NASA ARSET Working with Gap-Filled SIF Products

0:01 Speaker 1: Hello everyone, and welcome to the third and last session of this intermediate three part ARSEET training titled Solar Induced Fluorescence Observations for assessing vegetation changes related to floods, drought and fire impacts. I'm doctor Erica Potist, a scientist at NASA's Jet Propulsion Laboratory, and I'm also an instructor with the RSET program. Today's session will be focused on how to work with GAT filled sift products and it will consist of a theoretical session and a demonstration. Our invited expert is once again Jackie Ryan from NASA's Jet Propulsion Laboratory.

0:44 Speaker 1: So today is the last session of this three part training and it's focused on how to work with gatfield sift products. There is a homework associated with this training and it is open as of today. You can access it through the training website. The do date is November twelve, and we will be providing a certificate of completion to all of those participants who attended the live sessions. All three of the live sessions and complete the homework by the do date. The certificate of completion will be received or you can expect to receive it in about three months and this is a reminder of the prerequisite for this training, which is the fundamentals of remote sensing as well as a SIF rset training that we did a couple of years ago.

1:40 Speaker 1: It was a combined SIF and light art. So the prerequisite here is for you to review the SIF portion of that training. And here are the trainers for this session. I'm the R set training coordinator for this webinar series and we have an invited x once again returning is Jackie Ryan. She is a data visualization developer at NASA's Jet Propulsion Laboratory, and she will provide a discussion about the gap filled products and then she will do a demo to show you how to access and visualize these products.

2:25 Speaker 1: 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 question questions if we don't get to answer all of the questions, we will answer them in the Q and A document.

3:06 Speaker 1: So we compile all of the questions into a Q and A document and we will post that document on the training web page in about a week. So let's get started with session three working with gap filled sift products. Our invited expert is once again Jackie Ryan from NASA's Jet Propulsion Laboratory. She'll be presenting the rest of this session so cover both the theoretical portion and the demonstration. Welcome once again, Jackie. Great to have you back.

3:44 Speaker 2: Hello, and welcome to the third part of this art set course on solar induced chlorophyl fluorescence. In this exercise, we'll be discussing the use of gap filled sift products for vegetation analysis, using the EGOSIF data set created by a team out of the University of New Hampshire as an example. In particular, we'll investigate a case study involving the twenty nineteen Midwestern floods and their impact on agriculture in the corn belt. Here's a brief overview of what we're going to go over today and the main objectives for what we hope you'll get out of the Jupiter notebook code demonstration.

4:20 Speaker 2: Before we dive into the code, though, I'd like to provide you all with some context on sift measurements collected by other spacecraft. Thus far, we've been using OHSO two and OCO three exclusively in our exercises as a matter of course, without really discussing data sources from other missions. The ocomissions are not unique in collecting sif though depending on the needs of your own analysis, you might find it useful to incorporate data from other sources. In fact, the primary point of our discussion today is to highlight some of the limitations of exclusively relying on the OCO two and OCO three observations that we've used in the first two part arts.

5:01 Speaker 2: By using a gap filled product, we can obtain a broader spatial context for the impact of the flooding event in our case study. And you can see a satellite comparison view in the image on this slide. So the left side is twenty eighteen and the right side is twenty nineteen. The floods that year in twenty nineteen were extremely severe and caused the loss of three lives as well as significant damage to property across about a dozen states in the notebook. We'll just be focusing on the impacts to terrestrial plants rather than any of the economic or social impacts, and in specific we'll be focusing mostly on crop land.

5:46 Speaker 2: The basis for this analysis comes from the yin at Alia twenty twenty paper, and we'll be discussing their results a little later. Finally, as an extra point of comparison, we'll look at a bonus case study examining the impacts of the twenty twelve Midwestern drought on cropland in the Corn Belt. Before we dive into discussing other sources of SIFT data, I want to review why we're even using SIFT for vegetation analysis in the first place. If you're familiar with working with multi spectral remote sensing data for this type of work, then you might be aware that there are numerous types of vegetation indices that can be produced from satellite data.

6:28 Speaker 2: Some examples of these are the normalized difference Vegetation index and DVII, the Enhanced Vegetation Index EVII, near infrared reflectance from vegetation or nerve, and the Chlorophyll Carotenoid Index or CCI. All of these indices are produced by taking a ratio of observed spectral signals as a heuristic or photosynthesis, or at least for measuring the vegetation health in the environment. But the most direct measurement where actually interested in is gross primary production GPP, which is a measure of carbon fixation performed by an ecosystem.

7:08 Speaker 2: While GPP is the best possible measurement we could look at in vegetation analysis, at least from a remote sensing perspective, very challenging to observe from space. As an aside, there are some star based measurements that can measure biomass, like the recently launched Esis spacecraft itself called biomass, but this approach is relatively new and data coverage is limited. With all that being said, then, because of the challenges of using DPP directly, it is often necessary for us to use alternative measurements.

7:44 Speaker 2: The plot on this slide compares some of the metrics I've mentioned against GPP to show how well they are correlated. This data is for a boreal forest in northern Canada, so the correlation will vary somewhat depending on the biome type used in your own analysis or in the case study we'll look at today. For example, in this case, CCI is almost as correlated with GPP as SIF is, but CCI is particularly well stuited to studies of gymnosperm dominated biomes like this boreal forest due to the phenological changes and reflectants in the needles of evergreen trees.

8:24 Speaker 2: In other words, in the deciduous forest, evergreen broadleaf forest, or in cropland, we might not expect to see the same correlation between CCI and GPP. So I want to take a moment to look a little more deeply at this plot. Starting at the top, we can see that IF is the best correlated variable with GPP within R squared value of zero point nine to two. The second best, as I mentioned, is CCI within R squared to zero point nine. However, I also want to take a moment to point out NDVII with its R squared value of zero point seventy nine.

9:07 Speaker 2: You may already be familiar with MDBI from your own research or analysis. Thence it's easy to measure, and therefore it's commonly used across a wide variety of applications. In the case of our boreal forest here, even though it has good overall agreement with GPP, it does exhibit a hysteresis effect during winter. In other words, GPP tails off down to zero in late October or early November, but NDVII remains high into December. Since NDVII is mainly measuring a plant's coloration, it will tend to be a lagging indicator of vegetation health.

9:50 Speaker 2: Moor term stresses like those caused by floods, drought, or extreme heat might not show up in NDVII until a couple weeks after photosynthesis has stopped. IF, on the other hand, is more dynamic, meaning we're more likely to see rises and falls in activity in response to stresses even within a single day. If you recall Nick's explanation in part one. This is because SIF signal is tied to the photochemical quenching reaction, which in turn tells us how open the photosystem reaction centers are within plants across an ecosystem.

10:27 Speaker 2: So hopefully you're convinced now that SIF is an excellent metric for studying vegetation health and stress response. This is a relatively new type of measurement for spacecraft, with GOMEE II on METAPA back in two thousand and six being the earliest instrument capable of picking up the spectral signal directly. This wasn't an original use case for GOME, and the data set itself was derived later. The first dedicated SIFT data sets from space based remote sensing were actually developed in twenty eleven by Joanna Joiner, Luis Kwant, Christian Frankenberg and others using jacks as GOSAT satellite OCO two followed in twenty fourteen and improved upon the existing SIF record.

11:10 Speaker 2: More recently, we now have Tropomi on board Issa's Sentinel five P spacecraft, which launched in twenty eighteen, GOSAT two launching in twenty eighteen, as well OCO three, which launched in twenty nineteen as we've previously discussed, and most recently, the Chinese Academy of Space Technologies Gomong satellite, which launched in twenty twenty two. All of these satellites have publicly available data, with the exception of Gomong, and sometimes your analysis may be improved by combining or regridding data from multiple satellites into a single product, though it's useful to keep these in mind to know all the tools you have available when looking at SIF.

11:55 Speaker 1: SO.

11:55 Speaker 2: The first other satellite that I want to talk about is GOSAT. As I mentioned, it's been around the longest of all the data sets, but it suffers from load data availability in a spatial sense, as you can see from the plot here, which I've taken from the doty at all paper that we looked at in Part one. So ghos SAT here is up at the top, and then we see gridded rasters of Ozo two and Ozo three, very similar to the plots that we created in Part one. Like many other Earth observing satellites such as OZO two, got SAT has a sunsynchronous orbit, so we'll see a point at a given latitude at the same stolar time every single orbit.

12:40 Speaker 2: While GHSAT improved upon the resolution of GOM, it's still only about zero point one degrees. It's best used for ecosystem level analysis. That might be difficult to tell from the figure on this slide because the data has been gritted to a resolution of zero point five degrees. Alongside the GOSAT data in this case, you can find more information on the data from the two papers that I've linked the dois for, and you can also find the data itself in the data set link that I've included on this slide.

13:18 Speaker 2: PROPOSIF is another great resource to use in addition to OCIO two and three data. Most of the research papers that we've referenced in this course make use of Troponi data to augment other sources of ZIF, and for good reason. As you can see from the plot on this slide, the spatial coverage is quite substantial and retrievals are available at a sub daily cadence. Though we aren't covering its usage in this course, you're encouraged to take a look at the TROPOSIF data set on your own. One thing to note is that there's no direct comparison between OCIO two and three seven hundred and fifty seven nanimeter SIF and the SIFT from Tropomei, so it's best for you to compare only the seven hundred and forty nanometers SIFT data between the two instruments.

14:06 Speaker 2: And again, as with GOSAT, i've linked a paper describing the theoretical basis for the ZIFF retrievals from Tropo ME, as well as a DOI link to the data set itself. Let's turn back now to the OCO three gritted raster plot that we made in the first notebook exercise When you look at this plot, you can tell that for June twenty twenty, and in general, for any given month, there's relatively limited spatial coverage. Although the ISS orbits every ninety or so minutes, the low swath width of the instrument means that there's going to be gaps in coverage for some months.

14:45 Speaker 2: Additionally, because of the inclination of the ISS's orbit, there's a seasonal bias as to which areas get more daytime coverage during certain times of year. For the KNUS, this means that we get a lot of coverage or in May and June. You may have noticed this when looking at what dates SAM observations are available for during the second exercise. Another challenge that we encounter with any of the products that I've just mentioned is that their observational record only goes back about a decade.

15:18 Speaker 2: For OCO two we have eleven years of data, and at the most there's GOME which stretches back to two thousand and seven, but even that is not very long. For OsO three we have just six years of data, and even then, instrument anomalies and time spent powered off means that there are significant gaps where no data was being acquired. With all of these limitations in data availability for spacecraft observations, it begs the question and we do better? Is there a technique that we could use to improve the coverage of this data and estimate values that we have not observed directly using indirect measurements from other satellites, And, as you may have guessed already, the answer is that we can use a technique called gap filling or data fusion.

16:04 Speaker 2: There are many approaches to doing this, but the basic premise is that we need to train a model using a data driven approach to predict the value of a measurement. In our case sif the value we are predicting is called the label data, and the ancillary measurements we use to aid in predicting the value are called predictor variables. These can include land cover classification, surface reflectants, or meteorological data like vapor pressure deficit BPD, air temperature, and down welling shortwave radiation or DSR.

16:38 Speaker 2: That last one, DSR is similar to photosynthetically active radiation or PAR, and then it quantifies how much solar energy a plant can receive at a given time. So, for an example of how we could do gap filling. We could take MOTUS data, which has a resolution of up to five hundred meters per pixel and gives us all of the predictor variables that we need. We can associate those predictors with the observed OSIO two or OSIO three data use all of that to train a model. The model can then tell us what we expect SIF to be in places where we have no direct observations.

17:16 Speaker 2: Modeling technique that you use for this gap filling is a matter of debate and research, but the product we will be using in this exercise is based on a cubist regression tree, which is a data driven approach rather than machine learning. With that being said, there are definitely teams using machine learning to do gap filling with SIFT data, such as the works discussed in Junia's presentation in Part two. Now, let's talk about GOSIF, the data set that will be the star of the show for our code demonstration today.

17:52 Speaker 2: GOSIF was originally developed by Shing Li and Jingfung Shao at the University of New Hampshire in twenty nineteen by fusing OCO two data with modus. What they ended up with is a spatially contiguous global product at zero point zero five degree resolution or about six kilometers per pixel, and eight day temporal resolution, although there are also monthly and annual products. The reason we can only get eight day resolution out of the data is because MODUS is only able to produce a BRDF for bidirectional reflectance distribution function every eight days, and that's a required ingredient for determining the surface reflectance.

18:37 Speaker 2: Reflectance for this purpose requires taking many observations of the atmosphere with and without clouds and so as you might imagine, that can require several days of observations on average. Remember how I mentioned that the observational record of OCO two only goes back to twenty fourteen. An advantage of using MOIS for this purpose is that we can now estimate SIF all the way back to when Tara and Aqua were first launched in two thousand because of the observational record of MODUS. If you'd like to read more about the derivation of GOSIF, you can find the paper in the Doi on this slide.

19:17 Speaker 2: I've also included a direct link to the data store on the UNH website. Although we'll go into detail about downloading the products and the code demonstration. For this next slide, I want to talk briefly about some other gap filled products, and Junja has already gone over this research, so I'll only touch on it briefly here. But there are other products that you can consider using in your own analysis. None of these have the easy data availability that you would get with GOSIF, though, so you may need to reproduce their technique rather than going to a website and downloading data.

19:53 Speaker 2: Daying Jong and her team developed a diurnal SIF product zero point five degree resolution, using climate model data for predictor variables. If you recall I mentioned that Oucio three will observe the same location at different times of day due to its orbit, and this makes it a great resource for developing diurnal models, or models that can predict SITH at hourly cadence throughout the day. This can be used to demonstrate plant's response to heat stress in the afternoon. In many places where it gets really hot, like tropical rainforest, the plants will largely close up their leaves to preserve water and engage in non photochemical quenching in other words, heat production to avoid drying out.

20:35 Speaker 2: Despite having plenty of sunlight. This can be observed as an afternoon depression or drop in SITH values. The figure I've included on this slide shows that behavior and action, and you can see that in the final time bin for a given day over the Amazon rainforest. As the researchers pointed out in this study, there's a noticeable drop in SIF about equal to what seen in the morning and an increase in water use sufficiency. Another product to look at is called Ecooco three. The OCO three science team is currently working on a product that combines ecostress land surface temperature data with OsO three sand mode measurements as a means to quantify water use efficiency and water U sufficiency, as was mentioned in this other study as well, is a measure of how efficiently an ecosystem uses water to fix carbon.

21:35 Speaker 2: So this figure shows an example of an ecooco product. This is not a gap filled product per se, but it's another example of data fusion in action. Lastly, GEOSIF, which was also covered by GINGA, is a gap field product produced by combining the Korean GUK to a weather satellite for predictor variables do provide hourly data or estimating SIF in East Asia

22:08 Speaker 2: all right, so now we want to dive into our case study for today and consider how the researchers did their own vegetation analysis. The scientists, of course use SIFT to investigate impact, but they also considered terrestrial water storage ews from the Grace follow on mission, atmospheric CO two concentrations from the ACT America airborne campaign, as well as from OCO two itself, and GPP data derived from those measurements. The reason for using CO two concentrations in this context is that it's another way to gain insight into GPP behavior, since if CO two is elevated over a region of cropland compared to normal, then it indicates that plants are fixing less carbon than we would expect.

22:53 Speaker 2: Unfortunately, there were thick clouds over most of the corn Belt during this period, which somewhat limited the data availability during June and July, but the paper was still able to demonstrate early season growing conditions over cropland exhibited much lower than normal GPP, as can be seen in the second plot on this slide. This can also partly be explained by delay and planting, which is why we see an uptick in GPP later in the season. In fact, the crops were able to take advantage of the additional water in the soil later in the season, and the top plot of this slide you can see that TWS was two standard deviations above the mean in the flood here.

23:34 Speaker 2: Interestingly, we can also notice from the middle plot that non cropland areas were less affected by the flooding. Natural biomes are usually better adapted to periodic flooding due to the existing trees and root structures that help hold the soil in place. But let's turnover now to the syph analysis that the researchers did in their study. In their case, they used Tropomi SIF at seven hundred and forty nineometers basically aggregated to a county level, and took the difference between twenty nineteen and twenty eighteen for each month over the growing season.

24:14 Speaker 2: Reinforcing the findings that we looked at in the previous slide, you can see that the delayed planting date in this last plot over here up to five weeks in the case of Illinois, meant that the early growing season exhibited much lower than normal SIF. So if we look at the June data, we can see that the difference over here shows lower than normal SIF. On the other hand, the late season had higher than average SIF over here, one question to ask is do you expect to see similar results in the GOSIF data between twenty eighteen and twenty nineteen.

24:54 Speaker 2: This might seem obvious, but my hope is that it will build your confidence about using gap filled product in this context. Another question to keep in mind is you think that the later peak balanced out the earlier dip in primary production? In other words, what might the overall effect on yield be in this particular year.

25:18 Speaker 2: And to answer that last question, we can just pull up agriculture statistics curated by the USDA and see that cornyields in the US in twenty nineteen were about five percent lower than in twenty eighteen. When we're looking at this data, how was the SIF correlated with crop yields? As Junia issued in her presentation, the most thorough way of comparing the two is by incorporating climatology to build a rigorous estimate. In an oversimplified sense, though, we can consider that the total yield is tied to integrating under the curve of SIF or GPP.

25:52 Speaker 2: In other words, by summing up the values that we see in those metrics throughout the season. This mainly for annuals because the harvested crop completes its entire life cycle within a single season, as opposed to perennial crops whose yield can be infacted by previous seasons. Looking at this yield chart, you'll also notice there's a conspicuous dip in corn yields in twenty twelve. This motivates our bonus case study, where we'll be looking at the impacts of drought on crop production in that year.

26:23 Speaker 2: Twenty twelve was a severely hot and dry year in the corn belt, and there was no extra water storage to mitigate the effects like there was in twenty nineteen. Comparing twenty twelve and twenty thirteen, we can see that there was a twenty two percent drop in yield between the years. This is made slightly worse by the fact that there's an upward slope to the yield graph due to improvements and agricultural practices, crop protection, and better hybrids that get released year over year. So that is the upward trend line that I'm referring to here that conclude, it's the presentation portion of this part, and now let's turn over to the code demonstration.

27:08 Speaker 2: Now that I've explained the context of what goosif is and why it's useful for our analysis, I'd like to turn over now to the code demonstration. Today we'll be looking at the third notebook gosif dot IPIB, and as in the previous exercises, you'll first want to load up the Jupiter Hub environment using the setup script I've provided. So I'll jump out of this presentation now and you can see I already have a terminal window open, so I'll be using this terminal window to open up my Jupiter environment as we did with the previous examples.

27:50 Speaker 2: First, I want to navigate over to where I have the code. This may be different depending on where you cloned the repository on your own computer. And then, as before, I'll run the setupscript. Since I'm on a Mac, I'll use setup dot sh. If you're on Windows, you'll want to run the setup dot ps one PowerShell script. Okay, and I see that now my server is running, and I'll pull down my browser window so that you can view the Jupiter environment that I have open. I'll now want to navigate over to the third notebook.

28:41 Speaker 2: I'm just closing the other tabs so that we can focus on the task that we're going to be discussing today. So you should now see working with GOSIF data in your Jupiter environment on your screen.

28:58 Speaker 2: The first cell here that we're going to be running is just importing libraries as we've done in the previous exercises, So I'll run that cell now. Okay, with that out of the way, the first portion of the exercise is just going to focus on downloading gos of data itself, and to reiterate what was discussed during the presentation. This data set was produced by doctor Jing Fung Shao and Shingle at University of New Hampshire, and they took OCIO two data and produced a model that's able to estimate it at zero point zero five degree spatial resolution or about six kilometers per pixel on an eight day, monthly or annual cadence.

29:48 Speaker 2: And they used a number of other predictors to estimate these values in a gap filled manner. And this is the gap filled product that we were talking about from for I've already written a helper function called download and unpack GOSIF, which will make it easy for us to download data from the UNH data store. Before we go ahead and download the data, though, I want to pull up that site so that you can see where this is coming from. So here's the UNH website containing the GOSIF data set, and this is the cover page for it.

30:29 Speaker 2: And if I scroll down here, you can see down at the bottom of this web page it says download GOSIF and if we click on that link, that will take us to a data store like this. You can see that there's three different time cadences for our data. If I click on eight day, you're able to see a number of tiff files that have been packed using gzip, and these tiff files have dates associated with them. There's a year and a day of year, and these estimates go from the year two thousand all the way through to the end of twenty twenty four.

31:13 Speaker 2: So there's a wealth of information here for us to work with. You can see that each individual file is about ten megabytes. Okay, so let's return now to the notebook. The example file that I've selected for us to work with is from June of twenty twenty and as we've been doing in other exercises, I'm going to store it in the data directory within our notebook environment. We're going to create a new directory called gosif for storing this data. If you provide a day of year, you will get the eight day data.

32:00 Speaker 2: Since I'm providing a year and a month, the function will know that I'm asking for monthly data, and if I provide only a year, the function will download the annual file from the data store. Okay, so you can see that our progress bar shows that we just downloaded the file. It was nine point three megabytes and it unpacked the g zip to turn it into a GeoTIFF file. If I go into my data directory, you can see that there was a new gosift directory that was just created, and we have our tiff file right here.

32:41 Speaker 2: Now, the default Jupiter notebook environment will not be able to view this GeoTIFF file. Some operating systems can handle this natively, but generally speaking, geotif is a format that's intended for GIS software and it's not always the easiest to work with. If I turn over to my finder window, which I have over here, and I look in my data directory and I see my new gosif directory right here, I can take a look at this GeoTIFF that we just downloaded. Now, this doesn't look like very much right now, you'd see a lot of gray areas where we have no data.

33:23 Speaker 2: You can see white areas where there's desert region and no SIF activity, and then we see a whole bunch of black. And actually this is varying shades of gray associated with the SIFF value in the particular area. But it's very difficult for us to understand what we're looking at when we just look at the raw version of this GeoTIFF. Because of that, for the purposes of visualization, I've added additional functions that will convert this geotif to a P ANDNG. So let's take a look at that code cell that does the conversion to PNG.

34:09 Speaker 2: There are a few values here that are sort of like magic numbers, which actually come from the user's guide for the data. If I turn back over here to our data store web page that I showed you earlier, you can see there's an included PDF over here, and if I open that up, which will take a moment to load, you can see that the authors of the data set have included some helpful information for understanding what we're looking at. With the data one thing. Our units are watts per meter squared per micrometer stadium, which is the same unit that we've been using for sift throughout these exercises.

34:54 Speaker 2: However, the scale factor is one times tend to thes fourth, which means that this is a value that we have to multiply the raw values from the geotiph buy to get the values in the proper unit over here. Additionally, there are two fill values which are associated with data where we would not expect to find SIF activity in these files, so we'll need to filter out those values and treat them as a transparency when doing our conversion to PNG. Okay, So turning back over to the code, you can see that we're accounting for our two no data values with this data threshold right here.

35:40 Speaker 2: So we're only going to use any value from the geotiph with a number less than thirty two, seven hundred and sixty five. After we filter the values, we'll be scaling by the scale factor that's been provided, and that should give us a result that is in the wasper meter squared tridium micrometer unit that we expect. Okay, so I'll run this cell. So now that we've run the cell. We've converted our GEOTIP file that we downloaded from the UNH Data Store into a PNG with associated metadata in a JSON file.

36:28 Speaker 2: If you're not familiar with the geotip format, it contains georeferencing information embedded in the file itself. Because the PONG format does not support georeferencing information natively, we've included it in an ancillary JSON file for use later in our visualizations. Let's take a look at the files that we just generated to see what they look like. So I'll turn back over to my finder window and I'll open up this PNG map so you can see that this is a significant improvement in being able to view the data intuitively compared to what we were seeing with the GOOTIP.

37:12 Speaker 2: You could also accomplish this by viewing the information in a GIS program like qgis, but since we're in a Jupiter notebook, I thought it would be helpful to include these p andngs. Now, one thing you may have noticed is that there's no scale reference for the SIFF values in the PNG. There was one other parameter that I haven't mentioned yet, which is our vmax parameter, similar to the vmax that we used in other exercises throughout this course. In this case, vmax is scaled prior to the scale factor being applied, So a value of eight thousand here corresponds with zero point eight watts per meter squared storadium micrometer, and this is about in line with the SIFF values that we were seeing for June when creating our gritted raster.

38:14 Speaker 2: One thing to note about the ghost sift product is that its values are based off of the seven hundred and fifty seven nanimeter sift from OSIO two. So the scale of sift values that you would expect to see should match with true observations at seven hundred and fifty seven nanometers rather than the seven hundred and forty ananimeters which are common for peak sift values. All right, with our data preparation out of the way, we can get to the fun part, which is actually looking at our data on a map.

38:51 Speaker 2: I've written a helpful web based widget for doing that, and you can see that the setup for it is in this cell. What we'll be doing here is running the widget, which is actually a web page that will run locally on your computer that enables us to look interactively at this file. One thing you might have noticed when we pulled up the PNG is that there's really a lot of information in these files, so it's helpful to be able to zoom and pan around to really get a sense of every detail of the data. Let me run this cell now, and if I scroll back up, you can see that a map has opened within our Jupiter notebook.

39:44 Speaker 2: If you find that the aspect ratio of this makes it a little difficult to view the information itself, there's also a link here and this will open the map in a new tap. So with that open, we should now be able to look at the converted sift data from the geotif in more detail. You can see I can zum when using the scroll wheel or when using the plus and minus buttons. Over here, and if I zoom in on the United States, you can see this particularly active area in the temperate forests of the eastern United States.

40:24 Speaker 2: So over here, one of the details we can see is that this area of relatively lower sift is the US corn belt, and in June typically that's right at the beginning of the planting and growing season, so the props have not actually reached a mature enough stage to have a significant SIFT signal. Yet on the other hand, the broadly forests of the eastern United States show a significant amount of activity in this late spring early summer time of year. Let's see what other details we can see around the world in our SIFT data.

41:06 Speaker 2: One of the incredible things about this GOSIF data is that because we've done gap filling now through the GOSIF technique, we're able to actually view any part of the world for any timestep that has data available, Whereas when we were producing gritted rasters, we still ended up with gaps in the data due to areas that the orbit of the iss or Oco two did not cover during the time period that we aggregated over. One thing we can see in mid latitudes near the equator are the two tropical rainforests of South America and Africa, and these areas show moderate levels of SIF, lower than the temperate forest of the northern hemisphere at the same time of year, and that's consistent with the year round photosynthesis that we see in the rainforest regions of the world.

42:14 Speaker 2: On the other hand, in desert regions such as the Sahara Desert, the Mojave desert in the western United States and the deserts in Chile and Argentina have quite low values of SIF, which is associated with lower plant activity in those regions. One last detail all note here is that because of the thresholding that we did, we were actually able to filter out ice and water values which have no SIF activity and would just cover over important context in our map, and that was the point of the thresholding step that we did earlier.

43:01 Speaker 2: The last thing I'll point out on this widget is that the folder icon up here enables you to open other converted gosift granules that you may produce throughout this exercise or in the homework exercise, and this is really handy if you want to view other data later. You don't have to rerun the cell or before any additional code steps to visualize it. Now, the details for actually implementing something like this widget that we're using here are rather complicated, and it's not terribly relevant to the discussion of the data itself, so I'll leave it out of this demonstration.

43:48 Speaker 2: It's not necessary to create an interactive plot or anything this complex for viewing GEOTIP data. This widget was merely meant to take the place of loading data into GIS software like QGIS, which is what you normally might do when evaluating a new data set like this. Before we continue to the case study analysis portion of the exercise, I want to return to the previous cell and show you one more feature that we can use when analyzing these granules. I'll scroll back up to the previous cell. When we ran this cell originally, you may have noticed this commented outline.

44:29 Speaker 2: Similar to when we were creating gritted rasters in part one, we can add an argument that bounds the data to a specific georegion. In this case, the commented out arguments allow you to bound the data to the conus. So I'll uncomment this line, and if we rerun the data, we have converted the file again, and looking and finder, you can now see that our converted geotif only contains data over the knus. This can be useful when speeding up our processing later on in the analysis. If I return to the map and I rerun the cell, now you can see that the data has automatically been zoomed into the United States.

45:29 Speaker 2: And let me turn back over here and refresh the page. And when we open a file that has a bounded georegion, it will automatically zoom the view to that georegion that you've selected. This is handy for automatically setting up where you want to see the data instead of having to pan around and find it yourself. I will note, however, that I didn't change the file in this case, so if you are generating many PNGs in this manner, you might want to specify a new output filename so you don't overwrite previous data.

46:11 Speaker 2: And if I go back up to the previous cell,

46:16 Speaker 2: you can see that the file name is contained in the gosif PNG variable here, and we've specified it here to be the same as the geotif's file name, but with the PNG extension. So if you wanted to change it, you might add something here like underscore conus. All right, so I'll remove that for now. Another side note before we continue on is that the researchers also created a data set estimating GPP from GOSIF data using the linear relationship between the two quantities. Remember how in part two we showed the linear relationship between SIF and GPP in ideal conditions.

47:07 Speaker 2: If we go back to doctor Shou's website, and let me go back here and we scroll back up on this page. You can see a GOSIF GPP web page on the sidebar. Here you click that link. This is the cover page for that data set. And here's some information on the validation of the relationship between the GOSIF GPP product and tower GPP, showing that the results achieved by the researchers agree quite well with ground based measurements. So this can be another useful resource when performing vegetation analysis, although we won't cover it directly in this course.

47:54 Speaker 2: However, if on your own you'd like to look at that data yourself, you can do so by adding in the keyword data set equal to go SIF GPP underscore V two like so, and now when you download the data, it will actually use the GPP data from the UNH data store. One thing to note, though, is that the scale value and threshold values are a little bit different for that data set, so you'll need to tweak some of the other parameters yourself to get it to work. Again, we won't be covering that in this course.

48:41 Speaker 2: I just wanted to note it down if you did want to look at the GPP data on your own with that side note out of the way. Now we're going to use this product to perform our own vegetation analysis in the spirit of the in at Alia paper that I talked about in the presentation. To recap we'll be looking at how the twenty nineteen Midwest floods affected agriculture, especially the corn and soy crops that make up a majority of the cultivated land area in this region. This event was a big deal. It was the wettest January through August period in over one hundred and twenty five years.

49:23 Speaker 2: There were fields that were underwater, Farmers couldn't plant their crops on time, and the crops that did get planted off and drowned. The first thing we'll need to do to understand the scale of this event is to download all of the granules over the entire growing season, and we'll use the eight day product here instead of the monthly that we use in the first section to have the best temporal resolution for our analysis. You can see the instructions here in the explainer text that I've included.

49:53 Speaker 2: So we're going to download eight day average GOSIF from March through October of twenty eighteen, which will function as our control year, and twenty nineteen and twenty nineteen, as a reminder, is the flood year. In step two, we'll use the conversion function that we used before to get all of the data to use the same skull. In step two, we'll use the conversion function that we used in part one to convert all of the data to the same color scale from zero to zero point eight watsper meter squared stradium micrometer so that the data can be compared visually.

50:35 Speaker 2: Will also perform the bounding operation that I showed you to make the data only appear for the US Midwest, which is our region of interest. Will compile the map data into an animation, which is a helpful way of visualizing this and getting an intuitive, qualitative understanding of the event. Lastly, for a more quantitative approach, we'll be creating a time series to see the year of a year difference in sif within the region of interest. Okay, so let's take a look at our first cell here that's going to be performing step one as before, which is just downloading the data, and the date range that I've included is from day of year seventy three to day of year two hundred ninety eight.

51:29 Speaker 2: These two dates weren't chosen for any particular reason other than that they cover the growing season with a little extra and here you can see our function working. This will take a couple of minutes. Okay, so now our code cell has completed and we've downloaded all of the data over the growing season for twenty eighteen and twenty nineteen. Before we continue with the analysis, I want to mention that although looking at SIF is not comprehensive, if you feel confident about this material so far, you can also incorporate the estimated GOSIF GPP product that I mentioned and the XCO two product from the OCO two and OCO three missions similar to the SIF product.

52:55 Speaker 2: If you've been following along from part one, this would represent a more comprehensive analysis in line with the study. If we turn over to the study, I can show you a little more detail on what I mean by that.

53:13 Speaker 2: I touched on this in the presentation, but i'll show you the study directly now. In particular, I want to point out this plot, which not only compared SIFF from Tropomy and OCO two, but also considered the planted area to show the delay in crop planting in twenty nineteen and the delay in the peak and SIFF. One more detail to look at is the terrestrial water storage which I also touched on that comes from GRACE follow on data which I won't be covering, and the GPP data. Now, this GPP can be reproduced by downloading the GPP data from GOSIF and you can see how well the results match or maybe don't match, what was found in the study.

54:14 Speaker 2: Additionally, the researcher is considered XCO two. If growth is less in a region, then XCO two would be higher in general, and that is due to less carbon fixation from the plants over that region. So that's another reason to consider XCO two when performing this type of vegetation analysis. Let's return to the notebook. With our SIFT files downloaded, we now need to convert them to produce the animation. I've created a function called create GOOSIF Comparison Animation here which will take the files that we just downloaded, perform the conversion function, and create an animated JIFT file that actually shows the animation itself.

55:12 Speaker 2: So let's run that. Now, this was not actually part of the study itself, but I find it instructive to animate the data over time in a side by side graphic. Okay, so now the cell has run and our animation is displaying down here some sort of description of the I'm going to wait for it to loop back around. While you could look at the granules individually in the map widget that I showed you before, it's easier to compare them in this animated view in my opinion, So let's take a look at what's happening right now.

56:02 Speaker 2: You can see that SIF is peaking over here on the left in twenty eighteen, while it remains relatively muted over here on the right. Subsequently, later in the growing season, SIFF reaches a somewhat lower peak in August and September, whereas SIF is relatively low over on the left in our control year. You can play through this animation multiple times. If you missed it the first time, you might notice new things each time that it plays through.

56:35 Speaker 1: Well.

56:35 Speaker 2: This is useful for getting a rough qualitative sense of the scale of the event. It's helpful for us to also include a quantitative analysis using a time series to actually see the difference in peak over time in this region. Another thing that you'll note about this graphic is that there's a red polygon over the mid West, and this polygon identifies the rough shape of the corn Belt region. I didn't use any specific heuristics for creating this polygon. It was just based on an interpretation of where cornfields roughly lie in the US Midwest, so it's not comprehensive.

57:22 Speaker 2: In general, you might include aggregating all SIFT data within a given state, However that might include some wild regions within your analysis. An alternative would be to use something like the USDA's Cropland Coverage product to specifically filter to data that is only cropland. But that is a little bit more complicated than what we'll be doing in this analysis. So for creating our time series, we'll actually be averaging all of the sift values with in this polygon, and the percentage of planted regions within this polygon is quite high, so we should mostly be seeing a cropland signal.

58:16 Speaker 2: If you're curious about the data points or the spatial points in this polygon itself, that is in the corn Belt dot GJSON ancillary file over here. So I want to point out one last thing before moving on from this animation and looking at our quantitative analysis stick a look at this region down here in Missouri and a little bit of Arkansas below the red polygon. This corresponds to the Ozarks region, which is mostly deciduous broadly forest, and unlike planted cropland, the wild regions like this experienced a lower impact on SIF and actually had a slight uptick in SIF activity during the flood year.

59:11 Speaker 2: This area was more resilient due to the established root structures and the selection pressure on species to be more adapted to flooding. For instance, in floodplains of rivers you might see plants like cattails which are more resilient to flooding. As I said in the yin ad Alia paper, the researchers noted an increase in productivity of zero point zero four pedagrams of carbon in twenty nineteen compared to twenty eighteen across all land types in the seventeen states included in the study, So the effect of wildland regions on the overall average across all seventeen states actually resulted in a slight increase in primary productivity.

59:58 Speaker 2: On the other hand, and the cropland regions experienced a zero point two to one pedagram of carbon drop in productivity in June and July in the flood year, with the corresponding increase of zero point one four pedagrams of carbon in August in September due to the later peak. And remember we saw that peak in August in September in our animation where that dropped over here. So just doing the simple math, that resulted in a net loss of zero point zero seven pedagrams of carbon overall in the flood year over cropland regions specifically, Let's see if we can see that same trend in the quantitative analysis.

1:00:49 Speaker 2: So let's take a look at the cell that we have down here. Again, we're using the corn Belt dot gojson ancillary file, which is red polygon that you saw in the animation, and we'll be looping over all of the geotyph files which contained the raw data itself, and taking the zonal average over the region excluding any no data values. For each of those spatial averages that we acquire, we'll be able to create a time series one line for twenty eighteen, which is our control year, and one line for twenty nineteen, the flood year.

1:01:37 Speaker 2: Blistic Look at that plot, all right, here we go Another thing that I've done to help with our visualization is that to make the plot a little less busy, I've only included data values over the specific region where plants were growing most vigorously during the growing season. The researchers found that the reduction in SIF as you can see in the red line here for the flood year, represented about a fifteen percent reduction in crop yields, and that's huge. It represents perhaps billions of dollars in agricultural losses.

1:02:19 Speaker 2: Doing some quick math, we can see that our twenty eighteen peak in this case of zero point five eight lots per meter squared scradian micrometers around mid or early July of twenty eighteen is about thirteen point seven percent higher than the similar peak of zero point five to one or so in twenty nineteen. So even just using a roughly drawn polygon rather than the more rigorous approach of filtering by crop cover, we get a result that is fairly close to what we see in the study. On the other hand, and the time series also shows that later rise in productivity later in the season, whereas in twenty eighteen we see a steady drop off towards the end of the growing season.

1:03:15 Speaker 2: This later peak in the twenty nineteen flood year corresponds with the several weeks later planting date that farmers used to mitigate some of the effects of the flooding. As I mentioned before, if you perform this analysis again for some of the natural vegetation outside of the crop areas, you might see that some regions actually did better in twenty nineteen because of the extra water stored in the soil. Nature can be resilient in ways that some agricultural systems aren't. So that covers the primary portion of our twenty nine nineteen analysis.

1:04:01 Speaker 2: But I want to cover another case study of the twenty twelve drought which occurred in the corn belt, and the motivation for that is apparent if you look at the yield plot for corn and soy in the United States. So let's turn over to that, and I touched on these graphs briefly during the presentation as well. So looking at our corn yield data, we can see that in twenty nineteen, while there was certainly a drop and yield that's noticeable, it's only about five percent. And one thing we can notice about that is that even though we saw a fifteen percent drop in the peak sif that didn't result in a perfect fifteen percent drop in yield in the crop, so the relationship is not quite that simple in this case.

1:04:59 Speaker 2: The motivation for looking at twenty twelve, as I said, is that there's quite a conspicuous dip in yield in that year, and this was due to a lack of precipitation and extreme heat in twenty twelve over the US Midwest. If we look at the soy yield charts, we can also see that while twenty nineteen saw a drop in yields, it's not quite as severe as the drop that we saw in twenty twelve. Another thing that you may notice about this plot if you're not familiar with agriculture is that there's a consistent upward trend line, and this is due to changing agricultural practices and improvements in hybrids year over year that enable increasing yields.

1:05:56 Speaker 2: Because of that, when performing analysis of agricultural regions, it's important to choose control years that are consecutive years if possible, because of the increasing yield over time will also show up as increasing primary productivity and sif over time. Let's go back to our notebook and change our analysis to look at the years twenty twelve and twenty thirteen. The reason why I won't be using the year before of twenty eleven is because that year also saw a moderate drought and a drop in productivity in yield.

1:06:38 Speaker 2: So I'll go back up to the beginning of section three, and here we can see two variables, year A and year B, and instead of twenty eighteen and twenty nineteen, I'll instead put in twenty twelve and twenty thirteen, and we'll use the same time range to cover the growing season in different parts of the world. You may need to select different dates or when studying different plots with different planting times, especially if the crop is a perennial or planted in a different season of the year. So now we can see that we're downloading the twenty twelve and twenty thirteen growing season data.

1:07:31 Speaker 2: I find this example especially interesting in contrast to the twenty nineteen case study, because we're going to see sift crash for a completely different reason, essentially that there was no water instead of too much water. Let's run the animation cell again and take a look at what we can notice qualitatively about the data. In this case, the left side plot is going to be our anomalist year with twenty thirteen as our control. As we can see, the sift peak is quite low in the anomalous year compared with twenty thirteen.

1:08:31 Speaker 2: Even the unmanaged land in the ozarks down here, like we mentioned in twenty nineteen, is relatively low in twenty nineteen as compared with twenty thirteen. Although there was an early peak. There are a patchy regions here, suggesting that there was some difficulty in the growing region down here in the temperate forest. The amazing thing about GOSIF is that we can do this analysis and have confidence in what the data is telling us, even though both of these years were before the launch of OCO two.

1:09:10 Speaker 2: In the first place, you'll notice OCO two launched in twenty fourteen, and both of these years are before that launched. Eight. Lastly, let's examine the time series plot of these two years. So one thing I'll point out is that the labels will not render very well here since I've tuned it for the previous example. But you can hopefully still see what I'm talking about in this plot. Even though twenty thirteen lagged somewhat in the early season compared with twenty twelve, which could be for a variety of reasons.

1:09:49 Speaker 2: The big story we should see in this plot is just how muted the peak and siff was was in twenty twelve. It was more like a plateau at levels that we would normally see in the earth growing season. This much more dramatic crash in production resulted in a direct crash in yields, as we saw in the yield curve over here for corn and soy. I don't want to dive too deeply into the economics of this, but if you look at corn prices in that year, you'll notice that they also spiked, no doubt driven by the scarcity of new corn supply from that year's poor harvest.

1:10:28 Speaker 2: At any rate, I hope that these case studies have really illustrated the power of sift data and how we can realize its full potential by leveraging gap filling techniques. Whether it's through floods, droughts, heat waves, or even things like volcanic eruptions affecting sunlight. Sift can capture all of this information because it's measuring photosynthesis.

1:10:54 Speaker 2: Even though our analysis of these two case studies has been far from comprehensive should give you an overview of how gap filled sift data can be employed if you want to take it a step further on your own. One homework challenge to consider is this, and this isn't an official homework, how would you validate the top line gross primary productivity figures presented in the yin at Alia paper abstract. Remember that there is a GOSIF GPP product that I mentioned earlier. So if we took that data added spatially averaged GPP values similar to the way that we spatially averaged the sift values for each time increment over the corn belt and over the growing season, we compared the sums of those values between twenty eighteen and twenty nineteen.

1:11:43 Speaker 2: What might we notice about those sums? While the difference in the two sums probably won't line up exactly with the figures from the paper due to differences in the spatial aggregation technique, do you expect the percent difference should be comparable? Try this out your own if you're curious and want to go even further with this exercise. So that concludes the code demonstration portion of this part, and thank you for your attention.

1:12:15 Speaker 1: Thank you very much Jackie for that great presentation and demonstration. So now what we'll do is a summary of our session today of the salient points. And here's a summary of today's session. SIFF is an excellent metric for assessing vegetation health and stress response, as we saw with the demonstration on the case studies of the impacts of floods and droughts. SIF can be measured from several satellite platforms, each with its own trade offs. OCO and OCO three provide high quality SIF measurements, but they have limited spatial coverage and a relatively short observational record up to eleven years.

1:13:04 Speaker 1: Gap filling or data fusion, uses data driven models to predict SIF values and extend spatial and temporal coverage. And GOSIF is a gap filled product, so it's a gap filled SIF product. It's a globally continuous SIFT product with zero point zero five degree spatial resolution that's about a six kilometer spatial resolution and an eight day, monthly and annual temporal resolutions extending back to the year two thousand and Finally, floods and droughts cost vegetation stress, which reduces SIF.

1:13:47 Speaker 1: And here's a contact information for anyone that has more questions about the material that was presented today, you can reach out to Jackie Ryan or to Karen Ewan. And with that we reach the end of not only today's session, but this webinar series. I like to thank all of the participants and all of our invited experts, and we are now ready to start the question and answer session.

1:14:22 Speaker 1: The first question, I want to use SIFT data to analyze forest health and want to check and I like to check it relative to air pollutants and their impacts on forest. I like to have some suggestions related to this.

1:14:39 Speaker 3: Yes sometimes say is yes, so you can use as IF to analyze the forest health and check the impact of air pollutants on forest. So there are many studies that using SIEF to assess the impact of climate anomalies on forest health, such as that what Jackie Show today is on Cropland there's many studies using SAFE to look at the impact of drought, heat stress on the forest. But you can also use the SAFE to look at air pollutants. But the challenge is really to how to isolate the impact of air pollutants from other factors so that such as heat and drought because they could happen at the same time.

1:15:21 Speaker 3: In the document, I put the example of using safe to invest the impact of air pollutants on our health and I will put more references on this document. You can check it out later.

1:15:36 Speaker 1: Wonderful, Thank you very much. The next question number two, I get this error. Does anybody know why permission air access denied?

1:15:50 Speaker 2: Yeah, so that was hopefully just due to having the folder opening Windows Explore. I'm not sure if that fixed that particular users problem. I will note that Windows is a bit less comprehensively tested than mac os, so if you are having an issue or I can look into it further.

1:16:16 Speaker 1: Yes, please feel free to email Jackie contactor directly if you continue having issues. The next question number three, I want to know the difference between the SAMs and target acquisitions acquisition modes of OCO three. Can both modes be collected together for a site, for example, a flux tower site.

1:16:41 Speaker 3: Yes, so these two modes cannot be collected at the same time, but they can be collected at different times but over the same site. So basically you can collect like today using SAM, but not that they use a target and sam's covers much broader area. So it collects observations over about eighty kilometers by eighty kilometer area within two minutes, where target mode covers a much smaller areas about twenty five by twenty five kilometers, so it's much smaller than SAM and it's also primarily used for validation purposes, so not those target observations normally over teacon sites which are surface remottencing that are used for validation of the satellites.

1:17:27 Speaker 1: Great, let's move on then to the next question number four. Swift data has higher values in July August of the year. What factors drive these values?

1:17:40 Speaker 3: So, as a Jackie said in the in the training in the core in the in the course, the SIEF is a really indicator of plant productivity while plants growth need light, energy and water. So these quantities are higher in July and August in Northern hamshere. So that's the reason that SIEF has higher values in July and August in the North hamsphere. In the South Hampshire is a different case. So in the South hamsphere is the opposite. So CIF would have higher values in January February of the year when the when you have the highest light availability and also the temperature is warmer.

1:18:27 Speaker 1: Very good, thank you.

1:18:28 Speaker 2: There's a new question in the chat, what are those specific differences between NDBI and SITH values. I can try to answer that or maybe Ginger, do you want to answer that.

1:18:44 Speaker 3: Jackie, please give a try. Yeah, please go ahead, Okay.

1:18:47 Speaker 2: So, as I mentioned in the presentation, and DBI is primarily a measure of the color of plants. It's created by taking a ratio of specific reflectance bands from you can use a variety of sensors. It's kind of agnostic. SIF is a very carefully estimated measurement that comes from high resolution spectrometers, and it's more directly tied to the physiological response of the plant. Because of that, you get a much faster response of SIFF rising and falling in response to changes in photosynthesis activity in the plant, as compared with NDVII, which is considered a lagging indicator.

1:19:47 Speaker 2: So, for instance, if you had a prolonged drought where plants stopped photosynthesizing, you might be able to observe that up to a week before you would see a similar fall off in and DBI.

1:20:02 Speaker 1: Great, Thank you very much, Jackie and John GI. If you'd like to add anything else. Please feel free.

1:20:10 Speaker 3: Now I think Jackie clinging really well.

1:20:12 Speaker 1: Excellent, wonderful, thank you. Okay, So we have another question coming in. Are there plans to expansive coverage via additional satellites?

1:20:23 Speaker 3: I can, I can start, you can add. So there is another satellite called FLAX which will be launched later of this year early next year. They the flag satellite will is the will be launched by Italy Italian Space Agency. That's just main variable, mainly observable is STEFF. So that would increase observation coverage of SETH and other satellites I mean may not have SEF as primary observables, but SEF could still be retrieved from those satellites such as Temple, which our team members are working to to retrisive from Temple.

1:21:08 Speaker 3: Temple is a geostatious satellite over North America. If the team member could be successfully retrisive from the satellite, that could be dramatically increasing the observation coverage of SIF over North America. So those are two cell two sllies I would I know that could potitionally increase the SIF observation coverage. Jackie, do you have anything to add.

1:21:37 Speaker 2: I'll also note that improvements in gap filling is an area of active research. So there are many groups right now, including the ones that I mentioned in the presentation, that are trying to take the observational record that we do have and estimate SIF using other more highly available instruments. And actually this was focus of some research that I did earlier this year that is as yet unpublished, to take data from the Ghost satellites, which are geostationary weather satellites operated by Noah and estimate SIF by training a model with OCO three and the Ghost data.

1:22:32 Speaker 1: Great, thank you both. All right, if you have additional questions as you go through the material, please feel free to get in touch with any of the instructors throughout this training series Dr Yunjilu, Doctor Nick Parazu, Jackie Karen, or any of the r SET team members. So with that, then we reached the end end of this training series. I think it was a great training. I'm really excited to see so much interest in SIF and having a large community be interested in these sorts of data sets. So I first of all like to thank all of the participants again for your interests, for your enthusiasm and this topic and for all the great questions that have been coming in.

1:23:24 Speaker 1: I'd also, of course, I'd like to thank the r SET team, Brock Blevins, Jonathan O'Brien, Sarah Kutshaw, Sue Monty, Salwyn Hudson, Odoi, Maria Marabido, Melanie Foye Cook, Natasha Johnson Griffith, and others from the r SET team that help put this together and make these trainings possible. And finally, I would like to thank our the SIFT team who did an incredible job in putting all of this material and demonstrations together, especially today Speak or Jackie Ryan who's been phenomenal in presenting not only the theoretical part but also on the very instructive demos.

1:24:10 Speaker 1: So I'd like to give the opportunity to Jackie for any final words before we close.

1:24:16 Speaker 2: Go ahead, Jackie, Yeah, just thank you for attending and for your attention. And the last thing I'll mention is that we will be posting the homework hopefully this week. The only required portion is the eleven multiple choice questions for receiving their certificates. However, I also wrote three optional exercises if you want to pursue any of the analysis that we did in these three code demonstrations further on your own, and those exercises focus on ext what we looked at to view different areas of the world, such as the sugar growing region in San Paulo region of Brazil.

1:25:09 Speaker 2: So yeah, thank you.

1:25:12 Speaker 1: Super thank you Jackie and doctor Leu. Would you like to say any words before we close.

1:25:19 Speaker 3: I just want to thank the whole team for putting this together. This is fantastic, and also thankful the participants from all around the world for your interest and for your active participation. And feel free to reach out to us if you have any questions during our research and go luck with everything. Thank you, thank you very much.

1:25:41 Speaker 1: And finally, Karen Ewan, who's been instrumental in really getting this training or making this training possible. So Karen, would you like to say any words before we close?

1:25:57 Speaker 4: Thank you to all the participants for joining us for today and the whole series. And I do know that there is a survey that our set does provide.

1:26:08 Speaker 5: Please fill that out. This is how we are able to know what to put together and provide to all the participants because your feedback. We really do look at that and we work with great people like Jackie and Gingy to be able to put together these sessions for you. So we look forward to your feedback. Thank you again. We have a great team, and thank you to our set team. Thank you to Gingy and Jackie. It's really an honor to work with you all.

1:26:42 Speaker 1: Thank you very much, Karen. So yes, I just wanted to stress what Karen said that there is a survey that will be sent out. We ask you to please fill out the survey. It is important to us and to the OCO mission right to understand what you're interested in, how you're finding the use of the data, what you would like to see in the future, what are some of the challenges that you're having. I mean, all of that is really really helpful feedback for us and for our set and for the OCO team. So please fill out the survey.

1:27:21 Speaker 1: As mentioned, the homework will be posted on the training web page today. The do date is November twelve, I believe, and the certificate of completion will be provided to those participants that attended all three live sessions and complete the homework by the DO day. So with that, I hope to see you all in the future. Again. We hope to have another SIFT training sometime in the future. Maybe the next step something a little more building on what was presented this time around, so please let us know what you're interested in.

1:28:03 Speaker 1: Thanks everyone, and wishing you all a great day.