is that physical risk will will show up in an asymmetric manner for a variety of reasons across developing countries now does that mean we should disengage from those developing countries entirely because they have higher risk concentrations on the physical end I worked in emerging markets for years and my contention would be no you know one of the things that we need to understand is what we can concretely do to actually build resilience and enhance the financing so that we don't have those physical impacts in the future without those deeper understandings of what is actually driving vulnerability what actually is a proactive adaptation response it becomes an exercise where we just look at risk and don't actually think about the things we can do about it as financial institutions yeah yeah that was actually very good perspective right that you don't just want to estimate risk to make decisions but you actually can cause change by actually helping people address the vulnerabilities and that that's very very interesting so going back to to Hagen can you tell us a little bit about what you see going more into the things we're closing now that what you think would be like the future let's say the perfect future in our for on core in terms of both the response and planning that you mentioned before in terms of how the technology would evolve and how the operations would run well I think in many ways I just realized this to be honest in many ways you know as an operational manager in a grid company we really look at the world in our grid very similar to how Jared described how he looks at his investments so if I you know in many ways however we augment however we invest in our grid and for that matter even how we operate and run our facilities today is very much how Jared would look at his portfolio of investments so all these factor that he describes are exactly the same factors that we have to look at with every investment decisions that we make on our grid and for that matter it's a life cycle decision that includes maintenance, life cycle what is happening down the road as well as up front from an asset perspective so in many ways the drivers to make good or better investment decisions in a changing and uncertain world are very much the same and so if you're asking me what would be the perfect world look like is that I have data on everything and the perfect algorithm that eliminates that uncertainty so that we can investment decision with a high degree of certainty without the risk of stranded investments or under investment in the areas where we should have done more so more data and more intelligence to turn that data into reliably actionable information so that's the perfect future that I pray for but obviously that is impossible but what we're dealing with is we are making progress in that direction on one side I think we all recognize it's not just climate change it's changing customer expectations it's changing technology it's the changing society how we use electricity how we depend on electricity you know 20 years ago if you lived in the country you had a different attitude towards your electric service than you have if you live in a metropolitan area you know was in a connected life so that there's a lot of changes around us that are going on that are driving these uncertainties and driving the change that we need to adapt to as we provide any infrastructure service or make investment decisions and so what we have to we are making progress there on one side there is the volatility because of the intermittency of renewable generation of all the change that is going on around us so the uncertainty is increasing and makes our investment less attractive and more uncertain and we know it's less quantifiable than maybe in the predictable world of the past on the other side we're producing more data we are having promising technological advancements that now can address that uncertainty and the volatility that we fail and so we really need to watch both sides predict and understand these changes that are going on around us with open eyes and then really not just make investments in our infrastructure itself but also recognize the value of information and data that allows us to make more informed decision on those assets so in the past you know there's an intangible assets called data that was just a cost center now we're recognizing data to be more of an asset because it allows us to address and minimize risk and uncertainty in our investment decisions and so that's the future we're moving towards the volatility will continue to increase the uncertainty but we're also producing more data and more technology that allows us to manage and handle that volatility and that's a little bit conceptual so but I think it finds its expression in those three categories that I outlined earlier we're making better planning and investment decisions we are smarter how we can prevent and harden our grid so data and technology will help with that and we are smarter and better in how we respond to disasters by having data available having people with mobile devices out there having satellite images out there so so in all three categories we're getting better and so each company needs to gauge itself how are we making progress how are we exposed to these uncertainties and then how with our data investments with our intelligence that the insights that we're gaining uh how can we address that to bring that risk back down to to stay attractive for our investors thank you so I'll give a also bread and cherry that chest to to wrap up as we are running out of time so I'll please bread if you can also mention a little bit more I like your you mentioned about 5g and how this is it going to impact the the kinds of decisions that we make so you can mention that in your final course I think Coggan summarized very well it's just I think you've got to look at the interrelatedness between these physical networks whether it's the 5g gas electric and also transportation networks you know the rail systems and the road those interdependencies between those have to go into a risk assessment to vitality of industry and community behind those networks and the way you look at them and so Coggan said it perfectly that to do that efficiently and effectively and I think the term he used reliable information trust it because if you're in the finance area or you're an actuarial in the insurance industry it's about trusted data and you know that's what we're very focused on and working with companies like IBM is how do you get the trusted data across these things to make good decisions with jaren yeah I think just in terms of you know high level summary of some of where we are and where we should go you know are the journey for financial institutions really I think started you know very recently on fiscal risk so it was really I think accelerated and brought to a lot of people's attention you know six or seven years ago um and so this is a it's a relatively new field and when that all kicked off you know the focus was on awareness how do we build awareness of physical risks as a factor within financial institutions and I think many of us most of us are painfully aware of physical risk dynamics being real affecting uh affecting financial markets in a variety of ways in creating systemic risk and I think where we're at right now is trying to find out new ways to use the data uh and and uh broader universe of resources including engagement with companies to move quickly from awareness to impact thinking about what we can concretely do to make change how do we execute actions that are going to build real resilience and what is going to be the unique role of financial institutions in connection with uh corporates around the world and municipalities and sovereigns in building that resilience over a longer period of time. Thank you very much for the great pleasure talking to you three today I hope everyone else has enjoyed as well so I think Hendrik now you can go back to George. No thanks thanks Bianca and thanks Brad, Jared and Hagen for a really wonderful conversation a very insightful conversation and thanks for participating in our in our event. I'm going to turn it over actually to Hendrik. Hendrik opened our event and he's going to close close our day here. Hey Hendrik, those of you that weren't there in the morning Hendrik is our chief scientist in our future climate and I will leave it to Hendrik to close out our day. Yeah so I guess I have the impossible task to conclude today's sessions. I always like to look back and summarize what's sticked and of course I will not be sorry because we just heard too much but if you recall our first keynote speaker I think reminded us quite well about the human dimensions of the impact of climate change all right it's not just about the numbers it is it has real impact on people on businesses he also reminded us of that we can only be successful if we collaborate now the geospatial analytics and impact of this demonstration by Anne I felt may clear how impact modeling has become so easy so quick now it's never quick enough and easy enough but we're really catching up what Hagen described and Jared described with the increased volatility and technologies trying to keep up with addressing some of that they also show quite nicely how it has increased the accuracy of the predictions right and how it enables to understand the context of climate impact information right the vulnerability in these questions the presentation by Campbell and the chat with Campbell Bianca and so on hey reminded us that IBM is actually the leading weather forecasting company right using AI for a long time but it's not just about the weather right it's much more it's about the decisions you're making and we also learned how we can address that challenge of uncertainty which is inherent to forecast and but how we can actually use that information also to make better decisions the application discussants by Chitendra and Smith and Levent and of course a great panel we just had with Jared, Brad and Hagen also gave us a little insight on the real applications how to adapt to the impact of climate change so it's not just theoretical right it is about the right decisions it's about preparing tomorrow we have so we're not done so tomorrow we have an exciting agenda where we actually look at the role of the public sector and we'll kick off with Donald Wubbles who's a professor at URUC he is a presidential fellow he's a contributor to the intergovernmental panel for climate change so you cannot miss that right make sure you're coming back there will be a second keynote with a very intriguing title the climate dance in the triple helix by Lux Research also by an excellent speaker we'll then hear from two leaders from IBM Kumi and Centeno about our technologies and of course we're also going to conclude with an exciting panel tomorrow with government leaders so let me close by thanking all the participants thanks so much the speakers panelists for your contributions and we see you tomorrow thank you