Atmospheric Correction

Atmospheric correction is the process of removing the radiometric effects of the atmosphere from satellite or airborne imagery to retrieve accurate surface reflectance values. When sunlight travels through the atmosphere to Earth's surface and back to a sensor in orbit, it interacts with gas molecules, aerosols, and water vapor through scattering and absorption. These effects mean that the top-of-atmosphere (TOA) radiance recorded by a sensor is not a true representation of the surface below. Atmospheric correction transforms TOA radiance into bottom-of-atmosphere (BOA) surface reflectance — the quantity that actually describes the physical properties of land, water, or vegetation. Without this step, spectral indices like NDVI become unreliable, and multi-temporal comparisons are compromised.

Atmospheric correction methods fall into empirical (image-based) and physics-based (radiative transfer) categories. The simplest empirical approach is Dark Object Subtraction (DOS), introduced by Chavez in 1988, which assumes certain dark pixels should have near-zero reflectance. Physics-based methods use radiative transfer models like 6S and MODTRAN that simulate how photons interact with atmospheric constituents, requiring inputs such as aerosol optical depth, water vapor, and solar geometry.

Operational processors build on these codes. NASA's LaSRC uses 6SV to produce Landsat Collection 2 surface reflectance products. ESA's Sen2Cor processor for Sentinel-2 is based on libRadtran and produces Level-2A surface reflectance data. Both output Analysis Ready Data conforming to the CEOS CARD4L specification.

Accuracy depends heavily on the quality of aerosol and water vapor estimates, which can be unreliable over bright surfaces (deserts, snow) or near cloud edges. Thin cirrus clouds introduce spectrally broad scattering that is difficult to detect. Over heterogeneous terrain, adjacency effects — where photons scattered from bright neighboring pixels contaminate dark targets — can introduce errors. No single algorithm performs optimally across all surface types and atmospheric conditions.

The need for atmospheric correction emerged with Landsat MSS in the 1970s. Pat Chavez published Dark Object Subtraction in 1988. Eric Vermote and collaborators developed 6S in 1997 for satellite sensor simulation. The launch of MODIS in 1999 drove operational atmospheric correction at global scale. Sen2Cor was developed alongside Sentinel-2 (2015). Today, atmospheric correction is a standard preprocessing step performed by data providers before distribution.