Getting More From Your NDVI Sensor

Learn how NDVI is being used in current research and demonstrates how to overcome some of the NDVI’s limitations.

In this webinar, see examples of how NDVI is being used in current cutting-edge scientific research.He demonstrates how to overcome some of the NDVI’s limitations by tweaking the raw outputs from NDVI sensors.

Specific topics to be covered:

  • Background theory and measurement of NDVI
  • How NDVI can be used for canopy applications
  • Limitations of NDVI
  • Methods for fine-tuning the raw components of NDVI to overcome limitations

Next steps


Our scientists have decades of experience helping researchers and growers measure the soil-plant-atmosphere continuum.


Dr. Steve Garrity, METER Group environmental scientist


See all webinars

NDVI & PRI–Measurement Theory, Methods, and Applications

Master the basics of NDVI and PRI.


PRI—Why You Should Measure It

Learn the basics of the Photochemical Reflectance Index (PRI).


Leaf Area Index (LAI): Theory, Measurement, & Application

How to use PRI and NDVI to monitor environmental conditions that adversely affect plant growth.


Case studies, webinars, and articles you’ll love

Receive the latest content on a regular basis.


Thank you for attending today’s virtual seminar entitled Getting More from NDVI sensors, presented by Dr. Steven Garrity, Canopy and Atmosphere Product Manager at Decagon Devices. Dr. Garrity.

Hi, I’m Steve Garrity, the Canopy Product Manager here at Decagon Devices. Thanks for joining us for today’s virtual seminar. So before I get started, just to let you know, as I’m going through the slides, as I’m going through the seminar, if you end up having some questions, feel free to type those in. And then at the end of the seminar, we’ll take some time, and I’ll get to as many of those questions as possible. And any that I don’t get to, don’t worry, I will be sending you an email response shortly after the seminar is over.

Okay, so today, the topic of discussion is NDVI. And specifically, I’m going to talk to you a little bit about getting more from your NDVI sensor. So the figure that I’m showing here to start out with is a whole lot of different sampling platforms. And each of those sampling platforms has a sensor on it. We could imagine that that those sensors are NDVI sensors. And you can see then, that there’s a whole range of scales that we can sample NDVI across both in space, as well as in time. You know, we have sensors that are up in space, and they’re sampling the entire Earth’s surface. But we also have handheld small sensors that can be down, you know, measuring individual plants or even leaves. So just, you know, the broadest way that we can think of NDVI is here, so these data are actually obtained from an Earth orbiting satellite. And you can see that areas of high vegetation also have high NDVI values, and high NDVI values are represented by the dark green colors across the globe. Conversely, you can see that areas where it’s relatively dry and there’s not a lot of vegetation, have low NDVI values, and they tend to look fairly brown. So you can get a sense, even if you don’t know a lot about NDVI, you can get a sense that it is sensitive to the amount of vegetation cover that is present across the Earth’s surface.

So scaling down quite a bit now to probably where most of us work at the plot level or somewhere out in the field in a forest or in an agricultural setting. Let’s just take this example here and explore how NDVI might be useful to us. And so what I’m showing here, and what this figure shows is a successional gradient. So you can see that maybe we start at time zero with a bare patch of soil, maybe a few forbs or annual grasses that are growing. And if we leave that patch of ground alone for enough time, eventually, the vegetation will change. So we’ll go through a different set of several stages where eventually shrubs will take over, and if we leave it for a long enough period of time, we might even get a forest that grows there. So that’s over many, many years. Or it could be across very big spatial areas, where we move from grasslands to forests. You can also see that maybe we’re in an agricultural system, where every year there’s this turnover of vegetation. So we start with a bare field, bare soil, we plant the seed, eventually a plant emerges and it matures over time, and that cycle repeats itself every year. But there might be some room here for NDVI to come in and help us quantify the temporal variation or the temporal growth, or the growth of the canopy that occurs over time, as well as the spatial dynamics that occur across landscapes.

So backing up just a second, talking about where does NDVI come from. So the figure that’s plotted up here, it has a lot going on, but I’ll quickly orient you. So along the x-axis, I’m plotting wavelength, and this is wavelength of light within the electromagnetic spectrum. And so I’m plotting from 450 to about 950 nanometers. And that covers both the visible region or most of the visible region, as well as a portion of the near infrared. On the y-axis, then is percent reflectance. And so we can imagine that this is a typical reflectance spectrum that we would get if we were looking at green vegetation. And so you can see that line that’s continuous, that has an arrow pointing to it that says hyperspectral. That’s what we would expect to get from a spectral radiometer. And you can see that reflectance is typically low in the blue region, higher in the green region, again lower in the red region, and then it really shifts dramatically as we cross from the visible to the near infrared. You can see there’s a lot of reflectance in the near infrared region. So you can also see that there’s two vertical bars there, both of them labeled NDVI, so this gives you an idea of where a typical NDVI sensor is looking or is measuring within the spectrum. You can see that we have one band typically in the red region, and then the other band is in the near infrared region. And so what we’re trying to do here is capitalize on the large difference between what happens in the visible region of the spectrum with what happens in the near infrared portion of the spectrum. And it just so happens that plants reflect near infrared really, really strongly. And the NDVI takes advantage of this. So you can see in the two images on the right hand side of the screen, they’re both of the same area, the top image you can see is plotted or is displayed in true color, meaning it’s three bands, blue, green, and red, whereas the image below, of the same area, is what we call a false color infrared image. And so in this case, the three bands that are being displayed are blue, green, and then instead of red, we’re using the near infrared in its place. And so you can see that everywhere where it looks extremely bright, or is very bright red, that’s because there’s a lot of near infrared reflectance, which is very typical of green, or healthy vegetation.

So how do we actually calculate the NDVI? So you can see that the NDVI equation, we’re plugging in the reflectance, which is signified by ρ [rho], the reflectance in the near infrared, minus the reflectance in the red, all over the sum of those two quantities. And you can see that NDVI values typically range between negative one and one, with low values indicating low amounts of vegetation, and higher values indicating higher amounts of vegetation. So the way that we calculate the percent reflectance is we have to quantify both the upwelling radiation, so the radiation that’s striking the canopy and then reflected back to our sensor, as well as the total amount of radiation that’s incident on a canopy. So the ratio of those two give us percent reflectance in each of the bands. And the reason NDVI is formulated with red and near infrared is because the red really keys in on chlorophyll absorption, whereas the near infrared is sensitive to canopy structure and the internal cellular structure of leaves. So you can imagine that as we add more leaves to a canopy, there’s more chlorophyll, there’s greater amounts of structural complexity, so we would expect low or decreasing amounts of red reflectance, and higher and higher amounts of near infrared reflectance.

So a few common applications of NDVI that are pretty ubiquitous, people use NDVI to calculate Leaf Area Index. Related to that they use it to calculate the fractional light interception or F PAR of a canopy. Some people associate NDVI with biomass or yield of a crop. And just like the image I showed towards the beginning of the entire Earth, people also just use it to get a sense of general patterns of greenness or general patterns of where vegetation occurs or how much vegetation is in a particular location. So thinking back to this example that I’ve shown where we have very sparse vegetation scaling to a dense, mature canopy, you can imagine that along that gradient, whether it’s across space or time, there’s quite a bit of variability in leaf area index. And so you can see that the reflectance spectrum or spectra that I have plotted in the upper right, change depending on how much Leaf Area Index there is. So you can see at very low LAIs, there tends to be a fairly flat reflectance spectrum, but at the other extreme where there’s a lot of leaf area, we tend to have very strong absorption of red light by chlorophyll, and you can see that there’s also ever increasing reflectance in the near infrared region around 800 nanometers.

So there are a couple of limitations that people encounter when using the Normalized Difference Vegetation Index or NDVI. And they both tend to occur at the extremes of the spectrum. So anytime that there’s very low vegetation cover, so like I’m showing here, where it’s a corn canopy that’s young, and so the majority of the canopy is actually — or the majority of the scene is actually soil, NDVI tends to be sensitive to that soil, and so that can confound measurements of the amount of vegetation there. On the other extreme when there’s a large amount of vegetation, such as a tropical forest, NDVI tends to saturate when there’s a lot of leaf area index. So a tropical forest NDVI is not going to be very sensitive to small changes in LAI because LAI is already very high.

So is there anything that we can do to overcome both the sensitivity of NDVI to soil, as well as improve the sensitivity of NDVI to high amounts of leaf area index or LAI? One of the things that hinders us here is that we only have two bands to work with in the red and the near infrared regions of the spectrum, so we’re already limited by the amount of information we have. However, several solutions do exist, and I’d like to spend the rest of my time talking about a few of those. So you can see in this figure that it’s a transect, so this study was taking spectral measurements of these different vegetation indices across a transect where it was bare soil. So you can see as we move from dry clay loam to wet clay loam, we see a very strong response of NDVI just simply due to the wetness of the soil, and that’s something we wouldn’t want to have occur if we’re interested in measuring vegetation. We’re not interested in having an index that’s sensitive to changes in soil or changes in soil moisture. You can also see that there’s a few other indices that are plotted there, and they tend to have much lower sensitivity to variations in the soil across the transect.

So the first one of those is the Soil Adjusted Vegetation Index, or SAVI. And you can see that SAVI is similar in its calculation — the equation is very similar to NDVI. One of the things to note is that it incorporates the exact same two bands that the NDVI does, the near infrared and the red. The only thing that’s different there is it has this L parameter, and the L is just a soil adjustment factor with values that range anywhere from zero to one. And the way this works is that when vegetation cover is 100%, L is zero because there’s no need for a soil background adjustment. However, when vegetation cover is very, very low, that L parameter will approach one. And what people do because this is difficult to measure exactly how much vegetation cover you have there — usually you’re using the NDVI to tell you how much vegetation is there — but in this case, we’re trying to use SAVI, we’re trying to modify the NDVI so it’s not sensitive to soil, and we’re doing that by guessing beforehand what L should be. And it’s pretty common practice to set L to an intermediate value of 0.5. People have done this and it’s worked fairly well for them. And you can see here, that the Soil Adjusted Vegetation Index or SAVI tends to have much lower sensitivity to the soil background. So stepping through the figures in the upper right, you can see that along the x-axis, we have percent green cover, scaling from low to high. And you can see that as we go from low to high green cover, NDVI increases as we would expect. However, you can also see that the amount of noise or the soil noise that’s contributing to the NDVI signal is pretty high until we get to canopy covers pretty close to 100%. On the lower side, you can see that SAVI is also sensitive to changes in the percent green cover in a pretty linear fashion, but across that entire gradient of green canopy cover, you can also see that the noise contributed by soil is relatively flat and somewhere around 10%. So that’s a major improvement over the standard NDVI for using this vegetation index in areas where we have low percent cover of vegetation.

Okay, so that brings us to the next vegetation index, which is the modified SAVI. And so as I said, with the SAVI you can see in the equation there, there’s an L parameter that we have to guess at, and that’s not a really clean way of handling things. And so what Key did was he iteratively fit the SAVI equation or the L parameter within the SAVI equation, until he found a universal optimum. And not getting into the math, but just he was able to simplify that SAVI equation to where there’s no longer a need for the L parameter, and the only inputs that are required are the reflectances in the near infrared, and the red. And you can see the equation up on the screen. So this was a pretty significant advance because we no longer have to guess it L — it’s actually wrapped into the equation. And what I’m showing here in the upper right is that when Key compared SAVI to MSAVI, there is virtually no difference between the two indices in terms of their sensitivity to the amount of vegetation as well as their response to the soil background. So the advantage, no longer do we have to guess at what value to set L to.

Okay, so that takes care of the soil side of things. On the other extreme, remember, NDVI tends to saturate at high LAIs. So NDVI is useful if you’re kind of in that mid range of LAIs, as long as you don’t have strong soil effects, but as soon as we get to an LAI of above about four, we lose quite a bit of the sensitivity. And you can see that that loss of sensitivity in the figures on the right is primarily due to a saturation in the red band. So you can see we’ve got two different canopies here from which measurements were taken, a wheat canopy and a maize canopy. And you can see that near infrared reflectance tends to be sensitive across the entire spectrum of the wheat and the maize canopies, but the red saturates relatively quickly, and so you can see right where the red starts to saturate, that’s where the NDVI starts to saturate. So I will interject just this caveat really quick, so yes, NDVI is, it saturates at high LAIs. However, if your purpose is really to get at the fractional interception of light, NDVI tends not to have the saturation issue. So you can see here that F PAR, or the fractional interception of photosynthetically active radiation, is nearly complete far before NDVI saturates. And this is just because canopies tend to be pretty efficient at intercepting light. And once we get to an LAI of about four, most of the light has been intercepted or absorbed by the canopy, and incremental increases in LAI don’t significantly affect the the F PAR variable much more.

Okay, so getting back to a solution for the saturation issue with NDVI at high LAI. This is probably the most simple way to handle that problem. It’s called the wide dynamic range vegetation index, and you can see that its formulation is very similar to NDVI. The only difference is we have this ‘a’ parameter. And all a is is it’s a weighting coefficient that can be used to try and reduce the disparity between the contribution of the near infrared reflectance as well as the red reflectance. And so you can see that a is being multiplied by the near infrared reflectance to try and reduce its value and bring it closer to the red reflectance value, and in doing so, it balances out the red and the near infrared contribution to the vegetation index. And so as I said, a can range anywhere from about zero to one. And this graph here shows that as we use a smaller and smaller value of a, we get an increasingly linear response of the wide dynamic range vegetation index to Leaf Area Index, which is pretty logical. It’s a pretty easy thing to wrap your head around. So I think that the drawback here is that selection of a tends to be pretty subjective, and it’s something that you have to play with on your own until you find a value of a that is optimal for your solution. I think you can tend to err on the side of a very low value of a, simply because you’re going to get closer and closer to a linear response in the WDRVI to LAI as a decreases.

Okay, the next and final vegetation index that I’d like to talk about is the enhanced vegetation index or EVI. So the EVI was designed to enhance sensitivity in high biomass ecosystems, so when we have high LAI, trying to maintain sensitivity there, but it is also trying to reduce atmospheric influences. And so this was a vegetation index that was really created for the purposes of a satellite based platform where we have a lot of, from the satellite, there’s a lot of atmosphere to look through to the ground, and sometimes the aerosols in the atmosphere can affect the reflectances in the red in the near infrared regions, and cause spurious observations. The other thing that’s wrapped in here as well is a reduction in sensitivity of the index to soil. And so the EVI is sort of a, you can think of it as a solution to both extremes, the sensitivity of NDVI to soil, as well as the lack of sensitivity to high LAI, so it solves both these issues. But if you look at this equation, the two major inputs are near infrared and the red reflectances. C1, C2 and L are all parameters that can be estimated, but the blue band is something that has to be measured. And with a lot of the NDVI sensors that are out there, they’re just two band sensors, and so you don’t actually have that information in the blue. And oftentimes with the satellites, the blue band is relatively noisy and doesn’t always have the best quality data. So there are some problems with the EVI.

Those problems though, led Jiang to try and come up with a solution. And what Jiang observed is that there’s quite a bit of autocorrelation between the red band and the blue band, and so he decided that he would try to formulate EVI without the blue band and just using the near infrared and red reflectance in what he called the EVI2. And if you’re interested in the mathematics and the way that he derived this, I encourage you to go see his or read his paper. Here though, I’m just giving you the equation in case you’re interested in using it or looking into it in more depth. But the key figures using EVI so far, I’ve shown here. On the right hand side, you can see that when Jiang calculated his EVI2 and compared it to the traditional EVI, it’s nearly a one to one relationship. For all intents and purposes, EVI2 was equivalent to EVI. So that’s encouraging that we don’t need the blue band, we can just use the two inputs, the near infrared and the red band, to calculate EVI. On the left hand side was another study that followed up, just to see if EVI2 would perform as advertised. And so the study was looking at several different areas where there was a range of LAIs in both a burned area or a recently burned area, and an area that had not burned. And so what you can see is in the top left panel, NDVI has two different relationships with LAI depending on whether it came from a burned area or a non burned area, whereas EVI was less sensitive to that difference, the disparity in the background that was caused by the burned area. So with a burned area, you have soil, you have a dark background that’s going to affect the red reflectance, but the way that the EVI2 is formulated, it decreases that sensitivity. And so you can see that there’s just a single regression model that can be fit to all ecosystems, regardless of their of their history.

So in summary, NDVI has been around for a long time. It’s been around since the 70s, at least. It is commonly calculated from a lot of the satellite sensors that have and are currently orbiting Earth’s surface. There’s a lot of value that it can add. However, there are a couple of situations at the extremes where NDVI tends to perform poorly. What I’ve shown here today is that there are several solutions to that, and what’s great about the solutions is that everything I’ve shown today uses the near infrared and the red bands, just like the NDVI does. And so you can take an NDVI sensor, obtain the raw values or the values of the near infrared and the red reflectances, and reformulate them in one of these indices, or there are several others that are available that I haven’t covered today, to try and work around some of the limitations of NDVI.

After the seminar I’ll be posting or creating a webpage where I provide all of the references that I have talked about today, as well as going into a few others that I haven’t discussed. But if you are using NDVI, and it’s in one of these extreme cases where you have a lot of soil background, or you’re in a system with really high LAI, I would encourage you to look at and see how the near infrared and red bands can be used in some type of vegetation index to allow you to do the research that you are trying to do. Alright, thanks for joining us today.

icon-angle icon-bars icon-times