Monday, February 13, 2012

Channel Mixing and Red Vegetation

Q: Why does vegetation appear red when we look at processed Landsat imagery (or when we process our own TwinCam imagery?

The shortish answer: Because healthy green plants have a high reflectance value in the near-infrared (NIR) part of the spectrum. We can’t see NIR light normally, but the sensors in our cameras can. (Landsat and other imagers also ‘see’ NIR, as well as several other slices of the EM spectrum.)

Computers display color using three different channels, red, green, and blue. When we look for vegetation in TwinCam or Landsat imagery, we use MultiSpec to ‘map’ the NIR channel from the camera or sensor into the red channel that the computer displays. Since we have now used the red channel to show NIR data we will map the red and green channels from the camera to the green and blue channels on the computer, respectively. (We don't use the blue band here, so it isn't mapped to a channel.)

Vegetation looks red because the red pixels are showing the NIR, which again has a high reflectance value. (Remember that percent reflectance is simply the percentage of incoming sunlight that is being reflected back by the object in the image.) The green pixels on the computer are showing the red information from the camera. Plants absorb red light so these values are low. That means there won’t be much green in the vegetation on the screen. The blue pixels on the computer show the green information from the camera. Vegetation reflects more green light than it does in red or blue, so those values are relatively high. So, on the computer screen:

  • Red pixels show NIR information (high reflectance in vegetation)
  • Green pixels show Red information (low reflectance in vegetation)
  • Blue pixels show Green information (moderate reflectance in vegetation)
  • (blue information is discarded)

Below is a graphic representation of the channel mixing process using actual imagery from two cameras - one capturing visible light and the other capturing a slice of the near-infrared around 800 nm. (This is what we do when we process TwinCam imagery.)


A bit more information that you may find helpful:

Our TwinCam cameras have a sensitivity 8 bits per pixel - that is 28 or 256 integer values for each of the red, blue, and green channels. The values represent a grayscale range from 0-255, with 0 corresponding to 0% reflectance (black), and 255 responding 100% reflectance (white). This is called the Radiometric Resolution of the camera or sensor.

Based on the reflectance curve in the graph above, a typical pixel of healthy green vegetation might have the following values:

Normally, we would display the red, green and blue pixels values into the corresponding red, green and blue channels and our leaf would look greenish. However, we are interested in the NIR information, and to see those values we need to map that information to one of the other channels.

You can see from these values that the red channel has a considerably higher pixel value than the green or blue. Vegetation will appear red in these images. Some of the imagery and datasets we find online already have the channels mapped this way.