NASA ARSET_ Overview of GEOGLOWS for Monitoring and Predicting Flood Risk

Speaker 1: Hello everyone, Welcome back to this final session of the training on Monitoring and Predicting Floods using Earth Observations for Planning and Preparedness. Part three today is going to focus on overview of Group of Earth Observations Global Water Sustainability tool that is GEOGLOS for monitoring and predicting flood risk. Our guest speakers today are Rachel Megoffin from Aquaveo LLC and Riley Hales from Brigham Young University. We'll start with a brief review of what we saw in Part two. We had doctor Renito Franso.

Speaker 1: We provided an overview of Opera Dynamic Surface Extent or DSWX products which are derived from harmonized Lancet and Sentinel two optical imagery and Sentinel one synthetic aperture radar imagery. These products are DSWX HLS and DSWXS one. The products are near global and are available at thirty meter resolution. The temporal coverage or that TSWXHLS is from April twenty twenty three to present with approximately three day resolution and DSWXS one started in December twenty twenty three extend to present with six to resolution.

Speaker 1: We also saw how to select access and composite DSWXHLS and DSWXS one products. DSWX data products can be visualized using NASA Worldview and data can be accessed using nasal Data. We saw that multiday composite of these two products are recommended when there are persistent clouds. Flood detection more accurate with DSWXHLS for airy region, while DSWXS one for inundated vegetation and multiple DSWX data layers merged using QGIS can improve flood detection capability. Doctor Fransov demonstrated that so today in the final part will have overview of geoglows for Monitoring and Predicting flood risk.

Speaker 1: There will be one homework posted on twenty fifth June, that is today, and it will be due on ninth of July. The homework will be posted on the training webpage. A certificate of completion will be awarded to those who attend all live sessions and complete the homework assignment before the two date. So we'll start with today's session overview of geoglows for Monitoring and Predicting flood risk. Overall objectives for Part three are that by the end of this part participants will be able to identify the capabilities of Geogloss River Forecast System or RFS for global stream flow prediction and use Geoglose Hydroviewer to access globally available retrospective and predicted stream flow for selected reverse Our outline today is that Rachel and Riley they will be talking about Geoglows.

Speaker 1: They will provide an overview of Geoglows River Forecast System which is based on meteorology model, hydrology model, stream flow routing model and stream analysis for bias estimation and reduction. Then we'll talk about Geoglows RFS data access and there will be a demonstration of how to use Geoglose Hydroviewer. Once again, Please put your questions in the question box and we will address them at the end of the webinar. Feel free to enter your questions as we go and we will try to get all of the questions during the question and answer session.

Speaker 1: After the webinar, the remainder of the questions will be answered in the Q and a document which will be posted on the training website about a week after the training. With that, I want to introduce our speakers for today. Rachel Hooper macaffin is a hydrologic data scientist at Aquaveayo with a master's degree from Brigham University. She specializes in GIS and hydrologic modeling. For the past five years, she has contributed to the Geogloss River Forecasting System, starting as a student at Brigham Young University and continuing in her role at Aquaveyo.

Speaker 1: Through projects with Surveer and the World Meteorological Organization, she has helped lead workshops designed to build capacity for data management and flood preparedness across Central America, South America, and Africa. Doctor Riley Hills is a hydrology and river hydrolic researcher at Brigham Young University. He is the technical director of the Geogloss program, where he builds and operates a global coverage hydrological model called the River Forecast System. He has provided hydrologic capacity building and training in more than a dozen countries in Central and South America, Africa, the Middle East, and South Asia.

Speaker 1: RFS data is used by water and disaster management agencies in these areas for flood preparedness, early warning systems, agricultural planning, water quality monitoring, and other applications. So with that, we invite Rachel and Riley to talk about geoglows.

Speaker 2: Okay, thank you, Amita for that wonderful introduction. As she said, my name is Rachel Hubert macgoffin. I work for AQUAVEO and I've been working with the Geoglos Initiative for a couple of years now, and I'm really excited to share a little bit about our river forecast system, our hydrologic model. And so as I begin, you may be familiar with GEO, but if you're not, that's a group on Earth observations and it is a partnership of more than one hundred national governments in participating organizations that envisions a future in which decisions and actions for humanities benefits are based on coordinated, comprehensive and sustained Earth observations.

Speaker 2: And so this is a large partnership in geoglows kind falls under the umbrella of GEOO, and geoglows is focused on global water sustainability, and so GEOGLOWS leverages partnerships, data and resources to deliver accurate, open and accessible hydrological predictions on a global scale, playing a vital role in addressing the complex challenges of water resource management. And so I wanted to begin with this introduction, so you kind of understand that geoglows itself is a larger organization. Today, we're going to focus just on our hydrological model, and we're going to focus specifically on streamflow and so on the hydrologic global scale.

Speaker 2: We're going to focus on kind of what Geoglow's main flagship thing is is our river forecast system, our hydrologic model that produces streamflow. But if you want to learn more about geogos in general, you can visit our website www dot geoglows dot org. And so also, as I'm getting started, this hydrologic model, the River Forecast System RFS, as you will hear me refer to it, is slightly a paradigm shift from what you may normally see in hydrologic models. We often see these global entities providing dem data, land use data, meteorology data, and we're used to seeing those on global scales, and often what happens is those meteorology data sets are taken and then a lot of funding, resources and trainings are given to local organizations and then it becomes their responsibility to do the hydrologic modeling.

Speaker 2: And so each place may build their own hydrologic model based on these global data sets and then from there they take their hydrologic model and use it to do things like agriculture planning, water supply management, disaster preparedness. And so they're both doing the modeling and doing the local decision making. And this makes sense. They're getting to make their own local model. Until when some of that funding, some of those resources, some of those trainings go away, maybe the person who was trained to leave the organization, and suddenly we find that those hydrologic models may not be up kept, they may not be sustainable, and this approach may end up breaking.

Speaker 2: And so Geoglows seeks to solve this problem by shifting it and instead we take in our River Forecast system, we take this global dem land data, meteorology data, and we provide our own global hydrologic model. And so this is done on a global scale, and then we focus the trainings and the resources on specifically taking the hydrology data sets and applying them to local applications of things like agriculture planning, water supply management, disaster preparedness. And so as I talk about the River Forecast system, it's going to be a little different because instead of giving you links to GitHub or to code where you're going to go run the model yourself.

Speaker 2: We are going to focus on giving you the results from the model, giving you the time series, giving you the products, giving you the visualizations. So then hopefully if you are in a situation where you work for a national hydrologic service or for a local government, you can just focus on doing what you're mandated to do, where you take that data and apply it to the situations that you need. And so with that in mind, our global hydrologic model, as I've said, is the River Forecast System, and so this contains streamflow data for rivers globally.

Speaker 2: It has both a forecast and a retrospective component to it, and we have used this model to address hydrologic needs in countries around the world by providing global data and allowing for local bias correction. And so we have worked in Central and South America, We've done projects in Africa and the Hinalayan region, and so we've really focused on this idea of global modeling but then taking it to specific locations and having local applications. And I just wanted to mention if you have worked with Geoglows RFS before it has been upgraded to a new version in the last couple of years, and so it has new file formats, more streams, and a longer retrospective simulation.

Speaker 2: So even if you've worked with this before, you should listen because you may still learn something new. And again, all the information on this and everything I'm going to go over in this presentation are on our training site training dot geoglows dot org, and I'm going to put a plug in for that over and over again because I know it's really hard to learn all this information from one presentation, and so that's a really good resource to go to get more information. So before we get into how to access this data, I want to go a little into the nuts and bolts of our model.

Speaker 2: So this is going to be a bit of a tangent, and we will then go into how to actually access the data. But I kind of wanted to tie into the remote sensing theme and kind of explain the origins of how the river forecast system operates. And so we know that good meteorology modeling or good hydrologic modeling is based on good meteorology, and so without the meteorology, we're not going to be able to figure out the hydrology.

Speaker 3: And so a.

Speaker 2: Meteorology model is basically an energy balanced calculations driven mainly by solar radiation, and so we see three D grid of varying grids cell heights with many layers. These grids are a coarse resolution, so think like a quarter degree twenty five kilometers. So these are a bigger resolution. We're going to use smaller resolution on our streams when we get to our stream catchments, and I'll go over that. But the meteorology model has a courser resolution. And then there's energy transfers happening horizontally in columns and also vertically in layers, and so these type of models are going to be calibrated for atmospheric variables, usually not precipitation or water variables, and so you may be familiar with some of these models.

Speaker 2: We have the Global Forecast System GFS, which is the American model, and then we have ACMWF their Integrated Forecast System IFS is the European model. This is what we use for the River Forecast System. RFS is based on IFS the ECMWS data, and these models when they produce forecasts, they don't just produce one value to make a guess into the future. Rather, they do ensemble prediction, where they predict a range of values of different different possibilities of things that could happen, and they'll often produce different a short range forecast, a medium range forecast, the long range forecast, and so generally the best models are going to be medium range, and they're going to be good for twelve ish days in advance, depending on your location and a lot of different factors.

Speaker 2: And so Geoglows RFS uses the ECMWF one, We use the medium range, the fifteen day forecast, which is why our streamflow is also fifteen days. And we use the ERA five data for our retrospective data and the IFS data for our forecast data. And I just wanted to mention that while the data we take in is runoff data from EERA five, this does not come from a satellite, but EA five is a global atmospheric or analysis from ECMWF, and so satellites provide many of the real world observations blended into EERA five, including temperature, humidity, precipitation, soil moisture, and so even if the runoff isn't measured by satellite, there are satellite values and observations that are going in to improve the model as a whole, which should then hopefully improve the runoff values.

Speaker 2: And that's what we take in for RFS, and so some satellites including Noah fifteen through nineteen, Agua Aqua, and so if you want to learn more about it, go read the EI five observations documentations. But that's kind of how we tie into remote sensing.

Speaker 3: And so now that we've.

Speaker 2: Talked about meteorology data, I'm going to move to talk about hydrology models. And so we know that in hydrology we're going to treat everything as if it's happening in a bucket. So precipitation is coming down, evaporation is going up, we have groundwater discharge, we have infiltration, we have snow melt. It's all happening, but it's all happening in this bucket. And this is happening as calculations. So we get you know, precipitation plus soil moisture, but then we take away the of appo transpiration and whatever we have left is what our runoff is.

Speaker 2: And so water is happening only in a bucket traveling vertically, but then the water is going to leave the bucket as runoff, and we're going to need to route it to put it into channels and move it along the channels into our rivers. And so we're going to start and we're going to take those those weather forecast grids and then in a bucket is happening all our hydrology and we're left with runoff and we can convert that into runoff depths, so we end up with a grid of run runoff depths. And then we're going to take streams and catchments because we know that we're modeling streamflow.

Speaker 2: We want to get the amount of water in the stream and we can figure out.

Speaker 3: Where the streams are using.

Speaker 2: Elevation because we know that water is going to travel downhill, and we can use these runoff grits in these streams and catchments to create a basin runoff volume these tables with kind series, and from there we're going to be able to route the water through our streams to create an overall hydrograph which will represent our time series of our data in our stream. And so again, this runoff depth is happening on grids, but we have no flow between the grids, and instead we have these streams, and so our little stream catchments don't follow the grids, so we need to convert our runoff depth into the basin runoff vout volumes.

Speaker 2: So we take the grid lines and we intersect the basins and we can get the percentage of the basin that's in each of the different grids, and we can sum them all together, and so we end up with the depth over that polygon. Then we're able to take that and repeat it for every basin, for every timestep. So now we have the volume of water in every basin at every time step, and we're going to be able to use that in our routing process. And so next comes the computations where we have to figure out, based on this timestep, based on these volumes, how is how's my water actually moving through my river?

Speaker 2: And so we know runoff is going to follow a terrain slope until it enters a river channel, and then river channels concentrate the runoff into an open channel flow and we assume no losses once in this channel flow, and then we can put the water and calculate how fast it moves.

Speaker 3: And now there's lots.

Speaker 2: Of different hydrologic routing methods. The one we use in our model is most similar to Muskigum routing and so kind of to go over our model, we start with our satellite data and that gets meteorology and land surface model data, and then we have our digital elevation model which turns into streams and catchments. And now using these two things, we can get our runoff per basin. Those are our volumes on our little basins. Then we need to route our water through our stream. We actually use a Python package.

Speaker 2: It was actually written by Riley Hales who's going to be showing the High Reviewer later, and we use that to calculate how the water is moving through the street the streams and add in the different volumes as they enter into the stream beds. But then we don't stop there, so then we get river discharged. But we also as part of our process, make data distribution a high priority and so we take our outputs and we put it on web services, We make bulk downloads available, We put it in long term archives on AWS, so all this data that we have collected through the sprouting process is now available for distribution.

Speaker 2: We also create visualizations. In partnership with EZRI, we have a living outless layer. We also produce plots and maps, reports, all of these and then hopefully people can take all these things we've provided to do local applications. And so the goal is that while it is a global model, that by providing the data to distribution tools and the visualizations, that this can then be taken and applied locally as people need it.

Speaker 2: So there's three main products I'm going to talk about from RFS today. There's the retrospective stream flow data, which goes back to nineteen forty. There's the fifteen day stream flow forecast data, which is an ensemble of data available daily for fifteen days in the forecast. And then we also have our hydrography data all available for download as well, including our streamlines and catchments and some other products I.

Speaker 3: Will talk about as well.

Speaker 2: So to start with, the retrospective data is based on the E five data set. I kind of mentioned that earlier. We use the runoff value as the value that we wrote in and this is a deterministic result and so there's just one value, it's not a range of value. The retrospect is deterministic and we have about six point eight million stream segments and so this is available on an hourly timestep eighty five years of data on six point eight million streams globally, and it updates with a five day lag from present, so it's pretty up to date in terms of how recently it's available and also from the retrospective data.

Speaker 2: We use all these tools to create derivative products, so that way, if someone doesn't want to download the whole time series and do the computations themselves, we do have derivative products already available for download. And so this includes return periods like the two year, five year, ten year, twenty five year, and fifty twenty five year, fifty and one hundred year return periods calculated using the Gumbele distribution and the log Pierson Type three distribution, and so this is like your one hundred what the value of your one hundred year flood is, your fifty year flood, your ten year flood.

Speaker 2: We also pre compute all the monthly averages, the yearly averages, the daily averages, which this is really nice if you're doing projects where you need to access all this information. It can save on a lot of streams. It can save you time for me to download all the data and compute those yourself. We also have flowdoration curves both available, like the overall flow duration curve for the river, but we also have the monthly floatation curve available for each month for each river. And so these are some examples.

Speaker 2: Riley's going to show the hydro viewer after and he will show that there's also a lot of other plots in charge that can be made using the retrospective data, and you're able to compute those and download those. But this is just some examples of some of the derivative products that we have available for download. We also have our forecast data and so this is a fifteen day ensemble forecast. So I mentioned earlier that the IFS data is produced in an ensemble. They produce several different possibilities, and we take each of their ensemble values for their runoff and we route each one of them.

Speaker 2: So therefore we end up having ensemble of forecast members as well. And so this is a fifteen day forecast. It's available every day, there's a new forecast. It's at three hour time steps, and there's one control forecast and fifty ensemble members, so you will see fifty one different possibilities. And we also have forecast statistics available and that will help provide the uncertainty bounds. You can kind of see your range of possible answers you could get as well as what is considered to be.

Speaker 3: The most likely.

Speaker 2: I also want to mention that all our forecast data we save, and so we have about two years at this point worth of old retrospective data old forecast data available, and so if you want to go look at your forecast data from about a year ago, you could go look at that. And so for some reason you need old forecast data that is available as well. And then I also mentioned that we have our hydrography data available. So as a background, the RFS hydrography data is based on TEDx hydro which is from the Tandem satellite and it's delineated using taodem, and so we took the TEDx hydro stream delineations, but we did make some improvements to it to help our model run smoother and to fix some of the errors that were present.

Speaker 2: And so some of these will be applicable to everyone, and some of the improvements we made any will be applicable to anyone who wants to use the hydrography data. But also some of it was specifically for our model and what we needed to happen. So we combined some of the headwaters, like to order one streams we might combine with their order to stream to be one single basin. We combined some of the really tiny short streams as well as we're moving some water sheds and flat deserts where we're not seeing any flow ever in them.

Speaker 2: And so we did these things to help the model run smoother and to help make our numbers more accurate in the model overall better. And so the version that we ended up with of our hydrography is what we have available now for other people to download. And so this data, the hydrography data, is all the into VPUs, and a VPU is a vector processing unit which is basically just a collection of watersheds that we bundled together to make more efficient processing and data management. And so if you know your VPU, you can download just that VPU instead of needing to download globally all the streams.

Speaker 2: Because with six point eight million streams, that sends up to.

Speaker 3: Be a lot of data.

Speaker 2: So being able to download just your VPU of data is a way to make it a lot more efficient, and so ours is one hundred and twenty five globally, and you'll see that in the a US buckets. They're split into VPUs and that's how we manage the data. So you can download the actual stream center lines, and these are the center lines representing the flow path for a river segment, and every stream has a unique nine digit river idea. You might also hear this referred to as a link number, and this is what's going to be on the hydrography data.

Speaker 2: It's also the link number you'll use when you want to query the retrospective data or the forecast data. This is its identifying number. And so in the hydrography data you can see what the link numbers are. You'll also see that on the hydro viewer as well. And we also in our stream data have some additional attributes that might be useful to some people. So we have the stream order, we have the upstream area, we have what region it came from in TDX hydro and so, and like what the terminal link is the output for that stream.

Speaker 3: So that is all available in the stream package.

Speaker 2: And then we also have nexus points and lakes and so nex's points are just the points where the streams connect, so you can see all the points where we have the segments connecting with each other, and it'll mark and it'll tell you what the upstream and downstream rivers are at that point. And then we have the basins that we consider lakes in our model, and well these are all available on the AWS buckets that I'll show you how to access in a minute. And lastly, if you're looking at our hydrography data we do, if you are looking to put it in some sore of GIS software, the easiest way to do it is to take to use the EZRE webmap, and so in collaboration with EZRE, there's a publicly hosted Living Atlas layer.

Speaker 2: And so this layer is nice because it is done so that the thickness of the streams matches the amount of water in the streams, and then the colors represent if it's exceeding a certain return period. So you can see here that that little area where it's purple there that stream flow is at this point exceeding fifty years. And this is a ten day animated map, and so you can actually see those flow period those flow return period colors are representing what is currently happening because of this living outless layer, and so this is a really nice option to have for GIS.

Speaker 2: And so we start with our geoglows version two data and that is delivering AWS open data and web service sources. And so this is what I've talked. We've talked about the daily historical data, the daily animated maps, the daily new forecast. We also have our and this is all happening on a supercomputer. And then we're delivering all these products and they're being stored on AWS. They're being done stored as the Living Atlas layer with Ezer, there's an API, and so these are all stored. But then the hope is that all these stored data sets are then taken and powering local adaptable open source code and web applications.

Speaker 2: So the goal is that all this global data that we've provided can then be taken and applied locally. And so I now want to show you and tell you about how you can take all these things I've told you are available and use them.

Speaker 3: And so in order to do that, you will need.

Speaker 2: To access the Geoglosse data. Now, the easiest way to access the data is in the hydrid viewer, and this is going to be demoed in the second half of this workshop, and this is a web application that you're going to again it'll be demoed, but use you open it up, there's a map, you zoom into your stream, you click on your stream, and a bunch of charts and information and things you can download just pop up right there for you. And so it's really nice, it's really easy, and it's also generally easiest for that one if you only have one specific stream.

Speaker 2: And so let's say you have a whole series of streams and you're trying to query them every every day. The hydro viewer, where you have to click on every single one of the streams if you have two hundred streams, might not be the best option for you anymore. And so there is an option where you can query data yourself from the AWUS bucket. And so this means that you can write Python code to directly access the data and then it'll read into your Python code, which makes it easy from there to be able to manipulate it, include it in web applications, use it for what you need, analyzes, report, it's all available for you.

Speaker 2: And to help make that one step easier, we have the geoglows Python package that makes it super easy to query the data. And you don't even have to create the a to us bucket directly. You can just align a code to query everything using the.

Speaker 3: Googles Python package.

Speaker 2: And this queering the data is generally going to be the fastest way to download the data, especially if you're looking at a large quantity of data. Again, if you're looking at say one hundred different river segments, this is probably going to be your best way. And then of course we have our webmap that I've talked about that's the ten day animated streams layer showing forecast results. And this is free to use or view in any website or GIS, So if you're doing something with GIS software, this is a really easy way to integrate RFS.

Speaker 2: And again, I know this is a lot of information, and as I go through this, all all of this information is on training that geoglos dot org. So go ahead and take a look at that website to kind of refresh yourself on everything I've talked about. So I've mentioned that all our data is available on AWS buckets and they are available for open use, and this is going to be a really fast way to download large sets of data, and on our AWS buckets we have everything available. So we have metadata, routing, configuration files, we have the hydrography data I talked about, we have forecast data, we have retrospective data.

Speaker 2: We have some of those derivative products to the retrospective data with different averages, and so these are all available on AWS. And so you can go to the actual AWS bucket like you see here on my screen, and you can click on the subset that you want and you can click through and you can download the file. Now, I will say you will download a lot of data doing this because you're gonna have to download the whole file. So generally it is better to use the geoglos python package or to query it programmatically and then you can query specifically for the rivers you're interested in.

Speaker 3: And so the geoglows python.

Speaker 2: Package is a Python package that allows for data to be easily queried, and it defaults to querying the AWS buckets as its source, and so the retrospective and forecast values can be queried there if you want to use this. I'm not going to be able to go over all the options with the geogloss python package, but we do have documentation on how to use it and what commands are available.

Speaker 3: And one highlight of.

Speaker 2: The Geoglos Python package is that multiple rivers can be requested at once. You can request all of your rivers and it's also easy to write code to be able to check it again in a couple of days if you need to update it. And so I have a little code snippet here, I just want to show that it actually is like one line of code to get it all. And so you just import the package, you give it all the river ideas you want, and then it's as easy as geoglows dot data dot forecast ensembles with your river list or geoglows dot data dot retrospective with your river list, and that's going to spit you out a data frame with the daytime index and a column per river id, and so again it's a really fast and easy way to be able to get data for a lot of rivers.

Speaker 2: You also, if you are more comfortable with queering aid to bas directly instead of using the Python package, you can create aws bucket programmatically or use the command line if you are already familiar with this. And there's more examples of this on training dot geoglows dot org, so you can get the URI for the AWS bucket and using x ray you can just open that data set and select the rivers you want and get the data that way. You do need to make sure if you do this option that you need to make sure you create anonymously or it will throw an error.

Speaker 3: It is open for.

Speaker 2: Everyone, but if you don't queerate anonymously, it'll say you don't have permission. But as long as you create anonymously, you will have permission to access the data. I generally think the Python package is easier, but if you are more familiar with this way, or you have specific applications where you need to use the AWS buckets directly, this is an option of something you can do. And then again, our other way to access data, probably the easiest, most user friendly is the RFS version two hydro viewer.

Speaker 2: It has a ton of visualizations available for your river. It will it can do lots of things. You zoom in, you click on your river, you can see old forecast data, current forecast data, retrospective data, a bunch of different graphs. You can see the streams colored and Riley will talk more about this later, but this is a really cool tool we have for data visualization and for data accessibility because you can just download the data straight from there. And now I've kind of talked about our data, I've talked about how to access it, but I wanted to mention one more thing as I kind of start to wrap up this part of the presentation before Riley does the demonstration, and that is I've talked a little a lot about how the goal of geoglows is that well, it is a global model, we hope that it can go to local applications.

Speaker 2: And now, as a global model, often is bias exists on a local scale when you're going to be using a global model, and some of that bias comes from the runoff data that we input as the input into it arefs, is going to be other sources a bias that may occur based on your specific conditions of where you live. And so we are always working on ways to improve our model to get better numbers, better answers out. However, one of the ways you can get better numbers is a bias correction option, and so there is an option for local bias correction, which then enables better local.

Speaker 3: Results for the model.

Speaker 2: So this does require observed data at the stream segment, but then it will give you both forecasts and retrospective data, and it'll be corrected to match the volume of water we're seeing in the observed data. So if we tend to in the simulated data often over predict by bias correcting, we can correct the volume and our range of values will be more similar to what you may see in your observed value. And this is available to use in the Geoglows Python package. And we just have some functions geoglos dot bias dot correct historical and Geoglows dot bias dot correct forecast, and those are available for you to use and to perform your own local bias correction on your own data with your own data on the global data.

Speaker 2: And so the way this method works is so we take the simulated hydrograph that we are getting out of the RFS, and we're going to get the floatation curve for that simulation data, and then we're going to get the observed floatation curve from the observe data that you upload as part of the function, and we're going to use that observed data to correct the values of a hydrograph. So then the values of the hydrograph, the volume of it matches more similar to what we were seeing in the flowtation curve of the observed data.

Speaker 2: And so we also acknowledge that observed data, which is necessary to do bias correction, is not available for every river, and we have been working on ways to then try to improve the bias in areas.

Speaker 3: That may not have the subserved data.

Speaker 2: And one experimental process we have right now is called Saber stream analysis for Bias Estimation and Reduction. And what this does is it takes the bias correction methods that we have locally and it works to apply them globally. And so we've taken the data that has been given to us, whether and we have permission to use, whether from GRDC USGS, specific organizations or countries, and this is then included in the model. And we've taken the basins globally and they've been clustered and so like basins are similar to each other and so they have similar patterns.

Speaker 2: And then we're able to take flow duration curves from the gauged networks and the corrections we perform we're able to apply to other gauges, whether in a similar region or in a similar cluster in the same cluster, and so I won't go into the whole algorithm for it, but it's a way to apply the bias correction using regionalized base and data, and so all the streams can benefit from the bias correction happening in some of the streams. And so this data, the global bias corrected data, is available on the hydro viewer and all those time series and things, both for forecast and for retrospective data that we have available for our with our normal data is available for the global bias correction.

Speaker 2: But I do want to mention that this data is experimental. We are still evaluating it and working on lots of improvements in our model as always, and so just use this data at your own risk. It may improve some of your numbers, but it may make some of your numbers worse. And so just make sure you understand what you are using when you choose to use this data. And so that is going to conclude my presentation. That's kind of overviewing all of the river forecast system. Again, I know it's kind of a lot to take in all at once, and so I encourage you to go check out our way website, our training website at training dot geoglows dot org, where it's going to have all the information I talked about and more in addition to examples and code snippets to help you work through, and you can go look at the parts that specifically interest you.

Speaker 2: And then I'm going to let Riley take over and he's going to show you in real time how to use the hydro viewer and some of these tools I've talked about so that hopefully you feel a little more comfortable taking them and using.

Speaker 3: Them for yourselves.

Speaker 4: Well, thank you Rachel for that introduction to everything there is to know how geoglows RFS.

Speaker 5: I know that was a lot.

Speaker 4: I'm going to do my best to provide some concrete examples of everything that we just learned about. I'm going to start by giving another plug for the training sites. This is training dot geoglows dot org. And when you first land here on the left, you'll see and a column of about all the headings and there's River Forecasts system is one of your options. Part one about model formulation is a simplified version of what we just went over, but it does have a lot of extra links to other places where you can read more about the different components and journal articles and things like that.

Speaker 4: I'd love to draw your attention there if you have further questions, and I'd be happy to take them in the Q and A part as well. However, moving down to two available data where I'm starting to pick up on this this first entry data catalog.

Speaker 3: Has.

Speaker 4: If I scroll down just a little bit, there's this table of data sets, section and on.

Speaker 5: To highlight a couple of things here.

Speaker 4: First, all of this data is on an AWSS three buckets. There's a ABS has this program that makes it such that you do not need to have a user account. There's no tokens, no put in your credit card, no user name or password, nothing like that is required to access any of the data I'm about to show you. And that second is that there are two locations to find all of this data. So if you're the kind of person that wants to write code that obtain this, if you want to go brows what's there and refresh your memory, this table will give you an entry for every row.

Speaker 5: Is one of the things that I can provide you what file.

Speaker 4: Format it's in, what's the URL for you to go find it? The AWS region. If you're using it within the AWS ecosystem. So if if you're interested in obtaining the daily forecasts, you'll see this is in a ZAR formats and that's where you go looking for it. If you're interested in the hourly average flow historical simulation, this is the same thing comes in ZAR formats.

Speaker 5: Here's the URL to go looking for it.

Speaker 4: When you go to one of these buckets, all of these links will take you to well zoomed into the correct subdirectory within the buckets, but you'll end up on a page that looks like this. This farthest less column will be the different headings, and these headings will all correspond to something that is in the table on the data catalog portion of the training sites.

Speaker 5: So feel free to browse these as well.

Speaker 4: If you're a AWS guru, there should be old hats and feel free to go access this. No tokens or log in or anything like that is required. Moving on to the Hydroviewer. This is hydroviewer dot googlows dot org. There's a two main purposes of the hydro Viewer that I'm going to try to cover with you today. First, it's a visualization tool. We've got a lot of features in here to help you explore data graphically and through maps and all sorts of visual means. And second, it's a way to retrieve data and look at plots for those that are not native English speakers.

Speaker 4: You're of course welcome to use automatic Translate anywhere you want, but I'll start by up on the top right, we've got official human translated and verified translations in Spanish and in French, so if those that if you speak one of those language and have that preference, we do have a human translated version already available, so that the graphs and the map labels and anything that can be translated will be translated.

Speaker 4: Next is right next to that is there's this gear icon for settings. Now, this tool is definitely targeting a more scientific user, but even given all of the wealth of the charts and maps and everything that's available, there's some of those things are not on by defaults, So I'll draw your attention to First, we'll be looking at a simplified view of the forecast. By defaults, that's the recommended place I would start. However, there's this option that's off by defaults, which says show additional plots if you want to see the extra feel free to toggle that on, and as well as for bias correction, Rachel mentioned how we've done a lot of analysis to understand that bias is the primary error, and that is that it's correctable using gauge data, where it's a experimental method to make that apply to all river segments in the model, not just the ones where we have gauge data.

Speaker 4: Feel free to toggle us on, and I suggests the further reading and journal articles on the training sites.

Speaker 5: If you're going to try experimenting with this option, all.

Speaker 4: Right, let's start by looking at the visualization options. I'll frust drate your attention to the base map that we're looking at here. This is called the environment map, and it's got a couple of things that helped clarify looking at rivers. The blue lines are the rivers from TDX hydro that we've revise to do hydrologic modeling on. But these brown lines are watershed boundaries from hydro basins. And as you zoom in you'll see more those divisions appear, and there'll be other blue lined references and those all come from hydro sheds.

Speaker 4: However, if you have a preference for any others, you can of course change to imagery or topographic or any of the other options to be most friendly for all audiences, I'm going to keep it on a simple gray one so that there's the maximum amount of contrast for some of these streams. The first thing I'll point out about the streams themselves is that there's both different thicknesses and different colors that are apparents. As I zoom in and out, will see more streams. That's called a multi scale map.

Speaker 4: So I'll start by zooming in to this north part of South America, where I see some streams and a few different countries that are colored differently. The legend on the bottom left reminds me that those colors correspond to that's exceeding a return period, or in other words, that the yellow is the beginning of you should be worried about flooding occurring, And the farther you go down this list to exceeding fifty years, the more likely is this is a bigger floods covering.

Speaker 5: More area, and it is more rare in how big that event is.

Speaker 4: So I'll zoom in and we'll notice that once I zoom in here more lines where are parents here compared to zoomed out, I'll do that one more time to draw attention. From this level, I can see just the main stems, the biggest portions of the rivers, and when I zoom in, these streams densify. That'll happen two times until you get all the way zoomed in. The Next thing to look at is that this map layer is animated. If I come to the top left, there's a clock icon. That's why of our controls. If I toggle that on, a time slider appears on the bottom.

Speaker 4: Forecasts are generated every day at midnight, and so that by default you'll see that first time step that corresponding to midnight UTC time zone. Regardless of where you are in the world, it's always going to be midnight UTC is the first time step. Feel free to slide this by hands, or you can use the There's a play button and every couple seconds it will slide through the timesteps. And if I zoom in far enough, we'll see that the thickness and the coloring of the rivers is going to change as the slider progresses down the line.

Speaker 4: So for here, for instance, if I pick out just one spot at the headwater, there's some yellow here, but as time progresses, the flood is receding, so there's it's leaving that warning threshold and going back to the standard blue, which is within normal flow ranges. If I now pause this and jump ahead, the animation goes out for ten days, and jumping ten days into the future, you can see most of the yellow has disappeared and the thicknesses for some of those segments have gotten narrower.

Speaker 5: Maybe there's less water in them.

Speaker 4: Next thing that I will draw attention to on the maps is again on the top left, there's a there's controls to manipulate what streams are visible. This filter icon will prompt you to filter the streams that are visible.

Speaker 4: Rachel mentions that the one of the intended applications of all of this hydrology data is to support custom applications by country or by watershed, so the map is capable of having filters applied. There's a few that are the most common that I provide you some dropdowns for like for instance, this one river country means if I type in the name of a country like United States or let's say Brazil, I'm closer to Brazil.

Speaker 5: If I check that box and hit apply.

Speaker 4: The map will reduce itself to only showing me the river segments that intersect the boundaries of the country. Now you might say, well, if this is in the Amazon, that crosses into some other country boundaries. So if I go back to the filter menu, I might say, show me countries that are either in Brazil or who outlet outlets in Brazil. I hit apply, Then some more of the Amazon becomes visible, even though those segments are outside of Brazil, but they flow into that. Between those two controls you have a lot of ability to display rivers that are trans boundary.

Speaker 4: So if your application area is in a country that's as most are, where the rivers has trans boundary flow, those two controls here help you create targeted visualizations for whatever application area you have. I'll clear for now to restore that. However, you'll see that there's this enter custom sequel. This is an EZRI web map layer, and I've left you a link to documentation if you are familiar with writing sequel. I won't cover that here, But there are other attributes, like you could provide a bounding box on latitude longitude.

Speaker 4: You can provide specific river numbers that you want to be filtered this site's the visit the Ezraelving Atlas link here will direct you to the pages with information about what all the attributes are, and you can build your own filters here.

Speaker 4: The other main components of visualization, we offer some supplementary layers that provide additional context to this.

Speaker 5: Up on the top.

Speaker 4: Right of the map area, there's a layer selector and from the most of these come from the Noah twenty satellites veers. You can get the most recent imagery on thermal anomalies. You can get true color reflectants. One that I'll toggle on is this Veer's flood composits. This is using the most recently available Vier's imagery running through a classification algorithm, and the colors corresponds to different land use types or there's.

Speaker 5: Blue for water, for instance.

Speaker 4: And use this as an example to say if while looking at anything on this app including in most common perhaps even these supplementary layers like the flood composite. On the top rights there's a question mark icon which will provide you links to both the geoglows RFS data and our various websites for instructions, but also for instance, the Vier's flood composite layer. There's this link to a un University of Wisconsin websites that has the explanation for.

Speaker 5: What the colors mean on that layer.

Speaker 4: Another one worth drawing your attention to is hydro SOS Monthly Status. Hydro SOS is the WMO World Meteorological Organization Activity.

Speaker 4: These polygons that you're seeing correspond to hydro shed's basins, and there's a yellowish hand ish color which represents normal and for that the month that's being visualized, it was drier, slightly drier, or much drier than normal, it gets one of these shades of red, or if it's slightly wetter or much wetter, it gets the lighter blue or dark blue. Correspondingly, These maps are available for every monthly average going back for thirty years, so I see it or sorry. If I direct your attention to the top left, there's a control labeled SOS for hydro SOOS controls, and this one also will animate.

Speaker 4: It's not quite as responsive as the streams if you animate those, but the slider here can be manually dragged to whatever time period you're looking at, like for instance, I happen to stop on July nineteen ninety nine, and this is what the map.

Speaker 5: Looked like at this point.

Speaker 4: That concludes the main components of our visualization parts. I'm going to refresh the page and get back to the defaults view. So if I refresh the page and get back to the default view, we'll start looking at how you use the hydro viewer to extract data and get plots and look at things numerically. But I mentioned already that this map is multi scale, meaning that there's more rivers become visible as you zoom in. So if I start really zoomed out and I try and click in South America, the map is going to drag me into a much higher resolution or a much higher zoom so that I can see the rivers at their full detail being presented to me.

Speaker 4: And only when you're this far zoomed in will I be able to click on a stream and it will start pulling data for me. After you've clicked on a stream, you're going to see this modal appear. And there's two options at the top right for you to toggle between forecast data, which we're looking at now, and the other button takes me to look at the retrospective simulation data. I will start on the forecasts our main chart here. The top plot this hydrograph shows the predicted flow for the fifteen day period that forecast covers.

Speaker 4: I'm going to toggle off the return period by clicking on them for a minute so I can comment on the two main lines here, the black one labeled predicted flow, and this blue region labeled uncertainty. So the black line that's called predicted flow. If someone needs just a single number for what flow did we predict for a deterministic application, or you need to put this into an equation somewhere, this black line is labeled predicted flow.

Speaker 5: Is the median of the ensemble.

Speaker 4: As Rachel already mentions, this is a fifty one member ensemble and at every time step, so every three hours across the fifteen days, there's going to be fifty one different numbers, and this black line is the median at.

Speaker 5: Each of the time steps.

Speaker 4: It's the representative value that we would suggest you use. You can try averaging, you can use other means to reduce the ensemble down, but the black line that we put out is a good starting place is the median. The blue region labeled uncertainty. It corresponds to the twentieth percentile and the eightieth percentile at each time step, so that's sixty percent. So it's sixty percent of the data fall fell within that range. So it's more likely than not, according to our predictions, that the actual flow that will occur at whatever the lead time is will be within that boundary this blue shaded region.

Speaker 4: It's normal for the blue shaded region to get smaller as if you're looking at bigger watersheds, so there's more upstream drainage area. And that's because the for instance, at this flow right now, we're predicting seven thousand meters per seconds. This may or may not be biased to remember, we haven't done biased correction yet, but we're predicting seven thousand cubic meters per second on June sixteenth at three am. According to this and that, it's hard for there to be a very large window of uncertainty relative to larger and larger flows.

Speaker 4: But if you use this tool instead to look at a smaller tributary with a much smaller drainage area, it's normal for this blue area to become bigger because there's less certainty, and smaller uncertainties propagate into a larger relative amount of error that might be predicted. I'll turn the return periods back on. Now.

Speaker 5: These are the same colors that we're used for the map this.

Speaker 4: They go from yellow to orange, red all the way up to purple. The map stops coloring at fifty, but the chart shows a one hundred year return period as well. The return periods help provide context for how big the flow is. If you don't have a strong concept already of what's normal in a given river, you might look at four thousand cubic meters per second and say is that a lot or a little? And these return periods help you understands. At a two year return period, so theoretically, every other year, perhaps you might expect this river to reach this flow.

Speaker 4: So it's pretty common, but it's hydrologically we start to think of that as the beginning of If it happens just every other year, that's as a random guess, not knowing specifics about the river, that's when flooding would start to occur. The five, ten, twenty five and as they go up these larger return period values, as the flow is predicted to cross into one of those, it's more likely that the flow is going to cause flooding, and that the extent of how wide the flood will be will get bigger. Another way to look at that same information is in the table just below this.

Speaker 4: For each of the fifteen days in the prediction, you'll see the line up until about June eighteenth and on the graph is entirely within the Orange region. And this tables tells you specifically the percentage there are more ensemble members being plotted that are rather not being plotted, And this table tells you the percentage that are within each of these thresholds during each of the days in the fifteen columns.

Speaker 4: The last thing the comment on for the forecast is if you want to go back in time and look at what we projected in the future rather in the past. There's a date picker at the top. Rachel mentioned already that we have or the old forecast. We're in June twenty twenty sixth right now as of the day that we're presenting, the first forecast that we have is July one of twenty twenty four, so we're about to hit two full years of forecasts being stored. And I want to comment on refecasts. The record that you can go back and obtain is whatever we forecasted on that day with no deviations.

Speaker 4: Even if we found an error in the model later or going back, we determined that the model didn't do very well, and we could have adjusted certain parameters to make it perform better. We have not gone back and done reforecasts, so this past archive perhaps the farther back in time you go. There might be errors in your specific watershed which later we've discovered and addressed, but the forecast that you obtain for those dates in the past will not reflect those edits.

Speaker 5: There's no refecasting that happened.

Speaker 4: Last thing that I'll show on the charts is there's a tab to toggle over to the retrospective. Rachel mentioned that there's hourly average is the native resolution, and you can obtain that still. To keep the plot simpler, we just show the daily average and the monthly averages here. You can click on these to toggle them, and it's by default zooms into. Only a few years of data are visible. The slider on the bottom, you can drag the ends to shrink or grow the window that's visible. Normally, I'd suggest you keep this shrunk down to five or ten years if you want to be able to see what's really going on in the data, otherwise it becomes really spiky as it's all visible.

Speaker 4: There are some other plots here, and there's several more of these that are available if you go back to the settings and toggle on the additional plots. The only other one that's on by defaults at least at the time that we're in our workshop today, as this plots that shows the Hydross categories, and if I toggle on each of the individual categories here, these shades of blue and red correspond to the same thing that the maps show if we toggle on the Hydross layer, and by default, just whatever year you're in is visible.

Speaker 4: So that's twenty twenty six currently. But if you're studying this watershed and you're using remote sensing from twenty years ago or something, you might say two thousand and nine is the year i'm interested in, so you can scroll down finds that year on the legend entries over here, and that will appear on the map. Last thing to help you obtain data is on the top right. There's a download icon which will prompt you to download the CSBS. You can click on that and download either of the forecast data or the retrospective data or both and if you like the style of plots and you want to just borrow the plots that we provide in the viewer, every single one of the plots on the top rights of when you hover your mouse over, it will prompt you to download this plot as a p ANDNG and that's for all of them.

Speaker 5: A few other.

Speaker 4: Sort of informational items about the Hydro viewer is up on the top right. We have the ability for you to save rivers. This bookmark icon will show you by default, it's just a dozen or so of the biggest rivers in the worlds and you'll see the ID number that corresponds to that river and the name, and you can shortcut having to browse through the map to find that river by just hitting this chart icon and it will go load the information for that river, and you'll see that when opening up one of the rivers from that menu, the heart is is toggled here.

Speaker 5: If for.

Speaker 4: Your application, there's a river that you're gonna want to visit all the time, perhaps it corresponds to a gauged location, or it's a monitoring point, an infload, or a reservoir, something like that. I just clipped on another new river and I can hit the heart icon and it will auto populate the river ID and give me the chance to save it with a name, so I could say, maybe this corresponds to gauge one, two.

Speaker 5: Three, four, of course that's the name.

Speaker 4: Can be anything that's that's meaningful to you here and I hit add bookmark and the heart turns red. And now that's something that I can frequently go back and find without needing to browse through the map. If there's just a list of, you know, twelve rivers that you always want to check on, or that you're going to visit a couple times throughout the frequency of your projects, this is a good way to convenience yourself on not needing to browse the maps as many times and speed things up. That concludes my presentation on the Hydroviewer.

Speaker 4: Thank you so much for being with us. I'm looking forward to taking your questions and I'll turn the time back over to our hosts.

Speaker 1: Thank you so very much Rachel and doctor Hale's for your presentations and demonstration of Hydroviewer. This is going to be an extremely useful tool for flood prediction and preparedness. This brings us to the end of today's session. Just to summarize what we saw today, we learned about geoglows. It's a partnership initiative among one hundred plus national government and organizations to deliver accurate, open and accessible hydrological predictions on a global scale based on the ECMWF ERA five and integrated for class system data.

Speaker 1: The ERA five models include insitute and multiple satellite data assimilation. Geogloss uses a hydrologic model with inputs from models, gages and satellites to calculate runoff. Distributed vertically uses a river routing model to get river discharge and provides retrospective and forecast hydrographs for global river segments. Geoglose provides written periods, monthly and annual averages and maximums and flow duration curves from the retrospective data.

Speaker 1: Geoglos data visualization can be done through hydro Viewer and RGIS webmap and query and download data from Amazon Web service.

Speaker 1: This brings us to the conclusion of this training on monitoring and predicting floods using our third durations for planning and preparedness. Just to summarize this training, this was a three part training that focused on flood detection on land from remotely sensed optical and microwave data, imagery and flood modeling, and we were discharge prediction using weather, hydrology and river routing models. In part one we talked about global flood product. Second part focused on operodynamic surface water extent and today we talked about geogloss river forecasts system.

Speaker 1: So part one Global flood product was summarized by doctor Dance. LABA was based on Terra and aquamdis and Noah twenty and twenty one vers optical data. It uses red infrared and shortwave infrared reflectancies to detect water. Identification of recurring floods is based on twenty two year flood mask historical data. Also its global two hundred and fifty meters respecial resolution and near real time and twenty three years of archived data are available from Terra and AQUA multilook composites of one, two and three day to assign threshold based water detection is used.

Speaker 1: Terrent shadow and cloud shadow corrections are applied to remove false positive in water detection and composite product recommended by PERSIST for persistent cloudiness and longer lasting larger floods. We saw an example or several examples of how to look at worldview for data visualization of global flood product from different composites, and also how to access data from NASA Earth Data and Large Deck, as we reviewed earlier. Today, Part two was about opera dynamic surface what extent, presented by doctor Renato Franso.

Speaker 1: He described how operodynamic surface water Extent or DSWX has two products, one from optical imagery harmonized land set and Sentinel two data and based on SAR Centinel one sr set. So these two are DSWXHLS and DSWXS one products. Both have relatively high resolution thirty meters their near global Both DSWXHLS and S one products are available from twenty twenty three to present, HLS from April and S one from December. Temporal resolutions are also given here a three day for HLS and six to twelve for S one.

Speaker 1: We saw that DSWXHLS data should be composited over multiple days or combined with DSWXS one data when there are persistent clouds, and we also saw how to merge DSWX data layers in GIS for improved flood detection capability. There was a demonstration of how to visualize DSWX data in NASA Worldview and data Access to data. Finally, we had presentation and demonstration from Rachel Megafin and Riley Hales. They talked about GEOGLOWS, which is based on the acmw ERA five and Integrated Forecast System, uses in situ and multiple satellite data assimilation in these models and also uses hydrologic model and a river routing model.

Speaker 1: GEOGLOWS combines models, gauge data and satellite data and provides retrospective and forecasts. Hydrographs for fifteen day forecasts, written periods, monthly and annual averages and maximums and flow duration curves from the retrospective data also available from a hydro viewer. There's an example I've picked up. This is the cares in Ukraine. This is the same flood that we saw in Global Flood Product and this is the hydrograph or discharged time series for that and you can see from hydro viewer that you do see increased discharge during that period when we saw flooding in this Careson area.

Speaker 1: We also found out that RTIs webmap can also be used for visualizing this data and data can be accessed through Amazon Web Service. Overall, there are several benefits and limitations to what we learn in this training. For example, combining multiple satellite and incituo observations with models provides predictive capability for stream flow enabling risk assessment, planning and preparedness for floods, such as using weather forecast, ideology and river outing models. In geoglows in which multiple observations are used, then remote sensing observations both active microwave and passive optical from multiple satellites and sensors such as Terra and Aqua Modis twenty twenty one years, lands At eight and nine only Sentinel to MSI and sentinel ones are These are satellites and sensors used in Global Flood Product and Opera DSWX.

Speaker 1: They enable the detection of inundation and water logging during and after flood events, supporting response and recovery activities. SAR observations provide all weather capability to detect surface water, while optical observations cannot see through clouds and are therefore unable to detect surface water in overcast conditions. The special and temporal resolutions of these flood products vary by satellites and sensors, such as we saw that opera products are thirty meters, global flood product are two one hundred and fifty meters and resolution Temporal resolution varies from one to twelve days, and they carry uncerties due to various factors such as orange shadow, cloud shadow, and surface types.

Speaker 1: Monitoring flood potential using jeoglos alongside merged flood products from multiple optical inside data sets would greatly enhance flood related decision making activities. So the message is that we can use beneficial information from different satellites and sensors and models and combine them to be better prepared for flood related decision making. The homework assignment is posted today on our training web page and answers must be submitted via Google Forms and the homework is due on ninth of July and you will receive a certificate of completion approximately two months after the completion of this course if you attended all live webinars and if you complete the homework assignment by the due date.

Speaker 1: So we did exercise one and two after our previous sessions. Today the exercise will focus on using geoglows outfestream flow prediction using hydro Viewer. We will be looking at the same flood caase in South Sudan between fifteenth of May and thirty first of May, so you will be looking at streamflow or discharge using geoglows RFS. So about the homework exercises, we had three exercises counting today's exercise, first one on global flood product then a dynamic surface water extent. Both these were through NASA Worldview and today you'll be using hydro viheer to look at this South Sudan floodcase and some of the homework questions you will find based on these exercises.

Speaker 1: Once again, we want to thank all our guest speakers for their excellent presentations, demonstrations and contribution to this training and doctor Daniel Slayback, Doctor Renetto Frasso, Rachel mcaffin and doctor Riley Hales. We thank you all for your contribution to this training. It's the contact information for Rachel and Riley if you have any specific questions about Geoglows, and you can always contact us at our set with any questions that you have. Here is our set website with a lot of information you can find about different trainings and our set YouTube has recordings of all our trainings.

Speaker 1: For questions, comments, or to share how you have applied our trainings to your work or studies, please email NASA dot r SET at gmail dot com and join our mailing list to stay up to date on our latest trainings, and visit our contact page to subscribe. These are the resources we used in this training and share your thoughts. Within the day or two, we will send you an invitation to complete a short online survey and your feedback helps us improve the RSET program. Participation is optional and all responses are confidential.

Speaker 1: Survey data are crucial to helping us understand how to better meet your needs, and we want to help you use our observation data more effectively. We read every survey comment, so please share your thoughts. The survey will come from no Reply at Alchemera dot com, so this is not a spam email that you will receive from this address, so please we request that you complete the survey and your feedback is very important to us. With that, we want to thank you all for attending this training session and now we will go to our question and answer session with Rachel and Riley for geoglows.

Speaker 1: Yeah, thank you everyone for attending this training series and also our speakers for today, Rachel Meguffin and Riley Hale. We'll go to the question and answer session now, I'll read the questions and our guest speakers will help. So the first question is does you close provide data in coastal area areas or island nations and you can unmute and answer the question, thank you.

Speaker 4: Yes, we do have coverage on some islands, not all. The question or the answer that's visible here is that some of the foreseeings and data sets we use have varying resolutions. One additional clarification I can add on to that is that the dem products, the elevation products and things that are used to derive where the catchment boundaries are just we're not available on all islands. It is most places, but not all of them. I direct you to look around on the web viewer to see if the islands that you're interested in is covered.

Speaker 1: Great, thank you. A second question is can this geoglosse modeling be applied to a single basin.

Speaker 4: Yes, it can. The products that we use are all open source. The streams we provide the net forcings are from ECMWF and those don't require a license to obtain. You can either take our pre computed answer using these methods, or if you want to use the if you have some reason you need to run that yourself, or you want to apply some modification to any of the steps, then that you can just download the data from the relevant places. You'll have to do some additional work on your own to modify that odes to work on your area of interest, but everything you would need is open source to do that.

Speaker 1: Great, thank you. Question three is I'm wondering about the hydrological model used in geoglows WIF hydro lease flood.

Speaker 4: Yeah, so this question is asking, I believe about how we calculate the runoff volumes from the meteorology. This we use products from ECMWFS Integrated Forecast System i f S, and the land surface component of that is done with the model called h TESSEL and that is we take the answer straight from that. We apply modeling downstream of it, but our main source of truth for what the runoff depth is comes from that h TESLIL model.

Speaker 1: Next question is where we can download the hydrography shape file.

Speaker 4: Yeah, so there are several questions asking about where defines data for this one in particular, but also the others. I'll direct you to our training websites. That's training dot gogos dot org, and that's that'll be in this answer documents, and it's in the slides and in various places, and you'll find tables that direct you for whichever product you're looking for. We'll give you a description of the name and what file formats and so on and so forth for where you can download.

Speaker 1: That question file. Is there any bias correction too available?

Speaker 4: Yeah, So this can happen in two different ways. One, we provide code, and we've have papers describing a method where if you have your you want to bring your own gauge data, you can apply the method that we use to get our model data and your gauge data that you provide and calculate what the bias corrected flow would be. Alternatively, option two, there are some restrictions on what data we are and are not allowed to use for a broad global application, but we do have in the hydro viewer there's an option to toggle turning on using bias correction.

Speaker 4: This is also available in code, but go to the hydrid viewer and in the settings you can toggle on that you want to use the experimental bias correction and we will do our best guess of what the bias corrected flow is everywhere, using what gauge data we do have that is appropriately licensed and so on.

Speaker 1: Thank you. Question six, can you download the data from the living at last or is it just a visualizer?

Speaker 4: Yeah, so the living atless layer is really meant to just be for visualization. You can go browse all the attributes that are applied to the streams that's used for the styling and animation, but really it's meant just for visualization. There are plenty of other code tools and notebooks other things in the training to help you download if you need to get large volumes of data downloaded.

Speaker 1: Question seven, is there any approach in this method that estimates regulated basins?

Speaker 4: So this question also, I think has kind of two answers. Some of the most impactful regulations could be extracting water for irrigation or running a canal or something. It's really difficult to get information about what volumes of water are used there. We do not have a way to to make that, but there are other ways, like how lakes are managed or reservoirs or dams, those are sometimes able to or do have the information that we would need to make an estimate on that. The current model version does not do an especially good job of accounting for all those things, but Rachel and I are collaborating on doing better in the next model version.

Speaker 4: How will make the model accounts for the effects of reservoirs? So that's some some yes, and some know that we that we whether or not we're able to account for regulations.

Speaker 1: Great question eight is, since the ensemble forecasting model generates multiple possible forecast outcomes, is there a mechanism for incorporating subsequent observational data to update and define earlier forecasts, thereby improving prediction accuracy over time.

Speaker 4: Yeah, this is a really good question. That this is getting a data assimilation essentially. The biggest problem is that countries that have often the greatest need for this kind of supplemental global hydrology modeling often don't have very many gauges, and those that do typically don't have their data publicly accessible on the Internet, so that the lag time between when the measurement is made and when it would be accessible to me to try to do some sort of ingestion run A common filter or something like that is pretty long, so we don't primarily because it's just not feasible the computationally, the data doesn't become accessible on a short enough lead time to try doing that.

Speaker 4: However, as I said before, the code and data sets and everything you would need to go back in time and do a re forecast is all open source and available, So even though we don't do that at the global scale, that is possible, and you can check out our code and send some emails to chat more about that if you would like.

Speaker 3: Nice.

Speaker 1: Question nine is has geo gloves been evaluated or benchmarked against other operational flood forecasting platforms such as IFAs or Google flood up.

Speaker 5: Yes we have.

Speaker 4: Sometimes those are published, and I gave you a link to the publications there. That's again on trading that geobos you'll find a publication section.

Speaker 5: Also.

Speaker 4: We do similar kinds of benchmarkings and comparisons whenever we're participating in w MOO activities, including some that are ongoing right now, so I'll direct you to those publications to see if that gives you the answer that you're looking for.

Speaker 1: Question ten is does geoglows aim to serve as a flood forecasting tool or stream flow stetus and forecasting data provider.

Speaker 4: Primarily our use cases and that we've collaborated with country hydromed agencies has been for flooterly warning applications. We do have examples of applying this data and building tools or apps or other things around other applications such as water quality or irrigation or reservoir management, but those have are more in the minority to the majority of our collaborations have revolved around flood forecasting. We are looking to expand the number of pre made things that we provide on that on those other application areas.

Speaker 4: Though.

Speaker 1: Interesting elemon is how are the RFS simulations validated? Are the calibrator or validated using hydrometric stations?

Speaker 4: Yeah, so we have I've already alluded to this slightly that we have a lot of gauge data. It's growing. But the some of the problems in publishing about all of those gauge measurements is that they're not all permissively licensed for us to publish validation against that, that would necessarily mean we're giving away some of those gauge measurements and they're being put on plots and things, and that's not always allowed, So we do validate against a very large set of inciti measurements. Not all of them make it into publications, and not all of them can be shared, so there's some difficulty in how those gets share or how we're able to share that with other people.

Speaker 1: Question twelve, is there any thought to creating our library for jyo glows?

Speaker 4: Yeah? This I get this question occasionally. The geoglows main team does not have any our developers on it, so we're not in a great position to create that ourselves. We have a Python a job script job excuse me, job a script version. If you would like to take the lead in creating in our version, I'm definitely open to having that conversation.

Speaker 1: Question thirteen, is there any opportunity to get hand to hand training on setting the geoglows model model?

Speaker 4: So I would say yes, the I would direct you first to training dot geoglows, and there's some links here to a Google group and we have a YouTube channel with some webinars, So I'd hope that this covers a lot of the the questions and upfronts kind a direction that you're looking for, but you would need to reach out to us to discuss if you're looking for live trainings and things like that.

Speaker 1: Thank you. Question fourteen, how does geoglows differ from glow fasts?

Speaker 4: So this is a really good question, getting right to the core of how our model operates. What Rachel was presenting at at the beginning. We in rfs and glow fasts share the same ECMWF MET predictions that ensemble prediction the core. We're both built on the same meteorology predictions. As the farther you go down the line, though, we start to diverge on the different additional modeling steps that are applied. Because of that, that are met forcings at the core are the same, so we'll often show very similar signals.

Speaker 4: It's sometimes called or shapes on the graphs. If glow fast predicts it, we will probably also predict it, and vice versa because we share that same core, but because we use different modeling methods downstream of the mets and the land surface parts, they won't be identical, but they'll often be very complimentary.

Speaker 1: Question fifteen, incheoglows? Can we searched for a river by name?

Speaker 4: Yeah, this is another really common question we get. There's some difficulty in getting a really good authoritative source for what river names are. So in general, I would say no, you should use the ID numbers because that is guaranteed to be unique everywhere. For instance, there could be multiple rivers names the Colorado River or the fill in the blank. There's probably multiple rivers' names are sharing the same name. We are exploring ways to try to get some river name attributes applied to the streams so that you could do some searching and filtering, but that's not available just yet.

Speaker 1: Question sixteen Real Geo Clowse hydro Viewer and the Hissory Living Atlas eventually visualized trout conditions or below normal flows. I assume that pulling the data and Python would allow for this type of visualization in addition to flooding in local applications.

Speaker 4: So I think the answer is probably yes eventually. This is an ongoing area we're exploring right now. Actually, so I don't have an ability to promise specifically wet maps or charts or the timeline or anything like that, but I would say yes, we do want to have more on that front for now, though you'll have to download the data yourself and generate the visualization you're looking for.

Speaker 1: Wonderful, We'll take one more question, because we are almost at the at the end of our session time, how is the uncertainty computed?

Speaker 4: So the statistics for meteorology can get kind of complicated. The main thing for you to know is that the meteorologists would advise you to view the fifty one different predictions, which in turn we use to make fifty one different discharge predictions. Those should be treated as if they're equally probable, so that you can then do many many directions on how you want to analyze that and how you calculate uncertainty. I think this question is getting more at when on the graphs that we present, what's the blue shaded region?

Speaker 4: How is that uncertainty estimated? This is very unsophisticated, but it was done this way intentionally, so it's very simple and most readily understood by a big, wide array of audiences that have sometimes a great deal of statistical training and sometimes not very much at all. We take on every kid step, the values are sorted from max to minimum, high to low, and the middle sixty percents is that considered the uncertainty and that that checks out because the most extreme values tend to have very few of the members predicting that they happen, and the most common range of values sit in the middle.

Speaker 4: There are, of course, many other ways that you could go about estimating certainties. This one has chosen specifically because it's simple and it's understandable by a wide range of people.

Speaker 5: And the hydro viewer.

Speaker 4: Remove that option and not view the simplified forecast and view all the ensemble members and see different ways that uncertainty you can be presented. That's just not the default for the communication reasons.

Speaker 1: Great, thank you so much. So we have a few questions that we will be posting. We will be answering and then this question answered document will be posted on our training website in a week or so. With that, we want to thank our speakers once again, Rachel Megafwin, Ralei, thank you so much for your contribution, excellent presentations and demonstrations. This is going to be a very useful tool. So this brings us to the end of this training and we hope to see you at future our set training. So please subscribe to our listener or just go through the link and you will can subscribe to all the future upcoming trainings and stay up to date.

Speaker 1: At this point, I also want to thank our our set team, my colleague Sean McCartney, coordinator Natasha Johnson Griffin, and editor Maria Marabito. We also have instructional designer Kevin Fuel and our other coordinators brock Levin, Salvon, Salvin Hudson or Doy We thank you all and Sharry Morris so thank you for your help and thank you all for attending this training and we hope to see you soon. One last reminder, we have an exercise posted today using geoglows height of viewer and then some of the questions will.

Speaker 3: Be based on that.

Speaker 1: So you will have free question three exercises and one homework assignment and once you finish that and submit the homework by dud it, you will receive certificate in about eight weeks or so after the after today. So thank you all, and please do not forget to take the survey when you receive a survey link. We really look forward to your comments and thoughts for our future trainings, so thanks all once more,

The Story