 Hello, I am Jesse Goldstein, the GIS analyst with the Natural Capital Project, or NACCAP based at Stanford University. This video provides a tutorial of the seasonal water yield model, which is included in our Invest software suite. I'll cover decision contexts, model limitations, example applications, including from an actual NACCAP project as a case study, and conceptual model theory. I will also provide a walkthrough using actual inputs and outputs from the case study. Some questions that the seasonal water yield model, or SWY, can help to answer are, how much water does the landscape produce? From where on the landscape does this water supply originate? And how might land management or climate change affect these contributions? To answer this last question, multiple runs of SWY would need to be performed so that the results of the different iterations could be compared to one another. For example, varying the model inputs to reflect alterations in land cover types, crop management practices, the timing of planting or harvesting seasons, or changes in climate will affect the spatial and temporal distribution of the modeled results. Of course, reliable water supply is crucial to human and non-human life and health. SWY is appropriate to use in many decision contexts, such as for understanding the seasonal timing of stream flows, for ecological conservation and restoration goals, or for seasonal water quality analyses of pollution dilution potential, for investigating the effects of past or future land use or climate changes, for elucidating irrigation supply and timing for your agricultural hydropower and drinking water demands, and for assessments of flood risk reduction potential. For a NACCAP project I am working on in the tropical region of Latin America, the national and local governments are interested in understanding the timing and volume of water resources that could be expected to supply new aqueducts they are planning to construct. These aqueducts are intended to pipe water from a major river in order to serve communities that experience high influxes of tourists during the dry seasons. The tourists contribute to predictable spikes and freshwater demand in particular months when supply is reduced. The government is also concerned about more frequent and severe flooding events in these same communities during the wet months. I will demonstrate how the SWI model has been useful in this nation and is being used to aid their development plan. Like all invest models, SWI uses a simplified approach. It estimates quick flow and base flow, but does not include many of the complexities that occur as water moves through a landscape. Quick flow is primarily based on curve number, which does not take topography into account. For base flow, although the model uses a physics based approach, the equations are extremely simplified at both spatial and temporal scales, which significantly increases the uncertainty of the absolute values produced. So we strongly recommend interpretation of the values as relative to one another across the landscape as an index rather than having high confidence in the absolute values themselves. While quick flow is provided at monthly intervals, base flow is only an annual output. Additionally, due to simplification, there's also uncertainty around flow routing, which is used to determine the upslope contribution to actual evapotranspiration on a given pixel. Additionally, a recent study from Germany found that SWI does not perform as well in snow-dominated or arid regions. Before I get too deep into model theory, I'd like to define some terms. It's important to understand that SWI differentiates between quick flow and base flow. Quick flow refers to direct runoff. That is water reaching streams or rivers during or shortly after rain events. Base flow is water that reaches streams or rivers later, as in between rain events or during the dry season, and can have residence times of months or even years. Here I present a basic schematic of Earth's water cycle or water balance. As water vapor in the atmosphere rises, it expands and cools, condensing to form clouds. This moisture is returned to the surface as precipitation, which runs off filling rivers, streams, lakes, wetlands, and oceans. But it also infiltrates soils, percolating through to recharge aquifers and groundwater generally. Soils act as underground reservoirs, absorbing water when there's excess and storing those resources until release during dry periods, contributing to base flow. Completing the cycle, the water is transferred back to the atmosphere by evaporation from oceans, soils, and other surfaces, as well as through transpiration from plants as they exhale water through leaves, stems, and flowers. And they open their stomata to allow for the diffusion of carbon dioxide during photosynthesis. Of course, for the purposes of invest SWI model, the cycle is simplified. Water precipitates onto the surface and evapotranspiration, or ET, returns it to the atmosphere. ET is evaporation plus transpiration. On a very basic level, precipitation minus ET equals water yield. An important point about transpiration is that vegetation type greatly influences transpiration rates. So some plant taxa are more efficient or more capable of transpiring greater quantities of water than are others. Of course, transpiration also depends on the quantity of water available. This is where we differentiate between potential ET or PET, which is ET under an unlimited water supply scenario, and actual ET or AET, which is the actual realized ET. AET is limited either by plant demand, PET, or by the availability of water. Root and soil depths vary by vegetation and soil type, influencing the amount of water available to plants. Looking more closely at our model, please imagine that each of the black outline cubes shown here represents a parcel of land or a raster pixel. Quick flow runs off during rain events, while local recharge infiltrates into the ground, contributing to base flow. It's all pulled downhill by gravity. Quick flow depends on the amount of precipitation, the number of different precipitation events in each month, and curve number, or runoff curve number, abbreviated CN. Curve numbers are assigned to each land covered class and are used to convert rainfall to runoff for these different classes. Higher CN values have higher runoff potential. For example, areas with clay soils and little vegetative cover have high CN values. Lower values mean that water is more likely to infiltrate into the ground. For example, areas with sandy soils and dense vegetation have low CN values. As I mentioned, the model recognizes that actual ET is limited by either potential ET or the available water quantity that has contributed from upslope. So local recharge is equal to precipitation minus quick flow minus actual ET. And total base flow, the amount of flow that actually reaches streams or rivers, is equal to the sum of local recharge of all upslope pixels. Here is another graphic representation of a raster illustrating the same conceptual model, but with a better depiction of slope. Local recharge is the potential contribution to base flow. Base flow is the total flow, which actually reaches a stream or river. Next, I will walk through actual SWI input and output layers used in our Latin American case study. SWI is one of INVEST's most data heavy models. For data inputs, the model requires an area of interest, identified by a watershed polygon or polygons, a digital elevation model or DEM, and a threshold flow accumulation value, which I'll explain shortly, a land use, land cover raster, accompanied by curve number values, and ET coefficients for each class, 12 precipitation and 12 reference evapotranspiration rasters, one of each for each month of the year, along with a table containing the number of rain events in each month, and a raster of soil hydrological groups. Optionally, a climate zones raster and associated table can be included as a substitute for the rain events table. Lastly, there is an option to enter a local recharge raster if you have one from a different model. This first input from our Latin American example is a polygon vector layer that defines the area of interest, or AOI. Because this model routes water yield, it is essential that the AOI is hydrologically complete, so be sure to avoid inputting any partial watersheds. The digital elevation model, or DEM, acts as a sort of master input file, in that all SWY results are resampled to match its resolution and are clipped to its spatial extent. The legend shows that the colors relate to elevation in meters above sea level. In this case, please note that the highest terrain is in the southern end of the AOI, so we expect water to generally flow from there down to the north. Another important input is the threshold flow accumulation value, or TFA. This helps the model to create streams by routing flow across the landscape. The threshold flow accumulation is the number of pixels required to flow into a given pixel before it is considered a stream. Focusing on a single pixel, entering a high TFA value, creates a coarse stream network. With a lower TFA, the stream network will be finer. Because TFA is expressed in number of pixels, choosing a good value very much depends on the resolution of your DEM. The correct way to estimate your TFA is by playing with different values and then checking streams generated by the model against the real perennial stream network. Here is a simplified version of the land use, land cover raster, or LULC, from our Latin America project. Of interest is that the high elevation area in the south identified in the DEM is classified as forest cover, whereas much of the remainder of the AOI is pasture. In your GIS, this layer's attribute table must include numbered codes that are associated with each land use class. And for each of those classes, the VAPO transpiration coefficients and curve numbers need to be assigned by including a biophysical table. This one is a simplification of the one we used in Latin America. The second column from the left in this table must be labeled LU code and match the codes for each class from the land use, land cover raster. Next, the effect of vegetation characteristics and soil evaporation are integrated into an ET coefficient for each month and class denoted as KC, followed by an underscore and then the number of the month on the Gregorian calendar with one for January through 12 for December. You'll notice that in this case, values for KC1 through KC underscore 12 are the same within each class. This is because we were unable to obtain these local seasonal data for the majority of classes in our AOI. Information on the timing of planting and harvesting seasons for different crops would have allowed us to vary KC values with seasonal specificity. On the right, we have CN, which stands for curve number or runoff curve number, and is used to convert rainfall to runoff for different land use classes. Within each LULC class, there is a CN value for soil hydrological groups A, B, C, and D. Remember, the model uses curve numbers to convert rainfall to runoff. Higher CN values have higher runoff potential. With lower values, water is more likely to infiltrate into the ground. The biophysical table is essential for accurately modeling SWY, and our team spent a lot of time researching and tweaking these values to adjust for factors like local climate. Here is one of our 12 precipitation rasters. This one is for October, the rainiest month in terms of total depth according to the data we collected for the AOI. We chose to highlight the rainiest month because seasonal flooding is a concern of our Latin American partners and other stakeholders in this region. SWY users need to prepare 12 of these layers, one for each month of the year, and store them in a single folder directory to be entered into invest. Notice the south of the AOI again as it experiences the most rainfall of any region within the AOI. Likewise, 12 reference ET rasters are required and must be stored together in their own directory, similar to the precipitation inputs. A rain events table or a climate zone table is needed. In this case, we use the climate zone table shown here. The first column is labeled CZ underscore ID, and its values 11, 12, and 14 each correspond to the three climate zones defined in an associated climate zones raster. The other columns contain the number of rain events or storms for each month in zone. Greater than 0.1 millimeters of precipitation qualifies as an event. If you do not have distinct climate zones, you can simply enter a rain events table that only has a column for each of the 12 months without the first climate zone ID column shown here. And then a single row of data for the number of events by month across the entire AOI. Here is our raster of climate zones using the Koppengeiger classification. These data are extremely coarse with a resolution of nearly 30 by 30 kilometers. But you can see in the legend that we've defined Koppengeiger climate zones 11, 12, and 14, the same ones that were in the previous table. Here I've overlaid the watershed polygon on top of the climate zone raster. When working with coarse data like these, it's important that you have complete coverage across your AOI and that you provide plenty of buffer and clipping global data to your AOI. Similarly to this optional climate zone layer, you can choose to input a recharge layer if you happen to have one of those from a different model. Here is our hydrological soil groups raster input. Soil groups are defined by the soil conservation service. You'll notice that in this case greater than 99% of the AOI is composed of groups C and D. These soil groups consist predominantly of fine textured particles such as clays have slow infiltration rates and high runoff potentials which translates into high flutterers in low lying areas during storm events. Finally, there are three required model parameters alpha, beta, and gamma, which are useful for calibration. The default values are one 12th for alpha, one for beta, and one for gamma. Alpha and beta represent the fraction of annual recharge from upslope pixels that is available to a downslope pixel for ET in a given month. The product of alpha and beta is expected to be less than one since some water from upslope may be unavailable either when it follows deep flow paths or when the timing of supply and ET demand is not synchronized. The alpha parameter represents the temporal variability and the contribution of upslope available water to evapotranspiration on a pixel and is a function of precipitation seasonality. Recharge from a given month can be used by downslope areas during later months depending on the subsurface travel times. In the default parameterization, its value is set to one 12th assuming that the soil buffers water release, that the monthly contribution is exactly one 12th of the annual contribution. Beta is a function of local topography and soils. For a given amount of upslope recharge, the amount of water used by a pixel is a function of the storage capacity. It also depends on the characteristics of the upslope area, the use of the upslope subsidies conditioned by the shape and area of the contribution area. Recharge from the pixel just above the pixel of interest is less likely to be lost than from pixels much further away. Gamma represents the fraction of pixel recharge that is available to downslope pixels. It is a function of soil properties and possibly topography. In the default parameterization, gamma is constant over the landscape and plays a similar role to alpha. We use the three default values for our analysis in Latin America. Before we examine model outputs, I want to provide some possible sources of global input data. Of course, whenever local data are available and trusted, it should be used. For watersheds, I recommend using invest the lineated tool. DEMs can be downloaded from the U.S. Geological Survey's Earth Explorer data portal or from ESA. Spatial precipitation data are available from WorldClimb or the UK's Climate Research Unit. Reference ET can be obtained from CGIR, the consortium of International Agricultural Research Centers. Soil groups are available from FutureWater's High Hydro Soil Database provided by the EU or from Oak Ridge National Lab's Distributed Active Archive Center's Global Hydrological Soil Group's data set or from the International Soil Reference and Information Center based in the Netherlands. For populating the biophysical table, ET coefficients are provided in the UN Food and Agriculture Organization's guidelines for computing crop water requirements in Chapter 6 of Irrigation and Drainage Paper, Number 56. Curve numbers can come from the U.S. Department of Agriculture's Urban Hydrology for Small Watersheds, Technical Report 55, which provides curve number estimates for different crops using methods developed by the Soil Conservation Service. The average number of monthly rain events can be obtained from local climate statistics or online resources like the International Water Management Institute's Climate Atlas Web Query Service or from YR, a product from Norway. The Copenhagen Climate Classification is available from the Climate Change and Infectious Diseases Group at the Institute for Veterinary Public Health based in Vienna, Austria. Data from all of these sources listed here are free and can be found by searching online. Once the inputs are prepared, we run them through Invest and if the model completes successfully, we can examine the outputs. Remember that SWI differentiates between quick flow and base flow. The model's outputs include 12 rasters of quick flow, one for each month, and a raster of each pixel's relative annual index of local recharge, as well as a raster of each pixel's relative index of contribution to base flow. These are real results from that cap's work in Latin America. Here is quick flow for October, our wettest month. The local recharge or potential contribution to base flow of the pixel is computed in the local water balance. Precipitation that does not run off as quick flow and is not evapotranspired by the vegetation on a pixel infiltrates the soil to become local recharge. Notice here the highest local recharge values are in the south of the AOI, which we recall is characterized by high elevation forests that receive substantial rainfall. Similarly, this output map shows that the southern region of the AOI is responsible for supplying high levels of annual base flow. Here is an example of some enhancements we made to the base flow results shown in the river in dark blue, which flows north from the mountains down towards the proposed aqueduct intake point, which is indicated by the water droplet symbol. The light blue line shows the upriver watershed that supplies the aqueduct. Here is the same output on the left alongside two inputs on the right for comparison, reminding us that the southern portion of the AOI is indeed higher, indicated by the yellow and red colors in the DEM, and forested, shown in green in the LULC. Also recall that within the AOI, the southern portion receives the most precipitation. This insight helped to shape our recommendations to increase protections in a national park outlined here in pink. This park experiences timber extraction. Using this analysis as evidence, the government is interested in investing in increased protections for this national park as a nature-based solution to water provision. They are even considering devising a payment for ecosystem services system where downriver users would compensate upriver land stewards to protect the forests in this park. This is seen as potentially a much cheaper solution to sustainable water provision than is investing in aqueducts which may not even receive flow during the dry season or prolonged drought. Lastly, we post-processed quick flow results for October, the wettest month to create a measure we call runoff retention. Runoff retention is equal to one minus the quotient of quick flow divided by precipitation. Darker blue here represents a higher percentage of runoff retention. The forests of this national park act as a sponge, retaining runoff in the wet seasons, thus mitigating flood risk downriver, and supplying crucial drinking water to downriver aqueducts, communities, and tourists when demand is greatest in the dry seasons. Invest seasonal water yield model provided the results that supported NETCAP's recommendation that the government protect the forests of this national park from timber extraction, both in order to preserve reliable water supply and to reduce flood risk. Okay, always remember to reference the invest user guide. We recommend reading the user guide carefully before using any invest model. Also, please check out our community forum. It is a natural capital community and support site monitored regularly by our software engineers and scientists. There you can search for information, post new questions, connect with ecosystem service practitioners around the world, and share the results of your investments. Thanks for listening and happy modeling.