Satellite Band Combination Matrix & Synthesizer

⚒ Theoretical and Mathematical Foundation

Optical remote sensing is governed by the fundamental physical principles of electromagnetic radiation (EMR) interacting with terrestrial materials. Solar irradiance passing downward through the atmosphere undergoes absorption and scattering by atmospheric gases, ozone, water vapor, and suspended aerosols. The radiation reaching the land surface is partitioned into reflected, absorbed, and transmitted EMR, described by the principle of conservation of energy:

The spectral reflectance rho(lambda) is the dimensionless ratio of reflected radiant flux to incident radiant flux as a function of wavelength lambda:

Because distinct surface materials (vegetative canopies, exposed mineral soils, artificial impervious structures, and liquid water bodies) possess diagnostic atomic and chemical structures, their spectral reflectance profiles differ substantially across the electromagnetic spectrum.

Diagnostic Spectral Signatures Across Wavelength Domains

Remote sensing instruments sample specific atmospheric transmission windows where atmospheric gas absorption is minimal:

• Visible Domain (0.40 to 0.70 micrometers): Governed by electronic transitions and pigment absorption. In healthy vegetation, chlorophyll-a and chlorophyll-b pigments absorb strongly in the Blue (centered near 450 nm) and Red (centered near 660 nm) wavelengths, with a minor reflectance peak in the Green (550 nm). Clear, deep water absorbs minimally in the blue-green spectrum, giving water bodies their characteristic cyan appearance.
• Red-Edge Transition (0.70 to 0.75 micrometers): A critical boundary where leaf absorption drops precipitously while structural scattering begins. In Sentinel-2 MSI, Bands 5, 6, and 7 explicitly sample this steep slope, providing early diagnostic sensitivity to vegetative stress, nitrogen deficiency, and chlorophyll breakdown before visual yellowing occurs.
• Near Infrared Domain (0.75 to 1.00 micrometers): Highly scattered by internal spongy mesophyll leaf cell walls, leading to high reflectance (up to 40% to 60%) in healthy canopies. Conversely, liquid water almost completely absorbs EMR above 800 nm, rendering water bodies pitch-black in NIR imagery.
• Shortwave Infrared Domain (1.30 to 2.50 micrometers): Dominated by fundamental molecular vibrations of water (O-H stretch bands at 1.45 and 1.94 micrometers), hydroxyl minerals, clays (Al-OH absorption at 2.20 micrometers), and carbonate radicals. Dry bare soils and artificial urban materials (concrete, asphalt, terracotta, crushed aggregates) exhibit elevated reflectance in SWIR-1 and SWIR-2, contrasting sharply with water-absorbing canopies.

Additive RGB Color Synthesis Theory

Standard digital displays reproduce color by stimulating the human eye's red, green, and blue retinal cones using three additive color primaries. A multispectral satellite sensor captures dozens of discrete wavelength bands beyond human vision. False-color compositing maps any three selected spectral bands to the display's Red, Green, and Blue channels:

By selecting bands that maximize spectral contrast between target classes, features that appear indistinguishable in natural color (such as dry bare agricultural fields versus impervious concrete pavements) become visually distinct.

Cross-Sensor Spectral Homology Between Sentinel-2 and Landsat 8-9

⚙ Step-by-Step Desktop GIS and Command-Line Workflow

Building multi-band composites from raw Level-2A data requires stacking individual monochromatic raster bands into a unified multi-band raster container.

Virtual Raster Compositing in QGIS Desktop

1. Import Surface Reflectance Bands: Drag the target single-band GeoTIFF files into the QGIS Layers panel (e.g. T33TXF_B04_10m.jp2, T33TXF_B03_10m.jp2, and T33TXF_B02_10m.jp2).
2. Open Virtual Raster Builder: Navigate to the main menu and select Raster -> Miscellaneous -> Build Virtual Raster.
3. Select Input Layers: Click the ellipsis button and check the layers in exact channel order (Band 1 = Red, Band 2 = Green, Band 3 = Blue).
4. Enforce Separate Bands: Check the box labeled Place each input file into a separate band. This is essential; failing to check this box creates an unstitched spatial mosaic rather than an RGB composite.
5. Save and Style: Execute the tool. Right-click the resulting layer, select Properties -> Symbology, and set Render type to Multiband color. Adjust contrast enhancement to Stretch to MinMax using a 2% cumulative count cut.

GDAL Command-Line Compositing Recipes

Using the GDAL CLI allows processing full scenes without graphical overhead:

# 1. Build a false-color urban composite (B12, B11, B04) from Sentinel-2
gdalbuildvrt -separate urban_composite.vrt \
  B12_20m.jp2 \
  B11_20m.jp2 \
  B04_10m.jp2

# 2. Translate virtual raster into a Cloud-Optimized GeoTIFF with Deflate compression
gdal_translate -of COG \
  -co COMPRESS=DEFLATE \
  -co PREDICTOR=2 \
  urban_composite.vrt \
  urban_composite_cog.tif

Python Rasterio Multi-Band Stack Generation

The Python script below stacks individual reflectance bands and writes a georeferenced 3-band GeoTIFF with standardized metadata:

import rasterio
import numpy as np

# File paths for Sentinel-2 bands (e.g. Urban composite: B12, B11, B04)
band_paths = ["B12.tif", "B11.tif", "B04.tif"]

# Read metadata from first band
with rasterio.open(band_paths[0]) as src0:
    meta = src0.meta.copy()

# Update metadata for 3-band Float32 surface reflectance
meta.update({
    "count": 3,
    "dtype": "float32",
    "compress": "deflate"
})

with rasterio.open("sentinel2_urban_composite.tif", "w", **meta) as dst:
    for idx, path in enumerate(band_paths, start=1):
        with rasterio.open(path) as src:
            # Read band, cast to float, scale raw integer DN (0-10000) to reflectance (0.0-1.0)
            data = src.read(1).astype("float32") / 10000.0
            # Clip residual negative or cloud glare outliers
            data = np.clip(data, 0.0, 1.0)
            dst.write(data, idx)
            dst.set_band_description(idx, f"Channel_{idx}_{path}")

print("Multispectral composite exported successfully.")

⚠ Common Pitfalls and Field Troubleshooting

Creating false-color composites requires strict radiometric discipline. Ignoring sensor calibration factors or spatial grids degrades analytical rigor.

1. Spatial Resolution Mismatches in Sentinel-2 MSI

Sentinel-2 features bands at 10m, 20m, and 60m spatial resolutions. Attempting to build a composite combining Band 12 (20m) with Band 4 (10m) without spatial harmonization can cause GDAL or QGIS to default to the lowest resolution (20m) or throw non-matching grid dimensions errors. Always resample 20m bands to 10m using bilinear interpolation before stacking.

2. Digital Numbers vs Surface Reflectance Scale Factors

Raw satellite imagery files store surface reflectance as integer Digital Numbers (DN) to reduce file size. In Sentinel-2 Level-2A products, surface reflectance is stored as integer values scaled by 10,000 (reflectance = DN / 10000.0). However, following the ESA processing baseline change in January 2022 (PB 04.00), a radiometric offset of -1000 was introduced to accommodate negative surface reflectance values caused by dark target atmospheric overcorrection. In Landsat 8-9 Collection 2 Level-2 data, surface reflectance uses a multiplicative scale factor of 0.0000275 and an additive offset of -0.2:

Failing to apply these specific sensor scale factors leads to inverted contrast, crushed black levels, or blown-out white pixels.

3. Cross-Sensor Band Number Confusion

A frequent operational trap is using identical band numbers across different satellite platforms. The table below highlights critical band index traps:

❓ Frequently Asked Questions