Keywords: materials, truth, atmosphere

Summary

This demonstration is designed to show a few different aspects of the DIRSIG model:

  • The so-called "leaf stacking effect" that occurs in the near infrared (NIR) region, where high reflectance and high transmission makes the apparent reflectance of leaf stacks (more than one leaf) higher than a single leaf.

  • The use of a Uniform Atmosphere model in a specially configured mode that allows the user generate output images that are effectively in reflectance units.

  • The use of the Leaf Area Index truth collector to extract how many leaves are in the leaf stack as a function of location.

Details

The scene is composed of a background over which a stack of successively smaller "leaf planes" is placed. These "leaf planes" have a material optical properties that approximate those of real leaves. The impacts of the spectrally varying transmission at different wavelengths is then observed.

Important Files

The Scene

The scene geometry is contained in the geometry folder, but is composed entirely from 2 GDB files: one for the background and one that is instanced for each leaf plane. The "background" plane is located at Z = 0 and the "leaf" planes are stacked above it in a pyramid type configuration. The first leaf plane is 10 x 10 meters, the second is 8 x 8 meters, the third 6 x 6 and the last 4 x 4. Each plane is 1 meter above the previous one.

images/z_truth.png
Figure 1. The Z truth image visualizing the stack of leaf planes.

The "background" material (ID = 1) is configured as a simple 2% Lambertian reflector. The "leaf" material (ID = 2) is configured with an emissivity (reflectance) and extinction (transmission) pair that resemble a generic leaf (see plot below). The leaf optical properties feature the green color and dramatically higher near-IR reflectence coupled with the higher near-IR transmission.

images/leaf_props.png
Figure 2. Plot of the simple leaf optical properties

Both materials are configured with the Generic Radiometry Solver because this solver handles diffuse illumination better than the Classic Radiometry Solver. To make the simulation a bit faster, we made a few changes to the default solver parameters.

  • The Initial Sample Count is set to 1 (to make sure we sample the Sun).

  • The Maximum Bounces is set to 1 (while developing this demo we found that you don’t need to worry about anything higher than the 2nd bounce).

  • The Full-Sample Decay Rate is set to 1 (this doesn’t really matter since we start with only 1 sample).

  • The number of Zenith and Azimuth sampling quads was changed to 6 and 12, respectively (the defaults are 8 and 12).

  • The Minimum number of quad samples is set to 1 (we want to make sure we sample every quad in the sampling grid at least once).

  • The Enable direct only flag was set to false (NOTE: this flag is not shown in the Material Editor and must be added by hand).

The Atmosphere

One of the goals of this demonstration was to show the user how to get DIRSIG output that is in effective reflectance units. The "trick" to achieving this is that the simulation uses the Uniform Atmosphere model with the following setup:

  • The Spectrally-constant, hemispherical irradiance is 3.141592654 Watts/cm2.

  • The sky fraction is 1.

This effectively sets up the illumination conditions as 100% diffuse and with an illumination value that will result in reflected radiances in reflectance units (ranging 0 - 1).

Note
 It is important to realize that this setup isn’t triggering some sort of special mode in DIRSIG. It is simply specifying the magnitude of the illumination conditions so that the surface leaving radiances (an integral of the hemispherical illumination weighted by the surface reflectance function) matches the hemispherical reflectance.

The Platform

The demo.platform file describes a simple 320 x 240 (QVGA) camera that is configured with 4 narrow channels:

  • A NIR channel (centered at 0.80 microns)

  • A red channel (centered at 0.65 microns)

  • A green channel (centered at 0.55 microns)

  • A blue channel (centered at 0.45 microns)

In order for the channels to output radiances that are effective reflectances using the special Uniform atmosphere setup, each channel radiance must be normalized by the response. This option must be enabled for each channel via the Output spectral radiance (normalized) option on the Options tab in the channel description.

The focal plane also has the Leaf Area Index truth collector configured. The collector was supplied material ID #2, which corresponds to the "leaf" material.

Results

The simulation produces a radiance and truth image pair. The radiance image below shows the increase in radiance (effective reflectance) as the leaf stack gets higher (more leaf planes) in the middle of the image. Also note that the radiance (effective reflectance) varies across surfaces because of shadowing effects. For example, the corners of the leaf planes have the highest values because they see the most sky.

images/rgb.png
Figure 3. An RGB visualization of the leaf stack.

The NIR band demonstrates the "leaf stacking effect", where the apparent reflectance of the leaf plates increases with leaf stack depth:

images/nir.png
Figure 4. The NIR band shows the "leaf stacking effect".

Because the "leaf stacking effect" is driven by transmission, the effect is less in bands with less transmission. The plot below shows the apparent reflectance across the center of the simulated image (horizontal profile) for the blue and near-IR bands. The "saw tooth" pattern across the profile arises from the diffuse shadowing of the nearby leaf plane hovering above the current one. The apparent reflectance in the blue band is constant with leaf stack depth, and matches the 5% value supplied via the input reflectance at the center (top) of the stack. However, the apparent reflectance in the near-IR band shows the impact of leaf stack depth. The single leaf layer has an apparent reflectance close to the input value of 42%. However, as the stack gets deeper we can see the apparent increase in reflectance that comes from the transmitted radiance of leaf planes below. At the top of the stack, the apparent reflectance is closer to 60% (rather than 42%) due to sub-surface reflections. The bowl shape at the top of the stack arises from the gradient in the diffuse shadow beneath the top leaf plane.

images/profile.png
Figure 5. A plot of the apparent reflectance in the Green and NIR bands

The "Leaf Area Index" truth map in the demo_truth.img file reports how many single-sided leaf planes are within each pixel. The data values range from 0 in the area around the leaf stack to 4 at the top of the leaf stack.

images/lai_truth.png
Figure 6. Leaf Area Index (LAI) truth image.