Introduction to Flash Drought Monitoring using SIF and Soil Moisture

0:00 Speaker 1: Point out that CO2 data is measured over land and ocean, while SIF is only measured over land. This is because OCO2 relies on reflected sunlight, and the oceans absorb too strongly in the wavelengths where SIF occurs. A few key features to note in the SIF data. First, dark green values depict regions with the highest photosynthetic activity, which includes tropical rainforests, temperate forests, and agricultural areas. In fact, agricultural areas have consistently shown up as the strongest SIP signal on the planet since SIP's record keeping began.

0:36 Speaker 1: The animation shows how plants wake up from winter dormancy at different times across northern latitudes, starting around March or April in mid-latitudes and May to June in high northern boreal and tundra systems. You may have noticed the spatial and temporal gaps in the OCO2 data. As noted with SMAP earlier, this lack of wall-to-wall coverage makes it difficult for users to apply the remote sensing data to their application needs. All satellites, including OCO-2, have to make some type of trade-off in instrument performance, for example, by sacrificing extensive spatial coverage for high spatial resolution or extensive spectral coverage for high spectral resolution.

1:17 Speaker 1: Researchers have leveraged OCO's small footprint and traceability to ground data alongside complimentary Earth observation data sets to bridge the gap between local OCO measurements and broader regional applications. Neural networks trained on OCO-2 sifts and vegetation data from other sensors now produce seamless maps that preserve spatial patterns while filling gaps. Such maps provide meaningful spatial resolution, visual clarity, and traceability to land management decisions. This map derived from NASA OCO-2 and MODIS data shows the intense glow of SIF in the upper Midwest U.S., reflecting the high productivity of crops and soybeans in the Corn Belt region.

2:04 Speaker 1: Farmers, along with foresters and other resource managers, need early information on drought development. Here we see maps of anomalies in SIF on the left and maps of drought from the U.S. Drought Monitor on the right for the same period of June in the year of 2021. These maps are produced completely independently from each other, yet there is very strong correspondence between regions with low SIF in red and the presence of drought, also in red. Growing evidence such as this and in the peer-reviewed literature is building confidence that OCO2-based SIF products can offer additional early warning information on flash drought of value to agencies such as the Drought Monitor.

2:44 Speaker 1: Such information can help producers and managers prepare for worsening conditions in agricultural, rangeland, and forested regions. Together, OCO2 and SMAP offer early warning and near-real-time tracking of flash drought. But how is this possible? This figure shows two observational time series, OCO2 SIF anomalies in green and SMAP soil moisture anomalies in orange. The time axis shows the number of weeks relative to the onset of flash drought. This time series represents a composite of thousands of flash drought events across the U.S. over a two-decade period from 2000 to 2020.

3:22 Speaker 1: The pattern shown here is remarkably consistent across different years, land cover types, and geographical regions. OCO2 SIFT data show that early on, unusual spring warmth and low rainfall prompt a plant growth spurt. This enhances the glow from growing plants while also depleting soils of water to support plant growth. This triggers a series of events in the soil and atmosphere that in many cases lead to gradual and eventually rapid and severe soil water loss. This gradual and rapid loss of soil water and development of flash drought conditions is tracked perfectly by SMAP.

4:02 Speaker 1: If soil water is not replenished by rain or mitigated by cool weather, land surface feedbacks can amplify and or prolong drought. As such, SIPP provides a critical and unique early warning signature for flash drought up to three months ahead of onset, much sooner than other remote sensing indicators and over a larger area than possible from boots on the ground. The combination with SMAP can support tracking and flash drought development and provide critical information to decision makers. The past few years have seen the development of a number of SIFT-based indices for flash drought.

4:36 Speaker 1: I'll provide a brief summary here of two of these indices. This figure depicts an idealized drought in its four main stages. On the left in green, we have pre-drought in timescales of months. Moving right in orange, we have drought development occurring over days to weeks. And we have the primary drought period occurring for weeks to months. And finally, drought recovery, which can take months to years. The dashed line represents a typical seasonal cycle of SIF, increasing in spring, peaking in summer, and declining during fall.

5:09 Speaker 1: The bars show SIF during a drought year with values exceeding the dashed line representing higher than usual growth and values below the dashed line representing lower than usual growth. The first drought index is the Z-score representing a standardized anomaly from the long-term mean. This was the index used in the previous slide and shows positive growth anomalies early on before drought onset. followed by negative or slower growth anomalies during the drought period. Another index is called the rapid change index.

5:44 Speaker 1: This shows the rate of change relative to preceding weeks and climatology. This shows slower rates of growth early on, despite the positive growth anomalies above, and in particular, a deceleration of growth rate leading up to drought onset. Higher rates of growth can be seen later on during drought recovery. The rapid change index is a very convenient and objective index for operational monitoring of flash drought and will be the index used in the flash drought demos to follow. We conclude with a summary of which drought indices to use.

6:19 Speaker 1: Traditionally, drought was seen primarily as a climatological phenomenon where low precipitation and higher than normal temperature were among the assessed metrics. SPI, PDSI, and SPEI continue to be powerful indices to monitor changes in forcing associated with meteorological drought. Agricultural droughts are, however, assessed in terms of their impacts on cropland above any other factors. This requires metrics more aligned with changes in the land surface, including soil moisture for hydrological drought, combinations of soil moisture and SIFs for agricultural and ecological drought, and combinations of all indices for flash drought and operational decision support.

7:01 Speaker 1: The USDM remains the gold standard for combining information from multiple sources, which brings us to our final takeaway. More information is always better for flash drought monitoring. No single process is responsible for development of flash drought, and therefore no single index is sufficient. Combining multiple drought indices and variables offers the best chance for drought detection and attribution. Here are some additional resources, including visualization tools and data hubs. Thanks for listening, and now I will pass things along to Jackie.

7:41 Speaker 2: Thank you, Nick. And now we're going to turn over to the code demonstration portion of part one. My name is Jacqueline Ryan. I'll be the RSET instructor for this portion of part one. And today we're going to be talking about flash drought indicators in our Jupiter notebook. And this will be the demo outline for the full course. So in session one, which we'll be doing today, we'll be going over flash drought indicators, as I said, starting with the solar-induced chlorophyll fluorescence data called SIF and the SIF rapid change index.

8:20 Speaker 2: And then in the second session of the Jupiter notebook, We'll be talking about soil moisture and the Soil Water Deficit Index, or SWDI. Looking ahead to session two, which will be the following week, we're going to talk about flash drought detection and looking at a case study of the 2023 to 2024 summer flash drought in South Africa. And in particular, in the second section, we'll be applying detection rules and then investigating the spatial distribution of flash drought. So for session one, the demo has a number of objectives.

9:03 Speaker 2: By the end of the session, you should be able to recognize how SIF data can be used as a drought indicator, retrieve a subset of GoSIF data, and use it to derive a time series of SIF-RCI over a selected region, and we'll also be retrieving a subset of map level 4 soil moisture data and deriving a time series of SWDI over a selected region. We'll investigate how the SIF-RCI and SWDI time series can be used to study the development of the 2017 Northern Great Plains flash drought, and then we'll compare the advantages and limitations of using SIF and soil moisture to study drought compared with meteorological data from related research.

9:51 Speaker 2: So we've already gone over this in the prerequisites of the course, which I encourage you to read if you haven't already done so. We won't be going through any of the specific setup states to get the Jupyter Notebook open. But to get started with the course code material, you'll be able to find it on GitHub at this address or using this QR code. And I'm just going to open that up in my browser right now. So here we can see the code materials for this course. And there's going to be a helpful read me here as well to guide you through the instructions about what we have in the course, as well as how to install it on your computer.

10:39 Speaker 2: And this will be a set of course materials that runs locally on your computer. And we did that because there's a number of data sets that you need to retrieve as part of this course and we wanted those files to persist on your computer so that you can run the course code later and even on your own regions of interest. And so if you prefer to follow along on your own, these installation instructions should be able to help you get set up. One of the first things we're going to do as a prerequisite is that you'll need to download Python if you don't already have it or if you prefer to use Anaconda, you wanna make sure that you have Anaconda installed on your computer.

11:28 Speaker 2: The course code will work with either of these forms of Python, but you will want to have some form of Python on your computer. So I've already gone ahead and done that for my machine. So we'll skip ahead to step two. And in this step, we're going to be downloading the course materials What you want to do is to go to the code button at the top of this page, this green button right here, and you'll click this menu and you click download zip. And that will actually download the course materials into your downloads directory or wherever you have it set up.

12:11 Speaker 2: And this zip file, you'll want to unpack. into wherever you wanna keep the course materials. I've already gone ahead and done that. So I won't show you how to do that on your computer, but you'll want to do that on your own. The next thing you'll want to do as a prerequisite is to make sure that you have a NASA Earth Data account. And we'll need this account to download any NASA data that we're going to be using throughout the course. So I already have one of those. And I'm going to go to this webpage and that's urs.earthdata.nasa.gov.

12:52 Speaker 2: We have a link in the README as well. And I already have an account, so I'll just log in. But if you don't have an account yet, you want to click this register for a profile button over here on the right. And I'm just going to log in and this will tell me some information about my account.

13:14 Speaker 2: So the next step in our installation process is going to be a little more technical, but I'm hopeful that you'll all be able to follow along with the instructions that I've provided in the readme here. And what we're going to want to do is open the terminal application if you're on macOS or the PowerShell application if you're on Windows. And you're wanting to navigate to the directory where you unpacked that zip that we just downloaded from GitHub. So I'm gonna open the terminal application because I'm on macOS.

13:51 Speaker 2: And I start off in the home directory. And so in my case, I installed these files into the rsit, flash trial rsit directory. And so I'm gonna navigate there now. And then the next thing we'll want to do is run the setup script associated with the specific environment that we're running in. So that's going to depend on whether you're using Anaconda or regular Python and whether you're running on Windows or macOS. and the instructions are included here in the README in step five. So in my case, I'm using regular Python on macOS, and I'm going to run the setup.sh script that's been included with the course.

14:42 Speaker 2: So I'll go ahead and run that now. And that's going to install the dependencies for the course into a virtual environment, and then load up the Jupyter Notebooks in my web browser. And don't worry about the course installing dependencies into your global environment. It won't do that. It will just do it into a local environment that's only used for this course. And so I've already installed all of the dependencies associated with the course. So it's just going to open right up in my web browser for me.

15:16 Speaker 2: And now we can actually get into the content for the course. So in this part, Part one, we're going to be talking about flash drought indicators. And we'll be looking at two main remote sensing data sets. And we'll be using these data sets, not directly, but processing them into a form that will be helpful for us to detect and predict flash droughts. And both of these datasets are global in terms of their spatial coverage, and they also have a relatively long observational history, so we can study a lot of different retrospective cases all around the world.

16:02 Speaker 2: The first dataset that we'll be looking at is one for our solar-induced fluorescence, and this is the GOSIF dataset produced by University of New Hampshire. The second dataset we'll be looking at in the second section is a root zone soil moisture model data set that comes from the SMAP L4 global three-hourly nine kilometer.

16:26 Speaker 1: And We already.

16:30 Speaker 2: Discussed a lot of the details of these datasets and the way that they've been derived in the lecture portion of this course. So it's assumed that you have that context when we go into this course or when we go into this code material. So why are we specifically using these datasets? Because there's many ways of looking at drought, but in particular, we're using an approach that relies on SIF and soil moisture. And this comes from an impact-centric framework developed by Koshan Mohammadi and Guilin Wang at the University of Connecticut.

17:11 Speaker 2: Traditionally, when we looked at drought, we were looking at climatological variables such as precipitation or temperature. And we might define drought as a deficit in precipitation and higher than normal temperatures. But agricultural droughts are assessed in terms of their impact on cropland above any other factors. And so in order to determine whether precipitation deficit actually matters to the crop growth depends on the crop type, the existing soil moisture, and the climate of the growing region being studied.

17:52 Speaker 2: So what we're going to try to show in this course is that it's more effective to use land process-based metrics like SIF and soil moisture to detect flash drought than it is to use our more traditional measures like temperature and precipitation. So let's go to Section 1. And in this section, we'll first look at our solar-induced chlorophyll fluorescence metric, or SIF. and then we'll be converting it into a drought indicator known as the SIF Rapid Change Index or SIF-RCI. So we already talked about SIF during the presentation, but to reiterate, SIF data can be thought of as a measure of the photosynthetic activity of plants in aggregate.

18:41 Speaker 2: And the RSA program has already produced several courses on this data if you'd like to learn more about this measurement. In particular, last year, there was a course about using SIF observations for assessing vegetation changes related to floods, drought, and fire impacts. As most of the instruments capable of measuring SIF from space are limited in terms of their spatial coverage, for instance, the swath width, and the length of their observational records, we're going to rely on a gap-filled data set called GoSIF produced by Xing Li and Jingfeng Xiao at the University of New Hampshire.

19:20 Speaker 2: By gap-filling the data, GoSIF is able to combine OCO-2 observations with a secondary source like Terran Aqua MODIS to both fill in the spatial gaps and allow for the SIF measurement to be extrapolated past its observational record in the OCO-2 mission. So OCO-2 only goes back to 2014, but we actually have GoSIF data that covers from 2000 to the present. And Mohammadi and Wang used the GOM2 and contiguous SIF or C-SIF gap-filled products in their own research as data sources. But GoSIF is publicly available for free, and it offers the same eight-day temporal resolution that was used in these papers.

20:11 Speaker 2: And so it's just a convenient data source for CIF that we can use in this context. In the first code cell below, we're going to download some Go CIF data from the UNH repository and unpack it into the GeoTIFF format. So I'm gonna go ahead and do that now. And to save time, I've already downloaded this data. And what will happen is that when we rerun the cell, It won't download anything if you've already downloaded the data. And if I look at the data directory over here, I'll see that there's another directory called gosif, which actually contains the files that were just downloaded.

20:56 Speaker 2: In particular, we downloaded from 2017, January 1st through to about October. and that's the amount of data that we'll need to use for the case study that we're looking at today. I've actually downloaded a lot more data than that, but you don't need to download all of this data onto your computer for this course. So, since these are GeoTIFFs, the first thing we're gonna do is actually just look at the data in GIS software so that you can understand what we're dealing with. And this will give us a helpful intuition about the measurement before we process it any further.

21:40 Speaker 2: You'll notice that the file names over here have the date included in their file name, and that would be the year and then the day of year after this GOSIF name. So let's actually open up GIS software now, and I prefer to use QGIS, so I've opened that over here. And this is just what it looks like when I open a completely new window of QGIS. So I'm going to start a new project that's completely blank. And I'm just going to switch to adding a base map so that we have some context before we load in our geotests.

22:27 Speaker 2: And so I already have the Quick Map Services plugin for CJIS, which will enable me to load a base map. And I'm going to pick the OSM standard. And then I'm going to switch the projection over to EPSG 4326, or equirectangular, which is the default projection of the GoSIF data that we'll be working with. Next, I'm going to go to where I downloaded the data, which will be in the directory that I have the course materials, and then the notebooks directory, and then the data go SIF directory that I showed you in the browser window.

23:13 Speaker 2: And in particular, going back over here, we're going to load this day of year 97 GeoTIFF for 2017 that we just downloaded. So I'm going to scroll down until I find that. and I'll load it over here. And when we load it in by default, the data will be in gray scale and we'll also have our no data values mapped at full opacity. So the next thing we'll want to do is open the layers panel and go to the transparency tab under the properties for this particular layer. So here are the layer properties, and I'm going to go to transparency, and I'm going to just add in the no data values for this particular layer, which are 32766 and 32767.

24:11 Speaker 2: And you can see that included over here in our instructions. Okay, so I'm gonna go back over here Whoops. And I'll click apply. And so there's two no data values. One is for water and the other is for ice. And now we'll see those are being mapped to transparency. So we're only looking at the data. Next, we'll want to apply a color ramp to this data. So we'll need to go back into our properties and go to the symbology tab. And we'll want to pick single band pseudo color for our render type. Band one is fine.

25:13 Speaker 2: We'll go to choosing our color ramp. and I'm going to choose the yellow green or YLGN color ramp over here. And then the last thing I want to do before clicking apply is to change the min and max values over here so that we have a greater dynamic range to look at the data. So these values unlike a reflectance index such as NDVI, have a unit, which is watts per meter squared steradian micrometer, which is basically a spectral irradiance. And so by setting the min and max values here, we're actually setting it to the true or a scaled version of the true units that we're looking at.

26:06 Speaker 2: And I'll want to pick between minus 500 and 4,000 over here.

26:15 Speaker 2: And this corresponds to values of negative 0.05 to 0.4 watts per meter squared straight in micrometers based on the scaling factor of 10,000 for this data. And now we can see a nice dynamic range across the Earth for the SIF data. And when we zoom in, we can see that for this time of year, which is late spring in the United States, there's a nice gradation where deciduous broadleaf forests, for instance, in the southeastern United States, show high levels of SIF marked by this dark green color. And then more arid regions, such as the Sonoran Desert in the western U.S., show lower levels of SIF, which is what we would expect.

27:09 Speaker 2: We'll be able to use this SIF data to assess the photosynthetic activity anywhere in the world, but in particular, our case study will be looking at the Northern Great Plains region of the United States and Canada. So turning back over to our Jupyter notebooks, as you saw in my screen share, if you followed the steps correctly, The data should look like this in your GIS window. So unlike other vegetation indexes like NDVI that rely on reflectance, the SIF values are an estimate of the average spectral radiance of plants due to chlorophyll fluorescence.

27:51 Speaker 2: And therefore, it's sensitive to photosynthetic yields. One of the key advantages of this is that the metric should not saturate under dense canopy conditions or high levels of photosynthesis.

28:11 Speaker 2: So even though SIF and NDVI both vary according to the season and vegetation class, SIF responds more linearly to changes in absorbed photosynthetically active radiation, or APAR. Hopefully, we'll be able to detect this and use this to our advantage in studying the rapid onset of events such as flash drought. So flash droughts are characterized, like I said, by their rapid onset. And so it's often the case that they have already begun by the time that plants turn brown. Reflectance metrics are essentially a measure of the plant's color And so, if we're just relying on color to determine vegetation stress, it might be too late.

29:03 Speaker 2: SIP itself is estimated by detecting a faint signal in nanometer-wide Fraunhofer lines that can only be resolved by very spectrally sensitive spectrometers, such as the ones on OCO-2. So even though we use an instrument like MODIS to extrapolate the data, we're relying on observations from very specific and small number of spectrometers to actually get the SIF signal itself. Well, negative values in this data set, as we saw, our theta range goes slightly negative, while these might not seem physically plausible, we occasionally do see negative values for SIF due to the retrieval uncertainty.

29:47 Speaker 2: So now let's go into our Northern Great Plains flash drought case study. And this event has been described extensively across a number of different papers, but the specific technique that we'll be basing our approach off of comes from the Mohammadi and Wong paper that I mentioned earlier. So this area that we're looking at is primarily a semi-arid steppe with grasslands as well as wheat and barley fields, and it also has a few solitary mountain ranges containing mixed needleleaf and broadleaf forests.

30:24 Speaker 2: In the summer of 2017, The flash drought that occurred sparked wildfires and resulted in an estimated $ 2. 6 billion in agricultural losses. So let's use the data that we downloaded to get a sense of the onset of this event. To get the most out of our SIF data's greater sensitivity to changes in vegetation health, we need to look at the measurement in relation to its past behavior. In particular, we want to look at SIF anomaly, which is the difference between the current SIF value and its historical mean.

30:58 Speaker 2: Since GoSIF data has a range that is extrapolated over the record of MODIS, this enables us to build a climatology over the past 20 years or so. To save time, we've already done that. And these data are included in the inputs slash SIF increments and planes directory. So if I go over here, and I go to inputs, you can see that I have a directory called CIF increments and planes. And this has a number of these files called go CIF, de CIF, CLIM, and then they have a day of year associated with them.

31:40 Speaker 2: And so these are actually our climatology values for the period 2001 to 2020. over this specific region of interest that we've identified marked by the coordinates here. And I'm not going to fully go into those GeoTIFFs just yet, but we're going to use them in this next cell here to actually look at SIF anomaly. So using the mean SIF values from these files and the Go SIF values that we just downloaded for each eight-day time step in the dataset, we can look at the SIF anomaly over the region of interest.

32:22 Speaker 2: And we're going to use a map viewer to actually do that. So in the code, we've included a HTML file called sifmapviewer.html. And we're going to load that up now. And this visualization can be viewed inside the Jupyter Notebook itself. But if you want to give it a little bit more room, you can click the link that's included here. and that will open it in a new tab. And so, the first thing that we'll notice is that when we hover over values, over in the top right of the window, we can see the actual value for SIF anomaly.

33:07 Speaker 2: And there's no real max value that SIF can have, unlike NDVI which goes from zero to one. So, Generally speaking, for this particular data set in this context, our SIF anomaly will range from about minus 0.15 to 0.15. And that has the units of watts per meter squared steradian micrometer that I mentioned before. And I've started off our look into this event in May 17th, which is a little bit after the spring planting event. And so this layer is showing the five kilometer spatial resolution that we get from the data.

33:54 Speaker 2: And we've underlaid the nadir BRDF-adjusted reflectance from MODIS underneath. And I chose that layer as the base layer for a reason. For one thing, this NBAR dataset is actually used to estimate the SIF values in the GoSIF model. And for another thing, NBAR is a great way to look at the cloud-free land surface reflectance. And so one thing we can see is that as we flip through dates, this is the estimate of land surface reflectance without the effects of the atmosphere at each of these days. One of the things that NBAR is not so great at is giving us the reflectance over areas that have a lot of snow or ice on them, as well as areas where the MODIS instrument wasn't able to get a lot of clear looks at the land over the eight-day period that's used for temporal aggregation.

35:03 Speaker 2: So we will see some gaps in the data, but the clouds have been removed at least. And this gives us a better sense of actually the underlying land cover that is driving the SIF behavior that we're seeing. And as we reduce our opacity, we can see that the major agriculture regions in our case study region of interest are over here in this part of Montana, as well as over here in the eastern portion, whereas these mountain ranges have primarily our needle leaf and broadleaf forests. Similarly in Canada, most of the farmland in this region is located in the eastern portion of the region of interest.

35:55 Speaker 2: So as we increase our opacity again, we can see that most of the positive anomaly, even early in the season before the blast drought officially began, is concentrated in the wildland regions where we have our forests. And in the grasslands and croplands, we already have neutral SIPP anomaly. So we haven't fully tipped into a negative anomaly yet, but the conditions are vastly different than our wildland ecosystem. And if we go earlier in the season, almost the entire region is showing a positive anomaly.

36:38 Speaker 3: So at this.

36:39 Speaker 2: Point, you know, around the spring planting, early in the season, there wasn't very much of an indication at all that anything was going wrong in this region. But very quickly, towards the end of May and going into June, all of a sudden we see the development of a negative anomaly starting from the eastern portion of the region. And we can see that with these red pixels that are developing over here. And as we go on, our positive anomaly over in the west is getting smaller, and our negative anomaly over here is increasing.

37:18 Speaker 2: And as we go further, we can see that all of a sudden, the entire region essentially has flipped into this negative SIF anomaly, which means that compared to normal, plants are photosynthesizing less. The vegetation is under stress. Going into July, we can see there's SIF anomaly improved somewhat, but it's still quite bad across the entire region. And at this point, we have almost no regions with positive SIF anomaly. And then by August, you can see the negative SIF anomaly continuing. And another interesting thing that we can note happening by this late August time period is these gaps in the end bar are actually caused by smoke due to wildfires in this region.

38:15 Speaker 2: So the MODIS instrument didn't have a clear view at the land due to the smoke, and that created gaps in NBAR. And likewise, we can actually see burn scars form here at this time step in September. So over here, I'll flip back. You can see these are actually burn scars created by wildfires in the region. So by August and early September, the flash drought was already in full effect and farmers were well aware that there was a drought. So if we're talking about detecting this event, We want a detection scheme that's able to identify these sudden increases in negative SIF anomaly as early as possible.

39:13 Speaker 2: Ideally, we would want to detect this sometime in June as these features first started to develop. So if we go back over here, I'm going to show you the SIF RCI and how it's actually able to summarize this information that we looked at in the map context. So the first thing we're going to want to do is create a time series of the average SIF value over the region of interest and plot it compared to the mean climatology between the 2001 and 2020 period. Furthermore, to make SIF more useful as a drought detection indicator, we want to represent the deviation from the expected seasonal growth cycle as a time series as well.

40:02 Speaker 2: And what we'll want to do to do that is first derive something called the SIF standardized anomaly, which again comes from the Mohammadi and Wong paper. This is a statistical metric that characterizes the z-score of the change in SIF between each of our eight day time windows. And then finally we'll derive SIF-RCI, but I'll get to that in a minute. So now let's look a little more carefully at these climatology files in our inputs directory. And these files have four different bands that we can use for calculating the values in the formulas below.

40:50 Speaker 2: So in band one, we have the mean change in SIF between our eight-day windows, which is denoted by delta SIF J. And then the next band has the standard deviation in the change in SIF between our J eight-day windows. So again, this is the change between the values rather than the mean and standard deviation at each eight-day window. And that allows us to characterize the cycle throughout the year, rather than just the value of each time step on its own. In comparison, we have the mean and standard deviation as well in bands three and four, which we'll be using just for comparison's sake.

41:50 Speaker 2: So in the next cell, we're going to create a CSV file with the various time series we intend to use as flash route indicators. First, we'll just take the spatial average of SIF over the region that we were looking at. And then we'll take the mean, which actually just comes from the climatology files. And then we'll get the z-score, the SIF standardized anomaly that I just introduced. And then lastly, we'll get the RCI. So z-score can be calculated using the formula here. So we'll want to take the difference between the j eight-day windows, between, so the SIF at time step j minus the SIF at time step j minus one.

42:47 Speaker 2: We'll subtract that from our mean change in SIF and divide it by the standard deviation. then we can use these z-scores in a piecewise recursive manner to derive the RCI. And I don't want to fully get into this formula since it's a bit complicated, but the way that you can think about it is that SIF-RCI accumulates positive or negative anomaly over time and allows us to characterize the severity of a particular anomaly. And you can almost think of it like a discrete derivative is what we're doing.

43:31 Speaker 2: And it allows us to better characterize sudden changes in the SIF anomaly. So I'll run this cell here, which actually is implementing these formulas and reading the region of interest to do our spatial aggregation.

43:56 Speaker 2: And finally, it's going to save the time series that we calculated into a CSV. And so this will be running.

44:11 Speaker 2: And you can see that it saved out our time series of SIF for the Northern Great Plains region. And if I take a look at that, We can see we have the dates over the time range that we're looking at from January through to October. And we have our values for SIFs spatially aggregated, our mean SIF for that time of year, the SIF standardized anomaly, which we call DSIF Zscore here, and then our SIF RCI value.

44:56 Speaker 1: And these values need to.

44:57 Speaker 2: Be calculated per grid cell and then spatially averaged, which is really important because we want to characterize any spatial heterogeneity in the data before we average it all together. And now we're going to plot the data over here. And there is quite a bit of plotting code that I don't want to fully get into here. but essentially you'll be able to see the plot and then I can explain some of the details of what we did with plotting. So it's gonna take a little bit the first time and there we go.

45:43 Speaker 2: So there's two subplots here. One is kind of summarizing what we looked at on the map from earlier. And then the second one is actually showing our SIF RCI. So over here, these are the spatial average of the SIF values, these values that we were looking at in QGIS. And this is for our 2017 year that we're actually studying in this case study. In comparison, these are our 2001 to 2020 mean climatology. So you can see there's a very nice curve here throughout the growing season that peaks essentially in the middle of the growing season and then tapers off towards the harvest.

46:36 Speaker 2: And this is what we expect to happen. And what actually happened is that the values were above the mean early in the spring, which corresponds to those positive SIP anomalies that we were seeing in May, for example, which is marked over here. And then all of a sudden, we reach a peak and we start to go below our climatological mean. And values with negatives of anomaly are marked by this gray region here where we're below the mean line. And so one of the things that we can notice when we look at this data is that even though we weren't below the mean here early in June, something is still going on here compared to the behavior that we would expect.

47:29 Speaker 2: And so if we compare the slope of this climatology mean with slope here, that's what we're trying to characterize with the delta SIF instead of just looking at SIF. We can see that this delta is falling below what we expect for the time of year. And that allows us to characterize behavior before we've actually fallen below the mean over here. And so the light colored line here, or the light emphasis line rather, is showing us our z-scores, which is a bit chaotic because in some cases the delta is much greater than expected, which gives us these very large values of the z-score.

48:23 Speaker 2: And in some cases, we're slightly below expectations for the delta or maybe right on target, which gives us these values. RCI is able to smooth out these fluctuations in z-score and give us something that shows us how impactful are these accumulations of positive or negative anomaly in our change. And that's what's denoted by this SIF RCI line here. And essentially, this allows us to really quickly summarize what's going on with the behavior of SIF. And as a really quick way of assessing potential flash droughts, this isn't the final detection criteria that we're going to use in the course.

49:19 Speaker 2: We'll discuss that in part two. But we can just say that when we have values below zero, that's going to be an indicator that we might be seeing a flash drought. And we have more confidence to say that the more consecutive eight-day windows we see where the value is below zero. And so in this case, this was a very strong signal in flash drought where we spent, let's see, 32, 40, 48 days with SIF RCI below zero. And the beginning of that excursion below zero is our early June time step, which, as I recall, was June 2nd.

50:05 Speaker 2: So here is when we began detecting this flash drought signal. And you can see that was driven primarily by this negative anomaly over here in the eastern portion of our region of interest. And another thing to note is that most of the negative anomaly hadn't shown up yet. So we were actually detecting this change where we had positive anomaly increasing to all of a sudden things are starting to turn. So here are some other observations that we can make. For one thing, why did we have positive SIF anomaly early in the season?

50:53 Speaker 2: There's a couple of reasons for that. For one, above average spring temperatures caused early snow melt, which led to moist soil conditions by late March and early April. And apart from looking at meteorological data, which we're not going to do here, and the SIF plot itself, we can verify the claim about snowmelt by looking at the Normalized Different Snow Index, or NDSI, against a different year in a tool like Worldview, which I'll open up over here.

51:22 Speaker 2: So here we have the NDSI from Terra underneath a base map of Terra itself. And on the right, we have 2018, which could be considered a year with a more normal snowpack in this part of the world. And then 2017, which is our case study year. And by late March, we saw that most of the snow over this region, especially the agricultural regions, there's the Fort Peck Dam that marks a major part of our region of interest. Most of the snow had already melted by this time of year. So that drove earlier plant growth in the season than would have occurred otherwise.

52:13 Speaker 2: Another thing we can note is that the yield for cropland will increase annually due to the development of new seeds, improvements in machinery and agricultural practices and similar factors. Because SIF has a mechanistic relationship to photosynthesis, it can therefore be thought of as a proxy for crop productivity. If we were to plot the SIF from 2001 to 2020 for a single eight-day time window over this region, we would expect to see a positive slope in the data. And we can see that illustrated here with our winter wheat yield, which is a major crop for this region of Montana.

52:58 Speaker 2: And over time, the yield for winter wheat is increasing. And so similarly, the SIF might logically increase over time compared to an earlier value for SIF in the climatological mean that we're using. Another major observation we can see is that the SIF values are persistently below the mean line from June until the end of the growing season, which is our gray region here. And this decrease in crop productivity suggests the occurrence of the drought, but it's really better illustrated by the RCI that we produced.

53:48 Speaker 2: Another thing that we can note which is a little more complicated, is that high SIFT conditions preceding a flash drought event might be an indicator of the flash drought itself. In other words, if the plants were growing particularly strongly earlier in the season, they would drive higher evapotranspiration that under a negative precipitation deficit would and low soil moisture conditions would cause the soil to dry out faster and cause the drought to develop faster than it would have if the SIF were closer to its mean or even below the mean early in the season.

54:41 Speaker 2: Okay, so one thing that we can note is that while SIF-RCI is pretty good at detecting flash drought. It might be a little bit unimpressive to you on its own. With the benefit of hindsight, we could see that the period of below zero RCI in the summer amounted to an agriculturally impactful drought in the growing season. But that is with the benefit of hindsight. So there were other points in the time series where SIF approached close to this zero value and then recovered. So if we were just looking at this point alone and then waiting for the next eight-day time window to make a decision, it's hard to say.

55:28 Speaker 2: I mean, this might go back up. We don't know. And so it's really helpful for us to look at multiple indicators to give us multiple points of analysis to have a higher degree of confidence for indicating flash droughts. And one of the data sets that we can use for that is soil moisture. So while SIF, and specifically SIF-RCI, can tell us that vegetation is doing less photosynthesis than expected, it cannot tell us why. Vegetation health can decline not only due to drought, but also from flooding, severe storms, fires, and even human causes.

56:10 Speaker 2: To attribute the decline in crop productivity to flash drought specifically, we need to look at soil moisture, which can be thought of as the water supply available to plants. Typically, many flash drought and long-term drought studies use evapotranspiration as a key indicator of water stress, since it accounts for the demand side of the water economy in an ecosystem. Evapotranspiration, however, is influenced by the soil moisture itself, along with atmospheric water demand and incoming solar radiations.

56:42 Speaker 2: While a more complete analysis of flash drought, including the one used by Mohammadi and Wang, would include a metric that accounts for ET, such as the evaporative stress index, the ESI, or the evaporative demand drought index, EDDI, they're not going to be explored in this demo. In practice, evaporative demand can be very sensitive to short-duration changes in moisture imbalance and can therefore flag too many flash droughts.

57:12 Speaker 2: So in this section, we're going to show that the trajectory of roots and soil moisture relative to the water storing capacity of the soil can effectively indicate flash drought

57:26 Speaker 2: without needing to look at evapotranspiration. It would certainly help to look at ET, but the soil moisture is an

57:36 Speaker 2: effective indicator on its own. So to do that, we're going to look at the SMAP mission. And SMAP measures the surface soil moisture, which covers the 0 to 5 centimeter depth, on the day and night sides of the Earth using its L-band radiometer. And it's been operating since January 2015. The radiometer data are collected at a 40-kilometer resolution with a two- to three-day average revisit time. And then they get interpolated to a 9-kilometer resolution. In particular, we're going to be using the SPL4 SMGP collection, which has got the long name of SNAP L4 Global 3-Hourly 9-Kilometer Surface and Root Zone Soil Moisture Geophysical Data.

58:29 Speaker 2: And that's quite a mouthful, so we'll look at the data in worldview to get a sense of what it looks like.

58:38 Speaker 2: And so these are our model-based estimates of root zone soil moisture, but they assimilate the L-band radiometer observations into the land surface model that's actually used. So the idea is that while the observations can tell us about the surface, the model is able to give us estimates of the root zone layer, which goes down to up to one meter. We'll talk about that in a little more detail in part two. But I just want to introduce the data set here. And we can see that these values are a ratio metric unit.

59:20 Speaker 2: And that basically just indicates how much water per cubic meter of soil is present. And we have these estimates for the whole world, and they're at a three-hourly time resolution. On world view, we view one day at a time. But effectively, this can tell us about the development of drought conditions that are agriculturally impactful.

59:54 Speaker 2: So turning back over to our notebook. There's a lot of information in each of the granules, but we're basically only going to be using the root zone soil moisture from these granules. And although there's no space-based instruments that can measure root zone soil moisture directly, we're deriving it in this case by incorporating meteorological forcing data, such as precipitation, and then simulating the infiltration of water into deeper layers of the soil. We're also incorporating the surface soil moisture observations, as I mentioned.

1:00:37 Speaker 2: Compared with a purely model-based root zone soil moisture, the assimilation of the observations into the SPL4 SMGP product improves the accuracy of the estimates, which is substantiated in the algorithm theoretical basis document. And so one of the challenges of working with this data is that each of these granules is pretty large because it includes a lot of data that we actually don't need to work with. The data set itself is over 40 terabytes since the beginning of the mission. And so to process one year worth of data like we need to do here, if we wanted to just download it directly, it would be four terabytes.

1:01:22 Speaker 2: And since I assume most of you don't have a spare four terabytes on your computer to work with, we have to be a little more clever about how we access the data. Luckily, there's a new technology called virtual datasets. It's going to enable us to stream the data for this analysis without having to store entire granules. This is similar to the OpenDAP technology if you've ever worked with it. In fact, the team at the NSIDC, which is the DAC that archives this dataset, has already created a virtual dataset for us to use that catalogs the data between 2015 and October 2025.

1:02:00 Speaker 2: And we're gonna be using that resource in the following cell. The basis for much of the code that we're looking at below came from a tutorial by Luis Lopez, who's also the developer of the Earth Access Library over at NSIDC. In the first cell below, we're going to be creating a local desk cluster which Dask is a multiprocessing parallel compute library, if you're not familiar with it. And that's going to enable us to run a lot of these processes that are very data intensive in a more timely manner.

1:02:38 Speaker 2: And depending on the number of CPU cores on your machine, you might need to adjust the n workers parameter right here on this line. I use an end workers of eight as a default, which works great on my computer, but if you have fewer CPU cores, you might need to lower this number. You'll need to manually authenticate to the NASA Earth Data System. So remember I had you all create an account earlier? We're going to need to use that account now. If no environment variables or netRC are found, you're going to be prompted to log in.

1:03:14 Speaker 2: And then this is going to open a connection to the VDS, which is located at this web address here. And nothing is going to be downloaded to your computer yet. The virtual aspect of the virtual data set means that the printout from the cell is just telling us what the contents of the data set are, and then we'll be using it afterwards. So I'm going to run this.

1:03:40 Speaker 2: and it's going to take a couple seconds to connect to this remote virtual dataset and set up the Dask client, and then it will tell us some information. So we can see we have three main dimensions in our dataset, time, y, and x, and our y and x approximately correspond to the lat and lon of the spatial dimensions, the data, and then time is telling us each of these three hourly time windows between March of 2015 through to October of 2025. Now, this map level four data actually goes through to the present day, but for virtual data sets, you need to catalog the data periodically.

1:04:31 Speaker 2: And so the particular snapshot catalog that we're working with only goes up to October, and you would need to create a new one if you wanted to look at data that's more recent than that.

1:04:43 Speaker 1: We show you how to.

1:04:44 Speaker 2: Do that in the appendix notebook, but we won't be going through it in this particular demo. So this VDS will be perfectly good for our use cases. So one of the things that's a challenge of working with root zone soil moisture is that because it's such a difficult to estimate quantity, We actually want to convert it into a form that allows for easier comparison between other estimates of root zone soil moisture. In other words, there's quite a significant amount of uncertainty in the soil moisture that we're estimating.

1:05:22 Speaker 2: And one of the ways that we can do that is with the soil water deficit index. And this indicator was originally proposed by Jose Martinez Fernandez and others in 2015. And it's basically an indicator that converts soil moisture, especially roots and soil moisture, into a ratio between 10 and minus 10 that gives us an indication of how wet or dry the soil moisture conditions are relative to the model that they come from its own estimation of what wet and dry conditions are. And the formula that we use for that is right here.

1:06:15 Speaker 2: And this theta is our actual estimates of RZSM. Theta FC is our field capacity, which is basically the wettest value that the soil can attain at equilibrium. And we can see more wet conditions than that, but the soil will tend to dry down or have infiltration to reach an equilibrium if it's above the field capacity. And likewise, the wilting point or the permanent wilting point is the theta WP, which is the estimate of the driest conditions that the soil can obtain. And so intuitively, one way you can think about this is that if the soil is really dry, we'll see a value of negative 10, which is our wilting point.

1:07:18 Speaker 2: If the soil is wet, but in equilibrium more or less, we'll see a value of zero, which is our field capacity. And anything above zero is actually wet. quite wet conditions bordering on flood. And in between there, values of minus 2 to minus 5 are moderate soil water deficit. And then minus 5 to minus 10 would be a severe soil water deficit. And so where do these values come from, the field capacity and will point? In the Martinez-Fernandez paper, They derived the constants from in-situ measurements by taking the 95th percentile soil moisture as the field capacity and the fifth percentile as the bolting point from a suitably long observational record.

1:08:17 Speaker 3: Well, we could do the same.

1:08:17 Speaker 2: Thing for our analysis. This map mission actually provides these values in a land model constant file, which is a

1:08:28 Speaker 2: These values come from the GMAO catchment model that's used as the land surface model for deriving the RZSM estimates themselves. Because the constants do not vary with time, there's only one granule we need to download from NASA Earth data

1:08:48 Speaker 2: to actually get those values. So in the following cell, we're going to download this granule and then produce a time series of SWGI by streaming a subset of this map level 4 data and then save the spatially average result in a CSV. And so this cell is going to take quite a long time if you haven't already derived the time series, which I have to save time here. but normally this would take about five to 10 minutes to run due to streaming the data from NASA Earth data. And so I'm going to actually run this below.

1:09:30 Speaker 2: Here we go. Okay, so this will run the code. And it's gonna take a little while.

1:09:38 Speaker 1: Okay.

1:09:45 Speaker 2: And in my case, the code is actually run and we've saved out the CSV of our time series. And if I take a look at that over here, we'll be able to see that we've temporally averaged the data into three day time windows. And that's kind of arbitrarily chosen. I know that it doesn't match up with the eight day time windows that we use for CIF, but it was chosen basically because If you think about the revisit time of this map observations that we're assimilating into the model, we're assimilating new data every three days or so.

1:10:24 Speaker 2: In between there, we're looking at the behavior of the land surface model. Now, this isn't exactly the way that the data set works, but that's just the theory there. You could obviously use a different time window aggregation if you prefer, but three day aggregation works pretty well for the context that we're using. And these are our SWDI values, which we can look at when we plot it in the next cell. And so here I'm going to plot our values, and this will show that early in the season, our values of minus two and above we're not really indicating drought.

1:11:12 Speaker 2: So we had plenty of soil moisture. And if you look at previous years or other years where there wasn't a drought, you can see that even though minus two might be like moderate deficit, it's really not that bad. But as we go past the period of the year where there was planting into the growing season, there's a steadily decreasing amount of root zone soil moisture. And it did recover slightly in June. And as we'll see later in part two, this was associated with a precipitation event that occurred during this time period.

1:11:53 Speaker 2: But then afterwards, this was not able to fully recover the soil moisture. We continued to decrease into this severe deficit, which is our minus five. And if we just use this heuristic or this threshold of minus five SWDI to tell us whether or not there was a flash drought, then we would only detect this event around July. But if we just use that threshold, we're leaving a lot of information on the table since we do see that there's this steadily decreasing SWDI over the entire growing season up to this point.

1:12:34 Speaker 2: And then another key observation that we can make is that the soil moisture recovered slightly after the harvest, or depending on the crop, some of the harvest does continue until mid-September. And this is associated with rains that occurred late in season. And so these rains we're essentially too late to recover the crop yields in this region, but it is an interesting thing that we can note in this data. So again, one of the things that we want to be considering as we move to part two is that when we look at SWDI, we want to consider the rate of decline in addition to the length of time in drought.

1:13:24 Speaker 2: And that's because unlike with RCI, We're not comparing this to a climatological trend over a period of 20 years because we don't have 20 years worth of SMAP data to work with. Small excursions below the minus three or minus five thresholds are not significant on their own. So we want to see that this value stays persistently below these minus three or minus five thresholds. Because again, precipitation will rapidly recover the soil moisture. And one of the things that causes soil moisture to draw down is evaporative demand.

1:14:08 Speaker 2: And so ET-based metrics like ESI and EDDI will swing between what they would indicate as drought or non-drought rapidly as weather conditions change. But roots and soil moisture is going to change more slowly in response to these changes in conditions. And so that makes it a better metric to show how vulnerable cropland is to sustain dry conditions.

1:14:34 Speaker 1: In the future.

1:14:35 Speaker 2: In other words, even though we might not be able to declare a severe drought from our SWDI data in June, it hasn't reached that severe level yet, we can see that Any further dryness that occurs, any further precipitation deficit that we would see in the forecast, the soil is vulnerable to drying out further at this point. And if you run the previous cell again for 2016 data, which was a non-route year, and plot the comparison against 2017, you would see that some soil dry down is expected during the growing season.

1:15:14 Speaker 2: So this negative trend on its own is not necessarily an indicator of drought. The crops in this region will draw down moisture from the soil even in a good year. And so we want to also detect the severity when we talk about looking at flash drought. So SWDI is intended to mitigate some of the disadvantages of not comparing against past behavior by normalizing the estimates of soil moisture against the water carrying capacity of the local soil type. So in other words, if we were just looking at RZSM on its own, it would be hard for us to say, is this root zone soil moisture that we're looking at, you know, this value of 0.356, is that bad?

1:16:07 Speaker 1: Is it good?

1:16:07 Speaker 2: It really depends on the local soil conditions. And without being able to look at the climatological record, the next best way that we can do that is by using these land model constants and converting the values into this ratio that tells us this is essentially good at zero down to bad at minus 10. And so how close are these results that we obtained in this notebook to this study that this analysis is based on? And figure 2b from the study contains a plot of SWDI as well as CFRCI and a few other metrics for this particular case study.

1:16:53 Speaker 2: And this golden yellow line here is our SWDI. And without looking at the correlation metrics, our R squared value for the data, since we don't have access to their time series, we can see just qualitatively that the trend of the data is pretty much the same, where we have this decreasing value of SWDI and then a recovery sometime in September. And while their values, SWDI is indicated on the right axis over here, are about two below our values that we obtained, we did use different data sets. And one of the things that we're gonna talk about in part two is that there's some challenges to using SWDI because we're relying so heavily on those land model constants to tell us what values of soil moisture are good and what values are bad.

1:17:56 Speaker 2: But there's also techniques for mitigating that that we'll get into. Another thing that we can see is that our SIF RCI, which is our dashed line here, it goes into negative values our axis over here on the left is showing us that zero is marked by the dotted line, goes into negative values in early June, essentially, or late May, which corresponds with what our line over here is telling us as well. And so the story being told by the data from the paper, as well as the data from our own analysis is remarkably similar.

1:18:40 Speaker 2: But again, this is the benefit of hindsight. If we were just looking at the first point where these values go into our negative thresholds for indicating flash drought, it's hard to say whether or not they would recover. We can see that they didn't in this case, but we need to do a little bit more analysis to know for sure whether or not this would be a flash drought. And that's why our detection criteria needs to account for a certain amount of time spent below these thresholds to be confident.

1:19:18 Speaker 2: Unfortunately, that eats into the window of time that we have to make decisions compared with existing metrics. In this case, the black line is showing us the US Drought Monitor, which only really started indicating that there was a major drought occurring by late June, potentially even early July. So if our way of reading this data is good, we do actually have a time advantage. But if our way of detecting flash drought from the data sets that we're using is not so good, then it's basically just telling us it's wet when it's raining, or in our case, the opposite.

1:20:00 Speaker 2: So the remote sensing approach that we're using needs to have a lead time of over the information that everyone else has access to, such as the USDM, to be useful. In this case, we've shown that to be the case, and we'll look at some other case studies to show that this particular type of analysis is able to be used anywhere in the world, even in cases where you don't have the US Drought Monitor. And remember, we obtained these results even without filtering by land cover class. You would even improve your detection capabilities by filtering out the deciduous broadleaf forests and the needleleaf forests in the region of interest that we were looking at.

1:20:48 Speaker 2: If we return to the SIF anomaly visualization at the beginning, remember, the negative anomaly was strongest over the croplands and grasslands. particularly in the northeastern portion of the study region. So I'll show you that again. So remember, over here, this is cropland here and this is grassland here. That's where our negative anomaly started. We filtered by land cover class. We would probably get an even better SNR in this case study. So the last point that I wanna make is that Flash droughts can be recovered via well-timed brainstorm just as quickly as they develop.

1:21:31 Speaker 2: So it's not just so simple to decide that an event is a flash drought as soon as these metrics fall below the threshold. The judicious use of remote sensing data can provide a useful lead time over traditional signals like the USDM that other farmers or the broader commodities market may not have priced in. While high-quality in-situ measurements, such as those for precision agriculture, might in most cases outperform remote sensing, we still get a wealth of information from looking at the remote sensing and model-driven data that we've looked at in this case.

1:22:09 Speaker 2: And the advantage over in situ is that we have something that is spatially and temporally contiguous to work with for our detection criteria. And so that's about all I want to go over today, and I'll hand it Back over to Erica for final thoughts and comments.

1:22:32 Speaker 3: Great. Thank you very much, Nick, for the theoretical portion and Jackie for that great demonstration. Before we begin the questions, I will do a quick summary of the most important points covered during today's session. In summary, flash drought is characterized by the rapid drying of land and unusually swift intensification of drought conditions. It often triggers or amplifies sustained drought, and it impacts ecosystems, crops, as well as human water management. SMAP provides daily mapping or near-daily mapping of near-surface soil moisture, that's Model integrated root zone soil moisture, so that's down to one meter.

1:23:23 Speaker 3: SMAP tracks the subsequent gradual and rapid drawdown of soil water, leading into full flash drought development. OCO2 measures SIF, and it's an instantaneous indicator of plant photosynthetic activity, which acts as an early warning signature up to three months before flash drought onsets. by detecting plant stress and decreased photosynthetic activity before leaves turn brown or lose greenness. SIF anomalies are quantified using indices like the Z-score, which is a standardized anomaly from the long-term mean, as well as the rapid change index, which is the growth rate change relative to preceding weeks, and it's optimal for operational monitoring.

1:24:14 Speaker 3: There is no single index or process that accounts for flash drought, but combining multiple hydroecological indices, such as the USDM, SMAP, soil moisture, and OCO2 SIF, you can then get the most reliable detection, attribution, and early warning capability. SIF Rapid Change Index, the SIF RCI, and Soil Water Deficit Index, SWDI, can detect flash drought onset earlier than traditional monitoring tools like the U.S.

1:24:50 Speaker 2: Drought Monitor.

1:24:51 Speaker 3: Contact information, here you can find the email addresses for Nick Parazou and Jackie Ryan should you have any additional questions about the material that was presented today. We've also shared our RSAT website page as well as our RSAT YouTube page. page or link. And if you have any questions or comments or you'd like to share how you've used our trainings in your work or for your studies, please send an email to the email address that you see here. And finally, to join our mailing list or stay up to date on our latest training, visit our contact page to subscribe.

1:25:36 Speaker 3: as a reminder there is a second session that's part of this webinar series it will be next