NASA ARSET Introduction and Demonstration of STREAM

0:00 Speaker 1: Hello, and welcome everyone to this training on monitoring water quality in lakes and coastal regions using satellite based tool for Rapid Evaluation of Aquatic environments or STREAM. My name is a meta meta from NASA's Applied Remote Sensing Training Program or our SET program, and we have a guest speaker today, William Wainwright from NASA Goddard Spaceword Center, who will be focusing on introduction and demonstration of STREAM. We will start with a brief overview of our SET program. Our SET is part of NASA's Applied Sciences Program, its capacity building program, and provides accessible, relevant and cost free training on remote sensing satellites, sensors, methods, and tools.

0:50 Speaker 1: Trainings include a variety of applications of satellite data and our tailor to audiences with a variety of experience levels. Specific are SET focuses on these thematic areas for applications agriculture, climatory resilience, disasters, ecological conservation, health and air quality, water resources, and wildland fires. SET trainings are offered online or there are in person trainings offered as well. Online trainings can be live and instructor led like this one or They can be asynchronous and self based, available from our set website.

1:29 Speaker 1: The link is provided here. These trainings are cost free as I mentioned earlier, and only use open source software and data. Many of our trainings are offered in multiple languages, especially their material is available in Spanish for many trainings, and they accommodate differing levels of expertise. Please visit ourset website for more information. We'll start with an overview of this training, monitoring what quality in lakes and coastal regions using stream now. It is well recognized that monitoring water quality is vital for planning and managing drinking water treatment, for public health and ecosystem advisories, and also for assessing health and productivity of fresh water and saltwater fisheries.

2:23 Speaker 1: Conventional INC two measurements of water quality parameters are expensive and have limited spatial and temporal coverage. As you can see here there's an example of Lake Aria in the US. Is the western part of the lake which is prone to have harmful alcohol bloom or ABS. And if you can see these dots they represent where INCIT two measurements are taken. As you can see, not the entire lake is covered, just a few locations are covered these. Some of these measurements are weekly. There's buis which provide continuous data, but there are very few.

3:04 Speaker 1: Also, there are so many water bodies in the world and long coastal areas, they all cannot be measured with in water samples and water quality parameters cannot be assessed that way. On the other hand, if you can see this lenset image on August thirteen, twenty thirty four where alcohol bloom occurred in lake around the same area, now you can see a continuous picture and also see pattern where intensity of HEB was low and high in the lake. So that way, remote sensing provides a cost effective way to assess water quality.

3:43 Speaker 1: Also, it can cover thousands of lakes and coastal waters with improved coverage. Currently there are several satellites flying with sensors allowing water quality monitoring. The table here is for your reference and provides information about the satellites, sensors, measurements they do, and their special and temporal resolutions. Just briefly going over the current satellites, it's lens AT eight, lands At nine, Tera, Aqua S and pp GPSS GPSS are Noah satellites. NOAH twenty and twenty one, Sentinel two A, two B, two C, and Sentinel three A and three B. These are European Space Agency satellites and the latest NASA mission base that is flying currently as well.

4:37 Speaker 1: All these satellites are in polar orbits as you can see, and these are the sensors flying on these satellites. All these acronyms are defined at the end of this presentation should be available to you in your dog so this is only only two mods weirs, MSI OLG and o CI. These are the water quality monitoring sensors. They have spectral measurements in optical bands ranging from visible, near infrared, middle infrared, and some also have thermal infrared bands. So all these can be used to monitor water quality parameters.

5:20 Speaker 1: They can be derived from these data top of the atmosphere reflectance in optical bands. As you can see, they all have different temporal and special coverage and resolutions. Thing to note here is that Lenset eight and nine and Sentinel two series they have the highest special resolutions it's intensive meters and the other sensors like motor spheres o g oci, they have low to medium resolutions ranging from two fifty meters all the way to one kilometer. On the other hand, Lancet and Centinal two they have narrow swarths, whereas Terra acquire S and ppgps as Sentinel three and Pace have broad swaths in thousands of kilometers as you can see here, so they all can be used for water quality monitoring at different special and temporal resolutions.

6:26 Speaker 1: Lancet each one has sixteen dairy revisit time, so with both of them it's about eight day, and Sentinal two series has about five day revisit time. All other sensors that we have there one to two day revisit time, so there are trade offs between temporal and special resolutions.

6:47 Speaker 2: As you can.

6:47 Speaker 1: See, our set has several trainings on water quality monitoring using remote sensing, and the links and titles are provided here. These are set trainings focused on water quality parameters derived from sensors like motives, WIRs, OLG and oci with relatively low to medium resolutions from two fifty meters to one kilometer. Before stream higher resolution data that is, at twenty to thirty meter resolution data, water quality parameters were not readily available and had to be derived by using algorithms developed from satellite data and in CITU measurements for the water body of your interest.

7:34 Speaker 1: An example of that can be found here in this training. So here is an example of chlorophyll a concentration from pace oci at one kilometer. I can clearly see that coastal areas and open oceans are resolved and covered nicely in pace oci, and bigger lakes can also be seen like Great Lakes and other bigger water body can be resolved. But there are several small water bodies that are not resolved by this medium to low resolution sensors. And that brings us to today's topic, and that is Stream for water quality monitoring.

8:16 Speaker 1: Stream, as we will see later, uses LENST eight and nine and sentinal two series data and derives water quality parameters at twenty to thirty meters special resolution. It's an interactive web tool and an API enabling monitoring or water quality in near real time since twenty eighteen. Using Stream, you can see that they can resolve smaller water bodies if the lake or water body is one hundred meters square or bigger than STREAM can resolve it. Currently, Stream focuses on inland lakes and coastal areas in the US and some lakes in certain countries are also available from stream.

9:05 Speaker 1: What quality data are derived by using an open source model and we will cover that as well. An example shown here again is pace OCI and stream chlorophile A concentration. This is in Montana, uh This is Lake Canyon Ferry and you can see that this is a fairly large lake and can be seen in both pace OCI data as well as sentinel to data. Thing to notice here is that there are several small lakes that we see here in stream. They are not visible here in pace OCI. Also you can see that being this being high resolution, you can see more details in chlorophil A concentration distribution in the lake.

9:52 Speaker 1: With that, our training learning objectives are that by the end of this training you will be able to identify the purpose, capabilities, and benefits of this stream tool for analyzing inland and coastal water bodies. Identify the process to use stream to monitor chlorophil A concentration, secut disk depths and total suspended solids in lakes and coastal waters. Identify the steps to use stream API to search and download these parameters for a specific time period. Examine time series of chlorophil A concentration SECHI disk steps and total susmcted solids using QGIS and identify how an open source machine learning model based on mixture density network enables users to estimate water quality parameters for any inland or coastal water body which is greater than one hundred meters square or so worldwide.

10:55 Speaker 1: Here are the prerequisites for the strain. First one is Fundamentals of remote sensing. This ur set training provides background about satellite remote sensing principles, different satellite orbits, different data levels that described in here. The second training is monitoring water quality of inline leaks using remote sensing. This training provides information about how to use satellite data together with institute data to develop statistical algorithm and then can be used with satellite data to drive water quality parameters.

11:31 Speaker 1: There will be two parts to this training. Today Part one will focus on introduction and demonstration of stream and part two next week on February seventeenth, will focus on introduction to a machine learning model to estimate water quality parameters based on satellite observations. There will be one homework assignment will be open on February seventeenth and it will be due on March tenth posted on our set website. Our certificate of completion will be awarded to those who attend all live sessions and complete the homework assignment before the given due date.

12:12 Speaker 1: We'll start with today's session on introduction and demonstration of Stream and specific objectives for today are that by the end of part one, you will be able to identify features of stream website and API for selection and display water quality parameters for selected leaks and coastal areas. Recognize how to download water quality parameters using stream API, and recognize how to display maps and time series of Chlorophila concentration, secy disk depth and total suspended solids for selected water body.

12:51 Speaker 1: Outline for today is that our guest speaker, William Wainwright, will start by providing an overview of stream. He will also have demonstration of making maps of water quality parameters using stream web tool and how to use stream API to find water quality parameters multiple days, and then we will use qgis to make time series of these water quality parameters. Just a note about asking questions. Please put your questions in the question box and we will attest them at the end of the webinar.

13:28 Speaker 2: Feel free to enter.

13:29 Speaker 1: Your questions as we go and we'll try to get to all the questions during the Q and A session. After the webinar, the remainder of the questions will be answered in the Q and A document which will be posted to the training website about a week after the training.

13:46 Speaker 1: With that, we introduce our guest speaker for today, William Wainwright. William is a senior scientific programmer with the Freshwater Sensing Group at Science System and Applications in Code with CAR Space Flight Center Code six nineteen, that is, in the Terrestrial Information Lab. William comes from a background of astronomy and physics, with a master's in estrophysics from Rochester Institute of Technology. Now, William applies his skill set to the development of a remote sensing water quality platform called Stream.

14:23 Speaker 1: With that, we invite William to talk about Stream.

14:27 Speaker 3: William, thanks, Amita. My name is will Waynwright. I'm the developer of Stream along with my colleagues Akash, Navide, Arun and Ryan. Together we make up the Freshwater Sensing Group at SSAI and as a team we developed the models behind as well as the platform itself of Stream. Stream is a water quality portal developed by the Freshwater Sensing group. We are motivated by real issues like public health impacts to aquaculture and the larger ecosystem, as well as changing health of water systems from events like harmful alga blooms.

15:15 Speaker 3: Our main goals as a group are to improve the accuracy of remote sensing derived water quality observations in both inland and coastal waters, as well as making those products more accessible. The natural endpoint of those goals is the tool that we call Stream. The data pipeline for Stream starts with the top of atmosphere satellite data. Our team has developed water quality models for many different satellites and sensors, but we chose to begin with Stream, supporting these two groups here lands At eight and nine and the Sentinel two family.

15:56 Speaker 3: These satellites all have the optical band its necessary for us to derive water quality measurements, as well as relatively high spatial resolution compared to some of the other water quality compatible satellites that are available. This increased spatial resolution does come at the cost of temporal resolution, though Lanset features an eight day revisit rate. Between the pair of Lancet eight and nine and Sentinel to A and B operate as a pair, giving us a five day revisit rate. Sentinel two C operates out of phase with A and B and gives you an extra overpass of your site between those five days.

16:43 Speaker 3: In order to retrieve water quality products, we do a few steps to atmospherically correct the top of atmosphere data that Landset and Sentinel two provide. The main engine of this atmospheric correction is a pair of mixture density networks. These are a type of neural network that predict all of the water quality metrics at once, as well as an estimated uncertainty. This methodology is particularly useful in our application, as we're trying to get a sense of how all of the key parameters operate as a connected system rather than just individually.

17:24 Speaker 3: After the first round of atmosphere correction, we're left with the remote sensing reflectants. Then we use a second mixture density network to estimate the water quality parameters, namely chlorophyl a, total suspended solids, and the SECI disk depth. If you'd like to learn more about the methodology and the models that we use, the next r set training will be a deep dive into these models.

17:51 Speaker 4: By Ryan o'she.

17:56 Speaker 3: What really enables Stream to deliver water quality estimates, though, are large training data sets like Gloria and airnet oc. These data sets contain thousands of in situ water samples that were co witnessed by satellite overpass from water bodies of all sizes and all makeups across many different areas around the world and across decades. The matchup between in situ and satellite observations are what allow for us to fine tune our models, and the sheer diversity of water bodies represented are what allows Stream to offer such effective water quality estimates for the wide array of inland and coastal waters across the US and more, and that leads us to the website and API where you can access these water quality maps.

18:54 Speaker 3: The main tool on Stream is the interactive map page, where you can see all of the day for a given day's satellite overpass at once. Normally, the data from these satellites are segmented into square tiles in a grid. Stream stitches these tiles back together so you can see the connected water systems in full context. The webmap has many other features that I'm excited to get into shortly in our demo.

19:25 Speaker 4: Aside from the map.

19:27 Speaker 3: You can search and download scenes from the archive, which stores the results of our daily processing since it began in June of twenty twenty four.

19:38 Speaker 4: These products on.

19:39 Speaker 3: The archive are full resolution and georeferenced, and they represent an individual tile in the satellite's grid. If you'd prefer to do things in a more automated fashion, you can use our Wristful API to perform most of the same fund that you're able to on the website, such as query and downloading product maps, and that takes us to our demo. In this demo, I'll show you more of Stream's features as well as an example of how to search water quality maps. I'll show you how to query and download scenes of an area of interest, in this case Pyramid Lake.

20:26 Speaker 3: Stream does have full Conus coverage though, so the things that I'm about to demo are applicable to many inland and.

20:34 Speaker 4: Coastal water bodies.

20:37 Speaker 3: If you have any questions, please enter them in the chat and I will address as many as I can at the end. All right, so here we are on the home page of Stream, and as you scroll down you'll see a little bit about Stream and some of the things that we talked about in our presentation, you'll also find video demo on using.

21:01 Speaker 4: Streams, map page and.

21:03 Speaker 3: Features, and some of the latest updates that we are excited to announce.

21:10 Speaker 4: Back up at the top.

21:12 Speaker 3: If you go on to the map page, you can get there from either the nav bar or this explore map button that takes us to the map page with nothing loaded in. And if we're looking for an area that we know, we can go ahead and type it into this location search bar. So here we'll look for Pyramid Lake and it's found it and taken us right over to Peermid Lake. Now the easiest way to search for data for Pyramid Lake.

21:48 Speaker 4: We're going to go ahead and zoom out a.

21:50 Speaker 3: Bit here because the satellite tiles are quite large, and then we're going to enable one of our satellite grids. So here we'll look at the Sentinel two grid and you can see those square tiles that the satellite breaks data up into that I was talking about. This lake is a good example because you can see it also falls on the border of two different tiling schemes. So here we've got the one starting with ten, on the right, we've got the one starting with eleven, and in the middle there's a bit of overlap.

22:22 Speaker 3: What that is is just different segments of the satellite grid, so these tiles can sometimes be represented by more than one satellite overpass.

22:35 Speaker 3: When you've identified the area that you're interested in and you've got your satellite grid enabled, you can just go ahead and click on that tile and it will bring up the search results for that tile. So here it's brought us to the search results for the sentinel tile eleven TKE and you can see the results as sorted by date of availability. And we want to look at an example of a harmful alghol bloom affecting pyramid lakes, So we're going to go back to October the tenth of last year. When you click on one of these results, it'll bring you to the archive page where you can download all of the results.

23:22 Speaker 3: So clicking anything in this middle column would download the product represented on the rate. So this would download the full resolution chlorophyl a map as well as some of the other products. The bottom you can also find the compressed archive for all the products. But on the left column we have the map link and so here I'm going to click on the map link for chlorophyl a.

23:47 Speaker 4: And that will take us back to our map.

23:50 Speaker 3: Zoom back to the tile we were looking at, but it will populate all the data for the day that we clicked on. So again we got here by clicking on the data for this tile, but you can see the data for other tiles that were observed on the same day are also loaded in. We can go ahead and turn off the grid now and we're going to zoom in. You can see as we zoom in, the map will increase in resolution, so we can see on October tenth, there's a bit of a powerful happening in the water, but it's it's not extreme and I wouldn't constitute a harmful alglorbim just quite yet.

24:39 Speaker 3: On the right panel, you can add additional layers. So when we were looking in the archive, I saw that the next day that had data was the seventeenth. So if we go ahead and select the seventeenth and add layer, it'll load another layer on top for the seventeenth, And now you can see on the seventeenth there's quite a bit.

25:03 Speaker 4: Of action happening.

25:05 Speaker 3: We want to step back and forth between them. We can scroll down to the active layer menu. We can see the layer for the seventeenth is on top and the layer for the tenth is on the bottom. You can drag to reorder the layers if you'd like, but you can also toggle and hide them to step between them as well as if you click expand on any of the individual layers options, you can change the color palette to.

25:32 Speaker 4: Your suiting.

25:39 Speaker 3: When you are looking at multiple layers, though, as you move your cursor around, you'll see that on the right the coordinates update to where your mouse cursor is hovering, as well as the live redout of the pixel value for each layer. So if we hover a particularly area here, you can see that on the tenth it was a pretty low chlorophyll concentration of two point four seven, but on the seventeenth it was high at fifty five point four to two, and you can get a sense of that as you move your mouth around to different areas.

26:21 Speaker 4: So if you want to access your.

26:24 Speaker 5: Data via the API, we can go back to the homepage and on the rate here we have the API documentation.

26:37 Speaker 3: This is in an open API document for the RESTful API that serves stream, so on the left you can see the various endpoints and details of how to access them. The parameters that they require and everything you need to get off the ground with scripting. But in the top right we also have a button to download the example script that I've created in Python that you can modify to get working with Stream a little easier. So here I've pulled up the API example scripts and at the top you can see some configuration values that you can change.

27:21 Speaker 3: So we've got a tile list, which is just a list of the tiles sentinel to or landst format. There's some examples here, the date range in the specified format for the start and end date of your query. And then the products list that you would like to download, and these are the shorthands that the API expects of. The valid options are specified here, but these are just your products for chlorophyl a, total suspended solids and the SECI disc depth. I've also tried to comment the rest of the script if it's something you'd like to dive into and tweak yourself, but for now let's stick to the examples.

28:04 Speaker 3: So it's very simple to run. You just need to run it with any version of Python, and you can see that it's found eight dates that stream has data for the tile that we've asked for in the date range that we've asked for. So this title here eighteen suj. It's a part of Chesapeake Bay and we're downloading data for most of January of twenty twenty six.

28:43 Speaker 3: It's also gone ahead and created a stream downloads folder in the location that we have this script, and it's downloading those full resolution product maps for the tile for Chlorophyla to that folder. And now that it's completed, we can go ahead and check and see that we have a new folder and it contains those products.

29:14 Speaker 4: That we requested.

29:20 Speaker 3: Now, if we want to look back at Pyramid Lake again, what we would do is I'm going to clear out the folder, going to delete the maps that we downloaded. So here I'm just deleting the folder to get the tiles out of the way so I don't get them confused between the next batch of tiles from Pyramid Lake. But you could rename the folder or even change the location of the download directory in the script. I'm going to go back to editing this script, so instead of eighteen suj, we want eleven tke. That's the tile for Pyramid Lake.

30:05 Speaker 3: If you don't remember or you need help finding the tile that you're interested in, remember that the satellite grid checkbox on the map page, load in the full satellite grid with all the title names in the middle of the grid. And then for the date range we were looking at October, we can download Let's just do the full month of October, and then we're going to go ahead and stick with chlorophyl a for the product because we're interested in looking at chlorophyl for the harmful algoa bloom. So we can save our changes, rerun the script the same way we just did, and it found seven products for Pyramid Week in October, so we can see it successfully downloaded everything, and if we look back at the folder again, we have our colorphal maps.

31:34 Speaker 3: So if you have any questions about the use of the stream website or the API or both, please go ahead and leave any questions that you have in the chat and I'll try to address as many as I can. At the end, I would like to briefly acknowledge the various grants and teams that have supported the continued development of Stream. I've also provided a QR code here for you to try Stream yourself. Going back to our example of Pyramid Lake, a MEETA will now show you how to use the products that we downloaded through the Stream API in order to create a time series.

32:19 Speaker 1: Thank you so much William for the excellent presentation and very informative demonstration of using stream webtool and API both and next moment to continue using stream to make time series of water quality parameters. So we'll download several images using stream API and use qgis.

32:41 Speaker 2: To make time series.

32:43 Speaker 1: We'll pick locations Pyramid Lake and also Chesapeake Bay. In Pyramid Lake, will follow what William showed and we'll have a few images to look at time series of chlorophyll a concentration and then pick a small region in Chisapeake Bay and look at total suspended sediment time series in the bay. So next time I will share my screen and show this demo. So first of all, I'm using the same script that William showed for Pyramid Lake. And as you can see, this is the same tile eleven Tkee from Sentinel two that covers Pyramid Lake.

33:28 Speaker 1: So choosen that, and I'm just choosing a few days so that we can make time series. To see the procedure here October twenty twenty five from fifth through twenty fifth, so just twenty days, and we'll see how many images we have, and then I picked LA as William showed, let's also look at what I chose in Chi's a big bay. So for that I went to the stream web tool and then entered Pig Bay. Here now you can see sentinel two grids, and this is the grid we want to focus on. It's eighteen suh. Here is where Puto mc river goes into the bay and brings a lot of sediment sometimes.

34:14 Speaker 1: And so we've chosen this as a case study that will look at this style, not to look at time series of total suspended sediments.

34:28 Speaker 2: So now I.

34:30 Speaker 1: Have this copy of this script for Chiesapeake Bay in which the tile is eighteen suh. It's a longer time period May twenty four to end of twenty twenty five. And I've picked product lists as TSS and when these both these scripts I executed just as how William demonstrated. I have a number of files in streamline downloads. These are tip files. I've saved files for Pyramid Lake in a directory and all these R files for Pchicipeke Bay. So now we will move on to qgis and see how to make time series. I have downloaded and installed qgs on my computer and once you have it, you can click on this qgis I can and it will open an empty project.

35:30 Speaker 1: Now, just a note here in the appendix of today's presentation there is a procedure and there is a link on how to download qgis and how to install qus, as well as different plugins for doing calculations that has been described in the appendix, so you can refer to that. Here we have opened this new project. First we'll start with go to web and a map through quick map services. This is open streetmap and we just load the standard streetmap so we get the geographical regions here. Now we can add data to this.

36:14 Speaker 1: So I'm going to start with Pyramid Lake and we have raster layers the data that we saved earlier, So say add raster layer and navigate to the directory where the data are. So this is streamed downloads and in Pyramid Lake directory, I have data here. I have four days and I'm going to load pick this to add to the raster and what I'm going to show here is how to let's let's close this and you will see that you can click here and go zoom to the layer. So now you can see Pyramid Lake. Here you can change colors by going to properties and symbology

37:11 Speaker 1: big single band pseudo color. This is maximum, minimum and maximum value, and you can change that to a reasonable range.

37:21 Speaker 2: So I'm just saying one to.

37:25 Speaker 1: See twenty five milligrams per meter cube, and I'm going to have instead of continuous, have equal interval, and I'm going to increase number of intervals here, label precision to two and inward color maps. So when there is red its higher values of chlorophyll. Once you do that, you can say okay, and you will see. This is for seventeenth the image that Williams shoot and this is what you will see. After symbology you can see this. One more thing I want to show before I add rest of the rashers and show you how to make time series is that if you want to find time series at one point, you will have to pick one point.

38:22 Speaker 1: If you want entire lake, you will have to pick entire lake by adding a vector shape file here. What I'm going to do is suppose you want to pick a site for either water atrol or for setting up an a culture site, or for fishing or recreation, and you want to know a particular area how chlorophill a consultation varies over that region.

38:48 Speaker 2: With time.

38:49 Speaker 1: You can add a shape file. So here we're going to go and say create layer. Add new shape file layer. Here you can give.

39:00 Speaker 2: A name, uh, say lake and save.

39:07 Speaker 1: Go to geometry type add polygon because you will be picking an area. And now you can just say okay, and now to add that polygon here, you will click on this big click on this touggle edit and polygon symbol and then start by clicking and adding a polygon. When you want to add the polygon, you can say control right click and you can give it ID, say number one, and say okay. Now you have picked a region over which you want to find time series for a number of days. Now, just to save some time, I've already loaded all the rosters in Project Pyramid time Series and we'll work with this.

40:02 Speaker 1: So we have chlorophyll A rasters for tenth, seventeenth, twenty eighth, and twenty fourth of October twenty twenty five.

40:11 Speaker 2: I've also added.

40:12 Speaker 1: A polygon where we will have time series values and I've named it region of Interest on ROI. So once let's look at the here. This is seventeenth, this is twentieth, and this is twenty fourth October, so you can see how it algal bloom started and it increased and then it went up and then bent down by down again. So we see that cycle here. Now, if you want to find time series of area averaged chlorophyll, you will want to use raster analysis and zonal statistics to find mean value for this polygon.

40:59 Speaker 1: Now, if you use this Zono statistics plug in, you will have to or tool. You will have to do it raster by raster. Here you we only have four rasters, but if you have many, then it's not very convenient to do one by one. So there is another raster. It is also raster analysis Zono statistics. But this is for multiband and this is this comes from an experimental plug in from Dimexian Labs. So you have to install this into your QGIS and procedure to do that is also provided in the appendix. There is a link where you can go and install this experimental plug in Dimexian Lab.

41:48 Speaker 1: So we're going to use this Sono statistics multi band to create time series area average value for this polygon. To do that, the first step is that you create a virtual raster. So you go to raster miscellaneous and click on build virtual roster. Here. What you're going to do is add all your rasters and make one raster in which all the images appear as different bands. So here you can pick your input rosters and you can arrange them by dates and say, okay, you want to place each input file into a separate band.

42:39 Speaker 1: And important thing is that there is no data value that you want to add here, so that when you calculate statistics, those points are.

42:50 Speaker 2: Excluded.

42:51 Speaker 1: And how do you find that. You can click on this layer and go to properties. Sometimes it's given in the information tab, or you here you go to transparency. No data value is minus three two seven six seven, So you can find out what the value is and then you can add that to this virtual raster formation. Then you can run this. And when you run this, it creates a virtual a aster which is already created here. If you click on this virtual a master and look at properties and go to information, you will see that all four images now they're stored in here as bands, different bands, and that will help us in finding statistics in just one go by using multi band zonal statistics.

43:54 Speaker 1: So now once we have that, we can go to the plug in sooner statistics here. This is the astra layer. Virtual layer would be an aster layer, and pick your region of interest polygon for which you want to find mean value and you can say ch L would be a column prefix. The statistics will be stored in attributeable for this polygon and it will have prefix CHL. You can select statistical quantities you want to make time series off here. Accounts is the number of points of pixels within that polygon that we pick.

44:45 Speaker 1: This is some of values mean, mediums and a deviation, et cetera. Let's fix and a deviation and maximum value. These three quantities for chlorophyll, I can say okay, and then you can click here to run this process and at the end you will get statistical quantities for all the days or all the bands in this case in the attribute table that is associated with this shape file. So you can right click and open attribute table. You will see that this is the prefix that we picked. Band one is day one, Day two, three and four means on aviation and mixing.

45:27 Speaker 1: These values are available here. You can save this file as I can click right click on this and export as saved features and then it's comma separated value or a CSV file you can save to the folder of your choice.

45:47 Speaker 2: I've already done.

45:48 Speaker 1: That and save this file as p l ts dot c s V. So these are the numbers that we got for four days, means, deviation and maximum value. I've just rearranged them in two columns so that it's easier to read. These are all the numbers, and you can separate means, deviation and maximum values in separate columns and then plot. But for simplicity, I've just plotted all of them here. Just use insert and then chart, and then you can say bar or here. You can choose from here and you will get a plot here. So what you see here is day one means and deviation and maximum value.

46:36 Speaker 1: This is for day two, day three, and day four. You can see day two and three. So seventeenth and twenty October quite a lot high intensity of fluorophila concentration about three hundred milligrams per meter cube. So this is a quick way to get data from stream, get them in qgis, and make time series. I take the same for Chesapeake Bay suspended solids. So I want to show that to you here. Procedure is the same. I have all a few rasters for May to October of twenty twenty four. Here I made virtual roster by adding all these days as bands in this raster and have a polygon that's sitting here at the where there's potomac coming into the bay.

47:40 Speaker 1: One thing to see here is if you click on different rastas, you will see that this is TSS and this is in grams per meter cube, so it's a scale is the same for everything. But in some cases you will see good coverage. Sometimes you don't see Some areas do not have data, and that is there are two reasons either maybe there is no overpass here now satellite data or sometimes there are clouds and you cannot see the surface anyway. But this is the area for which where I created virtual raster, used solo stakes statistics multi band and created.

48:33 Speaker 2: Time series for that as well. So the time series is here.

48:42 Speaker 1: Now I here I have mean values only of TSS, and this is for different days. You can have dates here and then you can make a proper time series. But this shows how it mean values of TSS varied in that polygon into the big bay with time. So this is a easier way to make time series from stream data and hope you can find it useful and you can explore this on your own as well. So that ends our demonstration for time series. So this brings us to the end of today's session, and just to summarize, we saw background overview and demonstration of Stream webtool and API.

49:40 Speaker 1: We saw that stream is based on lands set eight and nine and Sentinel two, A, B and C optical measurements to obtain water quality parameters including chlorophyl a concentration, total suspended solids, and secy disk depth in coastal estuaries and inland lakes in the US. It provides it's the water quality parameters at twenty to thirty meters special resolution, and the API allows search and download of multiple images of an area of interest. Then we saw a couple of examples of how to select water quality parameters in areas of interest using stream.

50:18 Speaker 1: We focused on Pyramid Lake and then also on a portion of Chesapeake Bay. William showed how to map the water quality parameters and how to access and download multiple water quality data using the Stream API, and then we saw how to make time series of area averaged water quality parameters using qgis so clear advantage of using Stream can be seen here. These two figure show chlorophyl a concentration in Pyramid Lake from a pasoci at one kilometer and sentinel to MSI from stream at twenty meters.

50:55 Speaker 2: So here.

50:56 Speaker 1: If your application requires with all of water for drinking utilities or for aquaculture site selection, you want to use high resolution data because it shows a lot of details and special variation of water quality parameters. Now next week, on seventeenth of February, we will have an introduction to a machine learning model to estimate water quality parameters based on satellite observations. This is the model that's been used to produce stream parameters. As I mentioned earlier, there will be one homework assignment posted on seventeenth of February on the training website and answers must be submitted via Google forms.

51:43 Speaker 1: The homework will be due on tenth of March. Certificate of completion will be awarded to those who attend all live webinars and complete the homework assignment by the deadline. You will receive a certificate by email approximately two months after completion of the course. Once again we thank William Wainwright for his excellent presentation and demonstration of stream and making the webtool and API available to the community. Here are some resources and links to important websites, and acronyms are defined for the satellite and censor table.

52:27 Speaker 1: This is the contact information from William Wainwright and you can always contact us at our set, our set website link and YouTube link they're provided here. For questions, comments, or to share how you have applied our trainings to your work or studies. Please email at NASA rset at gmail dot com and join our quarterly newsletter to stay up to date on our latest trainings. You can send an email here with no subject line and follow the instructions sent in response, and then you can have access to the newsletter.

53:12 Speaker 1: Thank you everyone. We'll start with our question and answer session. We have William wain right here and we'll start with the questions. Question one, is it possible to replicate this on another continent? Also? What processing scale is available?

53:31 Speaker 4: Okay?

53:32 Speaker 3: So I'll do my best to go through and answer as many questions as we saw. If by the end of it you feel like your question hasn't been answered, please ask again or ask what more details? And we have a Q and a document that will be sending out after all the trainings have been completed that will hopefully answer everything. So to get back to question one, is it possible to replicate this on another continent? It definitely is. The models that are behind stream. The aim in developing those models are to have a globally applicable model to any inland and coastal water body.

54:15 Speaker 3: They obviously do have varying performance in different regions, and that will depend some on the amount of insitu samples that we've been able to acquire and train on from that region. So if you want to improve the accuracy of STREAMS models, you can always try to help provide the Freshwater Sensing group with more institute data that we can train on and we can continually improve those models. Question two, What countries are covered by stream?

54:49 Speaker 4: Right now?

54:49 Speaker 3: Stream covers the continental US or Konis, Hawaii, Alaska, and then there's lots of many select water bodies and satellite tiles around the world, including but not limited to, areas in India, South Korea, South Africa, Benin, Ghana, Chile, Uruguay, Peru, Colombia, Mexico, and Cuba. A lot of these select tiles were born from partnerships with other research teams, but We're hoping to eventually be able to cover as much of the world as we can.

55:30 Speaker 2: Thank you, William.

55:32 Speaker 1: The next question is not directly relevant to this topic today. Is there any way to monitor groundwater data?

55:40 Speaker 2: It's TSAR.

55:41 Speaker 1: Is there any way to monitor seawater and groundwater intrusion rates in coastal areas like Sundarbans. Currently incitude data is being collected in very limited areas, but due to the complexity of the region, it's not very effective. I just want to point out here is that ARSET will have a training on groundwater in April, which you may want to check out and join, where we can editor some of these questions. This training is mostly focused on stream and water quality, so right now when we are not addressing this question, but we'll definitely think of it when we do their ground.

56:21 Speaker 2: Or to training.

56:23 Speaker 1: Next question is for William. Does the tool work well with touch screen based devices like iPads and so on?

56:32 Speaker 3: Unfortunately, stream does not work very well right now on mobile devices unless you enable desktop mode. It should work fine with touch screens on desktop mode, but the mobile.

56:43 Speaker 4: Display of the page is not great.

56:45 Speaker 3: That's something I'm working on improving pretty soon hopefully. But yeah, you'll find it works bested on desktops.

56:56 Speaker 2: Great.

56:57 Speaker 1: Can move on to next question as well, what is the special cover your products understream?

57:03 Speaker 3: So Landset eight and nine derived products retain the native spatial resolution of thirty meters. Sentinel two derived products the top of atmosphere level one data comes into between ten and sixty meter resolution. We resample everything to twenty meters, and then we also provide the water quality products at that twenty meter resolution.

57:29 Speaker 4: And that kind of ties into the.

57:33 Speaker 3: Another question of why not use kind of a fusion between those two satellites. If you're referring to like the harmonized Lancet Sentinel products, that unfortunately doesn't include all of the original bands from both sensors, and one of the bands that's missing from that harmonized Sentinel Landset product is actually pretty key in how we derive water quality, right.

58:02 Speaker 1: So I think that also takes care of question six. If I select a Sentinel two file, does that mean the data for that tile is estimated from Sentinel two?

58:12 Speaker 2: And so why not use fusion? So I think you just answered that question.

58:17 Speaker 3: If you select the tile, you will get results just for that satellite. There are other ways and We're in the future looking at something called domain adaptation, which will let us kind of harmonize the results ourselves from different satellites.

58:37 Speaker 4: But that's something that's still.

58:38 Speaker 3: Very much experimental and not ready for the deployment on stream.

58:45 Speaker 1: Also, I just wanted to add to question five when you talk about special coverage, these data are global lends AT eight and Sentinel two. They both have global coverage right now. What quality parameters are produced for the US and for a few selectric ties as William mentioned, but the model that will be presented next week that can be used to get aut quality parameters anywhere where you can see resolve lakes and coastal areas exactly.

59:16 Speaker 3: And next week's training with Ryan O'Shea will talk more about how to apply the models that we use to any tile by not just LANDSTT and Sentinel, but also Pace, Sentinel three and many more. So if there's a satellite that you want covered or a region that you want covered that isn't currently supported on stream, you almost certainly will be able to do it with the models, and you'll learn how to do that next week.

59:48 Speaker 1: Yes, next question is is base data on stream web.

59:54 Speaker 3: Pace is not currently supported on stream. But as I was just saying, the models that.

1:00:04 Speaker 4: Stream uses do support PACE, so we do have water quality data four PACE, it's just not currently on stream.

1:00:15 Speaker 2: Yeah.

1:00:15 Speaker 1: The next question is does it extract data from Sentinel to or LENST as soon as they are available, or are the water quality data uploaded to stream at set period when a considerable amount of products are generated.

1:00:31 Speaker 3: We automatically download the scenes in bulk from Copernicus for Sentinel two and USGS for LANSA eight and nine pretty much as soon as they become available at level one. Now, typically they become available at the end of the day that they were observed, and given that we're processing between six hundred and eight hundred tiles per day, it can sometimes take well into the next day to finish downloading and processing all you know, eight hundred of those tiles. We don't have any specific order that we process them.

1:01:10 Speaker 3: We just asked for data. It becomes available to us from the data providers, and then we process it as we receive it.

1:01:20 Speaker 2: Great.

1:01:21 Speaker 1: The next question also, you can at risk can I collect carring you previous data in the Gulf of America.

1:01:30 Speaker 3: Coastal regions of the Gulf are covered natively natively by stream. If you find that your site is not covered, you can either apply the models yourself or reach out to us about potentially including your region of interest.

1:01:49 Speaker 1: Question tense, stream gives model estimates, how do we get these standard errors of these those estimates? How do we get these standard errors for the model est mates of chlorophyl TSS and SICHI depth.

1:02:04 Speaker 3: Yeah, we're working to add uncertainty as an additional product to the archive as far as how we derive that air and kind of the training and validation in general. That's something that'll be covered a lot more in part two of the training next week.

1:02:23 Speaker 4: But you're right that we don't We don't provide.

1:02:26 Speaker 3: Air maps on stream currently, but we would like to soon provide uncertainty in the archive page at least.

1:02:37 Speaker 1: Great question eleven, is there any way that I can improve the temporal resolution of the data? I mean, if I need daily data, how can I get it?

1:02:47 Speaker 4: I'm gonna I'm going to jump back to question kind of. I thought of something else as well.

1:02:51 Speaker 3: In the case of training and validation, I think the standard air means a little bit more because we're comparing against institute, so we actually have a known.

1:03:04 Speaker 4: Thing to compare to.

1:03:06 Speaker 3: When that model is then applied to near real time data, we don't have it suits you validation. So the uncertainty from the daily derived products on stream is more of just the model's best guess at how accurate it thinks it is. So it's it's still valuable because it gives you a measurement of effectively how close the water quality parameters that you're looking at match up with something it was trained on. So if you have a high uncertainty from streams daily processing, that will mean that we're probably giving you water quality estimates for a water body that doesn't look very much like anything we've seen in training.

1:03:54 Speaker 3: So that's what an uncertainty map for the daily processing would mean.

1:04:02 Speaker 2: Okay, can go to the next question as well.

1:04:05 Speaker 4: Ye, so is there any way I can improve?

1:04:12 Speaker 3: So one of the trade offs of the increased special resolution of Landset and Sentinel two relative to the other water quality satellites available is that decreased temp resolution. In the future, we're gonna try to cover more satellites with an aim of increasing temporal resolution though.

1:04:34 Speaker 2: Great.

1:04:34 Speaker 1: Yes, so as we saw earlier, you know, other satellites like Taraqua SMPBGPS as base. They do have daily data available. The resolutions are not as high as sentinel two in lens a day, so there's there's always a trade off between temple.

1:04:53 Speaker 2: Special resolutions exactly.

1:04:57 Speaker 1: Next question is is this calibrated? How acurator of the results? Is there any way to calibrate the data.

1:05:04 Speaker 3: So the models behind Stream are pretty heavily calibrated and validated using data sets like Gloria and aironet oc, but they're also supplemented with lots of individual institute data sets from research partners. The accuracy of the result can vary depending on how many samples that we've had to train against in that region. The best way to help calibrate data for Stream as a whole would be to help us expand our.

1:05:37 Speaker 4: Institute data set.

1:05:38 Speaker 3: If you find the performance is lacking in a region of interests, you may be able to help us improve the performance in that region by providing insitu data. You can also run the models locally to assess performance on an area that STREAM doesn't cover.

1:06:00 Speaker 2: Right.

1:06:01 Speaker 1: Next question is are these values of different parameters only for surface water or do they represent subsurface water in the water bodies.

1:06:12 Speaker 3: That's a question I don't fully know The answer to my best understanding, although this may be a better question for Ryan o'sh next week, is the water quality parameters are derived from the remote sensing reflectance at the water's surface, but the properties derived from that reflectance are assumed to be indicative of the water the water column as a whole.

1:06:38 Speaker 4: Again, now that's my understanding, but.

1:06:45 Speaker 3: Please if someone knows better in the chat or you know, next week, Ryan will probably be better be able to better answer that question as well.

1:06:54 Speaker 1: So yeah, we can talk to Ryan more. But these parameters are for surface water, as we are using satellite would look at top like water, leaving that that's what is used, which has information about water column. But the the values that you see are for surface water.

1:07:16 Speaker 4: Yeah.

1:07:17 Speaker 1: Question fourteen, what is the model evaluation for the model used for estimating chlorophyl A and TSS for SENTILEL and that of lendset. Is there a reference that can be cited for this information?

1:07:33 Speaker 3: Yeah, The name of the model and specific is aquaverse. We have a paper and I'll try to include a link to the Q inside the Q and a document. Yeah, but if you search for aquaverse you should see a paper about our models. And then also Ryan O'shay's GitHub has some of the the MDM tutorials.

1:08:04 Speaker 1: So next question is interesting, does stream work in seasonal bodies of water around the world?

1:08:13 Speaker 3: If I'm understanding the question correctly, are those water bodies where there's only water prevalent during certain seasons and then.

1:08:23 Speaker 4: They dry up?

1:08:24 Speaker 2: Or yes, I think that's what it means.

1:08:28 Speaker 3: Yeah, the way we identify water pixels to process is not exclusive of those types of water bodies, which means that if there's water in an image, it should be processed by the dream and report water quality. But yes, stream should work in those seasonal water bodies if you have examples. I've never personally looked into it, but I'd be really interested to see how those look.

1:09:01 Speaker 2: Great.

1:09:05 Speaker 1: So question sixteen, is it possible to have a quick look at the data used in the estimation of glorophile A and other presentations?

1:09:16 Speaker 4: So go ahead.

1:09:20 Speaker 3: The Gloria and air onet oc data sets, I believe are publicly available, but you can learn a bit more about some of the data and products used and the estimations and the other predictors.

1:09:36 Speaker 4: And next week's session just.

1:09:39 Speaker 1: To add that Arsett did a training in which Gloria observations or measurements were covered, and that link you will find on the slides where gloria data was mentioned, so you will find link to the training where you have more information about gloria data. Next question is can we get information about other water quality parameters? For example, if need data regarding nutrients, can I get that? So William, we can add to this. But to my knowledge, we are looking at optical properties of water. So if it changes color of the water and if these spectral bands can see that, those are the parameters they are derived.

1:10:28 Speaker 1: For nutrients like nitrogen or phosphorus, you'd probably have to have some satistical measure model to use that to connect institute data with satellite data and then indirectly derived nutrients based on that.

1:10:44 Speaker 2: I believe.

1:10:46 Speaker 3: Yeah, the limiting factors really does the parameter you're interested in change the optical properties of the water. There are some parameters that are not on stream that can be derived, like color, dissolved organic.

1:10:59 Speaker 4: Matter re set on.

1:11:01 Speaker 3: That's something that the models can produce, but we don't currently offer on stream. But there are many properties that people are interested in, like you know, dissolved phosphorus or oxygen that can't really be measured from space because they don't really measurably change the optical properties of the water. Yeah.

1:11:24 Speaker 1: The next question, please clarify the data range is available for these analysis. What is the earliest data available to analyze? So online web tool will have data from twenty twenty four onwards, correct William.

1:11:40 Speaker 3: Yeah, so stream began daily processing, so downloading data the day it became available and then automatically publishing it in June of twenty twenty four. You made these themes from before then. Most of those were manually processed with collaborators, you know, wanted a time series for a specific tile or two from twenty eighteen onward, so you'll see individual tiles covered across many years before twenty twenty four.

1:12:12 Speaker 4: We're trying to fill in.

1:12:15 Speaker 3: Coverage that began after June twenty twenty four, at least back until June of twenty twenty four, to make everything synchronous. But right now, full CONUS coverage for Sentinel began at the start of twenty twenty five, and we've recently fixed our landset downloading, so full CONUS for Landset began.

1:12:39 Speaker 4: Twenty twenty six.

1:12:41 Speaker 3: Oh, okay, select regions for Lancet have been available since June of twenty twenty four though, and well I think I said full conis for Sentinel has been available since the start of twenty twenty five, So right now the coverage can dep a bit on what tile you're looking at, but that should be permanized pretty soon. Everything should go back to at least June of twenty twenty four.

1:13:11 Speaker 1: Great next question is when analyzing tts from satellite data, do we focus on the highest value the lowest value are a different metric like the average? I would say that average, so range and average they both should be looked at. But you know we can get more information about it. Well, we have a couple of more questions million if you can address we have a few minutes. Also, does the model only use land set and sentinel or there are other hydrological and minological parameters used.

1:14:00 Speaker 4: There are some additional parameters, namely in speed. There's also the sensor.

1:14:10 Speaker 3: Angle, but that's still kind of the part of the satellite data.

1:14:14 Speaker 4: We have wind speed, water vapor, and there's one more.

1:14:19 Speaker 3: Oh ozone and NO two concentration are also used to do the atmosphere correction. The aerosol contribution is taken care of by the MDN, but the rest of it, the really correction and the gashous absorption, we need the aforementioned parameters.

1:14:41 Speaker 1: Correct question twenty two. Is it possible to use HIMA? What did you stationary satellite with stream?

1:14:51 Speaker 4: It's not currently possible to use other satellites with STREAM.

1:14:55 Speaker 3: I would have to check if that satellite is supported by the models, namely aquiverse, and I think we skipped over twenty If we apply STREAM to a region like Jordan that is not explicitly listed, what type of local data would most improve model calibration and confidence in the results.

1:15:24 Speaker 3: Any in situ chlorophyl A, tss SECHI, disk depth and SETAM measurements would help. The ones that are most likely to help the most are institue measurements that coincide pretty closely with satellite observations in terms of time, But any institu data will help.

1:15:54 Speaker 1: Going down to question twenty three for the nutrient level analysis, since these are not optically these are not optically inherent properties of water, indirect methods such as empirical method. Emial method is the only way.

1:16:13 Speaker 4: If you're able to.

1:16:16 Speaker 3: Like empirically show that it correlates to something that is optically.

1:16:25 Speaker 4: Adherent to the water.

1:16:27 Speaker 3: So, in other words, if you're able to find an optical property that's a good tracer for the property you're actually interested in, then that's one way to do it. If you're not able to demonstrate that something like chlorophyla or TSS is a good tracer for your nutrient analysis, then to my understanding, you would have to find another way to measure it.

1:16:55 Speaker 1: Next question is how soon would the rest of the world be covered with this small especially North Africa region.

1:17:04 Speaker 3: We're able to add individual tiles pretty readily, but if you're talking about full global, inland and coastal coverage, that's an order of magnitude more processing than we already do, and we don't have the processing and storage available to do that yet.

1:17:30 Speaker 1: So that is one of the reasons why the model is open source and made available to anyone who wants to apply it in their own.

1:17:39 Speaker 3: Leg and our hope is that the model will at least be relevant in globally, which is why we continue to retrain and revalidate it with each new institute sample that we get. But as far as stream, I think global coverage is less of a priority than increasing temporal resolution currently for us.

1:18:10 Speaker 2: Right, thanks.

1:18:12 Speaker 1: Next question is can we use stream for monitoring water quality in reverse?

1:18:17 Speaker 3: If landst thirty meter resolution or Sentinel at twenty meter resolution can resolve your river or other water body, then yeah.

1:18:32 Speaker 1: The next one does the bacteria account like ecoli and fecal quliform also be found out through stream?

1:18:43 Speaker 3: I think that's kind of similar to the question about nutrients. If you can trace it with a water quality parameter like ESS or chlorophyla, then potentially, but it's not something we report on directly.

1:19:06 Speaker 1: Next question is what are the top five features for the model from the feature importance analysis? Are they the same for Lancet and Sentinel.

1:19:17 Speaker 4: Top five features for the model?

1:19:22 Speaker 1: We can address it next week as well.

1:19:24 Speaker 4: Yeah, I'm not sure, so I'll refer that to Ryan probably.

1:19:35 Speaker 1: Next question is is there any training for applying the model for individual location?

1:19:43 Speaker 4: There is.

1:19:44 Speaker 3: Next week's training with Ryan O'Shea will show you how to run model yourself on a tile of interest.

1:19:57 Speaker 1: Twenty Question twenty nine. When Stream is applying in regions without extensive institute data, what level of confidence is considered sufficient for decision making or policy applications.

1:20:11 Speaker 3: That is a really good question and I think the answer to that does depend somewhat on you know, the decision making person or body.

1:20:24 Speaker 4: But I mean, first of all, when you have remote.

1:20:34 Speaker 3: Remote sensing derived water quality, you're never going to be as accurate as in situ sampling. So I think the best use for remote sensing water quality in a decision making capacity is to guide use of your limited insitu resources. So if you see things from remote sensing that make you want to follow up with in citu data, I think that's the best use of the remote sensing for like actual decision making and policy applications.

1:21:06 Speaker 1: Yeah, I might add to that that depends first of all, if you have institute data, and then if you can compare with what remote sensing is showing that gives you somewhat some confidence or air bar for remote sensing parameters, And then how much tolerance your decision would have. What range can your decision tolerate? That also is important to know.

1:21:29 Speaker 4: That's a good point.

1:21:33 Speaker 1: And the next question is can stream be used for a river that is less than thirty meters wide but very long.

1:21:42 Speaker 4: It can definitely be.

1:21:45 Speaker 3: Used, but you may see a reduction and within them because some of the land from the area will have been averaged into that pixel.

1:22:01 Speaker 1: So you so accuracy may be less in that case.

1:22:04 Speaker 3: So yeah, and depending on how much land was averaged in, it may actually be filtered out as not water.

1:22:14 Speaker 1: I will also add that if it's less than thirty meters wide, even thirty meter is you need at least three pixels clear in water to get good water quality.

1:22:27 Speaker 2: Estimates from remote sensing.

1:22:29 Speaker 1: So that way, I would say there will be a lot of contamination from land if.

1:22:33 Speaker 3: You Yeah, it's not an ideal use case, and if you do get results from stream, it won't be nearly as reliable as if it were a much wider water body.

1:22:48 Speaker 2: Yeah.

1:22:50 Speaker 1: Next question is do you use Do you use any of the water processing software such as eco light CENT to corel light, to gen C, to our CEC for converting toa surface level reflectance to water living.

1:23:07 Speaker 3: We don't use any of those in specific, but we do regularly compare our performance to those. We use our own in house atmosphere correction that.

1:23:19 Speaker 4: Is a mixture of.

1:23:23 Speaker 3: Raley correction and Gassess absorption correction, and then we use a mixture density network to do the aerosol contribution.

1:23:32 Speaker 4: So as a as a package all of that atmosphere correction is done.

1:23:37 Speaker 3: Just by us, but we do compare it pretty regularly to some of those other processors.

1:23:44 Speaker 1: Yeah, just to add, we will have some information on this next session because md and model will require water living reflectancies. So you could start with any level of satellite data and use any of this atmospheric correction models to convert CoA to surface level reflectancy remote sen collectancies, and then to attend to m d N. Next question is what model was used, machine learning or physical model? Which particular ones were evaluated before the best was chosen.

1:24:27 Speaker 3: So we use machine learning for the aerosol contribution because that's something that's been proven to be pretty difficult to do with the physical model.

1:24:38 Speaker 4: We do.

1:24:40 Speaker 3: Raley correction with a physical model because we found it to be more performant than if we also let the make sure density NetWorld take care of the Raley contribution, so we did.

1:24:53 Speaker 4: We did evaluate it.

1:24:54 Speaker 3: For for the case of ralely, but for aerosol contribution much have to use the mixture density network to get good performance.

1:25:11 Speaker 1: Question is what are the most common misinterpretation or misuse of stream outputs that users should be aware of.

1:25:21 Speaker 4: My answer to that would probably be.

1:25:25 Speaker 6: Cloud pixels can still frequently show up. We do our best to remove cloud and land pixels before we create the water quality maps, but it's not hard to find examples where you know, some cloud pixels may have made it through the cloud masking, and naturally we'll throw off the results in that area, so you'll see what might appear to be really high chlorophyll a data that it's just because there's cloud. You should now for the last couple weeks of data be able to find RGB true color composite maps that should help you see what the scene looked like on that day and be able to see for yourself what the cloud cover looks like.

1:26:15 Speaker 4: But we still have to.

1:26:19 Speaker 3: Kind of backwards propagate those true color composite maps for the rest of the stream data.

1:26:25 Speaker 2: SEP great. Thank you all.

1:26:30 Speaker 1: I think we are almost at the end of this session. If there are any more questions, we will try to address them later on, but we hope to see you next week at the same time on seventeenth of February. Thank you all for attending this session. Thank you to our guest speaker once again, and all the r SET team for their help with this training. Yeah I think there's one more question. Can it monitor turbidity in lakes?

1:27:09 Speaker 4: I think rabidity correlates to.

1:27:14 Speaker 2: Seki depth. Yeah, yeah, so you.

1:27:20 Speaker 1: Can look at sechi depth and then you know that will show increase or decrease turbiitity. If you actually want to to relate it with turbidity units, then you probably need that conversion, a relationship between those two in the water body of your interest. Okay, thanks everyone, We'll see you next week now, thank you.