Chapter 01

Satellite Image Pre-Processing

Raw satellite imagery requires systematic pre-processing — radiometric calibration, atmospheric correction, and geometric rectification — before analysis.

SNAP (ESA)QGIS Semi-Automatic ClassificationGoogle Earth EngineENVIPython (Rasterio)

Collection 2Landsat ARD↗ Pre-processed free

~30 min/tileSen2Cor Speed↗ Automated processing

Theory & Foundations

Pre-processing converts raw sensor measurements (Digital Numbers) into physically meaningful values (reflectance, temperature) that can be compared across dates and sensors.

Radiometric calibration converts DN to Top-of-Atmosphere (TOA) radiance using gain/offset values from metadata. TOA radiance represents energy reaching the sensor including atmospheric effects.

Atmospheric correction removes scattering and absorption effects to derive surface reflectance — the "true" spectral signature of ground features. Methods: DOS (simple), FLAASH (physics-based), Sen2Cor (Sentinel-2 specific).

Geometric correction ensures pixels align with real-world coordinates. Orthorectification corrects for terrain displacement using a DEM. Essential for multi-temporal analysis.

Cloud masking: QA bands in Landsat and SCL band in Sentinel-2 flag clouds, cloud shadows, snow, and water. Machine learning methods (s2cloudless, Fmask) provide improved masking.

In-Depth Coverage

Radiometric Calibration

Converting raw digital numbers to physical units.

DN → Radiance using gain/offset from metadata
Radiance → TOA Reflectance using sun angle
Level-1 products: Calibrated but not atmospherically corrected
Level-2 products: Surface reflectance (atmospherically corrected)

💡 Pro Tip: Always use Level-2 (surface reflectance) data for vegetation indices and time-series analysis.

Atmospheric Correction Methods

Removing atmospheric effects from satellite imagery.

DOS (Dark Object Subtraction): Simple, no atmospheric data needed
FLAASH: Physics-based, uses atmospheric models
6S: Radiative transfer code, most accurate
Sen2Cor: ESA's official Sentinel-2 processor
LaSRC: USGS Landsat surface reflectance code

Key Techniques

Radiometric calibrationAtmospheric correctionCloud maskingOrthorectificationBand math

Data Visualization

Data Processing Level Usage in Research (%)

TOA Reflectance25

Surface Reflectance55

Raw DN5

BT (Thermal)15

Pre-Processing Pipeline

⚙️CalibrateDN → TOA Radiance/Reflectance
🌫️CorrectAtmospheric correction → Surface Reflectance
☁️MaskRemove clouds, shadows, snow
📐RectifyGeometric/orthorectification

Practical Workflow

Download Level-1 imagery with metadata
Apply radiometric calibration (DN → TOA)
Perform atmospheric correction (TOA → Surface Reflectance)
Apply cloud masking using QA bands
Orthorectify using DEM if needed
Clip to study area extent
Validate against known reflectance targets
Create analysis-ready composites

Software & Tools

SNAPFreeDesktop Software — ESA Sentinel processing

Sen2CorFreePlugin — Sentinel-2 atmospheric correction

Video Tutorials

▶ Atmospheric Correction in SNAPRUS Copernicus · YouTube · 45 min ▶ Image Pre-Processing in GEEGoogle Earth Engine · YouTube · 30 min

Real-World Application

USGS provides Collection 2 Landsat data as Analysis Ready Data (ARD) — pre-processed to surface reflectance, saving researchers hundreds of hours of pre-processing.

Real-World Projects

USGS Analysis Ready Data📍 USA

Landsat Collection 2 pre-processed surface reflectance.

Impact: Saves researchers hundreds of hours of pre-processing.

Case Study

ESA's Sen2Cor processor converts Sentinel-2 Level-1C (TOA) to Level-2A (surface reflectance) with simultaneous scene classification, serving as the standard pre-processing pipeline.

Problem-Based Learning

P1 Compare NDVI between two dates with different atmospheric conditions

Pre-monsoon and post-monsoon Landsat scenes have very different atmospheric haze levels. Raw NDVI comparison would be misleading.

Solution: Apply atmospheric correction (LaSRC for Landsat) to both scenes. Validate surface reflectance against known targets. Calculate NDVI from corrected data.

Landsat Collection 2 Level-2GEEQGIS

✅ NDVI change map showed true vegetation response to monsoon, not atmospheric artifacts. 15% difference between corrected vs uncorrected results.

Knowledge Check

1. What does atmospheric correction produce?

2. Which is ESA's Sentinel-2 atmospheric correction tool?