 Hello, my name is Perrine Hamel. I'm a faculty at Nanyang Technical University in Singapore, and I'm one of the developers of the Urban Flood Risk Mitigation Model within the INVEST suite developed by the Natural Capital Project. So today we're going to do three things. The first one is we're going to learn why one would use the INVEST Urban Flood Risk Model. Number two, we're going to learn how the model works in theory and then how the model works in practice. So let's get started with why using the Urban Flood Risk Mitigation Model. We want to highlight first that there are several types of flooding. One of them is the fluvial floods or also called the river or river in floods. And these are essentially when the normal river level exceeds the bank levels and there's water overflowing the banks. So this is flooding coming from the river itself. And the second type of flooding is coastal flooding or storm surge. And this is essentially due to large waves of storm surges that reach the coast where people live. So this is a second type of flooding coming from the ocean or from a lake shore. The third type of flooding is fluvial flooding, which can be flash floods when there's very intense precipitation and more generally fluvial or surface water flooding. And this is actually the type of flooding that we are going to focus on today. Essentially, when there's a lot of precipitation in a short amount of time, and either the soil is saturated or they reach the infiltration capacity, which means that there's too much water produced at the surface of rural or urban areas. And this leads to local flooding. So again, this is the third type of flooding that we're going to focus on and that the model is designed to represent. So some important points about the urban flood risk mitigation model. I want to highlight that it's one of several urban water related services to understand where it fits in a broader picture of ecosystem services. So other water related services include some water runoff retention over longer timeframes. And this is important for water quality, as well as the long term water cycle in an urban watershed. So the difference between the model we're looking at today and that one is that we are today looking at extreme events or storm events when there's a large amount of precipitation for a single event. The other one is more for long term and water quality questions. The other related model or service is the groundwater recharge service, which is essentially looking at how much the groundwater is recharged through infiltration in pervious areas. There are a lot of alternative models and that's the second point I wanted to make. That's important. And some of them you may or may not know and called I tree, which also looks at several ecosystem services and swim the storm water management model and music and heck rise other hydrologic or hydraulic models models that are essentially designed to represent how water flows in that case in urban watersheds. So it's important to remember that invest is one option, among many others, and very often these other modeling options would be more appropriate if you have a strong focus on this urban water related services. And that's because these models are more sophisticated and will allow the user to reach higher level of accuracy. So why why would anyone use the urban invest model in that case, two main reasons. One is that it's consistent in its format, the type of outputs and inputs are consistent with other ecosystem services in invest. And so that's important when one wants to look at trade offs and synergies for different ecosystem services assessments. The second advantage is that there are low data requirements like any other invest model. This one was developed with data scarce environments in mind, which means that you will always find at least global data products are very easily available data products that one can use to run the model. Of course, if you have better data sets for soil data, for example, it's always better but there's a minimum product that we believe is reasonable to ask from users. So some applications just to give you a flavor of the type of model outputs the model produces. This is a map of the flood water retention in Singapore. So we're actually looking at which areas are retaining a lot of flood water, so do not contribute to flood hazard, as opposed to areas that do not retain a lot of water in lighter shades of green. So this is an output at the pixel scale, you can see it's very fine resolution. And we can also aggregate these data at the sub watershed level, like in this map here, and we can also focus more specifically on these areas that are flood prone. And we know there's a risk of flooding. And it's one way to determine how important and valuable this service is. This is because that's an area where there has been flood in the past and we know that it's important to protect or enhance and increase the flood retention service. So this is the type of map outputs that one can obtain with the model. The second example I wanted to share with you today is one in Shenzhen that some of our colleagues from the Chinese Academy of Sciences have developed with us. And so on this slide here, we're showing two maps, one for the flood water retention service, the one we are looking at today, another one for sediment retention. And this was part of an application of the model where we wanted not only the biophysical values for a suite of services, but also the economic values. And this is part of a larger initiative and project looking at calculating the gross ecological product for cities and more generally for provinces in China. So this is another application where we can translate the flood water retention service into an economic value that can be used in such applications. So let's get an overview of the model to understand how it works. The first step of the model is essentially to look at the supply of the service and by supply we mean the ecological functions that provide the impetration and this retention service. And so that's essentially during a storm how much of the obvious water will not be converted to direct runoff, which will increase the flood hazard. The second area of component of the model is the reduction of the water volume in flood prone areas. So that's the connection between the biophysical component and the reduction of flood water where there's a risk so where the value of this service is actually assumed to be the highest. And finally there's the valuation step which can be calculated optionally as the avoided flood damage for a given user. So that's essentially the three steps that I'm going to detail in the next slides. So for the supply, the model uses the SCS curve number method, which is a widely used approach, an empirically based approach that calculates the direct runoff. So if we look here at the graph in inches or in millimeters as a function of the rainfall depth, same thing in inches or millimeters. So this relationship between direct runoff and rainfall depends on what we call the curve number. This curve number is different for different types of land use and cover. So for example, for service versus impervious, and it also depends on the soil type and the antecedent moisture conditions, which is one parameter that the model currently ignores. And so to summarize, it's mainly the land use and cover, which is one of the inputs and the soil type, which is also a model input as provided as a raster. So the model will use these curves to calculate the runoff provided or generated on each pixel over the landscape. And the next step is simply to aggregate. So for each pixel, this peak flow or essentially the runoff generated by the storm event is calculated. And then these individual pixel runoffs are summed up to contribute to the watershed runoff. And so this is an important point for this model. There's actually no routing, which is a simplification. So this I mean there's no direct routing from one pixel to another with perhaps some possibility for infiltration. But there's essentially these processes are ignored in the model. So it's simply the sum of all these runoffs, which is acceptable for this type of model, essentially assuming that everything contributes to during this launch event. So there's no timing difference between the pixels further in the watershed than those really close to the flood prone area. And the runoff redemption is simply the storm volume. So essentially all the water that this storm event, this precipitation brought minus the peak volume. So what is actually coming out of the watershed, what runoff is generated. So the difference between the two is what has been retained by some of the pixels. Next, we move to the value of this service. So we saw that we can focus on flood prone areas and in particular flood prone watersheds to understand where this service might have the highest value. And if we do that, we can also look at the potential economic damage for flood risk. So the model does not explicitly look at the flood depth. It's only saying there's that much runoff that it's generated, but we don't know if it's going to translate into a large or small flood in terms of the depth of the flooding. But still, if we aggregate these values at meaningful units and areas, we may want to look at what is the potential economic damage if there is an important flood hazard in one area. It's value, the value of retention is actually highest if these buildings and the potential damage is higher. So that's essentially what the model is doing here. We're looking at the potential economic damage of an area and summing economic damage for every single building that we had had the footprint for and then looking at the potential service which is simply multiplying runoff retention and potential economic damage. So this economic damage for all the infrastructure in an area. So if we summarize the model inputs that we just went through this first the climate input which is this storm depth and I wrote here design storm depth because often it's a choice that the model user will make to focus on maybe a five year return return period which is the relatively frequent event versus a 10 year 50 year perhaps 100 year return flood so statistically type of event that occurs every 100 years. These can be obtained from engineering manuals and and what we call typically intensity duration frequency curves and essentially looking at these frequencies so if we select a 10 year flood and then looking for a typical duration that one might select as the time of the flood and so how much time it takes for a drop of rain to reach the outlet of a watershed and this could be taken as the duration. This data are also important as we've seen and so it's essentially the hydrologic sort of group that we need for this model, which is a categorization of so it's according to the infiltration capacity. And these are simply ABC or D and they will be codified in the model and as 123 for one for a to four for letter D. The same thing this can be from local soil data sets or global data sets. And finally, on the on that side, the land use and cover is an important input, and as part of the land use and cover which is a buster. And to provide these curve numbers that I mentioned, which for each type of land use and cover in on our landscape in the in the raster. We, there's one specific curve number that's associated with with it. And these curve numbers are actually per hydraulic soil group so there's one for soil group a, and one for soil group B, C and D. The same thing literature and because this model is quite widely used, and this often some existing studies using this data, or the US Department of Agriculture has also some default data that one can use. Other inputs include the watersheds or sub watersheds so the user can define if they want to aggregate at a course or find scan flood prone areas, which can be based on historical information and observation. They can be available from some of the local government agencies as well. And finally, these buildings and damage cost for each building, which is an optional input if one wants to go to this valuation step. And so with all this input data, the model will run and produce essentially these two outputs. One is the raster of runoff retention, as well as actual runoff. So we saw that the runoff retention is the difference between the input precipitation and the runoff generated on each pixel. And so this will be one output as well as the runoff retention itself. And then the shape file will aggregate these values to look at the runoff retention index, as well as in cubic meters so both values are always as a proportion of the input which is the precipitation. Or also as a volume or a depth in millimeter. The potential damage cost as well if this is if the input was entered by the user and this potential service which is the product of runoff retention and damage costs. So before going to the application and hands on demonstration of model. I wanted to give you a sense of some of the limitations and also the next step so the largest limitation of the model is essentially that we cannot look at the actual depth that are predicted or estimated in an area. And so in terms of valuation, it's quite limiting. So I wanted to share here some potential next steps in terms of model development or alternative ways using other models that one can go about quantifying this these values. So can is which stands for so automata dual drainage simulation is one model that we have tested with with some colleagues from the University of Exeter, and to get at this question of flood depth in a spatially explicit way. And so the model produces maps of water depth and velocity as a function of time and using relatively simple inputs elevation surface roughness and infiltration as a raster. So it's still much more greedy than invest in that sense and and it produces relatively rapid simulation, but need calibration so that's one of the downside of a lot of these more sophisticated models where you need to stronger hydrologic modeling background and often need to calibrate the model. The beneficiaries and valuation step of that model can be, as I mentioned, to look directly at the property values as a functional of water depth. And so essentially overlaying the inputs, the outputs of the biophysical model and the flood depth with information on infrastructure or population affected. So as an example for a test golf course actually in Minnesota. And my colleague Ben Jenke ran the model and obtained this map of flood depth here on on the site for a 100 year storm and for the Twin Cities metropolitan area. And so essentially all the darker colors represent higher flood depth up to three three meters. And, and this is the information that we can overlay with the population of our infrastructure data. So that's what this might look like if we look at the building information in gray or information on sensors block so population income age, perhaps more vulnerable people to flood. And so by overlaying this information, we can get a finer and understanding of the value of flood risk mitigation. So that's essentially what I wanted to share with you today. And we'll just go through a live demonstration of the model to make sure that this is well understood where the inputs go and how one goes about interpreting the model outputs. All right, so this is what the user interface. He looks like. And so these are all the inputs for the model that we, we explained earlier, and essentially running the investment all means entering all these data with data pre processed by the user and then running the model. So the first is the workspace and that's essentially anything that the user decides for where you want to write the outputs. And same thing for the results suffix anything that the name of the results. And so we can call it test today. What a shed vector. So this is the watershed that is needed also water sheds. And so you see that this check green checks will appear as as I speak. And for depth of rainfall. This is the design storm depth. And so for Singapore. We use here 105 millimeters based on the water utility guidelines. And we also look at the land cover raster so here we're going to use a raster that we've pre processed. So hydraulic and bio physical table so we can take a quick look before entering this input. So this is what the high, the physical table will look like. And can see that it's actually missing some information so I'm going to use a different one. But that's the curve numbers for ABC and D that I was mentioning for each land use land cover code. And and that's about it we're not going to use the this optional input so we just click run and wait for the model outputs and the model has completed successfully and so we can now check the model outputs by opening the workspace and look at the different outputs that the model has produced. So to visualize them and I will bring up here QTS and so we can start with the shapefile. And look at the type of outputs that the model produce I'm actually going to include them all. And I just wanted before we go into these outputs. This is also an important log file and with summary of all the inputs, and also if there's any error. So you'll find some of the important information to debug the model. But so focusing first on our shapefile, we get look, we can look at the different attributes. And so the model produce the runoff retention index as a percentage and a value in cubic meters. As well as the flood volume so this is the actual runoff, total runoff, not the retained volume. And so of course if we want to represent this in a more visual way, one might want to graduate and represent this as a function of the potential so classify and apply. So now we see proportionally which watersheds are actually have the highest retention value. And so behind this data essentially these data sets here. And so the, this is the first step of the model, this is the runoff produced by the model. And this is the runoff retention so if we subtract this runoff value from the precipitation value. And this is the retention as a percentage so you see these are values between zero and one. So that's essentially it for the for today and thanks for listening and of course I wanted to reiterate as well that all this information is in the model and users guide that one can find on the natural capital project the invest users guide. Thank you and have been running the model.