Module 14: Remote Sensing

14.3 Spectral Bands and Resolutions

Understanding a satellite's band inventory — the key to extracting meaningful information.

Lesson 70 of 100·14 min read

Key takeaways

Sensors sample the spectrum in discrete bands, each designed for specific applications.

Understanding band centre wavelength and width guides interpretation.

Four resolution types (spatial, spectral, temporal, radiometric) trade off by satellite.

Introduction

Every Earth-observation satellite is characterised by its bands. "Sentinel-2 has 13 bands" is the headline; knowing what each band measures and why it exists is the analytical depth. This short lesson covers the taxonomy.

What a band is

A band is a wavelength range the sensor captures. For each pixel, the sensor measures the total energy arriving in that range. Each band becomes one 2D image, co-registered with the others.

Two parameters define a band:

Centre wavelength — the nominal wavelength.
Bandwidth (full width at half maximum, FWHM) — how wide the sensor's sensitivity curve is.

Narrow bands distinguish subtle spectral features but collect less energy (noisier). Wide bands collect more energy but average over many features.

Four types of resolution

Spatial

Pixel size on the ground. Sentinel-2 has 10 m, 20 m, and 60 m bands. Spatial resolution determines the smallest distinguishable feature.

Spectral

Number and width of bands. Multispectral sensors have 3–15 bands; hyperspectral sensors have hundreds (e.g., EnMAP, PRISMA, AVIRIS).

Temporal

Revisit time. Landsat-8 alone revisits every 16 days; combined with Landsat-9 and Sentinel-2 in Harmonised Landsat Sentinel (HLS) every 2–3 days. MODIS revisits daily. GOES every 10–15 minutes.

Radiometric

Bit depth per pixel. 8-bit (0–255) is common for natural-colour imagery. 12-bit (0–4095) for Sentinel-2. 16-bit for high-dynamic-range scientific sensors. More bits = finer distinctions between similar radiance values.

Trade-offs

Spatial ↔ temporal: higher resolution sensors have narrower swaths; same satellite revisits less often.
Spatial ↔ spectral: adding bands adds data volume; more bands often sacrifice some pixels.
Radiometric ↔ temporal: more bits = more data = slower downlink.

A sensor's design explicitly chooses these trade-offs. Military tasking satellites prioritise spatial at the cost of revisit; weather satellites prioritise temporal at the cost of spatial.

Common band combinations for display

Display any three bands as red/green/blue to produce a "false colour" image:

True colour — Red/Green/Blue bands. Natural.
False colour (infrared) — NIR / Red / Green. Vegetation appears red.
Urban — SWIR / NIR / Red. Built-up area stands out.
Agriculture — SWIR / NIR / Blue. Highlights crop health.
Geology — SWIR / SWIR / NIR or similar. Mineral discrimination.

Band math

Combining bands with arithmetic produces indices:

NDVI = (NIR − Red) / (NIR + Red) — vegetation vigour.
NDWI = (Green − NIR) / (Green + NIR) — water.
NDBI = (SWIR − NIR) / (SWIR + NIR) — built-up.
NBR = (NIR − SWIR2) / (NIR + SWIR2) — burn severity.
NDSI = (Green − SWIR) / (Green + SWIR) — snow.

Module 14.4 dives into spectral indices.

Pan-sharpening

A panchromatic (pan) band is a single high-resolution band covering a wide visible range. Many sensors (Landsat, WorldView) have pan at 2× the spatial resolution of multispectral. Pan-sharpening merges the pan's spatial detail with the multispectral colour:

Code

out = colour_from_multispectral × brightness_from_pan

Algorithms: Brovey, IHS, Gram-Schmidt, ICA. Result: 15 m panchromatic + 30 m colour → 15 m pan-sharpened colour.

Spectral libraries

Reference reflectance spectra for known materials let you match imagery against a "library":

USGS Spectral Library — minerals, soils, vegetation.
ECOSTRESS Spectral Library — NASA's unified library.
SPECCHIO — community repository.

Used for:

Target detection (find spectra matching a known signature).
Classification validation.
Calibration.

Hyperspectral vs multispectral

Multispectral: 3–15 bands. Most current satellites.

Hyperspectral: 100+ bands at 5–10 nm spacing. Example missions: EnMAP (2022), PRISMA (2019), AVIRIS-NG (airborne). Applications:

Mineral identification.
Exact vegetation species.
Agricultural nutrient mapping.
Pollution detection.

Data volumes are vastly larger; processing more specialised.

Self-check exercises

1. Why does Sentinel-2 provide bands at 10 m, 20 m, and 60 m rather than all at 10 m?

Detector physics and data downlink constraints. The 60 m bands (atmospheric correction, cirrus) don't need fine spatial resolution — they measure atmospheric properties that vary on large scales. The 20 m red-edge and SWIR bands could be higher resolution but would require larger detector arrays and more downlink bandwidth. The design optimises for application-appropriate resolutions.

2. A false-colour (NIR / Red / Green) composite shows vegetation as bright red. Why?

The NIR band (plotted as red in the composite) has high reflectance over vegetation due to leaf cell scattering. The red band (plotted as green) has low reflectance because chlorophyll absorbs it. The green band (plotted as blue) has moderate reflectance. Result: vegetation is dominated by the NIR "red" channel — hence the bright red appearance, which is the most sensitive visual cue for vegetation extent and health.

3. Hyperspectral vs multispectral — when is hyperspectral worth the extra complexity?

When you need to distinguish specific materials with subtle spectral differences — mineralogy, individual plant species, pollution gases. Multispectral bands are too broad to separate e.g. calcite from dolomite. For land-cover classification, vegetation monitoring, and most broad-scale analyses, multispectral is fine and much easier to work with.

Summary

Bands = discrete wavelength ranges captured by the sensor.
Four resolutions (spatial, spectral, temporal, radiometric) trade off.
False-colour composites and band math unlock information beyond natural colour.
Hyperspectral deepens spectral resolution at cost of complexity and volume.