 Can you hear me? Checking the mic. Checking the mic. Good morning, everybody. All right, so I'm going to talk to you today about coupling natural human systems at the decision-making scale. And I'm going to do two things. I'm going to talk about coupling models in the first half. And then in the second half, I'm going to say why we're trying to do this at the scale of the decision-maker. And the second half is really in collaboration with my student Benjamin Minin, who's up here at the front. So I flew here on Monday from Southern Ontario, University of Waterloo, which is about 100 kilometers southwest of Toronto. And this landscape in Southern Ontario is incredibly human-dominated. There's about 13 million people in that area. There's about 52,000 farms. And competition for land is really intense. Housing market is really expensive. And there's a lot of drive to increase and sustain profitable farming for small family farms in the area. And this isn't just true for Southern Ontario. This is also true around the Great Lakes, and in particular around Lake Erie in Michigan, Ohio, Pennsylvania, New York, and the US. Now the basins around these lakes are shown in this slide here. But Lake Erie in particular, shown in yellow, has this basin which incorporates the majority of those farms in Southern Ontario. And the runoff from those farms is leading to massive eutrophication of Lake Erie. So we're getting massive algal blooms, which is creating a lot of toxicity. It's leading to fish die off. City of Toledo has had to shut down the pipes from extracting water from Lake Erie on multiple occasions. And part of my research program is to try and mitigate these issues, like many people in the room today. And so through that program, we're trying to quantify the impacts of best management practices. These are things that farmers are doing in their activities, try and reduce the amount of runoff, the amount of nutrient flow to the hydrological system. And we're trying to identify critical locations for applications of these BMPs, learn how to better incentivize farm adoption of the BMPs to generate more sustainable land management practices. But of course, all of this really requires decision-maker buy-in, requires data, and of course it requires models to prove that mitigation practices work. It's like a good modeler. I went back to my modeling process and I thought, okay, we're gonna formalize what system we're working in, use a conceptual model to foster discussion, operationalize that in code, create a computational laboratory, work through some results and keep iterating, improving the system that we're working in. But what we see is we got two camps of work that are ongoing. Got a group of people in the human systems. We've got a group of people in the natural systems. Some cases we get interaction between those groups or at least in terms of the modeling that's represented. Okay, so we'll have a lot of human system modelers that will do a really crude representation in the natural system, maybe using an inventory approach or benefits transfer approach, where they'll say, we have, for example, land cover change, then we'll have average value of carbon storage associated with that land cover change. So if we lose forests, then maybe we're losing temperate system, 9.8 kilograms of carbon per meter squared, something like that. It's a really crude representation of what's happening. On the natural system, there's a lot of prescription going on. We're prescribing how human activities are changing that system, but there isn't this dynamic feedback between the systems and interaction that's happening. There's very few people that are doing that. And some of the separation is put in there, partly because of the funding initiatives that are out there. Okay, so in Canada, we've got the social sciences and humanities research council. They provide 388 million towards social sciences and 1.1 billion to the natural sciences. So social sciences is only getting about 26% of that funding. We're seeing similar outcomes in the US and these are just numbers that I'm pulling off the internet so don't hold me to them exactly. But in the social sciences, we're getting about 250 million. And in the natural sciences, we're getting one and a half billion to math and physical sciences, 906 million to geosciences and engineering. And so the social sciences really getting about 8% of that. Now there is those initiatives coupled with natural human systems, which are coming out in there about 18 million this year in funding, but that's only going to fund six projects, maybe seven. So there's this embedded disparity between social sciences and natural sciences. And then there's this focus on these different funding agencies which are kind of keeping us separated. And so one of the ways that we can start to harness this is that we've got specialists in the human systems, specialists in the natural systems. And instead of having these crude representations of interaction between them, we can harness those specialists and their work through coupling models together. When I talk about model coupling, what I mean is a coordinated communication between models. And so this figure on the right is showing on the x-axis, the frequency of communication and the degree of coordination on the y-axis. And the majority of this coupling that's taking place is in the center row, which is obviously hard to see from the back of the room, but it's a unidirectional flow from one model to another, kind of like our prescription that's happening in the natural sciences. Okay, and this is true for most of us in many cases. This is my research, I do this all the time. Here I've got a human dominated landscape in Ohio with forest cover and ecosystem models that wanna represent ecosystem function like carbon storage or point-based. We multiply, it represents the vertical system very well. We will run those models and then we'll multiply them out against the area that those models represent, kind of like in a hydrological model, we'd use HRU and apply that to the landscape to estimate evapotranspiration, carbon storage, run off those sorts of things. But we know that's not the case. We've got patterns that affect how ecosystems behave. And in this case, I measured air temperature treatments and humidity from the edge of the forest, the interior and reparameterized the models along different swaths of edge effects to show that we get massive amounts of carbon storage differentials depending on the size and shape of the patches. We can transfer this to the human system as well. So here's some work that we did where we were coupling an agent-based model to BioNBGC ecosystem process model to investigate how land management practices on residential landscapes affected carbon storage when we're adding to the landscape through irrigation of fertilization or removing coarse woody degree, grass clippings and those sorts of things. And these are giving us massively different carbon trajectories. So many of us are doing these types of unidirectional impacts. But what we want to get to is the upper right where we're having bi-directional interaction between our models and feedbacks between our models. And this inclusion of the feedbacks and how we represent them really increases the non-linearity and the variability in our results and gets us closer to understanding and getting insights about the systems that we're actually interested in. And so this was the focus of a CSDMS workshop back in 2016 where there was about 30 of us here that were working in this area, putting our minds together and sharing our experiences. We published this paper in our system dynamics last year, making comparisons about the different types and styles of coupling that we did in our research programs along with how the feedbacks were represented in those different initiatives. And then we also identified a number of lessons that we learned as a community. The first lesson was about remembering modeling as an iterative process. And so I highlighted that already, but in this case, when we're starting to couple models together that haven't been coupled before, there's data that we haven't typically collected. And so one example is that when we combined the agent-based model with BioNBGC in the residential landscape, we realized that residential landscapes are really void of any data about carbon storage and the soil and the vegetation on those properties. So that fostered new proposals and new papers that we're exploring in quantifying carbon storage and residential landscapes that we're now bringing back into the modeling process. Part of that is also leveraging the sensitivity analysis of our models. So when we couple them together, we don't always know what parameters or variables are gonna be really important. What processes are really gonna affect the feedbacks and interactions. So we need to use sensitivity analysis to identify those pieces to foster where those additional research efforts are gonna go. We also noted creating a common language, as you've all heard the standard names convention that CSDMS is putting forward, that's incredibly important, not just for communicating between human and natural scientists, get on the same conceptual page, but it also feeds into the next one in terms of making code open access. If we can interpret that code easier because we have the standard naming convention, it's gonna help us out a lot more in making progress. Making the code open access here is about making sure that we provide hooks and leverage points between our systems. So for example, in ecosystem models, there's often mass balance equations that are provided, but those are just inherent in the natural system. We don't provide hooks and leverage points for us to add fertilization and manipulate some of the nutrient cycling activities that are going on in there. And so if we do that, then we can foster the greater collaboration between natural and human sciences. Ensuring consistency is about making sure that when we have models that may represent similar processes, that we understand how those processes are similar or different from each other. So if we have a natural system model that has something to do with wood harvest and we have a human system that has something to do with wood harvest, are those actually teaming up with each other? Do they mean the same thing? We have to come to some sort of resolution as to which one we're gonna move forward with when we couple those models together. And then of course we're always dealing with different spatial and temporal issues when we're trying to combine models together. So we have to reconcile those. So here's a picture of some figures from some work by Tom Evans back in 2013 that was looking at coupling carbon and land use models with natural systems models on the left, human system models on the right, temporal scales on the bottom and spatial scales on the y-axis that demonstrate that the natural system models are typically acting at a finer temporal resolution and a coarser spatial resolution than the human system models. Human system models are often starting at an annual time step. And they're often acting at a one kilometer to a hundred kilometer kind of missile range of activity that's taking place. And reconciling this is a necessary issue in order to move forward. But what I think we're missing is an opportunity here which is to scale the decision-maker. We're making decisions sub-annually. Farmers are making decisions sub-annually when you're mowing your lawn, clipping your trees. Our interactions with the environment are sub-annually. And so I think there's an opportunity for us to better represent the relationship between humans and the natural system at that decision-making scale. So part of my research program is about moving forward with that representation through the combination of drones, fieldwork and simulation models. And so I constructed a conceptual model, an agent-based model of agricultural land management systems, where we have individual actors in the real world, agents represented as virtual actors in the model. And we want to represent hundreds to thousands of these across the landscape in Southern Ontario. So don't worry about the specifics in this diagram, but that's just basically the human model. And most of the times the way people inform these human models are through social survey. So we've been conducting social surveys over the last year and a half where we're trying to use those data to empirically inform the characteristics of the agents, the behaviors of the agents, how they interact and decide when to adopt or stop using BMPs, those sorts of things. And what we're trying to bring to it is the addition of drones to validate and calibrate natural process models at the scale of the decision making. Okay, why drones? Because farmers love drones. They get all excited about new technology. It opens up that kind of communication portal. For me, with them, when I don't have a relationship with them in the past, students love drones. They want to get jobs and they think this is going to be a new way forward. Plus it's kind of exciting to be flying drones and of course little kids love drones, right? And so what we want to do is go out on demand with the drone, collect really high resolution data and see how that's going to work with our natural system models at that scale of the decision maker. But before we do that, we have to lend some credibility to it and determine how that drone performance and accuracy is going to measure up. So as a little aside, I'm going to mention that we went out and looked at profiling stream banks with a terrestrial laser scanner and compared that to drone and manual measurements to validate that the drone is actually going to provide us highly accurate data for representing natural system processes. So here we are on one side of the stream doing the TLS scanning on the other side of the stream with some Leica equipment. Then in Ontario, the government uses what's called the Ontario Stream Assessment Protocol to go and measure transects across the province and different streams that they're interested in measuring. So we also went out and conducted that with here's my student Omar Genich and then I was flying a UAV doing imagery collection that we're using in structure for motion. And so we put all this together in the generation of point clouds from the TLS and UAV. And this is showing the vegetation on the stream banks which we then have to filter out to create the terrain. And then we can make some comparisons between our manual measurements, the TLS and the UAV. And so what we identified was that the manual measurements were having highly systematic under-representing in terms of the slope of the bank, the height of the bank. And so we were having challenges in using those data moving forward for calibrating the models. So we can measure change detection in the stream of about 66 centimeters, which is quite coarse. Our UAV performed incredibly well relative to the TLS. So we're getting average errors of four centimeters with a 95% confidence interval, we can detect change of 14 centimeters and throwing in a host of different site conditions in some statistical models. What we really identified as the limiting issue is the accuracy of the ground control point locations. So in this histogram on the bottom left, these are where we have poorly located ground control points. We're getting errors in the 30 to 40 centimeter range. And if we have highly accurate ground control point locations and we're getting accuracy below five centimeters. So we kind of proved to ourselves, okay, we can do this. We can use UAVs, we're competing well with the industry standard for 3D surface reconstruction, which is coming from the terrestrial laser scanning. And so now we can start to use these data to collect them and start to validate and calibrate in-stream erosion models in this case or other types of natural system models in our other research. And with the drone, we can cover such an extensive amount of area at such a high resolution that we can start to map large spatial extents, these corridors across the province and start to establish relationships between land management activities and in-stream erosion processes. So there's a bunch of stream monitoring and research teams in Ontario that are really excited about this opportunity to go beyond transects to actual spatial coverage of the stream network. So take that back to the conceptual model that I mentioned, this agent-based modeling framework. And what we wanna do is we wanna minimize the impact on farmers. So we wanna go out, we wanna do a rapid agricultural assessment where we throw a few drones up in the sky, collect as much data as possible, conduct a social survey, get out of there in four hours or a day and then go back and start to parameterize our models. On the side, we're also doing lots of some kind of combinatorial work to decide when and how to collect these data, different flight patterns, timing of the year and that sort of thing. So I wanted to show a picture of Ben because he's been instrumental in this. You can see him clearly in the center with the TLS scanner right there, measuring an erosion plume down here. The specifics of his work are on the poster, strategically located by the coffee machine. And so he's out there doing the TLS scanning, I'm doing the flying and we've got a huge situation of ground control points in the landscape to reconstruct the surface. So we've got a 40 acre field that's located on the Nith River. The Nith River is one of the strongest contributors of nutrient flow to the Grand River, which then goes into Lake Erie and leads us back to our neutralification problem. And so it just blows my mind every time I look at this picture and I visit the field is the volume of erosion that's happening on this one farm field. This is 40 acres. And before we got there, the farmer had just instituted four berms on the landscape to slow down that water flow, collected, kind of create some deposition points. And it's been a huge amount of deposition that takes place. So it's a little bit hard to see back there, but there's massive deposition. These are some of the cones funneling the water into the ground, into their tile drainage system. And when we initially went out, there were about five and a half feet. There are about three feet or two feet in some places. And he's gone out with hundreds of bucket loads with a front end loader to redistribute that soil back uphill in the landscape prior to tilling that field. And so I know this is a lot. I'm trying to figure out how much is this to picture. And I just had my third baby when we bought a minivan. And I didn't know I would love minivans as much as I do. I should have had a minivan in college. I'm literally not joking. I love minivans. So I thought, and the reason why I love them is because I can pack so much in them and it's not stressful to then pack the car and make everybody happy. So I've got all this space and I started thinking, how much soil could I fit in that minivan? And so Ben's measurement is that with the TLS, we're getting 26 meters cubed just from this one plume of four massive plumes on this one field. That's almost five minivans full of soil. All right, and the UAV data just for that location is matching up pretty well. We prefer better, but it's doing pretty well there. And so part of that process when we want to start moving this towards the modeling, the hydrological and erosion models is that they all involve some level of flow accumulation. All right, so on the left, we have a 10 meter DEM that is distributed across the province. And here's what the flow accumulation pathways look with that DEM. And on the right hand side, we've got the five centimeter DEM that we created from our drone data. And the first thing you'll notice is that we get these kind of horizontal transfers of flow accumulation, which is happening because of the tillage practices, the contour tilling that this farm was doing. And the difference between the left and the right side is the difference between interacting with the farmer and the farmer ignoring you. If we take data out from what's shown on the left hand side here, it's not gonna agree with any of their experiences. You're gonna think, why are we telling them what to do or how to do it or have any interaction with them at all? What they see is what's happening on the right hand side. They know where erosion's happening. They know that they're encountering erosion, but this enables us to start to interact with them and do a better representation of how their management activities are actually influencing the natural process. And this gets really exciting when we start to link these with agent-based models of water packets moving on the landscape. So we can actually model packets of water, drop them on the landscape, see where they start, where they end up, and we can start to see where nutrients originate or pollutants originate and reside afterwards. So we've been using, this is just in the region of Waterloo, we've developed a simple model with Manning Coefficient's land cover and some other variables in there to see how the water is moving through the landscape and comparing that against hurricane hazel data, which is matching up really nicely. But that's the direction we wanna go. We wanna start dropping these on the farm field and see where those nutrients are moving and how is that gonna relate to different fertilization applications. We're also using these data to calibrate and validate crop models. So we're working with DSAT Crop System Epic. This is with my student Omar Genich and we're comparing the crop yields that we're generating with those models to the combined data, which is shown in the center on the right with the drone data that we're collecting in terms of NDVI, LAI, plant heights and those sorts of things. When we go into the field, we wanna maximize our impact in terms of data collection. So we also wanna look at things like pollination services that are happening around the periphery of the farm. So we've got another project where we're looking at mapping individual plants where we focused on milkweed and we can do that. It's amazing at a half a centimeter resolution we can actually identify individual plants and start to look at how the management activities are influencing those pollination services in proximity to those fields that are cultivated. And so where does this take us? The farmers know what's in their field. They don't really want us to tell them what to do or how to do it. And so if we come in abrasively like that, it's gonna be a problem. What we wanna do is maybe highlight what they already know in a different way. So we wanna show them that, yeah, we know that you know where the erosion's happening, but maybe you don't understand how much erosion is actually happening. How is this affecting your bottom line now and into the future? And then we can start to work with them to come up with alternative approaches to mitigate the issues and increase their decision-making capacity to respond to natural processes. The most fun part about this, as many of you know, is interacting with the farmers, transferring that knowledge back and forth with them. And now we're like a sounding board. They come to us with questions all the time about what we kinda can't do with drones and with natural models. And so what I'd argue is that if we can't demonstrate the environmental impacts of human activities at the scale of the decision-maker, then changing, influencing, and adapting human behavior is gonna be much more difficult for us. And if we can't calibrate and validate our natural system models at the scale of the decision-maker, then the relevance of our models and our science to society is going to be diminished. And our ability to complete this representation of coupling and feedbacks between human and natural systems is gonna be diminished. And there we're gonna have less ability to change this complex system where we've got heterogeneous actors making decisions independently and in coordination with their neighbors and their networks in a heterogeneous landscape that are having cumulative impacts to affect regional issues like eutrophication of Lake Erie. So this leaves us with all kinds of different questions like what is the accuracy of our natural process models when we're applying them at that scale of how do land management activities affect these natural processes locally and cumulatively across the region and are there thresholds that we can identify that trigger changes in behavior so we can start to nudge them in advance before we have issues. And so I invite you all to come, maybe not everybody, but most people to either break it session and 3.3 tomorrow to talk about a couple modeling of human natural systems. What are our goals? How are we gonna move this forward as a team? And what we really need is to identify what are your needs as natural system modelers? What do you wanna get from the natural from the human system group to help better create those linkages? Of course this couldn't be done without the funding help of different agencies in my student team including Ben and this poster in the back for more details. Thank you. Thank you, Derek. I love the word decisioneers. I had not seen that before and I think it's such a cool way of thinking about the whole process of decision making and how to interact with decision makers. We have time for some questions. So if people are having questions, Allison is like very quick. We make it very clear that we will not have any location identification information available and the social surveys are all anonymous. We don't have any relationship with their information and their location in there as well. So far we've been lucky through our networks to be working with farmers that are interested in working with us. Although Ben has had some experiences going to some farms and being chased off the land. But so far so good. Good question. Nicole? I'm sorry but I do not remember ever using the word fire. Oh, so berms, sorry. If I say sorry and processes, can I blame it on my accent? This wasn't easy one. Yeah, I've got that question. They're putting soil berms and planting them with grass to stop the flow of water downhill. And so the water collects their pools and goes down the drainage pipe into the drainage tile system. Thank you. Is there one last question? We haven't. Right now we've just been using a questionnaire and just kind of what we learned from interacting with them. We've kind of approached it more, even though I'd say I'm more on the human side because I've been building agent-based models for 20 years and this natural system stuff is a little new to me in the last decade, let's say. We were kind of approaching it more from where are the locations that we want to collect this natural system data from with the drone and with the TLS. And then we're kind of hitting up farmers in those areas to work with. And in most cases they've been really, really receptive to it. And part of that is because we have the drone technology and they get excited to see what's going on. Some of them have even had DJI Phantom 4s and they want to compare theirs to ours and learn about the differences and that sort of thing, which has been fun as well and surprising. That's a great question. We actually started it in Europe. I post-doc with Mark Roundsville and University of Edinburgh and we had several sites that were modeling agricultural decision-making in Europe. And so I took that when I came to the University of Waterloo and kind of modified it for Southern Ontario and then we actually went through a third modification to really focus on BMP adoption as well. So we'll have a bunch of links to four surveys that were conducted in different countries in Europe as part of the kind of original processes for developing that to make a comparative analysis down the road. Thank you, Derek. Thank you guys.