NASA ARSET_ Overview of Global Flood Product Derived from NASA Optical observations

0:00 Speaker 1: Hello, and welcome everyone to this Applied Remote Sensing training on monitoring and predicting floods using Earth observations for planning and preparedness. My name is Amita Meta, and today's session will focus on overview of global flood product derived from NASA optical satellite observations and for that we have a guest speaker, doctor Daniel's Layback from NASA Goddet Space Flight Center.

0:29 Speaker 1: We'll start with a brief introduction to Applied Remote Sensing Training Program or our SET. Our SET is part of NASA's Earth Action Program and it provides a cost free training on remote sensing satellites, sensors, methods, and tools. Trainings are provided in a variety of thematic areas listed here agriculture, disasters, ecological conservation, health and air quality, water resources, and wildlent fires, and they are tailored to audiences with a variety of experienced levels. Our SET trainings are online and there are also in person trainings offered.

1:10 Speaker 1: They are live and instructor led, or there are asynchronous and self based training available from our set website. Most RSET trainings are bilingual or often they are multi lingual, and most of the trainings are translated into Spanish and so materially is a labl from our website. Our set only uses open source software and data and accommodates differing levels of expertise. We'll start with this training series monitoring in predicteen floods using Earth observations for planning and preparedness.

1:47 Speaker 1: Now, floods are the most common and widespread of all weather related natural disasters. Floods they last for a few minutes, like in the case of flash floods, to weeks, and long lasting floods placing a huge burden on communities. It is very well recorded that floods caused loss of lives, displacement of communities, long term impacts on human health and well being. They damage infrastructure and economies, and destruct ecosystems. Every year, a number of floods occur all around the world. For example, these images from World Meteorlogical Organization.

2:29 Speaker 1: Major floods occurred in Asia and in the United States in twenty twenty five. Some of the countries affected were China, India, Nepal, Pakistan, and the Republic of Korea. In twenty twenty five, in United States, Texas and New Mexico, suffered flash floods and that killed more than one hundred people. And you can read more about flooding at this site provided by World Mythological Organization. And because of that, it's really important to monitor and predict flood so that there can be better preparedness and response for a flood disaster.

3:14 Speaker 1: These two figures are taken from a relatively recent paper by young Men at All and what it shows is a number of events of different types of disasters here and it's between nineteen seventy five and twenty twenty two. What you see here is that the blue bars shown here they show number of flood events and other disasters are also shown in different colors. But the main point here is that overall flood frequencies are going up as we go through this timeline, and number of flood events is the largest in all disasters.

3:55 Speaker 1: If you look at this figure it shows different types of floods. You can again see that riverine floods in light blue, they are the largest number of flooding occurring, and then flesh and pluvial floods are also relatively large. Another thing to notice here is that there is overall increase in flood frequency but the rate of change has slowed down in later part last twenty years or so, and also to see that there is intra annual variability of flood events. As we will see in this training, remote sensing observations are quite useful in monitoring floods and so we will be focusing on flood applications and impacts based on satellite observations, specifically optical reflectancies and synthetic aperture radar backscatter data from a number of satellites.

4:55 Speaker 1: They've been used for detecting floods. Also, there are weather model and hydrologic models which use remote sensing observations in the models such as precipitation, soil moisture, digital elevation, and lend cover and they can support both flood monitoring and prediction through these models and they are quite useful for early warning, for enhance safety planning, rescue and relief operations, and planning infrastructure and economic impact mitigation. There is a NASA Flood dashboard that you can look at provides information on floods and mapping flood impacts.

5:39 Speaker 1: Overall training learning objectives are that by the end of this training you will be able to identify data sets in NASA Global Flood Product, which is based on remotely sensed optical observations. Use the web tool NASA Worldview to access and visualize flooded regions from globebal flood product available from multiple satellites starting from twenty twenty one to near real time. Then identify opera dynamic surface water extent data for flood detection derived from optical and SAR observations, and access and visualize opera dynamic surface water extent data for flood events using NASA Worldview.

6:24 Speaker 1: Identify the capabilities of Geoglows River Forecast system for global stream flow prediction, and use Geogloss hydro Viewer to access globally available retrospective and predictive stream flow for selected rivers. There's a prerequisite fundamentals of remote sensing that's available from this link on our set website and you can get basic information about satellites, sensors, center characteristics and that will be useful in following some of the concepts used in this training.

7:01 Speaker 1: There will be three parts to this training today. As I mentioned, it will be about Global Flood Product. Second part will be on opera surface water extent based on SAR and optical observations, and the final session will be on geoglows. So next two sessions are next week on twenty third and twenty fifth of June. There will be one homework assignment posted on our set website on June twenty fifth, which is the last day of the training, and the homework will be due on July ninth, and a certificate of completion will be awarded to those who attend all live sessions and complete the homework assignment before the given due date.

7:44 Speaker 1: We'll start with today's session overview of Global Flood Product derived from NASA Optical Observations. So the objectives for part one are that by the end of this session you will be able to identify NASA Global flood product derived from themoodly sensed optical observations, recognize the data sources, spatial and temporal resolutions and limitations of optical based flood detection, gain experience using NASA Worldview to access and visualize global flood product in near real time, and know where to access flood product.

8:22 Speaker 1: Files are available for download. Here's the outline for today. Doctor Daniel's layback will start with examples of flood cases using Global Flood product or near real time global flood product. He will provide history and background of global flood product, then overview of nat flood products, a flood detection approach based on optical imagery from modis on Terra and Aqua satellites and weirs on NOAH twenty and twenty one satellites. Then talk about flood product EVALUEUA and case studies of flood detection.

9:03 Speaker 1: He will also demonstrate near real time global flood products access in visualization using NASA Worldview. A note about asking questions. Please put your questions in the questions box and we will address them at the end of the webinar. Feel free to enter your questions as we go. We will try to get to all the questions during the question and answer session. After the webinar, the remainder of the questions will be answered in the Question and Answered document, which will be posted on the training website about a week after the training.

9:41 Speaker 1: With that, we want to introduce our speaker for today, Doctor Daniel S. Layback. Doctor Slayback is a research scientist in Biospheric Sciences Branch at NASA or Space Flight Center. Doctor Slayback's work focuses on the application of remotely sensed imagery to study a variety of Earth system elements and building systems to generate and deliver data products. His current activities include examining the causes and impacts of land cover and land use change in high altitude Indian peatlands, development and operation of a near real time system to generate global daily flood map products, monitoring and quantifying the evolution of a new volcanic island in Tonga, and building data services for query and distribution of petabytes of high resolution imagery.

10:40 Speaker 1: Previously, he has worked on a range of projects, including Indian leisure change, studies on monarch butterfly over wintering habitat, evaluation of environmental codilates, of vertebrate diversity in the Western US, local and regional defas water station assessments, examining trends in global vegetation dynamics, and calibration of EdVance very higher solution radiometer data products. With that mean my doctor slay back, Doctor slay.

11:12 Speaker 2: Back, Okay, Well, thank you, Amita for the introduction. You know, as she said, I'm Dan Slayback. I'm a research scientist at the Goddard Spaceflight Center. I've been working on the flood product here for about the past fifteen years. So I'm excited to show you all an overview of the product and hopefully help you be able to use it more more usefully for any work you might have. So launching right in, I'm going to give just a brief overview of the talk. I'll talk about the product's history, a brief overview of what the product is, the approach, in other words, how we generate the product, recent updates, the product we had drop up the last December, the evaluation, limitations, distribution, the archive, and future directions.

12:08 Speaker 2: And then followed following that will be a short demo of showing how to access the product and look at it online. So a brief overview and history. The main features are it's a daily global two hundred and fifty meter resolution product. It's based on twice daily observations. It is neural time. This is one of the key features of this product compared to others. So within three hours of satellite overpass and usually sooner than that, we have a product out and then it updates if there's a new satellite overpass from the other satellite.

12:49 Speaker 2: We're really just detecting water. We detect non flood water and floodwater, and we categorize it as flood based on reference water layers, so it's really worth or a water detection project. The limitations, of course, this is based on optical imagery, so cloud cover is often a problem, as you can see an example in the right. Although it did capture that flood quite well, we can't see through clouds unfortunately. Flash floods and small floods, you know, floods that are not on the ground very long, are very difficult to capture, both because we may not observe them because they're not there that long, and because they're often small and spatial extent, so the two indred fifty meter IXTL size can be a limiting factor.

13:34 Speaker 2: The brief history, so this all started over twenty years ago when Bob Brackenridge at the Dart and Flood Observatory was using the what was called at that time Motus Rapid Response imagery. These were image JPEGs made from Modus observations you know within you know, shortly after after acquisition. So he was manually taking those and making flood maps out of them, which was quite helpful to be able to map flood within hours of having a satellite observation. So back in twenty ten we decided to operationalize this so that it didn't require somebody to take their time and effort to manually grab these images and apply threshold and try to map flood.

14:23 Speaker 2: So that's when this project really started. And our initial product is for those of you that have used it for some time, was the what you see in the lower right corner. There was a graphic map like this for by tend to retiles. There were also geotip products you could download, and so that carried on for about eleven years or twelve years, and then we began around twenty twenty transitioning this into Lance operational production systems. The previous system was really a PI run system based on a PI server, and it was not terribly robust, but it worked for the most part, but it was it had limitations.

15:07 Speaker 2: So we transitioned fully into Lands and the mote Apps production system that generates all the Modus products. This is much more robust, gets out the door a little bit quicker. I need to advance there, and that went into production in early twenty twenty one, and then we discontinued the old product, which we call the legacy product. At that point, about a year ago, we introduced a beer's product eventually take over from the Motus product. Is the Motus sensors are eaching end of life. So now I'm just going to show a few examples of the product and action so you have a better feeling for what it looks like and what it can do.

15:54 Speaker 2: This is a flood product from Southeast Asia. Way back in twenty eleven, major flooding, particularly Central Thailand, Central Cambodia. The Lake in the middle of floods pretty routinely. That the flooding in Central Thailand was more unusual as a significant event. So this is quite scaled out major flooding that we're able to observe. This is flooding from the Ukraine Coca Dam breach in twenty twenty three. The dam was breached about two or three in the morning on the sixth of June, and within twelve hours we had a product out the door showing this flood extent.

16:37 Speaker 2: So this shows you a couple of different things. You know, clearly we're just detecting water and calling it flood or surface water based on a separate reference water layer. You see the pixelation of the two hundred and fifty meter pixels here.

16:56 Speaker 2: So it's in particular in Kurslon City we're not detecting much flooding because it's an urban area and there's although the roads may well have been flooded, a lot of the buildings remain dry. So that is one problem limitation of the product. There's another example from the twenty twenty two Pakistan floods, major flooding in Pakistan that you all might remember. It was all over the news. This is in the Worldview app which I will demonstrate later. Again a major large regional flood. Here's a time series of an earlier flood twenty fourteen in Saying region, showing the daily product over ten eleven days.

17:39 Speaker 2: Where it is so you can actually see the flood moving downstream and slowly receding. Was the end of disciple. So again, these large floods are it's quite good at capturing these generally, Okay, So here is an overview of the approach. There's four basic steps. I'm just gonna review them quickly here and then i'll go into them in more detail on the following slide. So the first step is water detection. This is where we apply an algorithm to detect water. This is, you know, the key thing going on here water detection.

18:22 Speaker 2: Step two is what we call multi look compositing. This helps deal with some of the false positives we get from shadows and helps us, you know, sort of look around clouds as clouds move from image to image. So we combine several images the water detections from several images. The third step is terrain masking that gets rid of false positives in mountainous terrains where we have a lot of shadow from the terrain. And the fourth step is flood identification, which is simply comparing the result which we've gotten at that point to reference water mask and if the water matches, we call it surface water, and if it doesn't match, we call it flood.

19:04 Speaker 2: So I'll go into these in more detail next. Okay, So first step one water detection. So the first question is the data, the incoming data. So we have the MOTUS source imagery. MOTUS stands for the Modern Resolution Imaging Spectral Radiometer. This is an instrument that's been on two NASA satellites since the Terra satellite since nineteen ninety nine and the Aqua satellite since two thousand and two. They're both very much at end of life, so the missions will probably be ending in the next year, which is why we have this separate viers product, which I'll show you the details on the next slide.

19:44 Speaker 2: But the two products are really parallel. Everything's the same except the input data essentially and some differences from that. But everything else how they're generated is identical. So the key features of MOTUS it's two hundred and fifty meters global twice daily. With these two satellites, we can't see through clouds, so we are blocked by the cloudy planet, which it is often cloudy. On the bottom you see the data collect from ERA from the Terror satellite, and on the right from the Aqua satellite.

20:19 Speaker 2: You'll see you see these black wedges at the equator. These are swath gaps. These move from day to day so they are not always in the same place, but it can cause limitation of the data available equatorily. So that's the overview of MODUS. Here's Veers very similar. We also have this on several satellites. The ones we use are called now twenty and twenty one. They've been up for nine four years. GRED and seventy five meters resolution, so this is coarser than the Modus instrument, but also global with two satellites twice daily.

20:58 Speaker 2: They're also both in the afternoon, only about fifteen minutes apart. Like Modice it was more like three hours apart. And one of the nice features is we get rid of the equatory of swath gaps, so you don't have those areas of missing data along the equator. On the next slide, i'll show you the water detection algorithms. This is what Bob Brackenridge developed way back in the early two thousands and we've maintained it since that time. Is the key feature is the near infra red red threshold at the top.

21:36 Speaker 2: The other two conditions to sort of get rid of some edge cases, but it's really a threshold based approach. If you apply that threshold to a single input image like the Modus aquad image shown in the middle, you get simply you know, water in blue or no water in white, or no data or we didn't have any of the data in gray, which includes the swath gaps which see here, but also where the data is bad. In this case, the imagery was saturated at very bright clouds. That's much less a problem currently, but in the past you would see those issues.

22:14 Speaker 2: But we're certainly not detecting water under a bright cloud in any case, so that doesn't affect us. So on the next slide we move on to multi look compositing. So once we detected water, that's great, but we do have some issues with false positives from shadows looking like water in particular cloud shadows entering shadows. To deal with the cloud shadow false positives, we use this multi look compositing, but also allows you to collect imagery over one, two or three days because the clouds will move, they may move out of the way and you may be able to see the surface.

22:55 Speaker 2: So depending upon your timeline as a user, for how recent you need your input data to be, you might get a better product with the two or the three day products then with the one day because it may have been less valid yesterday. So the basic idea behind the multi look compositing is requiring multiple water observations for a pixel to be marked as water in the output to be carried through. So for a two day product, we in theory have four input images, two from each satellite per day, and typically the threshold although it varies a little, but it's typically half of the number of observations.

23:41 Speaker 2: So if we have four observations for a pixel, then we're going to require but two of those be detected as water in order to call the output pixel water. So applying that to this set of four images, we get that compositive water math on the right. Again, it's not telling us FUD or not, but it's telling us where we think we really most likely to have water.

24:07 Speaker 2: The next step is terrain false positive masking. So along with the cloud shadows looking like water, we have terrain shadows looking like water. We have two things we apply for this. We have computed terrain shadow masks, where we compute the topographic shadow at the midpoint of the month, so monthly for each month at the two different times for the satellites, and that gets rid of most of it, but there can still be some residual so we apply a general topographic mask based on the hand algorithm, which stands for height above nearest drainage.

24:43 Speaker 2: This essentially masks out areas where it is unlikely for footwater to accumulate based on the local photography, as there would be drainage draining it away of course, a flash flood might be there no temporarily, but we're unlikely to observe those due to the resolution and the time issues. I'll have a more detailed example of this in a few minutes. But the next step is, once we've applied the mont compositing and the masking, we compare the detected water to a reference water layer, and if it matches the reference water layer, we call it surface water, and if it doesn't match, we call it flood.

25:22 Speaker 2: And so we just classify the output very simply on that way. The reference water layer is based on a separate Modus product, a totally different algorithm than ours. It's a yearly product in the mod forty four W, so it's a very conservative approach and that generates the out. But now, as I mentioned, we have saw on some other slides, there's also a recurring flood layer. This was a new feature we added last December. It's the same idea. We're detecting water all the same, but we're comparing it now both to a yearly reference water layer and to a monthly recurring flood mask.

26:03 Speaker 2: And that recurring flood mask is based on analysis of our twenty two year product archive and any areas that are regularly recurring in October, for example. In this example, we're going to classify the output differently as recurring flood So this helps the user identify when the flooding, which may be extensive as you see here, is actually rather routine, or when on the western side of toy Saffy Can Cambodia here where it was actually not routine but also fairly extensive. So some details. I'm going to run through some detailed examples of what I just showed you, just to make a little more concrete.

26:49 Speaker 2: And again, the key issue we're dealing with is false positive masking. Shadows look like water. Cloud shadows look like water. Experience shadows look like water. We only have two bands at the highest resolution at two hundred and fifty meter resolution of three seventy five, and the beers in the red and the infrared, and water is very difficult to discriminate from shadow with just those two bands, so this is sort of the main limitation of using those bands. But we want the high resolution, so we use those so train shadows mostly in mountains in the winter.

27:27 Speaker 2: Our first cut approach is applying to computed terrain shadow masks and then the second cloud is to hand. I'll show that in a minute cloud shadows. The main approach is really just the multi look composity. The one day composite does not typically have a threshold greater than one, so any water detection will go into the one day product. And if you have clouds, you may likely have a lot of false posits, or you may not every cloud shadow gets detected as water, but it can be, it can be, it can be problematic.

28:00 Speaker 2: The two in the three day much much less, but I'll show you that as well. And so the key to all of this is looking at the source imagery and the world future, as I'll show in the demo, is a great way to be able to do this easily.

28:18 Speaker 2: So here's a detailed example. Looking here, we have a Modus aqua image from the Alps in winter. This is a bit of a mess. This is not ideal, certainly, but it shows some of the problems and how we're dealing with them. So we have clouds in white or pinkish. This is a false color composite out of symmetry. Snow is in Cyan, very bright. There might be some water here. It's sort of hard to see. Like Geneva is covered with clouds mostly. So if we apply our water detection algorithm to simply that single image, this is the result.

28:54 Speaker 2: So all the yellow is the detected water, and so there's some real water parts of like Geneva are detected. There's a lot of not real water terrain shadows next to the mountains, and there's a lot of cloud shadow next to the clouds that is not real. So this is you know, essentially useless at this level. You would have no idea what's real or not. If we composite this over two days, we get rid of a lot of the cloud shadows, allse positives if I fot back or you kind of see that there's some that are persistent.

29:28 Speaker 2: So it doesn't solve all of those with a two day The three day would do better, but it's doing a pretty good job, but it's doing really nothing for the terrain shadows, which if anything, only grow. But we get with two days of data, we're getting, you know, pretty good coverage of Lake Geneva. Not everything, but here's just the terrain to show you, you know, how how much terrain is involved in the Alps in Italy, France and Switzerland. Here is the terrain shadows computed from February morning and thirty in the morning, and then from one thirty in the afternoon to the sweep be relevant to aqua, the super be relevant to the Terra space craft observations.

30:13 Speaker 2: So you see they shift over time. If here's again the detective water, the composited two day detections water detections. If we apply those two terrain shadows, it does that, so we get rid of back and forth. We get rid of a lot of it, but there's still some residual terrain shadow along the mountain bridges. So here is the height the hand mask. This is the computed height above nearest drainage. It looks again like ghography more or less. Here's we threshold that by thirty meters. This was imparibly determined and that seems to generally work well all over the Earth.

31:02 Speaker 2: And once we apply that we it gets for most of that remaining noise. So most of the yellow here is what we is probably real water either you know, small lakes. I don't know that there was actually any flooding in this example, but any floodwater would also be appearing here. So we've cleaned it up quite substantially with this approach the multi look compositing for cloud shadow masking or to deal with cloud shadows. Again, we sum up all the water detections over the observation period, and so for every daily product, there's a one day composite, there's a two day, there's a three day.

31:46 Speaker 2: So you can compare those to one another and see which is best for the particular events an area of interest. The one day again only requires one water detection, so it will probably contain cloud shadow false positive if there are clouds. If there are not clouds, it is probably the best product to use because you are seeing today's data or the day of the product data. The two day is a good comprom a promise between one day at three day. You require two water observations, so that gets rid of a lot of the noise, but not everything.

32:23 Speaker 2: The three day really has a very few false positives, but it's extended over three day look back periods, so it's not going to it may not be as current. Okay, So you know this raises the question, of course, is the main sort of user complication of the product is which flood product should you use? And it really depends on the cloud conditions and your tolerance for false positives or false negatives. If you're really looking at an emergency situation, you only want the most up to date. You probably want to look at the one day, but you need to verify that you don't have cloud issues with that a good balance, it's the two day.

33:08 Speaker 2: The best approach is really to check the imagery, and I'll be showing you that in a minute. Here's an example that This is showing our not yet public our flood viewer. It's similar to Worldview, but it has the one day product available in it, which is why I'm using it here. But regardless, it's showing the flood products and you can look at the source imagery as well. So this was for a flood reported in southern Poland in September of twenty twenty four. If you look at the raw product, the pink here is the flood.

33:44 Speaker 2: There's a lot of flood showing up, so you might think, wow, you know, this is a significant event. Seems to follow linear type patterns, so those might be rivers, so this might be good, but you really should be looking at the imagery. So in this side we've clipped on the Modus Terra background image and impulse color and there's not much cloud. There's that cloud in the top north. But you might think, okay, this is pretty good, but you should keep looking. This is the one day composite. You never know what you're going to get.

34:18 Speaker 2: If you look at the Modus aqua image, we see a lot of cloud, and suspiciously, we see a lot of the flood falling directly in the cloud shadow. Not all of it, but some of it. So now you know you have something you need to be paying attention to. Here's the two days, so I will get rid of most of that flood falling in the in the cloud shadows. There might still be some, It's hard to know for sure, but this is one of the images.

34:51 Speaker 2: This is the aqua image again. But again you would want to look really at four images for the two day product, today's ra An Aqua and yesterday's Terara Naqua. The same thing applies for the beer's product. You'll just be looking. You know, it's twenty and twenty one. Just terror here's a three day composite for the same event. So now we really we drop off the amount of flood being reported quite substantially, and you know, maybe too much because you can you can barely see in some of those areas that probably more dark floodwater than it's being reported and red And if you again look at the imagery you'll find it two days ago.

35:31 Speaker 2: So the initial day of the three day composite, it was totally cloudy. So you're not getting any additional water observations by expanding to a three day but it is raising the threshold that of the number of water protections that had to occur. So this is a case where the three days certainly very conservative vestiment of flooding, but probably you know, it's missing some Okay, So moving on some recent updates with the product. Again, as I mentioned the recurring flood. This we introduced last December with released one point one of the product.

36:11 Speaker 2: The objective was to distinguish recurring flood from anomalous unusual flooding events, which in previous products for the last fifteen years, any unusual water was reported as flood and certainly it may well have been flood, but it may have been you know, un not as some flood the recurring annual seasonal flooding. So we to do this, we had we reprocessed the full twenty two year modus flood product archive or that doing that allowed us to generate this, and the temporal framework is using a rolling a three month window.

36:50 Speaker 2: We only require that three we detect flood within for three days within the window, but we require that happens within the third of the years, so seven of the twenty two years. So that seemed to be the best approach to sort of do this somewhat conservatively, but not too conservatively. But again, any user could take the final product and reclassify it with their own water masks. These are just data values in the in the flood product, so really simple gis operation to reclassify it if you have an improved annual or recurring flood masks for your region or zone of interest.

37:32 Speaker 2: And in this example for the Mississipi flooding, and you could see on the left or you'd see on the right, a lot of it was actually unusual flooding. That's what would have been read. Here's some examples. Here's northern California. This often floods in the winter, This northern Sacramento Valley. The previously we were constantly reporting lots of flood in northern California, and these areas were underwater, but they are all than underwater in the winter. So the new product is showing, you know, most of it's reported is recurring flood, not flood.

38:08 Speaker 2: There's the I think the same Mississippi example. Again, most of this was actually unusual flooding. There were some areas of recurring flood, but a lot of it was fairly unusual, so that was significant events. The severe product doesn't really look different than the modus. And again this is the one day here because there were no clouds or a few clouds that were complicating matters in this Punjab in September last September. Again mostly this was not recurring that dairy does flood quite regularly, but this level in this region, this was for September, this was unusual.

38:53 Speaker 2: And this is actually a custom composite generated from one week of data. So this is you know, things you can do with the raw data. You don't like the one or two or three day You could make a week long composite. It just added up in the gis essentially evaluation, so we uh with the initial the legacy product. So fifteen or so years ago we did an evaluation procedure to determine are we detecting you know, how often are we detecting flood? Is this working? Or you know what areas are problematic? We see differences between detecting flood and non flood water.

39:33 Speaker 2: The main caveat with the evaluation we did was that that we don't have vigorous ground true data sets for a global product of flooding that is very expensive and collect and it's always going to be biased towards accessible locations versus other areas. So we did a very fairly simple qualitative assessment using typically the Motus imagery itself or Lance adam injury that was available on the same day to determine, you know, is this really water and flood or not. So we picked fifty events, fifty flood events from the Dartmouth Flood Observatory Master List with global distribution, different line covers, et cetera, and fifty permanent water sites so not flood but just lakes rivers to evaluate whether we're getting those as well.

40:27 Speaker 2: And again this was connected for the legacy product, but the core algorithm here has not changed, so it's it's and we've looked at this for the the Lance product that was introduced five years ago. And there were no significant differences. So here are the results. So for the flood detection sites which are shown on the top and the permanent water sits in the bottom, you know, roughly a third for both set of sites it was too cloudy, so we don't count those. It was too cloudy to really tournament anything, so we don't know.

41:02 Speaker 2: But of the two thirds where it was accessible for the flood, you know, the rating the qualitative raiding given was good or better for two thirds of the case of sixty six percent and eighty five percent for the permanent water, so we know. You know, there are certain flood events that it does not work well on. These are typically small, limited in extent or limited in time, and which is obviously not a problem with permanent water, so it performs better with both sites. Here's some examples from that.

41:36 Speaker 2: So here's flooding in Buzzan, Herzegovina in twenty fourteen. Lance At eight. Image shows you quite clearly the flooding and our product is picking that up quite well. There's another example from kim Talk to the Mississippi River area. Again, you can see this in Motus itself. You don't even need lance at it's obviously flooding, and so our product is s working while here's an example where it didn't work well. Volcanic rock is often very dark and therefore like shadows often detected in some places detected as water.

42:12 Speaker 2: So we were initially routinely reporting flooding on the center of Hawaii Island, which is, you know, meta physically, biophysically impossible to have a lake on the top of the mountain like this.

42:27 Speaker 2: So we at the time we fixed this by editing the reference water map. But the hand mask entirely solves this problem. But this is this You will still see this in other areas where the hand mask is not helping, the volcanic dark areas, So there are some issues. Some areas of issues limitations of the product. Again, the clouds, as I over and over false negatives. We can't see through the clouds, so if it's persistently cloudy over an event, we're not gonna be able to see it false positives. Those cloud shadows detected.

43:09 Speaker 2: This swater Thames on the right shows both. So there's probably some flooding on the right part of the image under the clouds that we're not seeing. There's definitely flooding on the left that we are capturing and on the edge we may not be able to make it out, but there's false positives in the shadows of those storiated clouds on the edge. Two hundred and fifty meter resolution, we're not going to pick up small spatial extent flooding or flooding that's very rapid and not there when we observe it.

43:36 Speaker 2: Even though the two we have twice observations daily, they're at most three hours apart with notice, so there's twenty one hours when we're not having any observations. Land cover limitations under canopy, we cannot see the ground, so we're not seeing floodwater. In urban areas there's typically too many dry buildings, you know, even if the road, if the streets are flooded, and typically not picking that out. And changes in recent surface water extent, there's a tremendous number of new reservoirs being built all over the world, which you look at this product in detail, you start to realize and those will be reported as flood for a couple of years until they make it into the reference water layer, and then they'll be reverted to be to be reported as surface water.

44:24 Speaker 2: It changes in river course which you've been see in the tropics see some issues there. So but again, if you have a better reference water layer, you could certainly be classify the product accordingly for a local area. Here's an example that shows a bunch of these limitations in the same area and the same or two different products of the one day product on the left and the two day product and the right to have cloud obscuration. Of course you're not seeing things on the bottom and the right on the two day product.

44:56 Speaker 2: You know you can see the cloud and it's causing us not to report flood that we actually do report in the one day product. In the left, clouds shett for false positives are much more extensive in that circle area and the center left on the one day product than they are in the right. It's probably a few still on the right, but it's it's it's cleaned up urban areas non pen is see very little flooding there. Maybe they have really good drainage, but also probably just because it's an urban area, we are not able to capture it.

45:27 Speaker 2: So again, looking at different products, looking into source symmetry can tell you a lot about which product is the most relevant for your particular area of concern. So the archive we have now

45:45 Speaker 2: much effort and many requests released an archive of the product. We reprocessed the product back through the start of Terra, so there's twenty three full years where you have both Terra and Akala from twenty and three to twenty twenty five. Initial two years just have Terra. So the product still works, but it is you're you know, you're capturing less flood because you have half as many observations. We're working on evaluating that to provide more information about that, but it can be used. It's just you'll see a continuity between them ERA plus Aqua versus the Terra only ERA.

46:22 Speaker 2: The main difference in the processing was the reprocessing used final geolocation and final surface reflectance, so the neural time products take some shortcuts to get those products out the door quickly within three hours, and some of those are in the geolocation and some of them are in the calibration. So when we reprocessed, we're able to use all the final data for those. So this fixes minor problems you would see occasionally you would see a geolocation error causing just to report flooding in the river as if it had moved or the coastline.

46:58 Speaker 2: So those things just pear from the archive, which is nice that they were not common, but you can find them here and there occasionally. For twenty twenty six onwards, only the Modus NRT products are being archived. Currently none of the Beers products are being archived, but we're hoping to implement archiving for the NRT veers as well. Here's an example of you know what you can do with the archive. This is the recurring flood masks for this area in Cambodia to South Lake. This is a great example because the zum De scale you can see mass of flooding just wise in so many examples, but you know, from January to December you can see where you know, the areas that are typically or frequently being flooded.

47:46 Speaker 2: So if you if you detected large scale fooding in May, it's all unusual, and if you detect that same funding in November, it's probably mostly recurring flood Here's a separate example looking at the product archive for Lake Chad in Northern Cameroon, Nigeria and nichere and Chad. Just adding up the total number of daily water observations on the left for thirteen year period. This lake has been slowly going away over over time. But as you can see, the southern basin is pretty solid, the northern basin much less so, and the upper right.

48:28 Speaker 2: It's just looking at the simple number of years the water is detected. So this, you know, let's you look at some interesting things about water surface water change. Another example, the same idea for the Indus River flooding. Both of these examples are ignoring. These are just looking at water detections, so ignoring if it's classified as flood or surface water that you could certainly look at that separately. Distribution, so product distribution.

49:08 Speaker 2: The core product is generated in these ten by ten degree data files. There's two hundred and eighty seven of those spread across the Earth. Here's a map of some of them. The full map is on the homepage and then the user guide. And the core product is getting one HDF file pertile per day, and that includes all of the flood composits that one into two to three day. It also includes other ancillary layer layers with the number of water counts, some other things they go into generating the composites.

49:41 Speaker 2: If you wanted to create a custom composite, some of those layers might be of interest, or if you don't want to mess around with any of the filtering we've done, you could look at the water counts and work with those directly instead of the multi day compositing and all that stuff and the train masking the HDF files, you can fairly simply extract juts from those using GDL tools. There's an example and the user guide. We also do distribute geotip files for the flood layers within those files for the NRT products.

50:15 Speaker 2: And the flood layer is a data layer, is not an RGB layer that you've been seen in all the examples. Is simply has pixel values at zero, one, two, three into fifty five. So you throw those into a GIS and you'll have to assign colors to them to make it more intelligible.

50:35 Speaker 2: So for NRT downloads, there there are LANDS NRT download sites. These only retain data for about one week. The NRT neural time products from LANDS are only intended to be available for a short period of time because the goal is to get the nut led quickly and then for a longer older data you would you would look at the regular product, the science alreday products for downloading. You need a fore your data account. There's some browse instructions here which I will show in the demo later, so I want to belabor it here.

51:11 Speaker 2: You know the key things you need to know the collection which is six months for Motus, fifty two hundred for years, and they are distributed by day of year, so you need to know what the day of the year is, which you can google that and get a conversion from a month, month day to day of year quite easily. So we'll show this more later, so I won't talk about it too much now. The archive product downloads are from lads. This is only the Motus product of the reprocess and the narrow time from twenty twenty six forwards, similar type of approach, and I will also show this in the demo.

51:59 Speaker 2: There's also you can you the imagery in Worldview. Some of the examples I've shown you are in Worldview, like this one here. This is very convenient because you don't have to worry about the tiles. It's almost a globally you don't see the tiles, and you can compare it to lots of other thousands of other satellite image products that NASA has available, and you can do things of comparison compared one day versus another, or one product versus another. So it's very convenient. Because the lance this latest version of flood products started was again generation in twenty twenty one, those layers in Worldview only go back through twenty twenty one, so older products, older flood products will not show up in Worldview.

52:46 Speaker 2: We're hoping to rectify that at some point by processing those older products into the Gibbs imagery that Worldview accesses. Yeah, and I will show this as well in the demos, so I'm not going to stay on the slide too long now. And also when you start Worldview, you will see this welcome to Worldview page and there's an item there called assessing Floodwaters. So if you are not familiar with Worldview or not familiar with the flood product, I would recommend going through that. Other than other than that, you become the ad layers and you will get down to business of adding the flood product, and I will show that in the demos directly.

53:33 Speaker 2: Worldview also now has an events tab, and once had an events tabit it has a flood item in the events so you can filter on that. But sorry, this will show you all recently reported floods based on the gas database, and so if you've heard about a flood but weren't sure where it is, this might be a good way to find it. Many of these floods are too small for us to be detecting, like it can be quite small and still reported by g dacks. Whether we're going to have relevant imager is a question. It will also bring up other related If you put on one of these these events, that will bring up other relevant flood layers for you to look at, So that can be immediate to orient yourself towards flood events and available data.

54:18 Speaker 2: Finally, future directions. So this flood viewery which I showed in that example earlier, we're hoping to publicly release that. The main advantage is it adds the one day product we're both modus and beers, and the other advantages we're pulling in other non NASA flood products such as there's a Noah Birch Bason University flood product based on beers, and there's a radar based flood product from the Copernicus, the European Comperta system that is available in this as well. We're hoping to archive the viers neuralk time products.

54:57 Speaker 2: As I mentioned, we've also been wor being very hard on a machine learning update to water detection for Beers. This has real significant potential for improving the product. The detail of flood detection is greatly improved, and most significantly in my point of view, it rarely detects shadows as water, so all of the cloud shadow problems we have essentially disappear, and so it's so you can use the one day product without having to worry about that typically. So we're hoping to get that to roll that into an updated Beers product that is still under discussion and hopefully underway.

55:41 Speaker 2: We would also like to have a global alerts page to highlight active flooding.

55:47 Speaker 2: As you'll see when we when you look at worldview at a global scale, it's very hard to see flooding because foods usually are not that visible at this type of scale. So this would high areas of active flooding and then you can zume in and see the data directly. The main issue with this is the false positive, so making sure we're not alerting on things that are just clout shadows or and again getting the motus flood archived into Worldview, so you could look all the way back to two thousand flood events.

56:19 Speaker 2: That would be ideal. And the Sentinel three LLCI product we are hoping to expand and to look at using that product for flood detection. It has similar characteristics to Modus, but it is a morning overpass and typically places are less cloudy in the morning, so this would ideally replace you know, the morning overpass we're getting from terror provide additionally useful observations. But that's that's just you know, that's a wish SOS item you would like to move forward with. Okay, so that is the key presentation.

56:59 Speaker 2: I'm going to now move on to do show some demos using Worldview and the download sites, just to show you how to navigate through them. Okay, So here's just a brief cheat sheet for me for the demo and for h audience watching. So first I'm going to go to the product homepage just to show you a few basic things here. So there's you know, basic information about the product. Most importantly, there's a user guide, which is a PDF. I won't open it now, but I strongly encourage you to look at that. There's a summary of updates the details.

57:45 Speaker 2: There's also a summary of data access which I'm going to show you but call that info is also here there's a tile map which if you are downloading product for a specific area, you'll need to know what tile it is. This will you figure that out? This is also in the user guide, so it's there as well as here. And their ten degree you know they're on the grid, they're not weird increments of offsets from ten degrees. And finally, the FAQ section at the bottom, we have a lot of a standard set of vaqus so if you click on this you will get to this long list of questions about the products.

58:27 Speaker 2: So if you are unfamiliar with the product, or maybe even if you are, you might want to have a scan down this if you have questions, So just to introduce you to that. So now I'm going to jump into NASA Worldview, which is a great tool for looking at thousands of data products from NASA. And when you start it you will see again see this this pop introductory pop up. If you've never used Worldview, I would recommend going through this introduction to Worldview. I'll show you the basic tools, et cetera.

59:00 Speaker 2: Or the flood product. There's also assessing floodwaters tutorial, so I would recommend doing that if you if you're new to the product, I'm going to not do that and then interest of time and just show you quickly how to add the product directly. So the first thing you'll notice is when you start with View, it has the current up to date view of the Earth from Terra by default, the terror the motors sensor on Terra. You can click on Aqua several hours behind. It also has the veer sensors that we use for the flood product.

59:32 Speaker 2: So the source data is already being displayed by default in the interface. So that is great. So click on reference layers. They're not quite sure where you are then to add the flood product. We just click on add layers. You get this large menu with lots of options, but there's a flood option. Look on this. You'll see the opera products at the top, which somebody else will be talking about that for the our set training next, but for our global flood product. That here's the Beers product, there's the two and the three day players, and the Motus product the same.

1:00:11 Speaker 2: So I'm adding all of these in. There's some basic information on the right close this, so all of those have now been added. Probably you don't want to look at them all at the same time, so I'll just show one the two day Motus product. On the options here, I often turn off insufficient data visualization. It is basically showing you where you may have false negatives, where we're not reporting flood because it's too cloudy. But it also sort of obscures the background, so it can make it harder to see your sites of interest.

1:00:49 Speaker 2: So typically turn that off, but the news you would like. And again here at a global level, we don't really see a lot of flooding because it's usually not that large. Even if we go back to day however, you look closely, you will see something here in eastern Russia being largely reported in yellow, which is again recurring flood. So these are some rivers in Russia that are probably having a lot of spring snow melts still moving through them, I assume. And if we zoom right in, you can see here's the two day products.

1:01:31 Speaker 2: So we're looking at the Motus product. Let's look at the Modus imagery. Here's the Aqua image a little cloudy, the Terra image much less cloudy. Seems to be lining up in the river basin. If you wanted to compare, this a very nifty tool with world Views, the comparison tools. If you wanted to see if the flooding like this was happening last year, you can click on this. It's flits your screen. You have an A in a B tab is where we were on June seventh, twenty twenty six, So say and you can for each tab you can change the product.

1:02:09 Speaker 2: You can change the date. So let's change the date to a year ago. So let's make it twenty twenty five. You put in June seventh's just for embarrassing's sake. So a year ago the product was showing a lot less flooding than it is today. So this is a very nifty tool to compare one event versus the previous event. If you know the date for the previous event, you can of course browse dates and see what we have. You'll notice in the older a year ago data, we are not we don't have the recurring flood. That's because these imagery layers that are being displayed in Worldview were generated from the narrow time product and the year ago had not introduced that.

1:02:55 Speaker 2: So the other thing, but as website, let me turn off. The comparison that you can do is to try to determine if you have a good product where whether clouds are interfering At this site, I didn't see too many problems, but I'm gonna show you another area and it was undue six. Yeah, Western Australia where we have I'm recording some flund here and this is probably not It was seventh.

1:03:56 Speaker 2: There were just took a minute for imagery layers, so you can see here this is a very typical cloud shadow false positive problem. You see these red flood areas falling in the cloud shadows. You know, unless you knew that there was flooding in this region of interest, you would probably totally disregard this. But if you did hear there was flooding of interest, it you know there may be real flood. The two day product is showing this. If we turn that off and turn on the three day just a little more, you know, you still are getting some so's.

1:04:32 Speaker 2: It's not perfect, but it is definitely less than the two day But if you wanted to be sure, you should hear it looks pretty clear that a lot of this is falling in the shadows. Even check the other image, so Terra is actually looking very nice on this day. The two day product is also using data from June fifth, so let's go back a day. While fifth was not great, there's a lot of cloud and aqua, a lot of cloud shadows. So we have two days with clouds like this, where you have very bright clouds and very dark cloud shadows.

1:05:08 Speaker 2: This is gonna be a problem. Even in the two day product, you'll notice this herot image from yesterday, from the previous days is not nearly as problematic because it doesn't have sharp cloud shadows. So that's that's, you know, one way to look at the product and decide what you're seeing is real and useful for you or not. If you decide this is useful, you can get a link to download data from Worldview. If you quick on the data tab here, it will show you all the products that you had loaded up. And so for the flood, the two day window, this one is the main product.

1:05:49 Speaker 2: The second one is the geotip as it says there, So if you wanted to download the geotip the two day geotiph for this product, if I'm a it's telling you no Granules. That's because

1:06:01 Speaker 2: this is still twenty twenty five. If I did the comparison the other site, and we don't, the interface here is not linked to that data archived yet, so it's currently only linked to the neural Time data. So let's go forward to this year and now, if this was the product you were interested in, we're seeing five hundred grand was available. That's way too much data. We need to set an area of interest, so you can draw a box. Area available. Granules is too, so if you wanted that, you can click on here go through this opens Earth data search stick the tour for now and to use them over to Australia.

1:06:50 Speaker 2: You can see the tile. Let's see a preview of the data. You can see our box from the previous well few and the links to download the two data products. These are actually different dates, so day one five seven and day one five six. If you cook on the download button you will be allowed to download that year to directly. So that's how to download from Worldview, and again that only works for at the moment. That only will work for recent data, the past week's data the nuro time product. You can also just directly go to the nual Time download sites totally separate from worldview.

1:07:33 Speaker 2: Cook on, browse here, book on all data here. Here's where you need to know the collection number for six to one for the Motus product. It's fifty two hundred for the veers, so for Motus six one. And then you see a long list of many Motus products, most of which are my foot product. So you can use your browser search to search for what we call the short name, which is NCDWD. This is ALLID in the user bed and on the homepage, and you will see all these directories. The ones with F are geotips for specific one day, two day, three day flood layers.

1:08:12 Speaker 2: The one without the F is the HDF file so depending on what we WANTOK on those we'll see Even though this is neural time data, so there's no data in twenty twenty four, twenty five year, will only be data roughly the past seven to ten days, so one fifty nine will beach to day. One fifty eight would be yesterday. Look on that you'll see all the files available. It should be two hundred and eighty seven of these software one for each file. This being the tile number here, so that's the neural time download sites fairly straightforward.

1:08:47 Speaker 2: For the historical data, we have the LADS archived download site. So this only contains well, it contains the most reprocessed products from two five and the modus neural time products from the beginning of twenty twenty six to forward. It does not yet contain any of the BEERS products. So this is a little bit different, a little bit the same quick them find data. This loads a larger search interface, which you could certainly work with, but if you just want the simple interface that we saw before, click on online archive on the left and this opens again collection numbers for the motus we want six to one and you can search for mcd D. Here we have two directories just this one l three.

1:09:41 Speaker 2: These are all level fee products. This is the reprocessed historical so you click on that you see years two thousand to twenty twenty five. The NRT one is the neural time so this is only twenty twenty six from day one and forwards, so you click on any day. So again the listing all the HDF files in this space. Okay, so those are the download sites on NASA Worldview. And now I'm going to show uh how to load this imagery into qh I S. So firstly, in Worldview, let's get back to just a basic view. Uh.

1:10:22 Speaker 2: You know, Worldview is again not showing the raw the raw data files. It is an imagery product generated from the data files for the flood. It is showing as you see here blue for surface water, right for flood, et cetera. You can pull this directly into qg I S or r JS pro so you can work with it with your own data. And the way to do that, the way I remember how to do that is if you're in Worldview and you can become the information I have here and click on a PI access This will open. Then fell page on Gibbs Global and Global Imagery brows Services tells you all about these are the layers that are displayed in Worldview.

1:11:08 Speaker 2: World View is not displaying any raw data, and we want this would via JS applications. There's a sexual QGIS, RKS, et cetera. The key info here is really just the U r L for the w MS endpoint. So I'm going to copy that and then I'm going to move to HUGS to show you how to add that in. Okay, so here I've opened a new HGIS session PGS is free JS software. For those of you that I'm familiar with it, it's very very handy. And what we're going to do is add in a w MS layer for those gibbs ask the gibs service so that we can view the flood data directly in qgis with any other data you might have fig files or other energy.

1:12:01 Speaker 2: So the key is in the in this section on the left w S w mt S right click on that add a new connection. The only thing you really need is the r L which is what I copied from that API access page I just showed you and the name which with whatever you want massa gibs. That's what it is, and all the rest is you can be sleep it. It's free, open access. You don't need anycount And when that's added, you see this items show up become that you will see all of the many, many categories of imagery products available and gibs and you know this is what you can view in worldview.

1:12:46 Speaker 2: So let's add the corrected reflectance because that's always the key knowing what you're looking at. So I'm going to add the modus awquad seven to one. That's the false color and the like that for the flood product, and let's add the modus terra seven to one as well. So this again is the current state of data collected today. And then let's add the flood the flood hazard. There's the modus one day, that the modest two day. Okay, so those are all added. I'm gonna that up. You'll see them here. The one thing you cannot do here is turn different colors different category classes on or off, so you have to live with the gray insufficient data.

1:13:38 Speaker 2: But you can click them on on and off the hole. And here we're seeing both the injuries. So if we were finally, the other thing you need to do is turn on the time slider at the time the temporal controller, and then on here. And so this defaults I think the full range of your data. You can certainly make this go from six for example anyway, so you have less data to scroll through, and it tells you where your current date is. So just to show you another example, there's some flooding in southern Mozambique in January.

1:14:18 Speaker 2: It was fifth so change that directly. And so here we're just looking at the aqua image. We look at the one day flood product. It loads right up and you can see it. And then you can compare this to the two day product. The two days getting less flood, but it's getting quite a bit of it. If you were I had been looking at the one day and you were concerned about but whether you have the data or cloud shadows, it gives you. Men. Here once again we're seeing in the one day flood being reported directly in the modest shadows Terra doesn't.

1:15:11 Speaker 2: It's a little bit cloudy, but it's you know, these thoughts are not generating cloud shadow false positives. But if we click on the two day product off the one day, most of those are disappearing. So this again is where the two day comes in handy. But obviously if you you know, know the area of interest and what's going on, you can direct choose choose the right product after looking at the source imagery. So I think that's all I have for you here on the demo, just showing you how to get the syndic q I S and play with it initially, and of course if you have any other questions about the product or the how to use it or how to access it, the resources page and the presentation has a links for accessing help which is also all on the homepage.

1:16:05 Speaker 2: So ask it back to the media now, who will go over summary and continue the session. Thank you, Amida.

1:16:15 Speaker 1: Thank you so much Den for your excellent presentation and demonstration, especially providing information about how to visualize recent floodcases and how to download data. So again, b thank you for that. This brings us to the end of today's session and to briefly summarize what we saw. Doctor Slayback presented description of global flood product development approach. He showed that it's based on Terran Aquamotis Noah twenty and twenty one vers imagery uses red infrared and shortwave infrared reflectance to detect water.

1:16:53 Speaker 1: Has multilook composites available, so data from one, two and three day composites are available and based on threshold based water detection, terrain shadow and cloud shadow corrections are applied to remove false positive in water detection. Doctor Slaybek also showed how to select appropriate composite product for flood visualization or flood checking the flood based on true color imagery. If there are a lot of clouds present, then multi de composites more useful. Also, identification of recurring floods is available.

1:17:31 Speaker 1: It's based on twenty two years of historical flood mass data. Then we saw examples and demonstration of floodcases using Worldview recent flood places that we saw in Russia, Australia and Mozambique. Also flood data access and download using Worldview and lood Stack were demonstrated near real time and twenty three years of archive flood data from Terra and Aquar from twenty two thousand and three to twenty twenty five are available before that, so two thousand to two thousand and two data are available from Terra alone.

1:18:08 Speaker 1: In flood data can be downloaded as jutive from this laod's deck recent data so starting from twenty twenty one to present they are available from Worldview as we saw, and we also saw that flood data can be downloaded as jutive and analyzed in QGIS. Doctor Slabek also mentioned a few limitations of this product based on optical data, so it cannot see floods through clouds cloud shadow. Although they are removed, they can sometimes pose problem and there may be false detection of water because of that.

1:18:50 Speaker 1: It has medium special resolution of two hundred and fifty meters may not be adequate for resolving urban floods. It's based on twice daily observations and may miss some flash floods. Water under tree cover is not visible to satellites, and changes in surface water extents such as new reservoir changes in reserve river courses, they may not be in reference surface water mass data sets, so there may be some uncertainties because of that. We also saw future plants for flood product. There will be a release of new flood viewer in addition to the current setup in Global flood product.

1:19:33 Speaker 1: It will include Nova GMU and Copernicus glow Fast products. Flood products and currently, as we saw, weirs data are not archived, so they will be archived now will use machine learning. Update to water detection algorithms especially for weirs include global alerts to highlight active flooding at a global scale and include center in three old chip product. These are also optical products.

1:20:08 Speaker 1: So next week we will have a presentation about monitoring floods using opera surface water extent based on optical and SAR observations. Again, homework will be posted on our training website on twenty fifth of June and homework will be due on July ninth. It will be in terms of Google Forms, so you will be answering questions in Google forms. Certificate of completion will be awarded to those who attend all live webinars and complete the homework assignment by the due date. Then a certificate where email will be received in approximately two months after completion of the course.

1:20:58 Speaker 1: Once again, thank doctor Daniel Slayback for his time and for his very useful information and presentation about the Global Flood Product. Here is doctor Slayback's contact information if you have any additional questions, and you can always contact us at our SET if you have any questions. This is a link to our set website and our set YouTube. You'll find our recent and past training information on both these links. 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.

1:21:45 Speaker 1: Visit our contact page to subscribe. There's some resources and useful links based on today's presentation and here's the QR code you can access information about flood product and with that we thank you all for attending today's session and we will go to the question and answer session. Now, thank you, thank you so much. Will have our question and answer session with doctor Slayback. There are many questions, so we will address as many as we can and we will post the rest of them on our website.

1:22:26 Speaker 1: We'll start with the questions. There's one note. There is an exercise posted on the training web page that you can download. Exercise is not due anytime soon. It is your homework questions will be based on that, but it just allows you to explore a global flood product using worldview. So please download the exercise and work on the exercise. Okay, just a note here and thank you Jan once more for your presentation, and we'll just have a few questions here. I'll start with first questions. In the current flood detection workflow motives, envires use multil look compositing and terrain based false positive masking to improve reliability.

1:23:11 Speaker 1: Could similar compositing and conceptual masking approaches be extended in the future to detect other types of environmental animalies, for example, persistent reflectance changes, deby related spectral signatures or hydrologically mobilized materials. So that our observation products can support broader preparedness and environmental health assessment beyond floods, and you can unmute yourself and answer the questions, then, yeah.

1:23:43 Speaker 2: Thank you, Amita. Yeah, certainly you could. You could apply similar approaches to other questions. I think that really depends on what you're trying to look at and what your data sources are, and what you're what the limitations of your detection, you know, algorithm are. In our case, as I mentioned over and over, we have problems with shadows, so cloud shadows, train shadows so. And for that particular case, you know, the multi day compositing and the train masking and the hand masking or our approach to deal with it.

1:24:20 Speaker 2: But you know, in a more you know, if you have an algorithm that has less problems with things, you know, false positives or false detections, you hopefully don't need to absorb to those types of things.

1:24:31 Speaker 1: Ideally, yes, thank you. The next question is are these data available in Google or tenngin data catalog.

1:24:41 Speaker 2: I don't believe they are, but that is a good reminder to look into that, especially with the archive product. I suspect that would be attractive to them and relatively easy to get in since it's a fixed data set there. If you poke around there, there are some flood products that were generated using the same algorithm but a different machinery, not in the systems here. That was used to regenerate the product for historical flood events for a Nature paper that we've published in twenty twenty one, I believe, led by Brett Dulman.

1:25:18 Speaker 2: So that was looking at roughly nine hundred different flood events globally. So those those are in Google Earth Engine, I believe at least they were at some point, so you might see those, but it's not a global comprehensive data set.

1:25:35 Speaker 1: Yes. The next question is does the flood view covers all the places on the Earth and will there be options to choose and overlay composites from different products.

1:25:51 Speaker 2: If you mean the Blood Viewer, the new tool we're hoping to release that is global US. It's same as worldview. It shows full you know, not really Antarctica, but everything else.

1:26:09 Speaker 2: And yes, the part of the point of the of the flood Viewer or kind of in house viewer, is that it will bring in external flood products such as the one flood product from from the s A and the Noah, Noah of yours flood product, and so you can compare them directly with one another and see which one works best for your particular event. World we only have or mostly believe only has NASA products in it.

1:26:42 Speaker 1: Great question four is given that hand models successfully filter out steep ridge shadows, how does the system How does the system account for a sudden urban flesh floods where localized micro topography creates deep pool away from the natural drainage channels. Does it lead to false negatives? If yes, how can this problem be addressed and fixed in the future.

1:27:12 Speaker 2: Yeah, this this product is not going to work well for any kind of urban or micro topography, you know, small scale flash flooding, both because mostly because the two hundred and fifty meter pixel you're not going to in an urban area, most of that pixel is probably relatively dry rooftops or treetops and not the water on the ground. So you have the mixed mixed pixel problem. And then the timing or you know, the small smaller the flood extent, you know, flash flood constraint to a canyon or you know, canyons and urban areas between buildings, they move very quickly, so getting an observation is you know, difficult or unlikely.

1:27:56 Speaker 2: Perhaps you know, in the right conditions you might have a clear modus observation and that would be great, But then you have the two undred and fifty meter pencil problem. So yeah, the hand you know, hand isn't earlier problem there, but it would the hand mask would mask out many such areas because it is it's predicated on this idea that for two hundred and fifty meter pixel there's if there's sufficient drainage, that water is not going to stick around long. So if you were making a hand mask for a you know, say you had a global you know, one meter sour sensor daily that you can make a flood product from, you would well, you probably wouldn't have the same types of issues, but the hand mask would look very different at a different scale than it does forty to fifty meter.

1:28:46 Speaker 1: So the next question is on slight forty their difference is not significant? Correct? Also, what does composite mean in the slide showing the data for punjab in slide forty two. On slide forty D eight, how is the hand mask solving the problem?

1:29:05 Speaker 2: Okay, let me look slide forty was the recurrent flood example, I believe so, right, the difference there is just with the recurrent flood updates. So the detective water is the same. We're just reclassifying most of it in that Northern California example as recurring flood. But otherwise it's the same source water detection. So what was the data? And yeah, right, so that was a custom composite just generated by adding up all the one day products for one week to just bring out more clearly the flood extent for that event.

1:29:57 Speaker 2: Because each you know, each day, if you looked at each a you know, some days have clouds, some don't. So over same idea that we do with a three day or the two day product, but if you've extended over seven days, you can pick up you know more so that the point there was just to show that most of that flood event was actually not refurning but unusual for that time of the year. Aside forty eight

1:30:28 Speaker 2: Yeah on Hawaii Island. Hand solves the problem of of our product. Initially, you know, way back in the beginning reporting Central Hawaii Island as flood because it's it's that area is very steep, so there's not you're going to have water sitting on the side of Mana Loa volcano, it's gonna run off very quickly. So yeah, so hand very quickly solved that any any topography related problem. And here the problem was all that volcanic material is so dark that it gets detective as water.

1:31:06 Speaker 1: The next question is what does TMC mean? And it's too many clouds. Let's just read that. And next question is how do we determine the pre and post times during flood detection analysis?

1:31:22 Speaker 2: Yeah, I mean that is a good question. You really that depends on how you're coming at this. If you know about a flood event, ideally you have some other news type of data that's telling you that people are impacted or something's going on. If you're trying to do this simply from flood imagery or from imagery perspective, you know, if you have a daily product like cars and it's clear, you could be at least within a day, you could identify there in post flood times by when you're detecting water. But yeah, yes, the question, but yeah.

1:32:02 Speaker 1: It's a good question. And the thing is the in next two sessions you will see there are other tools and even a predictive model. So combining everything might be better to understand flood issues better. And question eight is if we use SAR data, would we so have cloud based question? Will it not mitigate the cloud based obstruction?

1:32:29 Speaker 2: Yeah, sor penetrates clouds. You do not have cloud problems, so it is phenomenal for that reason, and you will not have cloud shadows. You know, all of that goes away. It is, it is, it is great. The main limitation of stars that you don't have global daily covers and you and it's well, some of it is commercial. You have to purchase it sent on one from preparatas from the Europeans is free, but it doesn't have daily repeat. It's maybe five or six days the best. I think. It depends also there in the world.

1:33:06 Speaker 2: So for the right you know, if the timing works out, that can be a good solution. But if there's no stur data, then there's no star data. And the other problem with SAR is you have to typically have a free image with nope, without the floodwater present, and compare that, compare that to the post image. And there might be other ways they're doing it now, but that's the sort of a straightforward way. So you have to have an archive of dry dry line imagery for the site, which certainly probably exists for Centinel one point.

1:33:44 Speaker 1: Great and as proper in seted. Our set will have a nice our training starting July second. In session one is about uh star based flooding, so you may want to check that out. Next question is how does your VI's mL model behave on shadow water? Does it recognize water despite the shadow cast on it.

1:34:11 Speaker 2: That's a good question, and I'm not entirely sure since I'm not the developer of that model, but you could find out. I don't think it is a significant problem, but it's a good question to ask a confirm.

1:34:28 Speaker 1: The next question is if Moodys and Weirs can observe floods only one or two times per day under favorable conditions, how can these products support flash flood response when floodbeaks more occur may occur within one, two three hours.

1:34:45 Speaker 2: Yeah, similar to the earlier question, this is not going to work well for most flash blood situations, both through the timing issue and then probably the flash flood is smaller and spatial extent. Even if it's extremely dangerous, you know, working down a canyon, it may not be really that wide. So the two two hundred and fifteen meters pixel limitation as well. I mean, there's really very little ideal observational data to capture those events unless you've got you know, aircraft in the sty or something.

1:35:22 Speaker 2: Thanks.

1:35:23 Speaker 1: The next question is can these products detect urban plurial flooding or are they mainly suitable for riverine and large area flooding. What improvements are needed for dense urban area with complex training system.

1:35:40 Speaker 2: Yeah, urban areas are very problematic because they have a lot of dry objects rooftops, trees that are going to not be detected as water must be a very deep flood. So other than that, you know, the source of the flooding,

1:35:59 Speaker 2: whether privial or riverine overflow, doesn't really matter, but the urban problem is significant.

1:36:08 Speaker 1: The next question is can these satellite products distinguish or support analysis of compound flooding such as rainfall driven flooding combined with the river outflow, river overflow, storm search, dam release, or drainage system failure.

1:36:30 Speaker 2: Yeah, you know, we're just we're just detecting water here primarily, and then we're using fairly trivial methods to say if it's flood or or not flood or recurring flood, so you know the source of the flooding or the type of if there's a dam involved. That's sort of really up to the user to take the product and evaluate in more detail. Just bearing dams. As I showed in some of the slides, any new dam constructed in the past few years will probably show up as flood for a while until like it's thought up in the annual water mask, So that can be a little confusing mm hm.

1:37:18 Speaker 1: The next question is along the same line, which products can be used and how to look at riverine floods where small rivers are regulated with small dams or badges.

1:37:31 Speaker 2: Yeah, that's that's tough. It would depend on and how how you know small the flood flying surrounding the river

1:37:43 Speaker 2: mm hm.

1:37:45 Speaker 1: The next question is can satellite flood products contribute to early warning or are they mainly usually after flooding has already occurred. How can they be linked with rainfall forecasts and stream flow prediction systems such as you glows.

1:38:03 Speaker 2: Yeah, I think these are not going to be helpful for early one and they're just detecting water once it's on the ground. Flood models those other systems can certainly help predict when you might expect flooding based on rainfall forecasts and and serve models.

1:38:24 Speaker 1: So we are going to have a session on geoglows in session three and you will also see an exercise at the end where you are looking at geogloss prediction and UH one of one of the flood cases doctor Labat showed careson in in Ukraine. You can see that picking stream flow in that period and then you so if you know that, you can monitor that period and then keep looking at global food product when GEO clows up predicts stream flow, So you will have an exercise on that. And question fifteen is given the cloud comer limitations of optical observations during CEPR weather events.

1:39:08 Speaker 1: What are the best practices for using for fusing this product with SAR data within QIS and nearly all time emergency response?

1:39:24 Speaker 2: Yeah, I think you know, for a specific event where you have a you know, a person an analyst working the event, you know they should try to look at all the imagery available and see what's going to be useful. If it's you know, if they're looking world view and the notice imagery is clear, or the beers shown something and they can jump on that if they have sur data. That's typically the problem in A and A you know, yeah, an event when it when it's starting, you know, are you're actually gonna have SAR data.

1:40:00 Speaker 2: You might sent all one perhaps or night start perhaps, but if you have to order a new acquisition, that's going to take a day or two at least. So it's really I mean, this is sort of the problem with all of the different flood products out.

1:40:14 Speaker 3: There is.

1:40:17 Speaker 2: It's hard to know which one is going to be useful for a given event without looking at the details of that event, how cloudy is it, what's the data available, etc. So but you know, a human analyst can hopefully sort that out. An automated system that gets it's much trickier.

1:40:38 Speaker 2: But you know, I would just.

1:40:39 Speaker 3: Drop whatever you could into QGIS or look at it first in Worldview or the Flood Viewer when that's released, to give you a quick look at what's available and potentially useful.

1:40:53 Speaker 1: So again you will see a product next week which has both optical and SARD components that you can combine in q g S. So what what has modis thought about global flooding looking back?

1:41:15 Speaker 2: Yeah, that's a great question, and we're looking at that now with our with our pointy plus year r PAD. So we're hoping that the report on that in the paper. What you.

1:41:30 Speaker 1: Next question is, are there any methods or data sets available for floodplane mapping.

1:41:40 Speaker 2: Yeah, I'm not, you know, in particular familiar specifically with floodplane mapping, but I think there was a recent paper. We can look for the length and put it in the notes where they I don't know we're updating in some sort of global floodplane analysis. I believe, yeah, products like ours could be very useful for areas where flooding is routine and routinely detectable. I should say which, it's many areas, but certainly not everywhere. But yeah, otherwise, modeling, d MS all that kind of stuff, which has its own complications.

1:42:22 Speaker 1: Mm hmm. The next question is if optical and star based flood maps disagree, what would be the standard method for reconciling differences in the final flood product.

1:42:37 Speaker 2: Yeah. Again, I'm kind of repeating what I said earlier. If you have a human analyst doing this, I think they could very readily determine which one is more useful if you truly need to fuse them. That's you know, I don't know, that's that's a little harder. But if you're if you're trying to refuse on an automated basis, that's you know, we could get into trouble because one of the products might just have false positives or false negatives or you know, not really be useful. Fusing over the useful product is just going to diminish the quality of the overall output.

1:43:18 Speaker 2: So yeah, this is again the trick being able to look at things and see what's what's actually a good data product.

1:43:29 Speaker 1: M next question is how does this product differ from novah IR's flood Man.

1:43:39 Speaker 2: Yeah, they have a very different approach, and they have a paper out on it so you can read it in detail. I don't remember all the details offhand, but one of the differences is they don't just report binary flood that they report percent water or probability. I forget one of the two. So it's a you know, wonderful hundred or whatever it is, gradation, which can be useful in some drier areas. This seems to market up soil moisture than water, but certainly work working at.

1:44:21 Speaker 1: Okay, next question, and we'll take a couple of more questions and then we'll address the rest of them. Litron, how do I map flood extent when the ground stays wet the whole season? Pretending to working with Centinel sar in ge, I've also extracted flood extent from Sentinel one sar in ge and the maps came out empty. What happens when the SAR extraction is empty? Does it mean no flood or did my threshold method fail? Does the same flood principles apply?

1:44:59 Speaker 2: Yeah? Sorry, Flood detection is very different than what we're doing, and it can certainly be more complicated. So yeah, especially if the if the ground stays wet, I mean that is problematic even with optical I'm not sure that's our particular. I know there's a there's many different SAR algorithms for water detection and or flood detection, so they all have different strengths and weaknesses. But there's a lot of literature out there on that, and I'm not familiar with it personally. We're not doing a sour product.

1:45:38 Speaker 1: Yeah, yeah, So just to add a note to that. Our next session, which is going to be on twenty three, we'll talk about oper od enemy surface water extant product that has both SARA and optical components, and you may be able to ask more questions then even the nice you know, I'm going to keep moving down and take some questions later on. Next question is how much it's it's precision compared to ground reality.

1:46:14 Speaker 2: So the you know, the the source data is the surface reflectance product from motus or from beers, and their gelocation precision I believe is significantly less than a pixel, like maybe one hundred meters, but you could find that doctoration online for those products. Where it gets frickier is

1:46:41 Speaker 2: at the edge of the swath. When these sensors are acquiring data, the pixels get very blurry or extended because it's looking at an angle respect to the surface of the Earth. So those pixels and you will see this if you look in Worldview at any of the cryptic reflection status that's you check you switch between terra and A or the different viers depending on the day, it'll be sharper or blurrier. And that's because that source data is coming from closer to nater, closer straight below the satellite or towards the edge, so you have this an effect blurring effect on top of the pixel geolocation.

1:47:20 Speaker 2: So

1:47:23 Speaker 2: bottom line, the third part is that the neural time product is using preliminary relocation data and so that can be have errors. It doesn't often, but you will see sometimes a river will be shifted some distance away due to this the shortcuts they take to get the neural time data out within a few hours. That gets corrected in the final products. But for the flood product there typically isn't final later product, but for the archive that was reprocessed, all of those geolocation errors would have been fixed or where they might have occurred before.

1:48:03 Speaker 2: So yeah, maybe that's not a super cleayer answer, but you know, the pixels themselves are the gain location is fairly accurate for seventy five meter pixel. Great.

1:48:17 Speaker 1: Thank you so much for all your presentations and demonstration and answering all these questions. We have a few more questions, but right now we are already over our webinar time, so we'll address those questions later on. And we thank you all for joining to this's session and we hope to see you on twenty third for the next session of this training. And so please remember to download the exercise and work on it. It allows you to look at worldview and look at a global flood product. We want to thank our our set team here, our coordinator Natasha Johnson Griffin and our editor Maria Tito Cherry Morris, our coordinators broc Levins and Selwyn Hudson odoy Are, our learning management instructor Kevin Field and Susan Monthy, And we thank you all for attending today's session.

1:49:18 Speaker 1: And then once again, thank you so much for all your time and effort and contribution to this training.

1:49:29 Speaker 2: Problem. You're very welcome