Calculating Spectral Indices For Land and Aquatic Applications Using QGIS

0:01 Speaker 1: Hello, everyone, and thank you for attending the Calculating Spectral Indices for Land and Aquatic Applications in QGIS. We are NASA RSAT. My name is Justin Sain, and let's go ahead and get started.

0:18 Speaker 2: First, a little bit about RSAT.

0:21 Speaker 1: RSAT provides accessible, relevant, and cost-free training on remote sensing satellites, sensors, methods, and tools. Trainings include a variety of applications of satellite data and are tailored to audiences with a variety of experience levels. The thematic areas that we operate in are agriculture, climate and resilience, disasters, ecological conservation, which is us, health and air quality, and water resources. RSAT trainings are either online, such as this one, or in person. They're live and instructor-led or asynchronous and self-paced.

0:58 Speaker 1: As always, they are cost-free. They are bilingual or multilingual, as this one is done in both English and Spanish. We only use open source software and data, and we are happy to accommodate differing levels of expertise. To learn more, you can visit our website.

1:23 Speaker 1: Now we can begin our exploration of spectral indices. We'll start with an overview of spectral indices before moving on to a practical exercise where you will see the process of calculating spectral indices in QGIS. Spectral indices give us relatively simple calculation to measure properties in our remote sensing imagery.

1:45 Speaker 2: They're typically composed of just a.

1:47 Speaker 1: Few bands and aim to isolate differences in reflection and absorption across the electromagnetic spectrum. These signature differences can tell us things about our scene, such as the presence of chlorophyll in water or the health of vegetation. By the end of this training, you'll be able to describe data sources, data processing, and applications for indices related to land and aquatic environments. You'll be able to calculate a number of these indices in QGIS and you'll be able to identify which indices are appropriate for land monitoring versus those that are appropriate for aquatic monitoring.

2:27 Speaker 1: The prerequisite training for this course is the RSET Fundamentals of Remote Sensing. That training will give you the necessary context to understand how remote sensing works and how we can isolate spectral signatures related to the properties of our scene in which we're interested.

2:46 Speaker 1: This is a training in one part. Homework opens February 27th and is due March 13th.

2:54 Speaker 2: It'll be posted on the training website.

2:56 Speaker 1: A certificate of completion will be awarded to those who attend all of the sessions. This is one session, so that should be easy if you're here already, and complete the homework assignments before the given due date. Now, on to spectral indices in QGIS. Again, if you're hearing this in English, my name is Justin Fain, and I'm an RSAT trainer in ecological conservation located at NASA Ames Research Center in California, USA. If you're hearing this in Spanish, that's going to be done by my colleague Sativa Cruz, who is also a trainer with RSAT in ecological conservation.

3:38 Speaker 1: During this webinar, please feel free to enter any questions into the questions box. At the end, we'll go through and do our best to answer each of your questions. If we're unable to get to your questions during the live Q & A, we'll put all answers in the Q & A document and upload them to the training website about a week after the training. So let's review some remote sensing concepts. So to do a quick review of the information from the fundamentals of remote sensing, which are relevant to this course, Every surface on Earth, and in general, reflects and absorbs energy in different ways.

4:17 Speaker 1: This fact is easily observed by the fact that healthy vegetation appears green to our eyes. However, each surface also reflects and absorbs energy from the parts of the electromagnetic spectrum beyond what our eyes can detect. These differences in how surfaces interact across the spectrum are what we call spectral signatures. Understanding how changes in the spectral signature of the surface relate to the properties of the thing being observed allows us to create spectral indices which assign a numeric value to the property of interest.

4:52 Speaker 1: The electromagnetic spectrum is made up of all energy from the very long wavelengths of radio and microwave to the very short wavelengths of X-ray and gamma rays.

5:05 Speaker 2: Visible light, which is perceptible to.

5:06 Speaker 1: Humans, makes up only a very narrow portion of the electromagnetic spectrum. Remote sensing instruments, such as those on satellites, are often engineered to record light in a much broader range of wavelengths than the visible part of the spectrum. This means that we can use remote sensing imagery to see things that otherwise wouldn't be directly observable. In most cases, the source of our light for remote sensing is the sun. Solar radiation passes through the atmosphere, hits a target surface such as forest, water, buildup areas, and those objects reflect and absorb energy differently in different wavelengths.

5:48 Speaker 1: So this reflected solar radiation is measured by the satellite in various wavelengths. Each slice of the reflected radiation is recorded as a single image called a band. This is a simplified example of what happens with electromagnetic energy for green vegetation. The figure on the left shows incoming radiation in the visible and near-infrared wavelengths. Blue and red wavelengths are absorbed by green vegetation, and green and near-infrared wavelengths are reflected. This is primarily because of the chlorophyll in the leaves.

6:30 Speaker 1: We see vegetation as green because we only see the reflected energy in the visible wavelengths. We can see green, but we can't see infrared. Every surface on Earth reflects and absorbs energy in a different manner. For optical imagery, the sensors are capturing similar data that we can see with our eyes in the visible portion of the electromagnetic spectrum and some portions we can't see in the infrared. So each object on the ground produces a unique spectral signature of the amount of reflection or absorption at each wavelength.

7:04 Speaker 1: That's, of course, called the spectral signature. So we can obtain the reflectance values of different land cover types across the visible and infrared portions of the electromagnetic spectrum to create these signature figures like the one you see here. In the green, you can see a typical vegetation signature that has a small reflectance peak in the green portion of the spectrum and a very high peak in the near-infrared portion of the spectrum. Soil tends to have a steadily increasing reflectance value in the visible and near-infrared, and water tends to absorb most energy.

7:39 Speaker 1: Therefore, the reflectance is low throughout, with a small peak in the blue. In this figure, you can also see the bands of a typical Landsat image in gray. So with those bands, we receive one value of reflected energy based on the object being observed.

7:59 Speaker 1: So that brings us to spectral resolution, which is the ability of a sensor to define the wavelength intervals or ranges. When we use the word band, we're referencing the wavelength range and how the satellite is making the distinctions between different ranges. So each band represents a portion of the electromagnetic spectrum, and if you have the reflection of an object at different wavelengths, you can use band math.

8:22 Speaker 2: Or simple calculations to identify differences.

8:28 Speaker 1: When we have multispectral imagery, we likely have blue, green, red, near-infrared, and mid-infrared bands. So with satellite image, we can get one value for each wavelength range for each pixel.

8:42 Speaker 2: We can then stack the.

8:43 Speaker 1: Layers to view the imagery. We can also use the values for different layers at the same location on Earth to identify things like vegetation health. Here you can see the comparison of the bands of Landsat 7, 8, and 9, and Sentinel-2. So for Landsat 8 and 9, for example, those sensors have 11 bands, or wavelength ranges, where data is collected. The band number and the corresponding portion of the electromagnetic spectrum are shown here. Band 2 is the blue, band 3 is the green, band 4.

9:17 Speaker 2: Is the red, and so on.

9:20 Speaker 1: The gray parts of this figure are called the atmospheric windows, where the satellite can actually collect reflected energy from the ground.

9:29 Speaker 1: Different wavelengths penetrate differently in the water column, and this will influence the amount of information that sensors can obtain in aquatic environments. For example, wavelengths in the blue region are usually.

9:40 Speaker 2: Those that penetrate deeper.

9:42 Speaker 1: This is why we see the ocean as blue or dark blue, whereas those longer wavelengths, like the reds and near-infrared, penetrate less. The signal received by the sensor, either in orbit or airborne, is also affected greatly by the composition of the atmosphere above the target areas, as we can see in the figure on the right.

10:07 Speaker 2: When we talk about.

10:07 Speaker 1: Ocean color in this webinar, we're referring to.

10:09 Speaker 2: The remote sensing reflectance.

10:12 Speaker 1: Remote sensing reflectance is a function of the inherent optical properties of the material in the water. For example, absorption by phytoplankton, non-alkyl particles, colored dissolved organic matter, and water, and scattering in the backward direction by particles and other material. It also contains a correction factor to relate subsurface irradiance reflectance to radiance reflectance, as denoted by C.

10:39 Speaker 2: In this equation. Remote sensing reflectance.

10:42 Speaker 1: Is water leaving radiance corrected for the bidirectional effects of the air-to-sea interface and subsurface light field, and normalized by downwelling solar irradiance just above the sea surface. Remote sensing reflectance is the fundamental remote sensing quantity from which most ocean color products are derived, for example, chlorophyll, particulate and organic carbon, light absorption by carbon-dissolved organic matter, and suspended sediments, etc. Now, a bit more in depth about the spectral indices we'll.

11:21 Speaker 2: Be going over.

11:24 Speaker 1: Here is an overview laying out the spectral indices we'll be covering in this webinar. Don't worry too much about this table as we'll discuss each in a little more detail. First, let's look at the Normalized Difference Vegetation Index, or NDVI. NDVI is the relationship between the red and near-infrared wavelengths. The actual formula is specified here. The values of NDVI for an individual pixel range from negative 1.

11:54 Speaker 2: To positive 1.

11:56 Speaker 1: Any pixel between negative 1 and 0 means no vegetation, and pixels closer to 1 indicates the highest possible density of green leaves. The picture on the right shows that healthy green vegetation absorbs most of the visible light, only about 8% is reflected, and reflects a large portion of the near-infrared light.

12:16 Speaker 2: About 50%.

12:19 Speaker 1: Unhealthy, sparse, or senescing vegetation reflects more visible light, 30% is reflected, and less near-infrared light, only about 40%. You can see the difference in the resulting NDVI values below the image. The green vegetation has a value closer to 1, while the brown vegetation has a value closer to 0. So to give you a sense of how this calculation looks on the ground, imagine that each of these gray squares is a pixel, and we're interested in doing an NDVI calculation for one pixel, the red on the bottom right.

12:58 Speaker 1: The near-infrared value for this pixel is 0.1144, and the red value for this pixel is 0.0344. Given those values, we take the near-infrared value minus the red value and then divide that by the near-infrared value plus the red value, and we get an NDVI value of 0.538, indicating healthy vegetation.

13:29 Speaker 1: Then you can apply the same calculation across the entire image using the near-infrared and red values for each individual pixel and generate a new NDVI layer. So this shows you a bit more visually how those dir-infrared and red values are used on a pixel-by-pixel basis to examine NDVI over a large region by applying that same calculation across each pixel. Plant phenology is the annual dynamic of vegetation greenness and can be tracked using vegetation indices like NDVI. In the graph on the top, you can see the progression of vegetation dynamics as seasons change.

14:11 Speaker 1: In North America, early in the year, which is winter, there are little to no leaves on the trees, resulting in low NDVI values. When spring arrives, the vegetation greens up and NDVI increases until its peak in the summer. Then, as vegetation finesses and loses their leaves, the NDVI declines. The image below shows the difference in greenness in the winter versus the summer in North America. Now to look at another vegetation index, the Enhanced Vegetation Index, or EVI. So there are other indices such as the Enhanced Vegetation Index, or EVI, that could be useful.

14:54 Speaker 1: EVI is another measure of vegetation health. It's particularly useful in regions with high biomass because the NDVI can oversaturate or hit a maximum value in these regions, whereas EVI has a higher threshold and can identify more subtle differences in these regions. For example, the Amazon. There's also the Soil Adjusted Vegetation Index, or SAVI. The Soil Adjusted Vegetation Index minimizes the influence of bare or nearly bare ground when trying to assess vegetation. This one could be useful in semi-arid or arid regions, like the southwestern United States, where there's a greater bare ground or soil cover in relation to the vegetation.

15:42 Speaker 1: Here, the near-infrared and red bands are still used, but a correction factor is applied.

15:51 Speaker 1: Moving away from direct monitoring of vegetation, we have the normalized burn ratio, or NBR.

15:59 Speaker 2: We can use the reflectance.

16:01 Speaker 1: Properties from the satellite image to identify healthy vegetation versus burned areas. As we mentioned, in healthy vegetation, there's reflectance in the green and near-infrared portions of the electromagnetic spectrum, so that's why we see healthy vegetation as green. When we compare the healthy vegetation spectral signature to those of burned areas, we notice some stark differences, though. You can see this pretty clearly for the spectral signature curves of low, moderate, and high burn severity.

16:30 Speaker 1: Where healthy vegetation has this large peak in the near-infrared, bare soil and burned areas have much lower peaks in the near-infrared. And in the case of high-severity burned areas, there's much less response in the near-infrared. You can see that healthy vegetation has low reflectance in the shortwave infrared, but burned areas have high reflectance in those wavelengths. With these spectral characteristics, we can identify burned areas and distinguish it from healthy vegetation. To take advantage of this difference in spectral response between healthy vegetation and burned area, we use the Normalized Burned Ratio, or NBR, to map post-fire conditions.

17:10 Speaker 1: NBR uses remote sensing data, the near-infrared and shortwave infrared, to map burned areas and ultimately assess burn severity. You can see how this calculation is completed here on the side. Similar to NDVI, NBR is a unitless value from negative one to one. A high NBR value, one that is closer to one, indicates healthy vegetation, while a low value that's closer to negative one indicates recently burned areas and bare ground. NBR is a commonly used metric for identifying areas where vegetation has recently burned due to fire.

17:45 Speaker 1: Here you can see an example of NBR in action.

17:49 Speaker 2: The images show pre.

17:51 Speaker 1: During and post-fire NBR from the Mendocino complex that burned in California in 2018. Now, moving to the aquatic side of things, we'll take a look at the Normalized Difference Chlorophyll Index, or NDCI. So, chlorophyll is an indicator of phytoplankton biomass, as it's usually the main pigment of these communities. It can also be used indirectly to estimate nutrients in the water column, as typically waters with a high chlorophyll concentration are eutrophic, or high in nutrients, especially nitrates and phosphates.

18:34 Speaker 1: The overarching objective of the Mishra and Mishra paper was to improve the accuracy of chlorophyll A retrieval in turbid waters using simple, easy-to-implement intuitives, such as NDVI for vegetation, and a universal model. The concept of NDCI was developed using simulated data. However, field datasets acquired from the same geographic regions were used to further validate the NDCI and chlorophyll-A relationships. NDCI uses the remote sensing reflectance at 665 nanometers and 708 nanometers, emulating the medium resolution imaging spectrometer.

19:20 Speaker 2: Or MARIS, channels.

19:23 Speaker 1: Similar to other turbid productive chlorophyll-A algorithms, this index uses the information from the reflectance peak centered at 700 nanometers, which is maximally sensitive to the variations in chlorophyll-A concentration in water. Similarly, a wide spectral absorption peak between 665 nanometers and 675 nanometers is generally assigned to the absorption by chlorophyll A pigments. The authors selected the two spectral features centered at 665 nanometers and 708 nanometers to develop NDCI and to avoid the confounding influence of carbon dissolved organic matter and total suspended sediments on the water reflectance spectra at shorter wavelengths.

20:08 Speaker 1: Also, as both bands are closely located, they assume that the carbon-dissolved organic matter and total suspended sediment absorption is similar in magnitude. Based on the results from bio-optical modeling in this study, the combined range of carbon-dissolved organic matter and total suspended sediment absorption at 665 and 708 nanometers in the study regions are approximately equal and the difference between them can be assumed as negligible. Following the simplistic concept of NDVI applied in land vegetation monitoring, NDCI was developed by taking the spectral band difference of those 708 and 665 nanometer bands and normalizing the sum of their reflectance to eliminate any uncertainties in the estimation of remote sensing reflectance, seasonal solar azimuth differences, and atmospheric contributions of those wavelengths.

21:07 Speaker 1: To take a look at a case study, here you can see high NDCI values can be indicative of regions impacted by nutrient enrichment. In general, your NDCI will give an idea of the concentration of phytoplankton communities in the water column, but in relatively clear and shallow waters, it may be influenced by the presence of benthic vegetation, such as seagrass or algae. Now to look at the Normalized Difference Aquatic Vegetation Index. The Normalized Difference Aquatic Vegetation Index, or NDAVI, was developed by VIA et al.

21:48 Speaker 1: In this case, it's aimed at using higher spatial resolution data sets, such as those Landsat, and it focused on the blue and near-infrared bands. The idea behind it is that by using the blue band, which penetrates deeper into the water column, and the near-infrared, which has very little penetration in water, the benthic aquatic vegetation can be easily distinguished from other benthic components such as sand, mud, etc.

22:17 Speaker 1: Here's some data by a team from the NASA DEVELOP program where they explored different indices to follow potential correlations.

22:24 Speaker 2: With particular environmental factors.

22:27 Speaker 1: In this case, there's a unique band combination that allows seagrasses to be analyzed over time. photosynthetic organisms reflect a large amount of near-infrared light, so the NIR band and the blue band collectively allow aquatic vegetation to be assessed.

22:47 Speaker 1: Moving on to the floating algal index. This is the equation for the floating algal index.

22:54 Speaker 2: Note that it uses reflectance.

22:56 Speaker 1: Values in the red, near-infrared, and shortwave infrared regions of the spectrum, and is similar to other indices or methods used for chlorophyll. Here's an example application of FAI to detect sargassum with MODIS Aqua data. We can see clearly how the sargassum mats are shown in the image on the left, and how even with this multispectral data set, Still, we can see the big difference in the remote sensing reflectance, especially in the near-infrared region in the graph on the right.

23:36 Speaker 2: Here's data provided by.

23:37 Speaker 1: Our colleagues at the University of Puerto Rico Bio-Optical Oceanography Lab, where they applied the FAI to detect Sargassum mats in southwest Puerto Rico at much higher spatial resolution using Sentinel-2 data. A land mask was previously applied. The zoomed image on the right shows a chronic accumulation of sargassum close to or at the shoreline. The University of Puerto Rico colleagues, Dr. Roy Armstrong and Dr. William Hernandez, are using the FAI in combination with time-lapse photography to monitor the accumulation of sargassum floating mats each year and observe how these impact the marine life in the La Proguera National Reserve.

24:23 Speaker 1: Take a moment to look at the diversity of colors across this scene. The inherent optical properties of the material in the water give the water its characteristic color. Inversely, it's possible to use the color of the water we see from the image and infer the inherent optical properties of the water to get at questions like, what is the chlorophyll concentration? What's the clarity of the water? How much sediments are suspended in the water? These questions are the type that we can relate to some of these spectral indices.

24:57 Speaker 1: Now, I'd like for you to look at the plot on the right. The remote sensing reflectance spectra you see there correspond to the examples on the left. These are schematic illustrations of remote sensing reflectance spectra, not the actual spectra.

25:11 Speaker 2: Note the prism across the top figure.

25:13 Speaker 1: It shows you the colors of visible range and how those match up with the wavelengths along the x-axis. The near-infrared is also noted. These spectra are intended to demonstrate that materials in the water have different signature remote sensing reflectance spectra. Look at the sediment spectrum in orange.

25:31 Speaker 2: On the plot.

25:32 Speaker 1: It has a high magnitude across a broad range.

25:35 Speaker 3: Of the spectrum.

25:36 Speaker 2: It's also somewhat reflective in the near-infrared.

25:40 Speaker 1: In the image on the left, our eyes see a brightly reflecting target. Now, look at the carbon-dissolved organic material spectrum. has very low reflectance across the range, and when you look at one of those lakes, it appears black. Water absorbs strongly in the near-infrared, so it has very low values there, and while still quite dark, it does reflect more strongly in the blue. Now look at chlorophyll in green. It has a peak in the green region of the visible spectrum, between 500 and 600 nanometers, and then flattens towards the near-infrared.

26:16 Speaker 1: Note the small peak at around 680 nanometers. This is due to phytoplankton fluorescence.

26:24 Speaker 2: These spectra are.

26:25 Speaker 1: Hypothetical illustrations of what a pure pixel would look like for each of the examples at the water's surface. Now imagine mixing these spectra so that the peaks and troughs you see here are blended.

26:36 Speaker 2: That's what a natural sample would look like.

26:38 Speaker 1: And that's what's used by remote sensing algorithms to estimate concentrations of the different types of material in the water. The spectra that you're seeing here assume no effect of the atmosphere. In reality, though, the light that a sensor measures has traveled a long and complicated path before arriving at the sensor. Sunlight entered the Earth's atmosphere, traveled in a mostly downward path towards the surface. The light hit the surface, penetrated the surface, all while being absorbed and reflected across the spectrum in different ways to give the light its spectral shape once it's re-emitted from the surface.

27:13 Speaker 1: That remitted or reflected light then traveled upwards through the atmosphere, back out the top of the atmosphere, where it's finally measured by the satellite sensor. In addition to suspended sediments and carbon dissolved organic material, what are some of the other quality indicators satellites can observe or be used to derive?

27:34 Speaker 2: Some of those include.

27:35 Speaker 1: Sea surface temperature, Horothel A, salinity, the diffuse attenuation of light, and euphotic depth, or the depth where light attenuates to about 1% of the amount found at the surface. Unlike land surfaces, water itself absorbs strongly in the red, near-infrared, and shortwave-infrared regions of the spectrum. Even in highly turbid environments, water is opaque or black in the shortwave-infrared region. This can be used to our advantage when analyzing coastal regions as these wavelengths can be used for atmospheric corrections.

28:14 Speaker 1: Other water constituents like collared dissolved organic material absorb strongly in the blue region with practically no absorption in the red and other longer wavelengths. Because chlorophyll and other photosynthetic pigments absorb strongly in the blue and red regions and have a higher reflectance in the near infrared, pixels containing floating algae can be different from water pixels.

28:39 Speaker 2: This is the essence of using some of the.

28:41 Speaker 1: Spectral indices like the floating algal index to monitor the presence or absence of floating algae, such as sargassum or kelps in coastal areas. But since most of the indices that we mentioned today are dedicated to photosynthetic targets, first we'll touch briefly on one of the indices developed to assess turbidity in inland and coastal waterbodies. As kelp blades occur at or near the surface of water, indices usually developed for land purposes, such as the Normalized Difference Vegetation Index, or NDVI, can be applied to the study of kelps to estimate the health of these magnificent organisms.

29:21 Speaker 1: The graph on the right shows data from a recent paper where you can see a marked difference in reflectance of an unsubmerged kelp blade, particularly in the near-infrared, compared to one just under the surface or from a spectra collected in an area with sparse cover. Notice that even in the first centimeters of water, there's practically no signal in the near-infrared in the sparse or submerged reflectance curve. Another index, such as the floating algal index, which we've shown before, could also be useful to analyze the conditions of these kelps of the water surface.

29:54 Speaker 2: The floating algal.

29:55 Speaker 1: Index uses the red edge region of the spectrum and the short-waved infrared region and was specifically developed for MODIS data.

30:05 Speaker 1: Now to move on to the normalized difference turbidity index.

30:11 Speaker 1: So this was originally developed by Lecaux et al. in 2007 for water quality assessment in ponds and small inland water bodies. It used the Spot 5 Level 2 images at a 10-meter spatial resolution and has been applied with some success to other water bodies using other sensors, such as the Landsat 8 OLI and the Sentinel-2 MSI. The figure on the right is from one of the most recent NASA Ames Development Program projects, which aimed to study the impacts of water quality on the distribution of eelgrass in the South Slough National Estuarine Research Reserve in Oregon.

30:52 Speaker 1: Here, it was assumed that most of the turbidity in the slough was caused by sedimentation and organic debris. So in summary, the spectral index you choose will depend on the nature of the phenomenon you wish to observe. Some spectral indices are better than others when dealing with particular conditions, such as the influence of bare soils or turbid waters. Spectral indices provide quantifiable metrics that are useful in tracking the ways in which certain phenomena are distributed across space and time.

31:29 Speaker 1: And now, we'll move to our demonstration where we'll actually get into QGIS and look at a practical way to calculate some of these indices we've spent so long discussing. First, begin by opening QGIS. Now, click on the icon in the upper left corner to open a new blank project.

32:15 Speaker 1: You may notice that this right-hand window has disappeared. If that's the case, you can go into processing and activate your toolbox here or hit the gear icon. Now we need to find our data. I've stored mine on the desktop for ease of demonstration, but usually this would be a folder set up in whatever way you prefer to keep your projects organized. To load the data, I simply drag and drop the harmonized Landsat Sentinel data that I'm interested in to the map pane in the center. Here, I'm loading the HLS data, which covers a portion of the Mendocino Fire Complex in California in 2018.

33:11 Speaker 1: This scene includes Clear Lake, which gives us a great opportunity to look at vegetation, fire, and water targets all in one image. Since we've already discussed the calculations behind some common indices in the presentation portion of this training, we'll only focus on NDVI, NBR, and NDTI in this demonstration. That covers the domains of vegetation, fire, and water. As you'll see, we can achieve calculating all of these spectral indices using only the QGIS raster calculator tool. If you were interested in a different index, such as one of the others previously discussed, you would simply replace the equations that I'll be showing here with the equation for the index you wish to run.

33:55 Speaker 1: Now that we have our data loaded, I'm going to do a bit of cleanup to make things easier to read going forward. First, I'm going to select all of the HLS bands in my layers menu.

34:15 Speaker 1: then I'm going to group them. And now I will make that group mutually exclusive so that we're only seeing one band at a time. Now I can collapse all of these dropdown menus. Finally, I'm going to expand my browser window just so that I can see the band designations at the end of the file name. And with everything now looking nice and tidy, let's calculate some spectral indices.

34:52 Speaker 1: Now locate the search bar in the processing toolbox. Here you can type in raster calculator. There should be an option for raster calculator listed under raster analysis. Click on the raster calculator tool to open the window. We'll start with calculating NDVI, which you may recall uses the red and near-infrared bands. On HLS data, that corresponds to bands 4 and 8, respectively. Click the ellipses to the right of the input parameters field, and then select bands 4 and 8. Once those two bands are selected, click OK.

35:56 Speaker 1: Now back on the main raster calculator tool window, click the purple epsilon to the right of the next line. It's the symbol that looks like a backwards three. Here we can input our spectral index equation. Remember the NDVI is calculated as near infrared minus red all over near infrared.

36:20 Speaker 2: Plus red.

36:57 Speaker 1: Click OK to bring us back to the main raster calculator window. All of the other options here are optional, but if you'd like to save the results to your computer for later use, you can set a file name and destination using the ellipses to the right of the calculated input field. Click Run to execute the NDVI calculation.

37:35 Speaker 1: A result is automatically added to the map and we can see some interesting features. I'm going to apply a color scale to make things a bit clearer. As you can see, the areas of rectangular farmland have high NDVI values. This is what we would expect of green vegetation. The water body, Clear Lake, in the lower left is showing low NDVI values, which makes sense as water is a strong absorber of both the red and near infrared wavelengths. Most strikingly, we can see that there is a clear lack of vegetation to the north of the lake where the fire is already burned.

38:34 Speaker 1: However, A good deal of the fire damage is obscured by smoke, which is a weakness of NDVI for scenes where fire is actively burning. Smoke tends to have low NDVI values because it's such a bright target that the denominator of the NDVI equation is dominant. Also note that along the margins of this lake, there's a slightly higher NDVI value. This is most likely due to surface or near-surface aquatic vegetation. Now we can move on to the calculation of the normalized burn ratio.

39:13 Speaker 1: Remember from the presentation that the normalized burn ratio uses the near-infrared and shortwave infrared bands. Otherwise, the NBR calculation is identical to the NDVI calculation. Open the raster calculator tool again. Now, select band 8 and band 12 as your input layers.

39:44 Speaker 2: And click OK.

39:59 Speaker 1: Input the equation, paying attention to your parentheses. This is shortwave infrared minus near infrared all over shortwave infrared.

40:09 Speaker 2: Plus near infrared. and click OK.

40:58 Speaker 1: You can see the distinct dark of the burn scar north of the lake. Calculating NBR values before and after a fire, you can look at the change over time and create a secondary product called the Difference Normalized Burn Index or DNBR, which helps to measure fire burn severity over an area. And now, we'll conclude by taking a look at NDTI, the Normalized Difference Turbidity Index. As with the previous two indices, this is a simple combination of two bands. In this case, we're looking for the red and green bands, which are bands four and band three, respectively.

41:39 Speaker 1: Again, in the raster calculator, Select band three and band four and click OK. Input your equation into the prompt as usual.

42:37 Speaker 2: And run.

42:47 Speaker 1: Notice that even though NDTI is an index meant for aquatic environments, we've just calculated NDTI across a scene which is mostly land. This highlights the importance of understanding what the index is meant to represent before blindly applying a calculation and attempting to interpret the results. Here we can see subtle variations in turbidity across the lake, with few sites of the margin having significantly higher turbidity, especially along the southern shoreline. With that, we hope you have a better understanding of how to calculate spectral indices in QGIS using the raster calculator tool.

43:29 Speaker 1: Remember that these three indices are only meant to serve as practical examples. And simply by modifying the bands used as inputs and the equation being evaluated, you're capable of calculating a wide variety of spectral indices that are relevant to your area of interest. On behalf of myself and the whole RSAT team, I'd like to thank you for your attention and patience during this demonstration. So today we learned about the benefits of using spectral indices to study vegetation and aquatic targets.

43:58 Speaker 1: We talked about some of the most used spectral indices for land and aquatic targets. We demonstrated how to use QGIS to do simple calculations of spectral indices for land and aquatic applications. There is one homework assignment.

44:17 Speaker 2: Which is open today. You can access that from the training webpage.

44:22 Speaker 1: Please submit your answers via Google Forms. The homework is due by March 13, 2025. To obtain a certificate of completion, you have to attend this live webinar. Attendance is recorded automatically.

44:42 Speaker 2: You need to complete.

44:43 Speaker 1: The homework assignment by the deadline. And then you'll receive a certificate via email approximately two months after the completion of this course. Here is some of our contact information as well as links to our sister programs, DEVELOP and SURVEIR.

45:03 Speaker 1: on behalf of myself and the RSAT team, I want to thank.

45:05 Speaker 2: You for your attention today. Have a great day.

45:17 Speaker 1: Okay, so question one, does the version of QGIS we have matter for this? Not really, so I did this training in 3.34.14. Although, just after writing this up, I saw that there was a new update, so now there's 3.4 out, but it should work for any QGIS versions, the one that I have, the new one, and any of them that have the raster calculator tool.

45:53 Speaker 2: So, question two.

45:56 Speaker 1: There are four constants in the EVI formula. How are these calculated? Would there be any cases where these would change? So EVI is often used in highly vegetated areas, and it was developed to correct for atmospheric conditions, soil effects, and some canopy background noise. In its development, it underwent extensive calibration, and the constraints remain the same in general. In the formula, the constant L is used to correct for canopy background, and the C1 and C2 constants.

46:26 Speaker 2: Are atmospheric corrections.

46:28 Speaker 1: G is to allow for better discrimination between different vegetation types. So, question three. For dense vegetation cover, EVI is often preferred over NDVI due to atmospheric corrections. Are there cases where NDVI would still be preferable in highly vegetated areas? So NDVI is a simpler calculation that's appropriate for moderate vegetation cover. EVI also requires that you have a separate blue spectral band, which isn't always the case. It's more common, especially on older instruments, to not include the blue, but to have red and near-infrared.

47:16 Speaker 1: Question four, is NDCI useful in inland waters such as ponds or lakes, for measuring relative nutrient loads. The NDCI was originally developed for coastal and estuarine urban regions.

47:35 Speaker 2: You can apply it.

47:36 Speaker 1: To other environments, such as inland waters, to get an idea of the concentration of chlorophyll, but it doesn't give you nutrient estimations. Also, it was originally developed using MERIS imagery, which was about a 300-meter pixel size. over land and about one kilometer over the ocean. So the spatial resolution has to be taken into account, especially if the inland pond's relatively small, is smaller than that 300 meter pixel size.

48:10 Speaker 1: So when there's land or aquatic phenomenon, we would like to develop an index to measure, how do we select the right wavelengths to calculate? Yeah, so this depends on your specific use case and the sensors that you have available to you. The resource that I've linked here shared by another participant looks like a really great resource to.

48:35 Speaker 2: Look at all of the available.

48:37 Speaker 1: Indices, or at least most of them, and the applications of those indices. So that would be a great place to start. So question six, when conducting large-scale spatial analysis, we often notice significant noise in Landsat imagery, while Sentinel data appears to have less noise. What are the primary causes of this noise and what techniques can be used to minimize it, particularly when working with Landsat or Sentinel imagery?

49:08 Speaker 2: So for this, we're using.

49:10 Speaker 1: The harmonized Landsat Sentinel-2 dataset. So it's a mix of Landsat and Sentinel properties, but each sensor has specific properties unique to that sensor and platform. Without being exactly sure what kind of noise you're referring to in that question, there are some known issues, such as on Landsat-7, There was the scanline corrector failure, which causes data gaps at the edges of the orbital cloth. Also, Sentinel-2 just has a higher spatial resolution than Landsat, so its imagery appears to be more clear simply because it has a higher spatial resolution.

49:53 Speaker 1: Question seven, which vegetation index is most suitable for estimating nutrient content in vegetation, such as nitrogen levels? I'm not currently aware of any spectral index for nitrogen specifically. Normally, the things that we're talking about when we're talking about vegetation spectral indices is things like NDVI and the leaf area index. That is not to say that there isn't A nitrogen-specific spectral index, it's just not one I've come across before.

50:35 Speaker 2: So, question eight.

50:37 Speaker 1: How can the floating algal index be effectively applied to monitor algal blooms in small.

50:43 Speaker 2: Inland ponds using Sentinel-2 data?

50:45 Speaker 1: Are there any specific pre-processing steps or spectral adjustments required to enhance its accuracy in these environments? So Sentinel-2 has different bands than those that are used in FAI, so you may need to adjust for those factors. Do keep in mind that many, if not all, of these indices were originally developed after extensive research and into data collection to come up with the best approach or index. Having said this, you can play with different Sentinel-2 bands to see if there's a better combination that applies to your research questions.

51:18 Speaker 1: The other thing to be looking at is the bandwidth and band centers. And sometimes you can get reasonable approximations of bands on sensors for which the indices were not originally developed. Would the floating algal index be suitable for monitoring kelp farms or large macroalgaes?

51:48 Speaker 2: Yeah, absolutely.

51:49 Speaker 1: So kelps have a similar signal to that of sargassum. So as long as it's detecting kelp at or near the water surface and you're not running into issues with water attenuation, it should be okay to use. I am going to skip research specific questions, but thank you for those.

52:11 Speaker 2: We will be.

52:12 Speaker 1: Answering them in the doc to the best of our abilities, but The answer to most of these is really it depends, and you're more likely to be an expert in your specific study area in question than we are coming in sight unseen.

52:30 Speaker 3: Justin, this one, may I add a little bit on that? Sure.

52:34 Speaker 1: Sure.

52:36 Speaker 3: Yeah, most of the time, also be aware that for particular reasons, especially when you're dealing with water bodies data, the water column changes constantly, right? So usually... you will need to, if you're using a particular set of imagery or so, you will need to do a lot of in-situ validation. And for that, particularly for water bodies, ideally, you will need to be on site at least within one or two hours, plus or minus, of the satellite overpass so that you're sure that what you're capturing there the image is actually what relates to the conditions that are happening during that particular time.

53:36 Speaker 3: And if you leave it for another day or so and there's no overpass, probably the conditions in the water have changed completely. Thanks, Justin.

53:50 Speaker 2: Yeah, thank you, Juan. So on to.

53:56 Speaker 1: Question 11, what are the pros, cons of modifying a particular index to your study? The choice of index is dependent on several factors, including the research question, the data you have available, and the particulars of your study area. Each index has its particular strengths and limitations. With the example of assessing vegetation, you may choose EVI over NDVI in a case where the area is highly vegetated and you need that extra sensitivity. But as we said before, you may be in a situation where you don't have a blue band, so you would have to rely on NDVI instead.

54:35 Speaker 1: So there are a lot of considerations here. Also, as we said in the question eight, asking about the floating algal index, which is not designed for Sentinel-2, but can be reasonably used with Sentinel-2. You need to be considerate of the bands you have available and how closely those match the data for which these indices were developed.

55:11 Speaker 1: Question 12, yeah, I already covered this in brief. I'm curious to get the instructor's opinions on this index and site. It wasn't one that I had seen before, but it's definitely one that I will be using because it, just at cursory glance, appears to have a pretty good database of indices and appears accurate at first glance.

55:38 Speaker 2: So yeah, thank you for sharing that.

55:43 Speaker 1: So question 13. What's the maximum depth of water, for example, of seas that remote sensing can be applied to derive any water-related index? Would different depths interfere with the results? For example, if one field of underwater kelp is about 20 meters deeper than the other, would the reflectance values indicate one is healthier.

56:01 Speaker 2: Than the other due to.

56:02 Speaker 1: Different rates of light absorption? So remote sensing is usually applied to study the water surface. Part of the reason for this is that the absorption of most of the visible wavelengths happens rapidly in the water column. So the example we gave for kelp shows exactly this. Even slightly deep kelp branches can barely be detected. So most likely any kelp that's something like 20 meters wouldn't be detected at all. That is one of the limitations of looking down with airborne or satellite data. We can only really see the surface and a little bit of the subsurface in water.

56:41 Speaker 1: So when comparing vegetation changes over time, is it advisable to use the dry summer months or the rainy season months? So, yeah, that would depend on your vegetation of interest and the question.

56:55 Speaker 2: You're trying to answer.

56:56 Speaker 1: If you're comparing seasonal changes, the same vegetation, you may be comparing values between seasons. On the other hand, if you're looking for a year-to-year comparison, you'll still want to be consistent for the comparison timeframe. So if you're looking year-over-year, you want to make sure that you're getting imagery in the same part of the phenological cycle for your.

57:23 Speaker 2: Species of interest.

57:26 Speaker 1: You may choose the dry season if you're concerned about the influence of drier or drought conditions. Similarly, you might choose the rainy season if you're interested in things like looking at ephemeral pools. If you, for example, have an interest in the wildlife that springs up in ephemeral pools.

57:54 Speaker 1: What basic pre-processing of the imagery is needed for calculating the index? Does every index calculation require some sort of pre-processing or imagery product can be used directly without any pre-processing? Yeah, so there are different data levels provided. So level two imagery is going to be imagery that's already been atmospherically corrected. So when referring to Landsat data specifically, use level two images. They're processed at a higher level, generally considered more accurate and appropriate for spectral index calculations.

58:30 Speaker 1: So your level one products are the things, directly as they come from the sensor, level two products have had additional.

58:38 Speaker 2: Care taken to ensure that they're correct and corrected.

58:46 Speaker 1: Question 16, we don't have an answer in the doc yet, but can we calculate these indices using hyperspectral images? Yes, and there are indices specifically developed for use with hyperspectral sensors. If you wanted to do something like in DVI, again, you're looking at your band centers and your bandwidths to see make sure that you are getting the parts of the spectral signature, that spectral response curve, that correspond to the quantities you want to measure. So for example, with vegetation, you're interested in the red and near-infrared reflectance.

59:27 Speaker 1: And multispectral data Those are generally wider bandwidths, collecting a larger part of that section of the electromagnetic spectrum. With hyperspectral data, you're taking very narrow slices, so you need to be careful that you're selecting the correct wavelengths. Is there a possibility with one of these methods to find small water bodies such as water sources in forest, maybe with some data taken in winter and when most of the trees have no leaves.

1:00:03 Speaker 2: Located in Switzerland? Yes, absolutely.

1:00:08 Speaker 1: Again, because we are always looking down on the earth, your suggestion to look for these things in the winter when you're not obscured by tree canopy is a great suggestion.

1:00:23 Speaker 2: There are specific water.

1:00:25 Speaker 1: Indices that use the sort of dark optical properties of water to identify it, as opposed to dirt and vegetation. Also, maybe of interest in Switzerland, there are indices for snow as well, as you can imagine. Is there an option to calculate, for example, in DVI for a few images at the same time, or does it need to be done separately for each image? Yeah, so in QGIS, you'll need to calculate each image separately.

1:01:09 Speaker 2: To maintain the index values.

1:01:13 Speaker 1: However, you can use a platform like Google Earth Engine to process multiple images, which can be compared. For example, this could be useful on something like the NBR to get a multiple D-NBR, the difference normalized burn ratio, to see burn severity over time and vegetation recovery after a fire. In QGIS, there is a model builder, which allows you to create these sort of processing pipelines There is also an option to control QGIS and all of its underlying utilities with Python. So it really just depends on what you're comfortable with in taking this from a single calculation to more of an automated system.

1:02:06 Speaker 1: Unfortunately, the model builder was outside the scope of this training.

1:02:13 Speaker 2: But maybe that's something we can consider for the future. I would love to talk about that.

1:02:25 Speaker 1: When applying these indices using Sentinel-2 data, is downscaling or upscaling the resolution necessary for better accuracy? Is QGIS sufficient for processing, or is it recommended to use SNAP for optimal results? I will answer this in reverse order. QGIS is sufficient for your processing here, assuming you have the hardware necessary to do the up or down scaling. But the question of whether or not you need to up or down scale is really more dependent on your study site and what other data sets you would want to be using in addition to the Sentinel-2 derived spectral indices.

1:03:12 Speaker 1: Generally, this comes up more as a question of matching the resolution of datasets that are coarser than Sentinel-2, although you could also find yourself in a situation where you have very high resolution data, like something from a drone flight which is higher resolution than Sentinel-2 data. In that case, you would need to look at doing the upscaling or downscaling as appropriate to bring them into alignment. Are there any weaknesses or trade-offs when using NVR? Are there any other indices for measuring fire impact?

1:03:59 Speaker 1: Yeah, so there are quite a few other indices for measuring fire impact. My background is actually mostly in fire. So you have the burned area index, for example, which highlights burned area using the near-infrared.

1:04:19 Speaker 2: Spectrum, similar to how NDVI is calculated.

1:04:25 Speaker 1: The normalized burn ratio, which is designed to highlight burn areas in large fire zones, but also gives you the first step to a DNBR. And DNBR values can be used to assess your burn severities, as well as the repopulation of.

1:04:50 Speaker 2: Burn scars by vegetation.

1:04:52 Speaker 1: So you can look at the fire recovery. and succession in burn scars using a DNBR. And then the burning index is related to the fire behavior. Other things you can look at when looking at fire impacts are things that consider secondary.

1:05:20 Speaker 2: Qualities of the fire. So that's things like the thermal properties and the fire radiative power.

1:05:30 Speaker 1: So that would be something to look into there. Best index to differentiate vegetation types, namely woody shrub and herbaceous. So this is a little outside of the capabilities of simple spectral indices. Because what you're asking about becomes a classification problem. Of course, spectral indices could be.

1:06:02 Speaker 2: Used as an input to.

1:06:05 Speaker 1: The sort of classification algorithm. But in general, the best way to use a spectral index for something like this would be to look at the phonology of your separate vegetation types of interest. and see if there is a time during the year, during their phenological cycle, where you can capture differences. So say one has leaf out earlier than the other in the year, you can use the timing to separate these things rather than looking at their remote sensing properties. There is work that has been done and is being done to improve our ability to differentiate species using hyperspectral remote sensing data.

1:07:02 Speaker 1: As a plug for another RSET training, we have talked about invasive species monitoring with hyperspectral data, and that information may be relevant to this question. Is Landsat 8 or Sentinel-2 better for taking NDVI decadal and when facing clouds when creating NDWI, the normalized difference water index, by using Sentinel-2? The Landsat archive is longer than Sentinel-2. That is just a fact of.

1:07:46 Speaker 2: The history of these sensors.

1:07:48 Speaker 1: So the Landsat archive has imagery going back to the 80s. Whereas Sentinel-2 begins in 2015. So for decadal comparisons, generally Landsat would be the most appropriate.

1:08:03 Speaker 2: But Sentinel does.

1:08:04 Speaker 1: Have that higher spectral resolution. So for the last decade, 2015 to 2025 comparison would be better. The sensor is simply newer and has slightly better spatial resolution. That's going to depend on the sort of time scales that you're looking at, which sensor you end up choosing.

1:08:36 Speaker 2: Note also, you'd say.

1:08:38 Speaker 1: Landsat 8, but if you're looking back over a long period of time, you do run into a little bit of that issue where Landsat 7 had the scanline corrector failure. So there are gap-filling algorithms to help you get around that.

1:08:59 Speaker 1: For a multi-year NDVI time series, I would expect changes in the index due to seasonality. Could this be eliminated by comparison between a control area versus the area of interest? Also, analyze other influences.

1:09:13 Speaker 2: Like human impact.

1:09:15 Speaker 1: Hope that makes sense. Yeah, so if you're looking at a multi-year NDVI time series, you want to make sure that you're aligning the timing of the imagery you're using. So that's going to depend on how closely you can get imagery to that one-year mark to do year-over-year comparisons so that you're looking at the scene when it is in the same stage of its phenological cycle. Obviously, you would see differences in DVI, a measure of green vegetation, during the spring and summer where there is leafy vegetation versus winter imagery where you're not going to see as much green, if any, at all.

1:10:15 Speaker 1: What precautions should be taken when the water surface is not uniform with waves or granulations which reflect the light in a very direct.

1:10:29 Speaker 2: And intense way?

1:10:32 Speaker 1: Juan, if you would take this one.

1:10:34 Speaker 2: Juan, I'm primarily looking at land.

1:10:37 Speaker 1: Juan is our resident expert on all things ocean.

1:10:42 Speaker 3: Yeah, sure. Thanks, Justin. Of course, the, you know, waves will always be an issue when you're dealing with ocean data in particular or just, you know, big inland water bodies. And Usually it's not straightforward to deal with waves. A lot of times what people do is that, and it's not only waves, you also need to deal with sun glint effects also on the water surface, depending on the sun angle and other factors. So a lot of times with Waze in particular, an alternative is to probably look for another image from a different date where when win was not a factor and you might have a much clearer data set.

1:11:41 Speaker 3: You can also, an alternative is that similarly to what people do with cloud masking, you can mask the waves in that area and then work with the rest of the image. For Sunglin in particular, there have been algorithms that have been developed for different purposes. different sensors to deal with songlings and to eliminate most of the songlings from an image. So that's the short answer. It's quite a complicated process.

1:12:26 Speaker 2: Yes, thank you.

1:12:28 Speaker 1: And just a reminder, if you've put questions in, we will post this question and answer document. So if we don't get to your question or we're still answering it here, you'll have our full answers in due time.

1:12:44 Speaker 2: So I'm going to.

1:12:46 Speaker 1: Jump around a bit. Question 27, how distinct are spectral signatures of different vegetation species? Can they be used for tracking succession or the movement of specific species? As I mentioned before, we have recently done a training on hyperspectral data for invasive species monitoring. So that would be a great resource for you to look at there. But in general, hyperspectral data, it gives you the extra spectral resolution that you need to really get at these small differences in the spectral response curves of different vegetation species.

1:13:32 Speaker 1: How have these indices performed with PACE data? How much do they differ by sensor? Again, to do the RSAT plug, we recently had a training called Introduction to Plankton Aerosol Cloud Ocean Ecosystem, the PACE, hyperspectral observations for water quality monitoring. That happened near the end of last year, so I would refer you to that for more information about PACE. I will admit I am not an expert on PACE, but there are some really exciting things coming out with the PACE data.

1:14:12 Speaker 3: And Justin, also this one again, also remember that PACE is, for our audience, remember that PACE is a hyperspectral sensor. So as you mentioned earlier, you know, you have to consider that these have much thinner bands. than they are the usual multispectral sensors that we're more used to work with. So that is a factor to consider when you are exploring the use of data sets such as that of PACE with some of these indices.

1:14:51 Speaker 2: Yes, thank you.

1:14:54 Speaker 1: Lost my Q & A page. There we go. Question 31. Different articles suggest that SAVI is better than EVI. What's your insight in this construction? SAVI is the Soil Adjusted Vegetation Index. EVI gives you a higher top end. So that is to say that it is not as likely to become saturated by a high density of leafy green vegetation as something like NDVI. The Savi is soil-adjusted, so it's more appropriate for sparsely vegetated areas. So that's not to say that Savi is better than EVI or EVI is.

1:15:46 Speaker 2: Better than Savi. They each have their own specific use cases.

1:16:01 Speaker 1: Question 32, how does persistently poor air quality affect indices such as NDVI? And I'll answer that question first. This seems to be a question in two parts. So when we are using remote sensing imagery, of course we're always concerned about the effects of the atmosphere.

1:16:23 Speaker 2: On the indices.

1:16:27 Speaker 1: So we apply atmospheric corrections to try to get around some of these issues. But of course, we are limited by the windows of electromagnetic radiation that our atmosphere actually releases back to our sensors after reflectance. So anything that is in the air that is going to narrow or muddy our ability to see these spectral bands is going to be certainly a concern. And second part to this, can FAI be used to distinguish algae from water hyacinth or cyanobacteria?

1:17:15 Speaker 3: Again, some of that.

1:17:20 Speaker 1: I was just going to recommend Pace again.

1:17:26 Speaker 3: I would say that, you know, the first FAI was particularly one of the reasons behind, you know, using FAI for Sargassum and also for Kelps is that, you know, it uses the longer bands of the well, the red bands of the visible spectrum, right? And as we know, you know, salgassum and kelps, they usually have this brownish color. So it's easy to distinguish them with FAI. I would... I would recommend our participants to probably explore, instead of using that one, maybe using MDAVI, which we also covered, the Anomalous Difference Aquatic Vegetation Index, for distinguishing the water hyacinth in particular for cyanobacteria, though.

1:18:26 Speaker 3: because cyanobacteria, they have, you know, like a cyanine, and then it reflects around the 620, 625 or so nanometers, if I remember correctly, in the visible spectrum. There are specific indices to detect cyanobacteria in inland water bodies. One thing that we can do here, Justin, is that actually I was part of a paper that was published around 2014, 2015, so about a decade ago, but it's still relevant, where the main author is Dr. Ray Scudella from UC Santa Cruz. And he, actually in that paper, we explored, it was done in some of the inland lakes here in California, specifically in Monterey Bay.

1:19:20 Speaker 3: And he explored the use of, particularly of a cyanobacterial index. So we can obviously include the reference here for the benefit of our participants.

1:19:34 Speaker 2: Thanks. Thank you, Juan.

1:19:40 Speaker 1: Jumping back up to question 29, in addition to NDBI, what are other indices available to determine built-up areas for Landsat images? There are other indices that you can use, but it's going to depend on what your background against which you wish to determine built-up areas is. To wit, the hardest part of doing the examination of built-up areas is identifying the difference between things like bare soil and concrete. They do have slightly different spectral signatures, but for something like Landsat, where the bandwidths are fairly broad, sometimes you miss out on these little nuances that allow you to do that more fine-grained picking apart of the differences.

1:20:44 Speaker 1: Again, I would say that this is sort of getting into becoming a classification-type question, and you can use these indices as inputs into a classification algorithm, but that would be the way to go in an accurate discrimination between pixels which are mostly built up and those which are mostly your background, be that vegetation or bare ground, sand, etc.

1:21:13 Speaker 1: Any index or recommendation for soil organic carbon estimation? This is not my area of expertise. Because we're only looking at the surface of the soil and.

1:21:29 Speaker 2: We can't see any deeper, I.

1:21:35 Speaker 1: Think it would be exceedingly hard to get at soil organic carbon at depth. Typically, when we're looking at this for something like peatland, to look at the soil carbon in high organic carbon soil types like peat, We're looking at other indicators like the soil water content, which is something that we can get at using something like SMAP, which is the soil moisture active passive, which uses active and passive remote sensing to peer a little deeper into the soil. And with that, I think I've covered all of the questions that are in the document.

1:22:35 Speaker 3: There was, Justin, there was a question in the chat about when the homework would be available. I believe it will be available today. And then they will have until March 13th. to submit the answers. It's relatively simple, so it shouldn't take you long. But yes, it will be available later today.

1:23:02 Speaker 1: Yes, thank you for that. I hadn't been watching the chat.

1:23:10 Speaker 1: And so just finally, I want to thank you all again for attending and for your questions.

1:23:16 Speaker 2: And your patience.

1:23:18 Speaker 1: And of course, I would like to thank everyone else on the RSAT team who makes these trainings possible.