The general relationships provided in Chapter 5 relate pupil-plane radiance at a given remote viewing orientation to the apparent spectral properties of a material being viewed, along with a number of properties related to the radiative transfer from the material to the sensor. These properties include direct and diffuse illumination, atmospheric transmission, and atmospheric path radiance. The goal of remote sensing is generally to extract information from the apparent spectral properties, such as the reflectance or emissivity spectrum, of the materials being viewed. Therefore, the presence of these unknown illumination and atmospheric parameters is an undesired reality that must be dealt with to achieve this goal. Atmospheric compensation, also known as atmospheric correction or normalization, refers to image processing methods intended to invert relationships (such as those from Chapter 5) with limited knowledge of radiative transfer properties to estimate the apparent spectral properties of imaged materials. A by-product of atmospheric compensation is often the estimation of atmospheric spectral properties, which can be useful information in itself. In this case, the terms atmospheric retrieval and atmospheric sounding are often used in the literature.

Atmospheric compensation methods can be loosely separated into two categories: in-scene methods and model-based methods. Model-based methods use a radiative transfer model such as MODTRAN as a basis for atmospheric compensation. While such a model can fairly accurately capture the illumination, transmission, and path radiance characteristics for a specified remote sensing situation, a large number of atmospheric parameters such as gaseous concentration, pressure, and temperature profiles need to be precisely specified for a given situation, and these parameters are typically unknown. On the other hand, possible spectral variation of the atmospheric properties is constrained, making estimation of key unknown parameters a tractable problem in some cases. In-scene methods use some a priori knowledge of the spectral nature of materials expected to be present in a scene to guide estimation of atmospheric properties, without regard to a radiative transfer model such as MODTRAN. Because they do not require such sophisticated radiative transfer modeling, in-scene methods are somewhat simpler to understand and less computationally complex; thus, we begin with them first.