NASA ARSET Overview of OCO-2 and OCO-3 Observing Modes and SIF Observations

0:01 Speaker 1: Hello everyone, and welcome to the second session of this intermediate three part r SET training titled Solar Induced Fluorescence or SIF Observations for assessing vegetation changes related to floods, droughts, and fire impacts. I'm doctor Erica Potist, a scientist at NASA's Jet Propulsion Laboratory and also an instructor with the RSET program. Today's session will provide an overview of OCO two and OCO three observing modes and SIF observations, followed by demonstration on using the snapshot area mode data from OCO three.

0:40 Speaker 1: We have two invited experts, doctor Yunjilu and Jackie Ryan, both from NASA's Jet Propulsion Laboratory. This is the training outline. There are three sessions associated with this training. Today is session number two and it consists of an overview of OCO two and OCO three observing modes as well as IF observations. There will be a theoretical section and a demonstration on how to access and visualize OCO three SAM mode data. The third session will be next Wednesday, October twenty ninth. At the same time, there is a homework associated with this training and that homework will open on October twenty ninth, so during the last session, it will open up the homework and it's going to be due on November twelve.

1:41 Speaker 1: You can access the homework through the training web page, and participants who have attended all three sessions live sessions and complete the homework by the do date will be given a certificate of completion and just a reminder. These are the prerequisites for this training, the fundamentals of remote sensing, and a couple of years ago, we did an introductory training on SIF. It was a SIF and lighter to assess vegetation change and vulnerability. So the prerequisite would be a review of the SIF components of that training.

2:24 Speaker 1: In the next couple of slides, I'll provide an overview of today's session. So today's trainers are myself. I'm acting as the ARSET training coordinator for this webinar series, and we have two invited experts. We have doctor un g Liu, she's a scientist that NASA's Jet Propulsion Laboratory. She will provide the theoretical component of today's webinar. And we have Jackie Ryan, she's a data visualization developer at the NASA's Jet propulsion laborator as well, and she will be leading the demo. Here are the objectives for today's session.

3:11 Speaker 1: By the end of this session, participants will be able to identify the characteristics on operating modes of OCO two, OCO three, and other SIFT data sets, Recognize the synergistic use of SIF with other data sets, Recognize some of the applications of SIF. Recognize how the OCO three Snapshot Area Mode or SAM mode can be visualized using an open source Jupiter notebook as a means to evaluate land change due to fire impacts. Run this Jupiter Notebook environment from less than one and run the OCO three SAM notebook to analyze and visualize vegetation change due to fire across two case study sides.

3:58 Speaker 1: And finally quantify the CIF GPP relationship by comparing remote sensing data with OCO three or sorry remote sensing data from OCO three with Eddie flux tower data at the two case study sites. How to ask questions. Please write your questions in the questions box, which is located in the bottom right. There are three points. There's a menu that will pop up. Select the first option, which is Q and A, and there you can write your question feel free to answer your questions during the presentation and we will try to respond all of the questions during the Q and A session after the after this presentation, the remainder of the questions.

4:51 Speaker 1: If we don't get to answer all of the questions, we will answer them in the Q and A document. So we compile all of the questions to a Q and A document and we will post that document on the training web page in about a week. So today's session is focused on OCO two and OCO three observing modes and SIF observations. Our guest instructor is doctor Young Liu. As mentioned earlier, she is a principal scientist that NASA's Jet Propulsion Laboratory and she is an expert on SIF, so we're very grateful to have her UH lead this part of the training.

5:35 Speaker 1: Thank you very much, doctor Liu, and welcome.

5:42 Speaker 2: Thanks Eric for the introduction, and thank you all for joining the second part of the ARS training on SIF.

5:50 Speaker 1: UH.

5:51 Speaker 2: In this part of the lecturer, I will first give an overview of the Orbiting Carbon Observatory two and OCO three observing mode and they're a solo induced chlorophyll flowers safe observations, and then followed with a brief introduction to the level of three SAF products. In the end, I will and I will talk about application examples using these observations. So OCO two and OCO three both observes two main observables columns you two and in the safe observations using reflective sunlight, but they are in different platforms.

6:35 Speaker 2: So here the animation shows the columns you two from both OS two and OCO three, and OsO two is on a train, so it has poll to poll coverage depending on the seasons, and has fixed third like one thirty pm equatorial crossing time, and was launched in July twenty fourteen, so it has more than ten years of record, so it had The footprint size is pretty small, it's about one point three by two point three kilometers, but it has a very narrow swaths, so it only had eight footprints across track. It has about sixteen day repeace cycle.

7:13 Speaker 2: But OsO three is mounted on International Space Station, so it only has observations within fifty two degrees latitude bands. It also has a different coverage depending on the season, and observations also have different time of the day overpassing time, so this is very unique compared to OCO two so it has irregular repeace cycle and also irregular overpassing time and was launched in May twenty nineteen, so it has a little bit more than five years of record. So here I will talk a little bit about observing mods.

7:53 Speaker 2: OS two has three observing moss, so, nater, glint, and target mode. For the native mode on the left panel is looking straight down, so it provides the highest spatial resolution of footprint size on the surface, so it has better coverage over like partially cloudy or like the surf surface with high topography. But native observations doesn't have very good signal to noise racial over dark ocean surface just because ocean is highly absorbing of sunlight at natter mode, so this comes to the Glen mode.

8:32 Speaker 2: In the Glen mode, the spacecraft points instrument toward a bright Glean spot, so the solar radiation is reflected from the surface. So this increased the observation the signal to noise racial over the high latitude and also over the ocean surface. So over the ocean surface we have glint observations. And then for the targeting mode, the observatory locks its view on specific surface location and then retain that view of wealth lying overboard, so the target mode are primarily used for validation purpose

9:13 Speaker 2: besides the light nator like gland and the native observations. OSIO three has this various unique snapshot area maps for the SAM observations. O three scan this eighty kilometer by eight area within two minute, so it has a little bit broader observation coverage compared to target modes. And this plot showing the special stitutions of SAM observations focused on the self. The yellow location shows the location with a little bit with lower safe observation lower save values representing lower productivity, while the purpose sizes are the high STAF sites.

9:55 Speaker 2: And then you can also request additional self observations by requires by filling up online form OEO three same website. So I put the link for both the form and the OCEO THREEISM website here and between the request time and the first collection it takes about sixteen days. Here are two examples of ocoree SMS SIVE observations. The left is over Cairo, Egypt and the right over into China. So both sides are semi arid region with coverage. So you can see that the crop and the forest area has much higher safe value indicating much higher productivity compared to the background, so for both sides you can see that feature.

10:52 Speaker 2: These are two examples of safe malabs from the OTHO three website where it has all the columns U two and the SIF maps from the SAM locations. So the this the left shows the SIF maps over harber forest fluctower sites in the wintertime, while the right is at the same location but during the summertime, so you can see during the wintertime the safe value are relative much lower because of lower productivity, while the right has much higher safe values. And I can also see this spacial gradient, so it has much higher values over the top right corner, indicating higher productivities of the plants.

11:50 Speaker 2: So this animation displays monthly means safe values from Z two. It highlights this consistent of high productivity through throughout the year in tropical regions as well as this seasonal shift. You can see over the mid and high latitude, so during the burrial summer the midwest shows very high productivity primarily due to crop growth, and I can see that the high latitude there is no observations during the winter just because OS two using reflect sunlight and there is no light, so there is no observation during the winter time.

12:30 Speaker 2: Well, this animation presents uh BI weekly safe observations from O two and three together. Well, the SIF spatial patterns are very similar to the one that I just show, but the distinct something patterns of U two two three are much more obvious. Here you can see like the OsO three has its irregular spatial coverage, but OSU two had this fixed potopol observation coverage as mentioned earlier. So OsO three is mounted on the International Space Station, so it has this irregular orbit and also irregular local overpassing time, and this allows OsO three to observe SIF at different time of the day across its different overpasses.

13:19 Speaker 2: So this sixth plus display the local overpass time merried at the difference between the local passing time and the local solar noon over sixty day period. So this data shows that OSEO three a local overpassing time can range from approximately like ninety six to six relative to noon time. So because of this different overpassing time, the OSCO three observations can help us understand dironal cycles of plants productivity. This shows an example of SIF direnal cycle observed by OSIO three. This is over center Eta, California, although the observations at different times of the day were collected across multiple overpasses, but the patent generally shows a higher productivity during the noon and early afternoon, while lower productivity during the early morning and the late afternoon.

14:27 Speaker 2: And this example shows regraded target safe observations during target mode. The different footprints has very large overlaps, so we can take advantage of this feature and regrade the target mode observation to a much higher special resolution than the original footprint size. So this regraded product is at five hundred by five hundred meter resolution instead of the original one point three by two pounds three meter resolution. And this is over Caltach site in Los Angeles. On the left is the safe value in October nineteenth while the right shows the safe value in April twenty two for California.

15:14 Speaker 2: In October, we have much little water available, so the productivity is very low, shown here as low safe values. Well, during the spring we have much more water available, so that has higher productivity and shown here as higher safe values. And this is the special distributions of the target site across the globe, and those sites are also corresponding to a surface remote sensing teacon sites which are used for validation of the through two observation from two and three. So those sets are mainly concentrated over North America and Europe and some sets over East Asia.

16:06 Speaker 2: Both OS two and OCO three have very narrow swaths as I mentioned earlier, so this results in very significant spatial gaps between observation tracks. So this can become very inconvenient for a lot of data users. So in this next section, I will talk about the level three gap field SIF products that derived from OSU two oh three level two observations. I Like level two data products, level three had this wall to wall spatial coverage and also has much higher temporal coverage. This schematic plot illustrates a general process that has been used to generate level three SIF products.

16:55 Speaker 2: This proach different approaches. Generally just combine the satellite derived like vegetation disease such as e v I fraction of UH part or meteorological variables such as air temperature, vapor pressure, deficits SO moisture, and then combined with the machine technic machine learning techniques to derive a relationship between the input variables and the output variable SIEF, and then we can then we can use this trained model to predict the SIF values where we don't have SIEF observations. There are a few uh level of three SAF products that's based on those two level two SIEF, So GOSIEF is one of the most broader use products.

17:47 Speaker 2: It's generated by University of New Hampshire. So it has three temporal resolution eight day monthly and annually. It has eight zero point of five point of five spatial resolution. It also has a GP products corresponding to these resolutions, and SISIF was produced by University of Columbia that has four day temporal resolution and same special resolution as GOSSIF. Well, also two zero five products is generated by Cornell University. It has bi weekly temporal resolution and the same special resolution as gossive and uh CSIF.

18:34 Speaker 2: And this shows example of two level three products and also corresponding level two product so we can see the difference between the level two and the level three products. On the top shows annual means at one degree resolution and the right shows maximum SIEF and also at one day resolution. Well added added the bottom two panels shows that the corresponding level three products at point oh five resolution, so you can see like for the level two saf products still have some temporal gaps, has some has still had some special gaps.

19:14 Speaker 2: Well for the level three products it has this very nice world to wall coverage but retains this special characteristics characteristics at the level two products. For example, both shows the maximum silf values over the Midwest during this maxim in the maximum silf maps for the level three for OS two for three. As I mentioned before that because OS three is on International Space Station, so it has different time of the day at different overpassing time and this actually allows the community to generate diurnal levels safe products.

20:01 Speaker 2: So the left panel shows the hours from local noon for these three days, while this animation shows the hourly level of three safe products derived from zero three SIEF. So it has shown this very nice direnal cycles of three of the safe values and this data also available on this link. This shows example how the the level of three direnal safe products can help us understand the direnal cycles of GPP over the globe. So the the plus shows the difference between GBP in the afternoon and in the morning for two July months the top shows the dry July months.

20:52 Speaker 2: Well for the bottom shows the wet July and the wet and dry are defined at each pixel and we can see for either dry July or wet July, the afternoon GPP is lower than the morning GPP, indicating that the productivity is reduced in the afternoon due to the water stress. Well, in the dry July, the depression due to this water stress is even larger, which is very obvious in the southeast of US and also over the tropical region, indicating that the plants is much more water stressed during the dry year of the in July.

21:37 Speaker 2: So this indicates that level three the arenasive products can help us understand the direnal cycle of the plants for the synthuses. Another level of three direnersive products I want to introduce here is called geoseph. It combines three SEF and the greediness from a geostation satellite called GK two A so GK two A is situated over Eastern Asia and ocean any region UH. It has reflectance and short wave raidation and also viper pressure deficit derived from this satellite. Geoseph has hourly temporal resolution and at two kilometer spatial resolution and this the top left shows those three SIEF values and the top right shows the reflectance the near the greenest products from GK two A. But the bottom two kindels shows the level three SIEF in the morning and the level three SIEF in the afternoon, so it shows the morning SIEF has much higher values than the more than the afternoon SEF.

22:50 Speaker 2: This shows the similar feature as I just showed in the previous light that the afternoon productivity is reduced just because of the waters less especially during the summertime. On the next I will talk a little bit about synergistic use of OSCIO three STIFF and Eco stress data on the International Stapace Station UH. It has a suite of the cell emissions that can that observe the Earth. The eco stress is one of the satellites that observes of the surface temperature and also retrieved water stress.

23:32 Speaker 2: I will give a brief introduction of eco stress instrument. So the eco stress has special solution about seventy meters and except by seventy meters, so it has much higher special solution than OSIO three, But it also has very white swaths the swats with its close to four hundredometers. The plot on the right shows OSIO three stiff overlaid on top of the land surface temperature retrieved from ecostress and you can see that the higher sift values are on top of the lower temperature. This is because when plants for the synthesize, it releases water to the atmosphere and at some time absorbing heat so then cooler cool the lands surface.

24:26 Speaker 2: So this can really tell us how effective the plants converts water to productivity. So the community has defined quantity to measure this. It's called water use efficiency is defined as the ratio between the productivity. Here we can use to represent productivity and the evaper transparation which is amount of water released during for the synthsusis which can be derived from land surface temperature from ecostress and OSTIO three. Project team has generated this co located eco stress and also three data products and it's available on NASA Data Service and also I provide the link here here actual example using both SIEF and et from equal stress to investigate the direnal patterns of vegetation for the senses and the coupling between the water cycle and the carbon cycle.

25:32 Speaker 2: And the top two rolls shows uh the sief Direnal patterns and et direnal patterns over Amazon. So you can see that both thief and the evuption sparation increases towards the noon time and decreased toward the afternoon time. That's because we have more sunlight, so you have more productivity, but at the some time you release more water. But for what we use, efficiency is actually decreasing close to the noon time. That means the plant is more water stressed close to the noon time and also means it needs more water to fix the same amount of carbon at new relative to the morning time.

26:25 Speaker 2: So some example tells us that we can use SIF and IT together to understand the coupling between the water cycle and the carbon cycle at dinal dinal scale. At last, I will give you a few examples of using ZIEF to solve the to understand the climate impact on the productivity and how we use SIEF to improve the crop yield predictions.

27:00 Speaker 2: The first example is is. The first example investigates the response of the productivity to climate normalies. Is here is showing the response of this productivity to twenty twenty heat with in Siberia and the top final shows the temperature NORMALI is in March April May in twenty twenty and it shows that the temperature anomaly can be about five degrees celsius in this region. And in year twenty twenty Siberiation experienced the warmest spring in the record. Were corresponding to this warm temperature normalist we can see this increase of productivity.

27:50 Speaker 2: This is based on the gaossive GPP products. So this indicating the warmer temperature in the high latitude in springtime. Actually it promotes plants growth. This is because over the high altitude where it's normally very cold, so the plants grows is limited by temperature. So when the temperature is warmer, the plants actually grow better. So this tells us This example tells us that we can use SIF products and also the GP products to understand the climate anomalies on the plants grows. The second example and the show is how we can use uh SIF GPP products derived from U two and three to understand the drought impact uh So this study look at the drought impact the drought impact on the for the senses over Southeast Asia.

28:49 Speaker 2: So the left panel shows the lend crowd types over this region you can the green color represents the forest while the yellow represent the crop mainly cropland the middle panel shows the drought index averaged between September twenty fourteen and August twenty sixteenth, about two year time period, so red color means much more severe drought relative to this Corresponding to this severe drought, especially over the central part of this map with the gossive shows a large reduction in the productivity, especially over this crop land area.

29:35 Speaker 2: So this example indicates that gossive products either the derived level three products could help us understand the drought impact on the productivity. So the last example I want to show is how we use safe to improve the crop yield predictions. As we all know like it's very critical to to know the crop productivity for early warning of the food insecurity or for the economic planning purpose. Well, the crop yield is normally generated by the statistical method at the end of the harvest season, which can be very late for the planning purpose.

30:22 Speaker 2: Well, this study demonstrates that we can use silt to generate low latency crop yield prediction. So the left panel shows the crop yield prediction based on the climate data and also SIF also to SIEF data, and the middle panel shows crop yield prediction over Middle West US only use climate and land surface temperature, while the red panel shows the observation from statistical method So the comparison of these three plots, you can see that the SIF based prediction better capture this space distribution of the observation then the prediction only based on the climate.

31:05 Speaker 2: So this really shows the promise of generating the low latency crop yield prediction based on safe products from OZ two. So this come to the summary and some resources that you can check later on. So OCU two had this global observation coverage, well, OSU three has observation at a different times of the day that are really complimentary. So we have both Level two and the Level of three save products available and this can be used to understand the impact of climate predications on vegetation health and can also be used in drought monitoring and crop yield predictions.

31:50 Speaker 2: And here at least a few websites that I have talked about in the lecture, including OSAM website and also the products that I have mentioned throughout this lights. With that, I will hand it over to Jecki thank you so much.

32:14 Speaker 3: Thank you Ginja. Now we'll begin the second exercise of this SIFT training course, covering the OCO three snapshot area map mode and flux tower data by looking at case studies in two different regions of the United States. This is following up from our previous session introducing SIFT data and how to work with it. If you've been following along from the first exercise, you should have already installed the course materials and have the Jupiter Notebook environment set up. If you have not done so yet, please review the read me in the code repository or the instructions that we provided in part one before proceeding with this exercise.

32:53 Speaker 3: The first thing I'm going to do is open up my Jupiter notebook in my browser, and so I'll turn over to my desktop so that I can do that.

33:06 Speaker 3: And you can see I already have the Jupiter Notebook code open in my text editor, although you can use the terminal as well if you prefer. And I'm going to run the setup script that we ran in part one again and there's no harm in running this script more than once. And when I do so, it's going to open the Jupiter notebook in my browser. And we can see now I have the Jupiter server running, so I'll pull my browser window down to my screen and here you can see our notebook exercise one, and I'm going to turn over to the second exercise with our Jupiter environment open.

33:58 Speaker 3: We'll want to make sure sure that we're looking at the notebook exercise two. OCO three snapshot area map as you can see on my screen. So now that we have that out of the way, what is SAM mode. While OCO two and OsO three share the same spectrometer design, OsO three incorporates a new component, the pointing Mirror Assembly or PMA. This hardware addition enables the instrument to rapidly sweep a mirror back and forth inside the optics, expanding the observation swath. Within time windows of several minutes, OsO three can collect data over a nearly contiguous eighty kilometer by eighty kilometer region.

34:43 Speaker 3: This capability makes SAM particularly valuable for sift measurements, as it captures fine scale vegetation heterogeneity that would otherwise be missed due to spacecraft's long revisit times. In this exercise, we'll discuss what SAM data looks like, how to retrieve it, and improve upon the method that we used for plotting granules from the first exercise to incorporate footprint coordinates for each sample. Once we have a handle on plotting SAM sift data itself, we can talk about how SAMs are acquired repeatedly at different sites over time to build up a picture of vegetation dynamics in different seasons.

35:22 Speaker 3: The function of SAM observations is similar to that of Eddie Covarian's flux towers, a well established ground based measurement technique that is used by ecologists to study vegetation dynamics. We can compare data from these two sources to determine if we're getting the same story, so to speak, from the data in both cases. To accomplish that will correlate gross primary production in other words, GPP with OCO three sift values from SAM mode. The first site that we'll be using as a case study in the exercise will be the University of Michigan Biological Station in Northern Michigan in the United States.

36:04 Speaker 3: This site will provide a baseline for ideal conditions when studying a temperate, deciduous broadly forest, and we'll contrast our results here with the second site that'll be mentioned later. The site lies in a rural area on the northern tip of Michigan's Lower Peninsula and is comprised of about sixty percent vegetation cover with some ag cultural fields and small towns interspersed in the surrounding area. If you want to investigate a particular site on your own, one of the easiest ways to do that is actually with Google Maps.

36:37 Speaker 3: So I'll actually go to Google Maps so that we can pull up this site ourselves. And what I'll do is I can either input the coordinates of the site itself, or I can go to the rough location if I know where that is. In my case, I'll be inputting the coordinates of the site self. I've included those coordinates in our fourth notebook that we'll be discussing later. For now, we'll skip a specific discussion of that. So here's the latitude and longitude coordinates of our site, and I'll talk about where I got that information later, But first I just want to discuss what this location looks like.

37:25 Speaker 3: And so I've put this into Google Maps, and you can see that if I switch to the satellite layer, this is the location of the case study site that we'll be discussing today. When I zoom ount, you can see that this location is on the lower Peninsula of Michigan. Here's the upper peninsula. And when I zoom in to the highest zoom level, the highest resolution data that we have available, you can see that there's an interesting feature here, which is the tower itself. It looks somewhat similar to a cell phone tower.

38:06 Speaker 3: The next thing we'll do to investigate this site is turnover to the three SAM website itself to investigate this tower location in the mission repository, and the website that we use for that is included in the introduction paragraph of our notebook. So I'll go to this link here so you can see the URL is up here, and this website contains a map of all of the various sites that OCO three has studied using the SAM mode over the life of the mission.

38:50 Speaker 3: You can find a specific SAM site by heading to the curated SAM web page that we're looking at right now and searching for it by name or finding it on the map. The different colors for these pins represent different types, in other words, different local ecological conditions that can guide our expectations of what level of plant activity we can find at that location. Let's look at our University of Michigan biological station site just by zooming in on the map and looking around for the location specifically to see how that site is categorized.

39:28 Speaker 3: So if I click on this pin, you can see that this is the site that we're interested in. It's called SIF University of Michigan, USA, and the OCO three Database or SIF UMB is its short name. The type of the SIF site is SIF high, and so the classification that we've given this particular site is that it's a location where we can expect to find high sift values during the growing season. The nearby site that we see to the north over Sue Saint Marie with a blue square, represents a fossil site, and this is a location where we would expect to find industrial emissions, more urbanized area and increased emissions perhaps lower sift compared to a rural or wilderness area like we have in the case study site that we're using.

40:27 Speaker 3: You also notice that there are yellow pins and squares representing SIF low sites such as this one to the west, and this is where we would expect to find less dense vegetation in ecosystems like grassland or rural or wilderness areas. So let's turn back to the info bubble for the site that we're interested in, and we'll use that site name that was mentioned up here to actually search for the sam ac quisitions that were acquired at this site. And don't worry, we'll actually be plotting what these SAM acquisitions look like so you can see the actual data that we're talking about.

41:12 Speaker 3: But first we need to find what data we can actually retrieve.

41:19 Speaker 3: You can see when you click on the pin down below in this form, the site name gets populated for you automatically, and that site name should be sufficient to find the SAM acquisitions. So if we just type search,

41:38 Speaker 3: we can see a table with all of these SAM acquisitions at this particular site. The table of results describes the time of each observation as well as the number of soundings, which roughly corresponds to the number of samples we can expect to find in the data from each SAM acquisition. So here's our column with soundings, and you can see some of them have a value of zero and some of them have a much higher value. To choose the best possible example, we should therefore select a SAM acquisition with a large number of soundings to ensure that we find one with lots of sample points to look at.

42:17 Speaker 3: There are a couple to choose from in just this first page of results, so I'll choose the August eleventh, twenty nineteen example for no particular reason other than that it has a large number of soundings. When we click the plots button on the right side of the table, we can see that the OCO three team has already processed this SAM acquisition with a number of different products, so I'll click that now. The first product that we're presented with is the XCO two data, which is the primary product of the OCO three mission.

42:55 Speaker 3: However, when we scroll down, we can also see that there are a number of other different pieces of information, including a SIFT seven hundred and fifty seven nanimeter product, and this is the one that we'll be reproducing today. There are two important details we should note before heading to the notebook to actually reproduce that plot. Though. Number one, we'll want to note down the geographic extents of the plot data, which are helpfully marked along the X and yaxis. The extent of SAM coverage will vary slightly between acquisitions at the same site, so you will need to adjust these values a little bit for each date you want to visualize, or else use a larger bounding box.

43:40 Speaker 3: The other important detail to note down about the plot is the large red star on the map, and this marker represents the location of the tower site associated with the region. And recall that that tower site is the location that we were looking at in Google Maps. So let's look at this plot a little closer so you can see the exact details that I'm talking about. Remember, we'll want to note down this bounding box, the extent in latitude and longitude, as well as the location of our red star marker on the plot.

44:14 Speaker 3: Okay, with all of that out of the way, we can finally turn over to the notebook and begin running some code. As with the first notebook, in the first cell, will just import some necessary dependencies that we need, so I'll run that now,

44:36 Speaker 3: and with that out of the way, you can see I've also included the same information that I just discussed in a static form in the explainer text for the notebook. In this section. In the second cell, we will supply the data of interest and the data set, which is the same OCIO three sift data set that we used in exercise one. So again we're using the just disk downloader that we used from section one, and we'll get the granule for August eleventh. Now that we found the granule, we need to filter the data before plotting.

45:18 Speaker 3: To do this, we'll pull up the measurement mode and quality flag variables from the granule, and we know to do this by looking at section four point seven of the OCO three Data User's Guide. I won't pull up the user's guide for this particular exercise, but you can look at it on your own if you're interested. Measurement Mode three is referred to as an area map, and this is what we're actually going to be filtering the data on. We'll also filter to just select points with the quality flag of zero or one, which if you recall, are points that have a quality of best or good.

46:01 Speaker 3: Additionally, the last filter I'm using is the SAM extent variable that you can see here in the code cell itself, and this extent is based on the latitude and longitude extent that I asked you to note down in the SAM plot from the OZO three website. So, now that we've selected the actual filter, criteria that we want to use for creating our plot. The last thing to do is to run the cell itself. So an important detail is that we're using the same plotting function that we use for plotting any other siff granule from the first exercise.

46:55 Speaker 3: In this case, we get a plot that looks like this. Now, this is pretty okay. It looks somewhat similar to the SAM plot that's on the OCO three website, similar enough that we can tell that we're looking at the same data. So if I scroll down, I've included the SAM plot itself, and you can see that the shape of data points and the values that we're using, or the values that we're observing, are pretty much the same between the two, with some minor cosmetic differences, So that tells us that we're at least retrieving the correct data.

47:39 Speaker 3: On the other hand, one of the most crucial differences you'll notice right away is that the samples and the official plot are represented by quadrilaterals at varying angles. So let me scroll back down to that so that you can see it. And here we see the quadrilaterals representing each of the sift samples.

48:00 Speaker 3: One of the variables provided in the granule file that we haven't considered yet are footprint vertices. Each sample acquired by the instrument has an associated geographic extent of about two kilometers cross track and one kilometer along track. When using SAM mode, the samples or pixels of the data becomes skewed as the PMA moves, adjusting the view zenithangle as it goes. To characterize SIPH observations as accurately as possible, we should incorporate this information into our plots that we know the exact area covered by each SIFT value that we see.

48:37 Speaker 3: This is especially important in regions with a lot of spatial heterogeneity, such as where the transition between urban and forested areas is particularly sharp. One other piece of information will incorporate in our improved plot will be the tower site coordinate. I've written a function called plot sam that is included in this cell. So this next cell contains the function plot sam and this will actually incorporate the footprint coordinates to make our plot from a pere look more like the official plot down here.

49:21 Speaker 3: Specifically, the way that we'll do that is will use the geolocation footprint latitude vertices variable and the same one for longitude and will also use the map plot lib patches module to create a set of polygons containing these footprint vertices as the vertices of the quadrilaterals used to represent each sample, and don't think too hard about that, just taking the list of these coordinates and putting it into this patches dot polygon class that you can see here.

50:09 Speaker 3: Since these patches will be in the same order as the sift sample values from the granule, we can color map each polygon with its associated value using the set array method in our patch collection that we've created up here, and this will automatically handle assigning the correct value to the correct quadrilateral. Another point that we'll add is satellite imagery based map using the contextally module. For the scale of this plot, I've used a zoom level of nine, which I think strikes a good balance between the detail of the satellite data versus the amount of data that's required to actually display the plot.

50:56 Speaker 3: Each zoom level is about four times higher resolution as the pre one, two times in each dimension, So if you have a slower internet connection or you want to use less data, you can choose a zoom level of seven or eight, depending on your specific needs. And that comes at the trade off of lower resolution for the base map. The last detail that we'll want to note down is that our original plot used in equirectangular projection. For the updated plot, we're using web Mercator to match the projection of the official plot, and you'll notice the difference there in terms of the subtle difference in the shape of the data itself.

51:48 Speaker 3: So here you can see that the angle of this tale of SAM samples is somewhat different than the angle of the same feature in the official plot, and that's due to projection differences. So accounting for all of those cosmetic differences as well as the very important inclusion of the footprint vertices, let's see how our own version of the official samplot looks. So I'm running the cell. It's going to take about thirty seconds, and that's because we're actually downloading the satellite imagery to use for our base map.

52:31 Speaker 3: And here we can see that the code has completed and our result looks very close to the official samplot.

52:41 Speaker 3: Despite this, the important detail is that we are now incorporating the shape of the SAM footprints into our visualization. A single samplot on its own can be useful for context on a specific day and to get a sense of the sift behavior and your region of interest. But we can only truly build up a picture of vegetation dynamics around a site through many successive observations across seasons and years. So let's move to the second section now, where we'll be performing a more long term analysis of our UMBs case study site.

53:18 Speaker 3: Now we'll compare the SAM acquisitions to the flux tower data collected by instruments located at that red star on our plot at ecovariance. Flux towers measure carbon dioxide exchange between ecosystems and the atmosphere. Since plants take in carbon dioxide and release oxygen when performing photosynthesis, we can measure this behavior by tracking minute changes and gas concentrations around the tower site. The towers are also effectively weather stations, and the additional data they collect can be used to differentiate the motion of gases in the environment due to wind and other effects from the influence of primary production.

54:00 Speaker 3: Although some towers measure sift, this is somewhat uncommon, so we can't perform a direct sift to sift comparison between the tower and OSCO three For this particular site. When we look at tower data, the two variables were most interested in are net Ecosystem Exchange n EE and GPP gross primary production. In this exercise, we'll pull GPP values from the UMBs tower data from towers all across North America are archived on the Amerflux website and can be accessed for free with an account. In this demonstration, however, we're not going to guide you through creating an account or downloading the data yourself, since you can do this on your own later.

54:47 Speaker 3: That also means that we haven't automated this step, so you'll have to manually download data for any particular site that you want to study outside of our two case study sites.

55:00 Speaker 3: As I mentioned, we've already downloaded the tower data and included it in the CSV files in the same directory as the notebooks themselves, which you can see over here.

55:17 Speaker 3: Let's briefly pull up the amraaflex page for the UMBs tower anyways, so that we can see what the information that gets provided for a given tower looks like. And I'll do that by clicking on the link that we've included in the description text. Okay, so, looking at the Amerflux website, we can see that the coordinate that I am put into Google Maps at the beginning of this lesson is included right here. We've also got a description of the vegetation information around the site, which I also mentioned at the beginning of this exercise, a helpful picture showing context of what the tower and the surrounding area looks like, and we have links to download the data down here.

56:09 Speaker 3: AmeriFlux data is provided in a CSV format at monthly, daily, and even sub daily timescales of either one hour or every half hour. For our purposes, using daily tower data will be sufficient. And going back over to our notebook, this is the CSV file that we've provided in the course materials itself, and here you can see that file name in its full detail. Let's talk about the file name itself so that you can get a sense of what we're actually working with. So for a given CSV file with amraflux data, AMF tells us that this is a Maraflex data USUMB tells us that the site identifier code is for the UMBs tower that we're interested in.

57:08 Speaker 3: Fluxnet refers to FlexNet data as opposed to batim base or any other type of data set that might be included for this tower. The subset identifier tells us that this is only a subset of the variables. However, since we're only interested in NEE and GPP, that's sufficient for our use case. DD tells us that this is the daily time cadence data, and twenty nineteen to twenty twenty one tells us that this data only covers the years twenty nineteen to twenty twenty one. I've manually cropped the number of dates in this data to include data only from the lifetime of the OCO three mission.

57:54 Speaker 3: If you had downloaded this data yourself, you would get two thousand and seven through twenty twenty one worth of data. So let's take a look at what we're actually going to be doing. In this code block. We'll open the CSV with the GPP data and create a time series plot of its values for comparison. Will also plot the average SIFT value from the area around the tower site, which has already been pre processed in the JSON file that you can see down here in the included course materials. The code to perform this step is included in the optional appendix notebook that I briefly looked at the beginning of this exercise.

58:34 Speaker 3: Zero four underscore appendix dot IPIMB and that code has already been run for you to save time, since the search code itself to perform this PREUF processing can take upwards of two hours. For the time series, we'll use two different channels from the CSV file itself, corresponding to two different retrieval methods for estimating GPP from et ecovariance data. Interested citations have been provided on the details of these two techniques, and here are the sources for how this GPP data is actually derived.

59:11 Speaker 3: So let's plot the time series now and see what it looks like.

59:17 Speaker 3: All right, we only have a little over two years worth of data to look at here, and that's enough for us to observe the cyclical nature of the data. Remember, both GPP and SIF are proxies for plant activity, so we would expect to see a seasonal pattern in the data. Indeed, we can see that the GPP peaks in the growing season from May to October, and that there are dormant seasons in winter and early spring from November through April. The deciduous broadly forest we were studying at this site exhibits rapid transitions during leaf out in spring and when the leaves fall in autumn.

59:58 Speaker 3: In this case, we can all also note that the two GPP retrievals agree with each other pretty closely, represented by our two differently colored lines in the plot. Although there are some outliers where the two GPP retrieval methods do disagree, it's clear that the SIFT data follows the same phenological trend of high values in the growing season and low values in the dormant season. Before we move on, let's think about the differences in the GPP and SIFT data a little more critically. GPP is a measure of primary production and is therefore a direct measure of the carbon produced by plants through their photosynthesis.

1:00:37 Speaker 3: While SIF is a good proxy for photosynthesis activity, it only represents the light reactions in the plant, and when observed from space, only those plants representing the canopy rather than the entire ecosystem, are included in the measurable SIFT value that we get. Broadly speaking, observed sift will depend on the light use efficiency of the plants being observed, which can vary by species and plant physiology. For example, a crop like corn, a common C four photosynthesizer, has a different light use efficiency compared to SOY, which uses C three photosynthesis.

1:01:17 Speaker 3: Additionally, changes in plant health can have very fine spatial heterogeneity that affects the SIFT signal in some places differently than in other's. While a tower only observes a small area, the spatial aggregation technique that we use to get the SIFT values that are included in this plot average samples across a broad area of several square kilometers. If there were an ecological disruption like a flood or a fire, how might you expect the SIF GPP relationship that we're looking at here to be affected?

1:01:51 Speaker 3: Do you think that ecological disruptions can impact the phenological or seasonal patterns that we've noticed at the end of the exercise, use our second case study to run these plots again and see if there's any difference that we can note. Okay, so now that we've looked at the time series, let's try to correlate our remote sensing SIFT data with the tower based GPP data to see how related they are. You can see the general pattern of SIF follows the pattern of GPP in our plot, but we'd also like to quantify how well it does.

1:02:28 Speaker 3: So

1:02:33 Speaker 3: in in this last cell in the section we'll be performing that correlation. To do this, we'll create a scatter plot with SIF on the x axis and GPP on the y axis. For the GPP data, we'll choose the nighttime partitioned retrieval method from the tower, although you'll obtain a similar result using the data time partition values either of the lines on the plot. We'll add in a linear regression line using the sk learn module and compute in our squared value. Finally, to make the points easier to parse, we'll give each point a color based on the season that it came from.

1:03:19 Speaker 3: We've provided the option to use different marker shapes depending on the year of the data point, but to keep the plot simple, we've disabled that future by default. Let's run the code to create the scatter plot and see what we get. So here you can see that we've sorted data into seasons. We perform some simple filtering of our input data values, We create the plot here, and we calculate in our squared value down here. So again, let's run the cell and see the result that we get. It's quite a lot of code, which I won't be going through in this exercise, but there's plenty of comments, so if you're interested in the specifics of the implementation, you can review it on your own.

1:04:10 Speaker 3: After this exercise, and here we have the correlation plot. Our R squared result of zero point seventy three suggests that these variables are well correlated, and it's only a little less than the R squared value reported from comparing daily SIFT to daily GPP in the pirat at All twenty twenty two paper. That's the basis for this exercise. We can improve the correlation further by filtering the data on biome type when doing the preprocessing to create the JSON source file that we use for creating this plot in the first place, and that comes from the appendix.

1:04:52 Speaker 3: As I mentioned, as with the time series plot, we can see that there are some outlier points in the data. In particular, these points over here, the points far to the right of the trend line suggest times when the observed SIFT was higher than expected compared to the GPP value measured at the tower. We can use the samplot that we generated in section one to pull up context for one of these points so that we can observe if there's anything that we can notice about the spatial distribution of sift points that could indicate why we get this result.

1:05:32 Speaker 3: The summer data point, this yellow one furthest to the right corresponds to June TEWOD twenty twenty one, which we can note down by looking at the time series plot. Let's pull this up as a samplot, and we do so. In the last cell over here, we can see that while there are some large sift values near the tower indicated by the darker green, others are quite low, indicated by the light green or yellow. Additionally, there are gaps in the data caused by clouds and low quality data. The sift value we observed could have a variety of causes, such as our spatial aggregation, picking up agricultural fields in the vicinity during peak growing season, or due to surface and isotropy playing a larger role at the time of the SAM acquisition.

1:06:29 Speaker 3: Unlike NATER mode, where the view zenith angle should be at or near zero, the SAM mode can have varying view zenith angle values between samples due to the action of the pointing mirror assembly. In this particular case, it looks like we have low data availability, noted by the lack of data up here in the forested region directly around the tower, causing the cropland to the east over here to be overrepresented in the spatial act bridge. We can test this observation by filtering our data in the appendix pre processing step to only consider SIFT samples with a deciduous forest or mixed forest biome classification.

1:07:12 Speaker 3: If you were to do that, you would see that the observed SIF at this outlier point would be closer to the trend line, and our ur squared would be improved to zero point seventy six. Overall, that's left as an exercise to the reader. So to review what we've learned so far and deepen our understanding of the SIFT GPP relationship, we'll move to the second case study of this exercise. By creating the plots in section two over again, we should hopefully see some different results. This time, we'll be using the Metolius forest site in Oregon, also called us ME two, so I'll pull that up by searching for it directly in the Amerfleux website.

1:07:59 Speaker 3: I'll search for us ME two. If you don't know the site identifyer of your specific site, that's okay. You can find it by using the OCO three SAM website that I mentioned at the beginning of this exercise, and here we see the second case study site as opposed to the Michigan site. This location is an evergreen forest on the side of a mountain in the Cascade Range of the Western US. Because we are talking about an evergreen forest now, one of the patterns you'll notice is a less pronounced drop off and ramp up in activity during seasonal transitions.

1:08:44 Speaker 3: Another important detail to note about USMY two is that there was a major wildfire in this region of Oregon in August twenty twenty and the area around the tower itself was burned. Despite this, the tower managed to continue collecting data, so we have a valuable resource on how this ecosystem responded to the disruption. Let's turn back over to the notebook and run the plots again to see what we can observe. So I'll scroll back up to the beginning of section two, which is where we created the time series plot in the first cell of this section, and you can see we've already provided a line with the CSV file of ameriflex data for the us ME two site, as well as our pre processive values from OCO three that again come from the appendix notebook.

1:09:39 Speaker 3: I'll comment out the original lines that we used for producing our plots with UMBs site data, and I'll uncomment these lines to represent the USMB two site data, and then I'll run the cell again. We don't have a ton of pre fire SIFT and GPP data to work with. Although we could look further back in the GPP record if we wanted to, we've cut it off to the beginning of the OZ three mission for this particular exercise. This is still enough for us to see about one year worth of data before the fire and give us a picture of the phenological cycle and the matorial use.

1:10:25 Speaker 3: This is enough for us to see about one year worth of data. It gives us a picture of the phenological cycle in the Metolius forest region. As I mentioned, the dormance season down here is shorter and has less of a stark drop off compared with the deciduous forest that we studied earlier. This can also partly be attributed to the climatology, although you would see similar effects when comparing forests with the same climate classification. And helpfully, these two sites have a very similar latitude after the fire.

1:10:59 Speaker 3: GP ke heat drops off significantly, noted by this flatline down here. Nevertheless, it is possible to see a small peak in activity during the summer of twenty twenty one, with a larger peak at least in the nighttime partitioned data in twenty twenty two. This recovery around the tower can likely be attributed to serial growth in the region. Where the burn scar is filled in with grasses and opportunistic brush, the serial growth is sought. The seal growth is stronger in the subsequent year, possibly due to young trees taking the place of burned and logged mature trees affected by the fire.

1:11:43 Speaker 3: Now, let's look at the sift values that we see in this plot. While the initial sift values from twenty nineteen to twenty twenty agree with the DPP behavior around the tower, the two measurements diverge after the fire. In summer twenty twenty one and twenty two, we see peaks and sift to a similar level that we saw prior to the fire, a trend that is not reflected in GPP. When the wildfire burned through this region, not every area was as heavily impacted as the region directly around the tower site.

1:12:16 Speaker 3: Therefore, we can still observe some SIFF activity from unaffected strands of trees in the vicinity. Now, let's generate our regression plot for this example and see what information we can note. And again, as with the first cell that we ran, I'm going to uncomment these two lines containing the updated GPP and SIFT file. Okay, our r squared value is significantly worse than the Michigan site at just zero point one two one. In the absence of additional context. This might suggest that these variables are almost uncorrelated, But as we observed in the time series, there's still a seasonal pattern to sift.

1:13:05 Speaker 3: It just diverges from the GPP data due to the effects of the fire. Taking a look at the samplot where a specific date can help to make this more apparent. For the final plot of this exercise, let's pull up a SAM of the Metolius forest region before and after the fire to see what spatial patterns we can observe, at least qualitatively. For the pre fire data point, we can look at June nineteenth, twenty twenty, and then after the fire we can study May twenty eighth, twenty twenty one. We can plot those two SAMs ourselves using the notebook.

1:13:42 Speaker 3: However, you'll have to find the SAM extense manually by checking against the official OsO three plots. An easier way for us to do this is to just look at the official OsO three SAM plots themselves, since they've already been generated for us. So I'll return to the OCAO three SAM website and I'll find the us ME two site on the map. I know it's in the US state of Oregon, so I'll zoom in

1:14:14 Speaker 3: and the SIF low site that we can see here is referred to as Ecostress us ME two on the map.

1:14:23 Speaker 3: Let's now search for all of the SAM acquisitions. The first state we'll want to use is June nineteenth, twenty twenty, as I mentioned, and let's pull up the tab with the plots. If I scroll down to the SIF seven hundred and fifty seven nanometer plot, we can take a closer look at the spatial features that we note in the twenty twenty data. We note a pattern of growing season behavior in this region under normal conditions, generally higher in this season despite a somewhat lower average value compared with the GPP.

1:15:06 Speaker 3: But we can observe that the area around the tower has a somewhat lower value and low data availability. It's only when we average points from around the region of the tower that we achieve the value that you could see in the time series and the correlation plot. Now, let's take a look at the twenty twenty one data to see if we can note anything different about that particular acquisition. So I'll go to the next page of results and I'll go to May twenty eighth and let me pull up this plot. In this example, we noted a much higher SIFT value than the corresponding GPP value in our time series.

1:15:56 Speaker 3: Let me turn back over to the time series plot so that you can see what I'm talking about. Over here, you can see the time series for this site, and we're looking at May twenty twenty one, which is around here, and SIF is peaking for the growing season in this region. Despite GPP being very low around the tower site. Let's turn back over to the plot and see what we can notice. Data availability is still quite sparse like we saw in twenty twenty around the tower site, but in the twenty twenty one data we can definitely see that the value closest to the tower is relatively low compared to the surrounding area.

1:16:43 Speaker 3: There are sporadic samples in the forest in this region that have high SIPH These darker green values, especially higher up in the mountain range, that show that the spatial heterogeneity of this region is quite high. Suggests that the fire affected different regions of this forest in quite a fine pattern, with areas that were completely burnt in some regions and unaffected in other areas. You can investigate this relationship by analyzing more dates and even other sites on your own. And now that we've explained the basics of performing a sift GPP correlation, should be able to perform it for any arbitrary site within the Americ Flex data set.

1:17:34 Speaker 3: That concludes this exercise. And thank you very much for your attention.

1:17:45 Speaker 1: Thank you very much Jackie for that great demonstration, and to doctor Leeu for the great presentation. Next I will provide a summary of today's session. So in SAMMRIOCO two has global coverage. OCO three has observations between more or less fifty two degrees north south latitude and at different times of day. OCO three has an extra observing mode which is called the SAM mode, the snapshot area mode and level two and level three SIFT data can be used for understanding or addressing different applications such as understanding the impact of climate perturvations on vegetation health, drought monitoring, and crop yield prediction.

1:18:34 Speaker 1: And here's a list of resources for those wishing to dive deeper into the material that was presented today. If you have any questions about the material presented today, please feel free to contact doctr Jongliu, Jackie Ryan or Karen Ewan. And then just a reminder that there is a third session, a third and final session next week next Wednesday, October twenty ninth, at the same time, and that session will run Jupiter notebooks to compare how gat filled sift products such as GOSIF can be used for assessing the impact of floods and droughts on cropland through specific case studies in the Midwest.

1:19:28 Speaker 1: You'll be using Jupiter notebook tutorials to choose another region or timeframe of your choice and reproduce new products for another analysis. And you'll compare sift products aggregated in space and time using open source tools and how they can be used to study vegetation change across different regions in a variety of science and applied use cases. So with that, we've reached the end now of session two, and thank you again to our guest instructors and to the participants for their interest. And we've been receiving a lot of questions, so we will be starting the Q and A session now.

1:20:19 Speaker 1: So we've been compiling your questions onto a document and we will go through the questions and we will also edit this document and post it on the training page within a couple of days. All right, so let's just start from the top down. The first question, will snow an ocean and oceans show us black? Could we still see pine trees? Or will it appear black due to snow? So doctor Leo, if you'd like to go ahead and respond to that. So her response is that snow is highly absorbed, being at the near infrared band, so it does appear black, and snow can reduce the amount of light reaching the canopy and as a result, reduce the sift signal.

1:21:14 Speaker 1: That said, SIF signals from non dormant photosystems are detected in evergreen needlely forced during winter and spring in the presence of snow. And she has included here a reference article. If you would like to dig deeper into this great Let's go on to the second question. I'm doing a SIFT study for different biomes in India. Where can I get flux tower GPP data for different biomes to compare with SIFT generated GPP values? And her response is that you can find the flux tower data at the flux net website with the link that is posted here.

1:22:00 Speaker 1: However, the coverage in India is somewhat limited and you will want to check whether India has flux tower data that is not part of flux nets. So there are many flux towers around the world. Obviously there's the flux net network, but there might be other networks and you might want to check. Let's go on with the next question. Correlation between SIF and GPP tower data within each season does not appear to be strong other than in the fall. This seems to suggest that we cannot reliably interpret changes within a season, even in the summer outside the noise.

1:22:50 Speaker 1: So I will let maybe either doctor Liu, if she has resolved her tech is shoes, or Jackie respond to that.

1:23:06 Speaker 3: I do think that's a little bit of a better question for doctor Leo, but I'll try to answer it as best as I can, so towards the end of the notebook, I include some suggestions on how to improve the correlation. One of the challenges that we have with using SAM mode data is, as I mentioned, like, data can be very sparse around a tower for any given observation, which requires us to do the spatial aggregation. But at the same time, because of that, we're including a lot of points that aren't necessarily the same as the tower site itself.

1:23:49 Speaker 3: So typically the best comparison that you'll have is when a tower has photospec as well, which is relatively common. With that being said, you can still use the data, but it is, like I mentioned, recommended to filter by biome type.

1:24:16 Speaker 1: Okay, thank you very much for that response. Will go on to question number four. What is the difference between daily sifts seven hundred and forty nanometers and SIF seven hundred and forty nanometers? So I'll go ahead and read the response that's being typed. Daily SIF is the daily mean, SIF so it's the average value which accounts for changes in solar radiance throughout the day, While SIF seven so daily SIF seven forty is the average value, while sif seven forty is the instantaneous sief value retrieved from that specific band, from that specific seven hundred and forty nanometer band.

1:25:14 Speaker 1: The next question question number five, What are the specific characteristics of each retrieval band? Is there a preference between seven hundred and fifty fifty seven nanometers over seven hundred and forty nanometers.

1:25:38 Speaker 3: I'll let Ginger add any details if she'd like, But the seven hundred and forty nanimeter data will generally have a stronger sief signal, but it doesn't come from a measurement taken directly by the instrument, so it's interpolated from the seven hundred and fifty seven and seven to seventy one data. If you would prefer to use a measurement directly from the instrument, that's a case when you would want to use seven fifty seven instead.

1:26:19 Speaker 1: Okay, thank you, and let's move on. So it looks like that's the last question. However, I believe there have been new questions added. Let me take a look. No, it looks like that's the last question. Any additional questions, please go ahead and write them, or what we'll do is any additional questions that we receive. Oh, here's one, okay, so let me just include it in the Yeah, all right, So question number six, why do the result resulting data visualizations appear as sparse pixels that don't cover the entire study area?

1:27:08 Speaker 1: So, assuming we define the specific geographical boundaries, what methods are used to compensate for the missing data gaps in the final map products?

1:27:24 Speaker 3: Generally speaking, you can use gap filled products if the data availability is not enough for your use case. Other than that, there's not really a way for us to improve upon the gaps without synthesizing or estimating data from other predictors. So with SAM, the gaps occur due to bad data points or the presence of heavy clouds, for instance, preventing the estimate.

1:28:07 Speaker 1: Okay, thank you very much, Jackie. The next question regarding historical baseline data for regional studies. Since satellite imagery isn't available for periods like the early nineteen hundreds, could you recommend any key repositories for declassified aerial photography or historical topographic maps that are freely accessible for scientific use.

1:28:35 Speaker 3: Sorry, I don't know the answer to that question.

1:28:38 Speaker 1: Yeah, I think we'll have to take a look. Unless Karen or Young g if you have any insight into this, otherwise we will look into it and respond it in the document. Yeah.

1:28:54 Speaker 4: I think this is dependent certainly on the area you want to look at, the country you want to look at, because obviously aerial photography and the practice of it would be much later on. When you're saying the early nineteen hundreds, that would be difficult to get anything in the nineteen hundreds just because of the availability of the equipment to be able to do that and to do the surveillance. I would be hard pressed to find anything before in nineteen fifty I think. And also the historical topographic maps I would have to check when USGS or other agencies would have started collecting that, because that would all be produced during land surveys, so that would also be country dependent.

1:29:43 Speaker 1: Thank you very much, Karen. Okay, we have another question or more of a state. Okay, So it says the measurements have been directly or have the measurements been directly acquired from sensors. I'm not sure if this is a question. It says the lesson seems more advanced now.

1:30:10 Speaker 3: So yes, what we were doing in this lesson was comparing two different sensors. One was from space and one was on the ground. The continuous line plot that we looked at in the time series was the ground based data, and that's a real sensor. The sam plots also were real acquisitions taken over what we can consider as like one time point from the OCO three instrument.

1:30:48 Speaker 1: All right, so I think that is it in terms of questions. Just to remind you, there is one more session as part of this training series, and that session it will be next Wednesday. At the same time, we're going to be looking at gap filled products. It's going to be primarily a demonstration that Jackie will be leading, so please tune in. And as a reminder about the homework, there is one homework associated with this training and that will open up next Wednesday and it will be posted on the training web page.

1:31:29 Speaker 1: All right. With that, we've then reached the end of the second session. I'd like to give the opportunity to our our guest experts to say some final words. So Jackie, if you would like to say some words before we close.

1:31:47 Speaker 3: Yeah, I have two comments. There was actually another question that came in. Is it possible to integrate a second variable such as land service temperature or soil moisture with this data to understand the underlying changes or causes of observed changes in vegetation health and uh. This is an area of active research within our science community. And on the OSIO three website there's a recently released product that combines ecostress with OSIO three. So ecostress provides land surface temperature and it enables you to calculate water use efficiency.

1:32:35 Speaker 3: So if we do eventually do an advanced course, that might be a topic of discussion that we would cover there.

1:32:44 Speaker 1: Yeah, absolutely, thanks for bringing that up. And I think data integration is always a great interest in for our community, so there's certainly certainly something to consider. Go ahead.

1:32:58 Speaker 3: The second comment that I I wanted to make was a lot of you last time mentioned that you use Conda for your Python environment, and so I added a Conda setupscript for both Mac and Windows, so you should be able to install the course materials using Conda.

1:33:20 Speaker 1: Now, thanks, Jackie and Karen. Would you like to say some closing words.

1:33:28 Speaker 5: Just thank you to all the participants for joining us today and please join us next week because I think for those who want to work with our data, you'd be very pleased with what Jackie has put together and in this you can see she's very responsive real time and follow up for to help and support our users. So we look forward to seeing you next week as well.

1:33:52 Speaker 1: Thank you, Thank you very much, Karen again, thank you to our invited experts, to Jackie, john Gie and Karen for today's presentation demonstration, and of course thank you to all of the participants for tuning in, for your interest, for your questions and enthusiasm about this topic. And finally, I'd like to thank the r set team for making this possible. So with that, we will see each other in one more week. So until then, I'm wishing you all a good day. Bye bye,