A spectral vegetation index (SVI) is generated by combining data from multiple spectral bands into a single value. Usually simple algebraic formulations, SVIs are designed to enhance the vegetation signal in remotely sensed data and provide an approximate measure of live, green vegetation amount.
The rationale for spectral vegetation indices (SVIs) is to exploit the unique spectral signature of green vegetation as compared to spectral signatures of other earth materials (Figure 4.04). Green leaves have a distinct spectral reflectance pattern in the visible (vis) and near-infrared (nir) wavelengths. Reflectances in the blue and red regions are very low, with a slightly higher bump in the green. This is why leaves appear green to human eyes. In the near-infrared (nir), the spectral response of green leaves is much greater than in any portion of the visible. Other materials such as bare soil, sand, exposed rock, concrete, or asphalt, generally show a steady rise in reflectance (with no dramatic jumps) as wavelength increases from the visible to the near-infrared.
Most SVIs compare the differences between the red and near-infrared reflectances (if no red band data are available, a wider visible band can be used in place of the red). Red (or visible) reflectance is sensitive to chlorophyll content and the near-infrared reflectance is sensitive to the mesophyll structure of leaves. In a given image scene or pixel (picture element), the greater the difference between the red (or visible) and near-infrared reflectances, the greater the amount of green vegetation present. Small differences between the red (or visible) and near-infrared reflectances indicate a scene or pixel containing mostly bare soil or other nongreen materials.
Spectral vegetation indices have been found to be related to a number of biophysical parameters (variables) of interest to many researchers, including Leaf Area Index (LAI), percent vegetation cover, green leaf biomass, fraction of absorbed photosynthetically active radiation (fAPAR), photosynthetic capacity, and carbon dioxide fluxes.
Two widely used SVIs are the Simple Ratio or SR (sometimes referred to as the RVI or ratio vegetation index) and the Normalized Difference Vegetation Index, or NDVI.
The simple ratio vegetation index (termed SR or RVI) is calculated using the following formula.
Or, if no red band is available,
where NIR and RED (or VIS) are the response in the near-infrared and red or visible bands respectively. Ideally, reflectances (corrected for atmospheric effects) are used, but at-sensor values are sometimes used also. Figure 4.05 illustrates an example calculating the simple ratio.
If both the RED and NIR bands (or the VIS and NIR) have the same or similar reflectances, then the Simple Ratio (SR) is 1 or close to 1. SR values for bare soils generally are near 1; as the amount of green vegetation increases in a pixel (picture element), the SR increases. Note that the SR is not bounded; its values can increase far beyond 1. Generally, very high SR values are on the order of 30.
NDVI is calculated as follows:
or, if no red band is available,
where, as before, NIR and RED (or VIS) are the response in the near-infrared and red (or visible) bands respectively. Ideally, reflectances (corrected for atmospheric effects) are used, but at-sensor values are sometimes used also. Figure 4.05 includes an example calculation of the NDVI.
In the NDVI, the difference between the near-infrared and red (or visible) reflectances is divided by their sum. This normalization is used to minimize the effects of variable irradiance (illumination) levels. Unlike the unbounded Simple Ratio, the NDVI has a range limited to a value from -1 to 1. Data from vegetated areas will yield positive values for the NDVI due to high near-infrared and low red or visible reflectances. As the amount of green vegetation increases in a pixel (picture element), NDVI increases in value up to nearly 1.
In contrast, bare soil and rocks generally show similar reflectances in the near-infrared and red or visible, generating positive but lower NDVI values close to 0. The red or visible reflectance of water, clouds, and snow are larger than their near-infrared reflectance, so scenes containing these materials produce negative NDVIs.
When an NDVI image is calculated for scientific study, there is some preprocessing that should be done to reduce the effects of clouds, sensor degradation, sensor viewing angle, sun angle, and atmospheric effects (discussed below in Section 5.4). After preprocessing, however, the NDVI still is sensitive to external factors such as soil background that are most obvious in areas with sparse vegetation. Despite limitations, the NDVI's ability to compensate for some changing variables has made it useful for surveying phenology (seasonal dynamics) and other changes in global vegetation.
Existing spectral vegetation indices have been enhanced and others developed to improve upon parameter estimations in various applications. These include greenness, PVI (perpendicular VI), SAVI (soil adjusted VI), ARVI (atmospherically resistant VI), SARVI (soil adjusted, atmospherically resistant VI), and in addition to ratio indices, the DVI (difference vegetation index), linear combinations of visible and near-infrared reflectances, and second derivatives of canopy reflectance in two narrow spectral bands centered at 0.69 µm and 0.74 µm.
Ongoing research has shown that no single index is appropriate for retrieval of all parameters; use of an SVI must be matched accordingly to the variable of interest and with knowledge about the vegetation structure occurring in the area of interest. Using a variety of SVIs can be more effective. Utilization of additional spectral bands may help to further optimize the ability of SVIs to predict biophysical variables.
Spectral vegetation indices are designed to enhance the vegetation cover signal while minimizing the response of various background materials (existing around and under vegetation). This should provide a reliable measure of vegetation amount in an image pixel (picture element). However, the results are not perfect since factors other than vegetation amount may affect the values of SVIs. Variability in SVIs can arise from atmospheric effects, viewing and illumination angles (see Chapter 5 Section 2.4), sensor calibration, geometric registration errors, subpixel water and clouds, snow cover, background materials, image compositing, and landscape topographic features such as slope and relief. Several of these factors are discussed below.
Reflected or emitted energy received by a satellite sensor must pass through the atmosphere. The measurement values recorded "at sensor" may be very different from what left the ground surface, given that the atmosphere scatters and absorbs energy (depending on spectral wavelength, and atmospheric conditions which change through time). In general, atmospheric effects tend to lower the value of SVIs. Clouds and water that are obvious in a scene can be "masked" (ignored) or compensated for using computer routines, however sometimes the clouds or water cannot be discerned since they are smaller than a sensor's spatial resolution (see Chapter 5 Section 2.2).
Data values are also affected by the angle at which a surface or cover material is illuminated (illumination angle) and the angle at which a sensor receives or views the reflected or emitted energy (view angle). The height of the Sun changes throughout a day and throughout the seasons, so scenes imaged at different times may look different just due to changing Sun angles. For example, shadows will be longer when the sun is low in the sky.
As for view angles, many sensors collect data looking straight down at the ground (called a "nadir" view). Turning away from that nadir view direction ("off-nadir" view), the landscape appearance changes somewhat. Consider viewing a grass lawn: looking straight down, in addition to blades of grass, the soil underneath the grass can be seen (unless the grass is very thick and lush). Viewing the lawn at more of an angle, more blades of grass and less of the background soil are seen. Perhaps this explains the saying "the grass is always greener on the other side of the fence." On your side of the fence you have a nadir view of the grass, but looking over the fence to your neighbor's yard is an off-nadir view.
So measurements taken from nadir views record the response of bare ground or other materials between leaves and plants, while off-nadir views see more of the vegetation. Some off-nadir views also see more or less shadowing than other views. Varying amounts of shadow, and presence of nongreen materials, such as bare soils, leaf litter, exposed rocks, concrete, asphalt, sand, snow, water or ice, in a scene all can affect the values of SVIs generated from datasets.
Sensor calibration can also affect SVI values. At the time of a satellite launch, the radiometric and spectral response of instruments on the satellite are usually tested and documented (referred to as "calibration"). Over time, the response of sensors can change, due to detector degradation and other factors. However, it can be difficult to test a sensor's changes in response once a satellite is in orbit. Because of this, year-to-year changes observed in NDVI values for a forested area may not be the result of actual changes in the forest, but rather due to changes in a sensor. Attempts are made to monitor instrument changes by repeatedly observing locations that are assumed to have relatively constant reflectances over time (such as some deserts and bodies of water). New sensors can be designed with onboard calibration systems.
When analyzing remote sensing data and SVI trends and patterns, the various factors affecting SVI values must be considered. Observed changes can be due to complex interaction of all the materials in a scene, how that scene is illuminated and viewed, particular atmospheric conditions at the time of measurement, and sensor functioning.
Because of these limitations, NDVI and other SVIs are not perfect measures of vegetative biomass but rather can be thought of as reasonable "surrogates" for vegetation amount, and with careful analysis can be effective for monitoring global vegetation dynamics.