 Hi, so the Habitat Quality Model is useful in two key decision-making contexts. First, it can help assess the conservation status of habitat important for biodiversity by identifying areas vulnerable to anthropogenic impacts, and especially where habitat data, in terms of presence and absence, are scarce. It can also be applied in terms of assessing changes to species' habitat under different scenarios, such as future land cover or climate change threats. The Habitat Model does not treat biodiversity as an ecosystem service. Instead, it is used to assess the overlap and trade-offs among biodiversity conservation, ecosystem service provision, and other land use types. Habitat Quality is defined as the ability of environment to provide conditions for appropriate individual or population persistence. And Habitat Rarity is defined as the relative commonness of the habitat relative to the baseline land use scenario. Both indicators are used, I mean, both Habitat Quality and Rarity are used as indicators of status for biodiversity. So, generally, areas with large habitat extent and quality are likely to support higher biodiversity. The Habitat Quality Model depends on habitat suitability for the species of interest, which can range from a single species, such as a forest bird or even a larger category, like all mammals, or biodiversity in general, not being not species-specific. For instance, we can ask, does it prefer grassland open canopy forest or closed canopy forest? The model also depends on proximity and intensity of threats represented by two major categories, habitat loss and habitat fragmentation. And we can ask questions such as, how far away is the threat or how severe is the threat? So, habitat threats and degradation depends on the distance between the habitat and the threat. For instance, our cities as well as the relative weight of the threats. For instance, our cities more damaging than roads. Sensitivity of the habitat to the threats, such as is the forest more sensitive to road than grassland? Or how quickly the impact decays over a distance? Here's the model we can choose between linear or exponential relationships. And lastly, we can also account for accessibility or protection status, where the habitat types that are more protected are less likely to get degraded by illegal harvesting or hunting. But this does not help treat threats such like pollution, which can transcend boundaries. So this is a simple overview of how the model works. Let's say we have a bird species and live in a forested area that we designate in the model as habitat, using the language cover map. Then we have non-habitat in the landscape, such as agriculture, as well as threats such as roads and cities where people can come in and affect the habitat. And these threats can be linear or exponential by nature. So, and we can also have national protected areas that can help lessen those impacts. And using the habitat degradation map we can derive this habitat quality map. We can also apply the model to look at the rarer habitats, which are usually targeted for conservation, since they're more likely to disappear entirely along with the species and the processes in them. The rarity here is measured relatively between current or future language cover maps compared to a baseline land cover map, which represents an ideal historical state. And we identify those habits, those rare habitat based on the change in extent of each habitat in terms of number of grid cell between baseline and present or future conditions. Here, note that forest disappears between the baseline and current or projected land use cover. If we want to look for habitats with a high rarity score, here we see that most of the wetland disappears between the two land use cover maps. So, since it is more rare and seems to be disappearing, we probably want to protect that habitat. In general, habitats that are rare on both baseline and current or projected land use cover maps are likely more stable. Here is an overview of the input data needs to run the model. First, the model requires a land cover map with land cover classes. A sensitivity table will also be used for each land cover class. Threat raster maps, which are usually in geotiff formats and are used and must be represented on the same scale and metric. For instance, presence absence or kilometers or density. And the threat information is stored in a table, including their spatial impact and their weights. Optionally, the model can also incorporate a baseline land cover map, which would represent an ideal state or historical state of the landscape. It can also incorporate a future land cover map to look at projected changes across the landscape and how those would impact the habitat. And it can also incorporate an accessibility or protected area map, which basically tells us where legal protections is currently happening or physical barriers due to elevation, which can hinder threats impact. The sensitivity table, the threat table and the accessibility map are all scored between zero and one. And I'm going to explain how. So we assign values between zero and one because it is a relative ranking model. We do not quantify anything. So for suitability, zero means that the species does not like to live in this habitat, while one means that it is its preferred location. For sensitivity, zero means the habitat is not sensitive at all to the threat, while one means that it is very sensitive. Threat zero means inaccessibility, meaning that it's fully protected. People cannot enter the area. And one means that it is not protected and open for everyone. For threat zero means low impact on habitat, while one means very high impact on the habitat. So now looking at the structure of the workspace. So create a folder for the model and the input folder must be nested within that larger folder. And then within the input folder, we need to create another folder for the threat raster is where all the rasters need to be stored together in that folder. Note that at the end of each threat raster names, we include an underscore C to represent the current threat status and an underscore F for future threat status. Each threat names needs to also be under eight characters. So taking a deeper dive in the different input data for the model, the land use cover map is a geotip raster where each value corresponds to a land cover class and need to match the sensitivity table. The sensitivity table represent the land use land cover code which match the land cover maps so those two inputs are connected through that code. It helps us also define which land cover class is classified as habitat or non habitat, zero being used for not a habitat and one being used for habitats. And then we have the threads, which can be 123 depends on the study area or the question at hand. And each threat names must have L underscore at the beginning of the name that has nothing that has no implication for the model, but it is a relief from older model programming. All you need to know is that it needs to be included for the model to run. If we look at the weights, the sensitivity values under each thread. We noticed that those value basically represent the relative sensitivity of each habitat to each threat. So again, thinking of the relative ranking, one being highly sensitive and zero being not sensitive at all. It is also important to note that all the threads names must match the names of the threat maps in that nested folder. Then we have the threat table, which helps us list the threats, their weights and their spatial impact. The threat names must match. The threat names must match the maps of the threads I guess I just mentioned. And the table also include a max distance which is how far can that threat impact the habitat and these must be expressed in kilometers. If you use meters, the model will spend for a very long time and because the model will assume that those values are kilometers. The weights help measure the impact of each thread on habitat relative to other threats, one being the highest threat and zero being the lowest threat. The array column indicates the spatial impact of the threat. Is it either linear or exponential and can help indicates how quickly the threat decays over space until it reaches its max distance. The threats maps are rasters representing presence absence of the threat for instance here shown with presence of urban area or not same thing for roads and railways. They also represent density of threats such as nightlight or herders, which can be expressed in kilometers or density. And note in terms of GIS structure that all the no data in these rasters must be coded as zero. And all these rasters must be placed in the same folder nested in the input folder. You do not manually enter enter each raster in the model. We had the last optional input which is the accessibility layer, which represents different level of protection on the landscape with zero being completely protected and one being no protection at all. And also note that all the non defined areas are automatically treated as open, meaning assigning one value in the model. For the future land use cover maps. We can also include a future land use cover map which will present how the landscape is projected to change under current human or market drivers for instance. We can also include a baseline and use cover map which is used to calculate the rary index of habitat. And it's important to note that across all these land cover map if you put in the analysis, the values of the common land use cover that types must match across each raster. Once you have a new land use land cover type in the future maps, it needs to be assigned its own unique value. So, to set up the model, you need to download and install it using this link at the natural capital project dot edu slash software slash invest. Once you have downloaded, you can double click on the file and click next until it's completely and successfully installed. Once it is installed, you can open the habitat quality model. And the interface is going to look like this. So first, we need to set the, we need to point the workspace where we're operating. Then we need to include the current land use cover map in geotape format. Optionally, if it is part of the study or the research question, we can include a future and baseline land use cover map here for this example we didn't. Then we also need to point the folder, we need to point it towards the folder containing the threat rasters. And then the threat data table which is in CSV format. Here optionally we also included the accessibility layer for the level of protection across the landscape which is a vector so a shape file and the sensitivity table which is in CSV. And lastly, there's that half saturation constant and the default is 0.5. We also call a layman terms and what it does is basically adjust the spread and the distribution of the habitat quality values, but it doesn't change the ranking order of the habitat quality. So basically it helps us visualize the changes across the landscape more easily. First, the default is 0.5. And ultimately, you want to set this value to be equal to one half of the maximum degradation score. So the degradation score is obtained from the degradation map, which you obtain from running the model. So, therefore, to calibrate this value, you need to run the model twice. First, you run it with a default to identify what's the max score for the degradation score, and then you rerun the model using the half, the one half of that max value. So the model gives us two to three key outputs. The first one is the habitat degradation map. So here, which can help us, as I was just mentioning, set this house saturation factor. So here, looking at the habitat degradation map, we can see that the max value is 0.06. Meaning that for the second run of the model, we should set this factor 2.03 to help visualize the habitat map. The second key output is the habitat quality map, which range from zero to one, because it's a relative ranking system. And lastly, if we were to, if we had included a baseline layer, we would obtain a habitat rating output as well. It's important to note that all the values are relative to one another. We're not quantifying anything in absolute terms. So looking now at the habitat quality with the default saturation term, which is 0.05. So here, we can see some red spots in several areas, but it's really hard to tell if things vary. So looking at the scale, we see that there isn't much values between 0.1 and 0.99. So what does that tell us? We need to rerun the model and change that key factor. Once we do and set it to that half value of the degradation max score, now we start looking, now we start seeing some nuances across the landscapes. In every model, this model has limitations. The relative ranking is, the relative rankings are more important than absolute values here. And threats are treated in an additive fashion, when sometimes collective effects of threats are larger than the simple sum would suggest. The habitats are also using artificial boundaries because the model habitat is typically within a larger landscape and doesn't account for effects outside those boundaries. Sometimes one way to circumvent that is to improve the analysis is to model a larger area than the area of interest. We cannot consider patch size and connectivity when we know that small or isolated patches may not actually be suitable for species like large mammals. And feel free to contact us if you have questions or meet me during my office hours. Thank you.