Atmospheric correction accounts for the attenuation caused by scattering and absorption in the atmosphere. It transforms top-of-atmosphere (TOA) reflectance to bottom-of-atmosphere (BOA) reflectance.
The decision to perform atmospheric correction depends on the need, i.e. the envisioned usage of the derived EO information product and the nature of the underlying problem. This includes requirements to the accuracy of extracted biophysical information. Additionally, the decision and choice of methods depends on the type of remote sensing data available, the amount of in-situ historical and/or concurrent atmospheric information available.
An atmospheric correction is essential when biophysical or geophysical parameters (e.g. of water or vegetation) are going to be extracted from the remote sensing data. If the data is not corrected, the subtle differences in reflectance among the contributing image bands may be lost. This is especially relevant when biophysical information shall be compared to that of images from other dates.
However, some cases exist where it is unnecessary to perform atmospheric correction. For example, it is not necessary for producing an image classification product from a single date of remotely sensed data. If a maximum likelihood classification is applied that uses training data with the same relative scale for the pixel values, then, atmospheric correction has little effect on the classification accuracy. The same holds true for a post-classification change detection where the classifications of the two different dates were performed independently.
The process of (absolute) atmospheric correction requires a model atmosphere and in situ atmospheric measurements acquired at the time of remote sensor data acquisition as input. In situ data can be available from other sensors on-board the sensor platform.
Dark Object Subtraction (DOS) is one of the most popular empirical atmospheric correction techniques. This technique assumes that a black object has a reflectance value of zero. Yet, a dark object present in a satellite image will have a value different than zero because of the atmospheric scattering. This value is then subtracted from all pixels in a given spectral band.