< Back  ||  Home > Reports > Canopy Cover

Correlating Canopy Cover with NDVI

Measures of vegetation greenness, such as the Normalized Difference Vegetation Index (NDVI), are influenced by many factors. Variations in atmospheric conditions, the spectral character of soil and rock, as well as trees, shrubs and grasses all contribute to the greenness measurement in each pixel. The ability to isolate each of these factors makes it easier to interpret the meaning and variability of these greenness measures. Natural resource managers often focus on the growth and health of rangeland grasses. In order to extract more specific information about the greenness of the grass cover, we have created this demonstration paper to illustrate how the effect of tree and shrub cover can be isolated using high spatial resolution remote sensing data. This demonstration was created as part of Synergy III for the V Bar V Ranch study area in Northern Arizona.

High spatial resolution multispectral and panchromatic Ikonos satellite imagery of the V Bar V ranch were acquired on 4 April 2000 (see Figure 1). The multispectral Ikonos image has a resolution of 4-meters, and the panchromatic image has a resolution of 1-meter. Four individual Ikonos scenes are required to cover the area of the ranch. These scenes were orthorectified using ground control points extracted from 1-meter Digital Orthophoto Quarter Quadrangle (DOQQ) aerial photography. The four scenes were mosaiced to create a seamless, continuous image of the ranch. A false color composite image was also created (Figure 2). The false color composite combines near-infrared, red, and green radiation into a single image, showing areas of actively growing vegetation in red. In addition, a 4-meter resolution NDVI image [NIR – Red / NIR + Red] and 1-meter texture-variance image were created.

Figure 1. Ikonos 1-meter Resolution Panchromatic Image

Figure 2. Ikonos 4-meter Resolution False Color Composite

The four bands of the multispectral data, the NDVI, and the texture-variance images were all included in the classification. An unsupervised classification was run using ERDAS Imagine. The original classification contained 13 classes, which were later combined into just seven land cover classes (Figure 3):

1. Darkened recently burned areas and areas of dark shadow or standing water
2. Tree Canopy
3. Dark Canopy Understory
4. Open Ground (some grasses may be present)
5. Small Tree or Shrub Canopy
6. Bare Ground and Bright Gypsum Soil
7. Agricultural Fields

Final class identification was evaluated through visual interpretation of the false color composite image, the 1-meter panchromatic image, and 1-meter DOQQ data covering the ranch. The agricultural field class was manually digitized on-screen.

Figure 3. Final Ikonos Classification Image.

Canopy cover includes cover from both trees and shrubs, but not grasses. Discovering the percentage of canopy cover on the V Bar V Ranch will allow us to better interpret greenness measurements as they relate to rangeland grasses. For this purpose, the classified Ikonos image was condensed into a binary classification: canopy (tree and shrub), and non-canopy (Figure 4).

Figure 4. Binary Classification Showing Tree and Shrub Canopy Cover (Green) and Non-canopy (White).

Accuracy assessment of the 4-meter binary classification was done using the 1-meter panchromatic imagery. 200 points were distributed on the binary image in an equalized random pattern (equal number of points in canopy and non-canopy classes). The same points were distributed on the panchromatic image. A polygon coverage of the random points was then created in ArcInfo. This consisted of a 4 x 4 meter square centered on each random point. A 4- meter square on the panchromatic image encompasses 16 pixels, while it only encompasses one pixel in the 4-meter imagery. Using this technique we were able to evaluate the 16 1-meter (panchromatic) pixels in which tree and shrub canopy could be directly identified in relation to the 4-meter pixels in the classified image. A pixel was determined to be canopy if at least 1/3 of the 16 pixel square on the panchromatic image was inhabited by tree or shrub canopy. Anything less than 1/3 was considered to be non-canopy. The accuracy assessment performed in this way yielded an overall accuracy of 85.6%.

To understand how canopy effects NDVI measurements, we must compare the two on similar scales. Greenness measurements from the NOAA-AVHRR satellite are available at a resolution of 1-km. In order to relate percent canopy cover on the V Bar V Ranch to AVHRR NDVI measurements, we computed the percent cover within each 1-km AVHRR pixel covering the ranch. The binary classification, at 4-meter resolution, was overlaid onto a grid of AVHRR 1-km cells. The percent tree and shrub canopy was then calculated within each cell. This yielded a grid of percent canopy cover on the V Bar V Ranch at the AVHRR resolution (Figure 5). This information could then be readily used while interpreting multitemporal animations of NDVI across space and time.

Figure 5. Percent Canopy Cover Grid of the V Bar V Ranch at 1-km Resolution.