What do the multiscatter models ISAACS and DISORT do in FLAASH?

ISAACS and DISORT are numerical models used to simulate multiple scattering of a photon as it passes through the atmosphere. The magnitude of the scattering is dependent on the size of the scatterer relative to the photon's wavelength. Typical scatterers include aerosols like soot from car exhaust and smokestacks (primarily scatters red), water vapor (primarily scatters blue), or sea salts (wavelength effected depends on the type and size). Generally speaking, they scatter the shorter wavelengths (blue-red) much more than longer ones (IR). In fact, once you get to about 850 nm you really don't have to worry about scattering at all. Scattering has two primary effects on the image collected by a remote sensing instrument:

1. reduces the signal preferentially in the shorter wavelengths
2. increases the adjacency effect (where signal from nearby pixels can scatter into the path collected for another pixel)

As you might expect, the magnitude of the effects are related to the density and type of scatterers in the atmosphere. If its very clear day, there's virtually no scattering. A very hazy day and there's lots of scattering (in fact, that's why it looks hazy).

The FLAASH multiscatter models are characterized by the number of "streams", which refers to the number of different scattering angles that are modeled for each photon. The ISAACS and DISORT 2-stream are very similar, although the ISAACS is a lot faster because it is a simpler model. ISAACS is usually OK for scenes that are not hazy . The more haze you have, the more important it will be to correct for the scattering, but only if you need to use the visible wavelengths. If you are only using the IR wavelengths, then there is no need to use a complicated (and time consuming) multiscatter model because these wavelengths are not affected by scattering. Under special circumstances, or for special applications, it can be very important to choose a high-stream DISORT model. For example, if you have a very hazy scene and you're looking for high spectral contrast targets in the visible (for example, a white building next to a forest) then the adjacency effect is likely to contaminate your spectra, so you should use a DISORT 8-stream model.

Review on 12/31/2013 MM