Raw satellite imagery is not ready for analysis. Sensor effects, atmospheric interference and terrain distortion all separate the recorded numbers from ground truth. Preprocessing — radiometric calibration, atmospheric correction, orthorectification, and often mosaicking and cloud masking — removes these layers of contamination. Done well it is invisible; skipped or botched, it quietly invalidates every analysis built on top.
Radiometric and atmospheric correction
Radiometric calibration converts raw digital numbers into physical radiance, then reflectance, using sensor coefficients. Atmospheric correction goes further, removing the haze, scattering and absorption the atmosphere adds, to estimate surface reflectance — what the ground itself reflects.
Surface reflectance is the currency of serious analysis: indices become comparable between dates and sensors, and classifications transfer instead of breaking on every new scene. Modern “analysis-ready data” products deliver this correction pre-applied.
Geometric correction and orthorectification
A raw image is a perspective view distorted by sensor geometry, Earth curvature and — most of all — terrain relief: mountains lean, valleys shift. Orthorectification uses a digital elevation model and sensor model to project every pixel to its true map position, producing imagery that overlays correctly with GIS layers.
Positional accuracy is then refined against ground control points or reference imagery, with residuals reported. For engineering-grade work, this accuracy statement is part of the deliverable, not an afterthought.
Mosaicking, cloud masking and QA
Large-area products stitch many scenes into seamless mosaics: seamlines routed through low-contrast areas, radiometry balanced so boundaries vanish, clouds and shadows masked and filled from other dates. Each step has quality metrics — seam visibility, colour consistency, residual cloud fraction.
Production pipelines automate the bulk of this, but review remains human: preprocessing is where a trained eye catches the subtle artefacts automation happily propagates.
Frequently asked questions
What is analysis-ready data (ARD)?
Imagery already processed to surface reflectance, orthorectified to a consistent grid, with cloud and quality masks attached — so analysts start at the science instead of the corrections. Landsat and Sentinel programmes both publish ARD-style collections.
Why does my imagery not line up with my GIS layers?
Usually the imagery is not orthorectified (terrain displacement), or a coordinate system mismatch is in play. Orthorectification against a DEM plus a proper CRS check resolves the overwhelming majority of alignment complaints.
Related topics
Need this done professionally?
GLOBEIR delivers this work for government and enterprise clients.