The Remote Sensing Full Stack
As SpaceApps students, you understand the complete signal chain—from the physics of the photon to the perception in the brain.
The 7 stages of data flow from 'Origin' to 'Decision'.
- ORIGIN: Physics of light and electromagnetic radiation.
- INTERACTION: Reflectance spectral signatures of Earth's surface.
- CAPTURE: Sensor mechanics (Photons converted to Digital Numbers).
- TELEMETRY: Data transmission to Ground Stations.
- PROCESSING: Atmospheric and Geometric corrections.
- VISUALIZATION: Mapping high-bit data to 8-bit RGB displays.
- PERCEPTION: Human cognitive interpretation and decision-making.
1. Understanding Digital Numbers (DN)
A major challenge in image processing is Bit Depth. While your monitor typically only displays 256 shades (8-bit), satellite sensors record thousands of variations.
The Truncation Problem: When we compress a high-resolution sensor's data down to a linear 8-bit display, we can lose up to 75% of the subtle variation! This is why analysts use "non-linear stretching" to highlight specific areas of interest (AOI) like river depressions or subtle moisture changes.
The logical progression from raw DN values to information extraction.
Radiometric Correction
Removes "haze" and atmospheric noise so we see the true surface reflectance.
Geometric Correction
Aligns the image to a map coordinate system (like UTM) so every pixel has a real-world location.
2. Anatomy of a Digital Image
At its core, a satellite image is a grid of Pixels (Picture Elements). Each pixel has a Digital Number (DN) representing the intensity of energy reflected or emitted from that spot on Earth.
Comparison: True Color (Natural) vs. False Color Infrared (Highlighting Vegetation in Red).
True Color
Bands: Blue, Green, Red.
Looks like what we see from a plane.
False Color (NIR)
Bands: Green, Red, Near-Infrared.
Healthy plants appear bright red.
2. Band Combinations & False Colors
Since multispectral and hyperspectral sensors capture wavelengths outside our visible range, we use False Color Composites to map invisible bands (like NIR or SWIR) into the Red, Green, and Blue channels of our screens.
The same landscape viewed through different band filters (Landsat 8-9 example).
| Combination (R,G,B) | Name | Primary Application |
|---|---|---|
| 4, 3, 2 | Natural Color | Visual verification, bathymetry, general mapping. |
| 5, 4, 3 | Color Infrared (CIR) | Vegetation health, crop monitoring, invasive species. |
| 7, 6, 4 | Short-Wave IR (SWIR) | Identifying moisture content, burn scars, and geology. |
| 6, 5, 2 | Agriculture | Deep contrast between crops and soil. |
Sustainability Note: Using the SWIR (7,6,4) combination is critical for post-fire recovery monitoring, as it can "see through" smoke and highlight exactly where the soil has been dehydrated by intense heat.
3. Hyperspectral: The Spectral Fingerprint
If multispectral sensing is like seeing in 10 colors, Hyperspectral Sensing is like seeing in the entire rainbow—and beyond. By using hundreds of narrow bands, we can create a continuous curve of reflectance.
From a single grayscale band to hundreds of narrow spectral channels: each step adds more spectral information, enabling increasingly precise material identification. Source: Esri.
| Capability | Multispectral (e.g. Landsat) | Hyperspectral (e.g. EMIT, AVIRIS) |
|---|---|---|
| Bandwidth | Broad (approx 50-100nm) | Ultra-Narrow (approx 5-10nm) |
| Detail | Discriminates general classes (Forest vs Water) | Identifies specific materials (Oak vs Pine, Clay vs Sand) |
| Data Volume | Manageable (MBs) | Massive (GBs per scene) |
Notice how the hyperspectral curve (continuous) captures tiny 'wiggles' that reveal chemical composition.
4. Spectral Indices: NDVI
Instead of just looking at pictures, we can use math to highlight specific features. The most famous is NDVI (Normalized Difference Vegetation Index).
How the ratio of NIR and Red light tells us about plant health.
NDVI = (NIR - Red) / (NIR + Red)
Healthy plants absorb Red light for photosynthesis but reflect NIR light to stay cool. A high NDVI value (close to 1.0) means lots of healthy green leaves!
5. Navigating the Data Cube & Harmonization
In modern remote sensing, we don't just use one satellite. We use Data Harmonization to fill the "gaps" in our data.
Temporal Gaps
Landsat gives high-quality data every 16 days. Cubesats fill the days in between.
Spectral Gaps
Startups "upgrade" standard imagery by adding specific missing bands (Hyperspectral).
By combining multiple sources, we take advantage of all resolutions (Spatial, Spectral, and Temporal) to create superior info-products.
The Pro-Tip: In a hypercube, you can extract a "Z-profile" for any pixel. This is the 1D spectrum used for Spectral Unmixing—finding out what percentage of a pixel is soil, water, or vegetation.
6. From Sensor to Decision
Processing hyperspectral data requires a systematic scientific workflow:
Standardized Hyperspectral Data Pipeline.
- ATMOSPHERIC CORRECTION: Remove haze and sunlight effects.
- DIMENSIONALITY REDUCTION: Use MNF or PCA to find 'useful' data.
- ENDMEMBER EXTRACTION: Find the 'purest' pixels in the scene.
- MAPPING: Use Spectral Angle Mapper (SAM) to classify the whole image.
🎓 Knowledge Check
1. Why do healthy plants look bright RED in a standard False Color Infrared (CIR) image?
2. Which challenge is MOST unique to Hyperspectral data compared to Multispectral?