Raw satellite imagery contains systematic distortions that must be corrected before meaningful analysis. Preprocessing transforms raw sensor data into scientifically usable products through three main steps: radiometric calibration (converting raw digital numbers to physical units of radiance or reflectance), atmospheric correction (removing the atmosphere's contribution to isolate the surface signal), and geometric correction (aligning image pixels to accurate ground coordinates). Without preprocessing, pixel values do not represent true surface properties, multi-date comparisons are invalid, and data from different sensors cannot be combined.
Raw satellite data straight from the sensor is not ready for analysis. It contains a mixture of surface information, atmospheric effects, and geometric distortions that must be systematically separated and corrected. This preprocessing chain is the unglamorous but essential foundation of all quantitative remote sensing.
Radiometric calibration converts raw digital numbers (DN) to physical units. Each sensor has calibration coefficients that convert DN to at-sensor radiance (watts per square meter per steradian per micrometer). From radiance, dividing by the solar irradiance at the top of the atmosphere (adjusted for Earth-Sun distance and solar zenith angle) yields top-of-atmosphere (TOA) reflectance -- a standardized quantity that removes sensor-specific and illumination effects but still includes the atmosphere.
Atmospheric correction is the most scientifically important step. The atmosphere scatters incoming sunlight into the sensor's field of view (path radiance), absorbs portions of both downwelling and upwelling radiation, and alters the spectral distribution of light reaching the surface. Physics-based models (6S, MODTRAN, libRadtran) simulate these processes using atmospheric parameters (aerosol optical depth, water vapor column, ozone) to estimate and remove the atmospheric contribution. The result is surface reflectance -- what the surface would look like if there were no atmosphere.
Geometric correction ensures that pixels map to correct geographic locations. Satellite ephemeris data provides an initial geometric model, but systematic and non-systematic errors require correction using ground control points and, for accurate results, orthorectification using a DEM to remove terrain-induced displacement. The result is an image where each pixel has a reliable geographic coordinate, enabling overlay with other geospatial data and precise multi-temporal registration essential for change detection.