 Here we are. Hello. Welcome everybody to the next lecture which is about trees and how artificial intelligent gents might save them and two persons are here who will talk about this. One of them is Markus. Markus Fossey is an AI expert and intelligence architect at Birds on Mars and he's also active as community lead for buildings and transportation at Climate Change AI and the other person is Mirjan Riegel and she's a senior product manager at Technology Foundation in Berlin and what brings them together is the project Q-Trees and that's the project what hopefully you will save some trees. Hello. Yes, okay. Yeah, thanks for the introduction and thanks to you all for coming. I know you must be hungry, so I really appreciate that you stick around to hear our talk. Yeah, so Markus and I work together on a project called Q-Trees together with the Greenery Department in Berlin, Mitte the project is funded by Zug on behalf of the federal ministry for the environment and is also part of the German strategy for adaptation to climate change. Now in a nutshell, what is Q-Trees all about? It's an intelligent prediction model to help optimize the watering of city trees based on AI and I'm going to tell you in my part of the talk a bit more about the why we're doing this and then hand over to Markus who will explain the how and explain to you the model that we've developed. So Berlin trees have been around for more than 300 million years, right? So this is supposedly Berlin's oldest tree the Dicke Marie that is standing somewhere in Tegel and has is 800 years of age. They've seen a lot, right? They're very resilient plants. They can live up to thousands of years and one could think that they can they can handle big crises because that's what they've been doing in the past, but something has changed. The climate has changed and it's not a general claim. It's a problem on our very doorstep. I'm sure you've seen already a similar representation to this. This is actually a chart that is showing the changes in temperature in Berlin and Bruntenburg based on an on an average reference value between 1971 and 2000 and long story short temperatures are rising. So the the three hottest years have been recorded in the last four years and this is of course not not great news. Also the precipitation patterns have changed. So again, we're still in Berlin. Here you can see the average daily rainfall in millimeters between 2015 and 2000 and on the on the first glance you might think well in recent years it has rained more so that surely must be good for the trees, right? But there are some anomalies here for example the drought in 2018 which had a massive impact on trees experts are still seeing today. So let's zoom in on that here and here you can maybe see why 2018 was particularly problematic. So in February you can see that the rain was missing at a very crucial point in time. During growing season for trees in the build-up to spring and then in July on the other hand there was a lot of poor down heavy rainfalls and it rained 105 millimeters in only two days. The effect of that is that water just drains off the ground above the surface and trees can't actually benefit from that. In addition to that in winter it rained a lot and that's a point in time where trees also can't benefit because there's not much water needed. So to sum up we have direct effects of climate change, rising temperatures, changing precipitation patterns that lead to droughts, heatwaves, more storms and negative effects on the environment. That's not all unfortunately so in the urban context trees are also facing very city-specific challenges. I took some examples here. This is not a full complete list. Just some of the stress factors trees are dealing with, right? So for example, you have small tree pits, which means you have a high level of ceiling of soil surface, prohibiting trees from absorbing the water that is needed. You have also misuse of these tree pits, for example by littering, soil compaction due to construction sites. Then there's salt, I think about urine, road gritting in winter and then of course especially in summer, sunlight reflection, right? So for example from skyscrapers around the Europaplatz is a common example. This works like a burning lens and causes the trees even to dry out further. So yeah, you can see some of these effects of these stress factors on the left-hand side. There's a young tree that should be blossoming but is already dried out and on the right the quality isn't great but maybe you can figure out that the top leaves are already with it. So of course the greenery departments are already doing as much as they can to help save our trees and they do a lot of watering, but we asked ourselves isn't there a more efficient way to support this watering, this water management? And this is basically how the project came about. So Qtris is all about saving city trees with an intelligent prediction model to optimize these watering activities. So as mentioned at the beginning there are three core project partners on this project. So first of all we have the greenery departments in Berlin, Mitte and Neukölln and these are our botanic experts. They have heaps of knowledge on all things water management, tree maintenance, green spaces. Then there's us the Technology Foundation Berlin. We have a lot of expertise in running participative digital projects because we work at the intersection of civil society, administration, business and science, bringing everybody together and working on solutions that help make Berlin a more sustainable and better place to live in and we do the coordination of all the project partners across this project. And then last but not least birds on Mars, our AI experts, I'm sure Marcus can tell you a bit more about what they do, but it's it's an agency that focuses on how AI can come up with solutions that help society that inform interesting digital projects. So handing over to Marcus, who can tell us a bit more. Okay, okay, so I'm gonna tell you a little bit more about how we actually do this and so something that I have to learn as a student, or I didn't learn as a student, I had to learn then as a researcher and a data scientist. So when you're doing actual data projects, if you want to do successful data projects, you're not starting by looking at data and coding. And so this is also what we didn't do actually in this project. So we, before we used any visual code, Jupiter notebooks, whatever, we filled a lot of my robots. I don't know if you know my robot, this is like a big brainstorming tool and we did a lot of expert interviews. We did a lot of brainstorming with experts. We asked the greenery departments, but also external experts. So because of course, I mean, we all have plans at home and know how to water them, but we are not experts in watering city trees and we would never claim to say, hey, we can come with AI and solve it just by doing that. So that's why we first talked a lot with people, listened a lot to then figure out, okay, what could the solution be? And what does the solution look like? So like Miriam already said, I mean, trees on its own would not need water. But since we are making it so complicated for them here in the city and then the climate is changing and some trees that have been planted 50 years ago are now not used to this climate. They need to be watered and they actually watered, right? So the greenery departments go out and give trees 200 litres, 300 litres of water of each tree. And that's quite a lot of water and we are thinking now, okay, can this be a little bit more fine-grained? And what you can do then is also do some physical modeling because I mean you understand the ground, you understand the physics of the tree. This would work in theory, but you're kind of missing this feedback from what's actually going on. And like we said, it's very complex in the city, so that's why we didn't go this route. But instead we looked at sensors, right? So we're using soil sensors. Actually, we're using soil tension sensors, which are, it's a very detail, but it's basically simulating how a tree, how much water is there for a tree in the ground. And here you see Julia also from the city lab, looking at the sensors that we have there. Loravan sensors, they're sending the data somewhere that we can access them. But of course we don't want to put sensors everywhere, right? I mean putting sensors to every tree would not be very sustainable. This would be a lot of hardware, so it would not be environmentally sustainable, but also not from an economic standpoint. It's just too expensive. So what we want to do is we want to use machine learning to learn from all the relevant context information, make predictions for all the trees that we don't have sensor data for. And so what is the data that we're using there? First have a look at the sensor data. This is what it looks like. We are measuring this in three depth. And here you can actually also see that also in this march it basically didn't rain in Berlin, and this caused the sensor values to go up, which in this case means it's getting drier. And what data we're using then to make these sort of predictions, these sort of generalizations to trees that we're not measuring, of course weather information. Yeah, soil tension, it's a little bit too technical. There's actually two types of ways to measuring how much water is in the ground, and this is one way. It's a big debate and it's a little bit of a topic for the bigger crowd, I think. Yeah, but it's how dry or not the soil is. So we're using weather information. We're using this by one kilometer by one kilometer squares from the Deutsche Wetterdienst to then also have this information where in Berlin did it rain, because of course not everywhere in Berlin there's the same amount of rain. So this is quite useful. Then we of course need to know how much was where the trees watered. So this is where we have two sources. One is actually from the Greenway Departments. This is actually the only closed data we're using so far because they're currently only providing it to us, and it's also the least structured data, I have to say. I mean, if you've worked with real world problems, this is, I mean, it's Excel sheets, and they're not all having the same format and not always the same format, so it's, but it's data. And then we have the data from Geese and Keats. So this is another project by Technologie Stiftung. Some of you may know this project. So this is also where you can go out and if you water trees in your street, you can actually put it in there in this database, then we also have access to this data for our prediction models. So now we have the watering data, the weather data. What we also use, because we also need to know if the tree is very big or very small, because that also depends how much, yeah, how fast it's drinking basically. And so for that we can use Berlin Open Data. So if you know Berlin has an Open Data platform, it's actually not completely on the screen here, but it's daten.berlin.de. So there's a lot of open data from Berlin. It's, the user interfaces are very cumbersome, but so it's kind of hard to get the data, but it's possible and it's there and it's very cool data. So if we have data for 800,000 trees in Berlin, what type of tree is it? How old is it? And so on. And we can use this for our prediction models. And then we noticed in our, this is actually what we got from the expert interviews, that the shade is a very important factor. So right? I mean, you may not need to know this also from your trees at home, plants at home. So if they're in a shady place or in a sunny place, they have different water requirements. And this is true also for Berlin. So we calculated a shade index for a whole of Berlin. So I actually can show you what it looks like here in front of this building. And you can actually see that the trees that are in front of the building, so black means they're not getting, getting almost none of the sun that's available during a day just because of physics. So the sun is going around here. And the trees that are up there are getting a lot of sun. And so this is a very big influence. And so we're calculating the shade index and use this also to inform our model. Yeah. And this is basically what it then looks like. We put all of this information in a machine learning model. And here we're using random forests, not just for the sake of this meta joke of trees making predictions about trees. No, it's actually a very useful model if you have tabular data. So if you're a machine learning expert data scientist, you know this. So it's very useful. And then we can make a now cast basically predicting those sensor values where we are not measuring them. And then we try to predict that also for the next weeks ahead, which I mean, the accuracy mostly depends on the weather data that's available. So we are evaluating then within this and next year, how good this is. Okay. So that sums up how this model is working. I wanted just to share with you some of the lessons learned from this project, which is we are working with CCI also think also is applicable in a lot of other contexts. So when you want to use machine learning and AI in the context of climate change, you have to know that all these projects are really, really interdisciplinary because I've never met a person that's really expert in botany and machine learning or electric engineering and machine learning or I don't know, biodiversity and machine learning. So these are just too big like you would have to a lot of PhDs to get this. So that's why it's really important to collaborate to form teams to really talk to each other and listen to each other. Otherwise, you can't really solve these problems. I've seen a lot of also in my research career, a lot of work where people just solve the wrong problems. So that's why it's really important to actually do this. And then also, I mean, our title is a bit, I mean, AI will not save our city trees. There's a lot of other things that will save the trees, but AI can be one part of an overall solution strategy. So this is what I at least believe. So you have to realize AI is not going to be the silver bullet that's saving our city trees. Yeah. So I'm handing over to Miriam. Yeah, that's about it really. We'd love to hear your feedback though. Because as you can see on the slide, we already built something. So on the left is some screenshots from our prototype. So the solution being a web-based application for civil society. People who are interested in the topics want to find out more about the trees in their neighborhood and participate. So Marcus and I, after the talk, will be running around and would have a few questions for you and would love to show you what we've built so far. And yeah, on the right-hand side, it's just a teaser. So that would be or will be the second solution we're building as part of this project, which is an expert dashboard for the greenery department. So again, if you have subject knowledge, some ideas, input feedback on how such a dashboard could look like, again, to optimize the water management for watering city trees, then please come talk to us. And that's about it really. Last but not least, City Lab is hiring, Technology Foundation Berlin is hiring. Just putting it out there. So we're currently, for example, looking for a product owner to join our team in our prototyping lab. It's very exciting work. You get to work on a lot of exciting projects. So have a look. And I think Marcus, you're looking for people as well. Thanks a lot. Thank you. I don't know, it's over or the talking time is over, but we said as it's lunchtime now, if there are any questions, like for five minutes, it would be possible to do that. Otherwise, I think you are around here. Okay, so thank you very much. There is one question there which we could take, I guess. Thank you that I can ask a question. As far as I know, there's a very sad situation that there are so many trees in Berlin that are ill right. So that's also the reason why you're doing it, I assume. And is it also in your plan to like give the decision makers like to give them information about a trade-off, like whether to rather water this tree and not the other? Because I mean, what we're facing now is scarcity of money in order to water them. So would that also be a solution to save as many trees as possible? Thank you. So you can add on to that Marcus. Yes, so I mean, the idea being we're very aware that water will be a problem, the scarcity of water specifically, right? So optimizing the watering activities of the greenery departments is that priority. And then of course, our hope would be that it's easier for them to see on that dashboard which areas are in need of water in comparison to others, right? So hopefully with this model and the data we will be visualizing, it's easier for the teams they're sending to actually take these decisions and make priorities about which trees are in need and which aren't. And maybe just to add to that, according to damage, that was actually some data I wanted to share in the talk which of course I forgot, so I need to make that slide. So based on the Berlin administration, we know that as of last year, they have reported that more than half of the inner city trees are damaged. So it's really critical in terms of damage, but then also dying in general of city trees, right? So between 2011 and 2021, the city planted 30,000 new trees, but also 60,000 trees had to be cut down. And this is also not only but also due to drought stress, right? So very alarming numbers. Yeah, just to also quickly add my five cents, so we are not really doing an automated irrigation system right now. So in the end, it's the human beings that are still scheduling everything and sending out the people. So we are just providing them decision support, I would say. What they do in the end with that, it's something that we also have to discuss next year when we build this dashboard with them together, like what's the information that are really, really important. So we can show that, but maybe we can also show some thresholds. And right now they're also prioritizing already, right? So right now they take age mostly, so they basically start with the youngest trees and go towards the oldest trees. And we hope to use something like shading information, for example, to also fine-grain this a bit to basically say, hey, maybe take the first one in the sun, give them a little bit more than the shade a little bit less. So this is the basic idea of this project. Okay, I think we got one short last question. So I think you already answered it. The project is not finished, I guess, and you're continuing on it. What are your future plans? Because also in regard to what was said before, it sounds very reasonable to take this health information into account and stuff. Is this on now? Yeah. Yeah, so the project is still running. So as I mentioned just now, we have the first prototype out for the app for the civil society. And then as Markus mentioned, what's up next is the expert dashboard for the greenery departments. So this is why we're here today to gather lots of input and feedback also from you, because we're currently working on this and the project will officially run for another year. So September 2023, that's roughly the time range. But regarding the technical developments, it will be mostly focused on the sort of forecast for this now and the next days. So I mean, this is because if you ask for money to funding, you first have to do what you promised. But of course, we also hoped, I mean, machine learning models can also help you inform what's the sort of aspects for a certain tree that maybe like a certain location, the age, the shading and so on. So we hope also to get some ideas of this long term. Is there's a good position for trees? There's a good tree for this position, but it's not part of this project. So it will be touched maybe. But of course, we're also trying to see if it's possibilities to extend on this project, maybe follow up on that because we kind of can motivate this, but we can't invest a lot of manpower, women power to work on this problem here. Hey, perfect. Thank you for this bits and boime project, which fits really well to this conference. The rest of you enjoy lunch and have a good day.