Every pixel carries a truth invisible to the eye. Hyperspectral imaging doesn’t just capture color, it captures the chemistry of the world.
Intro
This post is based on one of our lab assignments in Applied Remote Sensing at Delta State University. The task was to explore the power of hyperspectral imagery, the kind of data that captures hundreds of spectral bands and reveals more about the Earth’s surface than what is visible to the human eye.
Unlike multispectral data that focuses on a few wide bands, hyperspectral remote sensing measures energy reflected across hundreds of narrow, contiguous bands. It can detect differences so subtle that two plants of the same species may show distinct spectral responses depending on their health or water content. As explained by Lillesand, Kiefer, and Chipman (2015), this precision allows for identifying vegetation types, soil minerals, and other materials based purely on their spectral behavior.
In this lab, I worked with hyperspectral data from the Hyperion sensor aboard NASA’s Earth Observing-1 (EO-1) satellite using ENVI software. The goal was to visualize, classify, and interpret these data to understand how materials reflect and absorb light.
Step 1: Preparing the Hyperspectral Image
The Hyperion sensor collects reflected energy across 220 spectral bands, each about 10 nanometers wide, covering the visible, near-infrared, and shortwave infrared regions. For this exercise, we used 152 bands after removing noisy and overlapping ones to ensure data clarity.
Using ENVI, I created a true-color composite from the following bands:
| Display Channel | Hyperion Band | Spectral Region | Purpose / Expected Appearance |
|---|---|---|---|
| Red | Band 29 | Visible red | Highlights soil and built-up areas |
| Green | Band 20 | Visible green | Displays vegetation in natural tones |
| Blue | Band 12 | Visible blue | Enhances water and shadow features |
The resulting composite provided a natural view of the scene, showing green vegetation, dark water bodies, and bright built-up zones. It served as the foundation for linking visible features with their spectral information.
As Jensen (2016) notes, each surface material reflects and absorbs electromagnetic energy differently based on its physical and chemical properties. Establishing a visually accurate true-color composite ensures that radiometric and spectral consistency are maintained before performing any classification.
Step 2: Understanding Image Storage Format
The EO-1 Hyperion image was stored in Band Sequential (BSQ) format. In this structure, each band is stored as a complete two-dimensional image, one after another. It is straightforward for display and band-by-band manipulation.
However, other storage types such as Band Interleaved by Line (BIL) and Band Interleaved by Pixel (BIP) can be more efficient for hyperspectral analysis. The BIP format, in particular, stores all spectral band values for a pixel together, which makes spectral classification and similarity analysis faster and more effective.
Determining the interleave type was done both through the ENVI interface and by manually inspecting the .hdr file. The header metadata confirmed that the image used BSQ format. While this format works well for simple visualization, the BIP layout would be preferable for advanced pixel-level spectral processing.
Step 3: Exploring Spectral Features
Each pixel in a hyperspectral image is a signature, a detailed record of how that surface interacts with light across hundreds of wavelengths. Using ENVI’s Z-Profile and Spectral Profile tools, I analyzed reflectance characteristics of individual pixels.
A pixel at Sample 944, Line 3975 represented cultivated vegetation. The spectrum revealed strong absorption in the visible red region (around 680 nm) and high reflectance in the near-infrared range, forming what is known as the “red edge”, a classic indicator of healthy vegetation (Lillesand et al., 2015).
The spectral profile was saved into a spectral library for use as an endmember, a reference spectrum that serves as a benchmark during classification. As NV5 Geospatial (2023) notes, maintaining a spectral library allows for direct comparison between materials, enabling more accurate identification and classification in future analyses.
To further explore variability, I analyzed three additional pixels: (931, 4083), (936, 4084), and (765, 4622). The first two showed strong near-infrared reflectance typical of dense vegetation, while the third exhibited a flatter curve associated with bare soil or sparse vegetation.
Comparing these profiles revealed that subtle changes in vegetation density, soil background, and moisture content significantly alter spectral responses.
Step 4: Classification and Endmember Analysis
With the spectral library established, the Spectral Angle Mapper (SAM) technique was used to classify the image. SAM compares the spectral angle between each pixel and known reference spectra (endmembers). A smaller angle means a closer match.
Using vegetation and soil endmembers, the classification effectively mapped crop zones but also showed some misclassifications where mixed pixels or canopy variations caused deviations from ideal signatures.
As Jensen (2016) explained, illumination differences can affect spectral values, but SAM mitigates this by focusing on the shape of the curve rather than absolute reflectance.
Step 5: Spectral Feature Fitting and Continuum Removal
To highlight specific absorption features, I used the Spectral Feature Fitting (SFF) technique, which applies continuum removal, a method that normalizes spectra to enhance diagnostic absorption patterns (NV5 Geospatial, 2023).
This process isolates key absorption features, improving mineral and vegetation interpretation. When comparing raw reflectance (EFFORT-corrected) to continuum-removed data, the latter displayed clearer and deeper absorption bands.
Pixel-by-pixel comparison between EFFORT-corrected and continuum-removed spectra showed that continuum removal minimized brightness effects and emphasized absorption depth and shape.
The improvement was clear: the continuum-removed spectra revealed patterns linked to minerals such as kaolinite and alunite, which would be less visible in the unprocessed data.
Step 6: Key Insights
Hyperspectral data are more than images. They are quantitative measurements of how the Earth interacts with light. Through this lab, it became evident that hyperspectral analysis is not just about visualization but about extracting meaningful physical information from every pixel.
Each step, from true-color composites to continuum removal, builds a deeper understanding of how energy, matter, and information intersect. The combination of SAM and SFF demonstrates how scientists can identify vegetation health, monitor environmental changes, and detect minerals with remarkable precision.
As Lillesand, Kiefer, and Chipman (2015) describe, hyperspectral sensing bridges the gap between physical composition and remote measurement. Jensen (2016) expands on this by noting that the value of such data lies in its ability to turn electromagnetic readings into actionable insights about the planet’s surface and atmosphere.
Outro
This lab reinforced a fundamental idea: remote sensing is about seeing the unseen. The EO-1 Hyperion data allowed me to visualize the invisible, to read the spectral language of the Earth.
From creating true-color composites to performing SAM and SFF analysis, I learned how hyperspectral imagery connects physics, ecology, and computation. These methods are indispensable in environmental monitoring, precision agriculture, and resource assessment.
Hyperspectral remote sensing turns light into knowledge.
References
Jensen, J. R. (2016). Introductory Digital Image Processing: A Remote Sensing Perspective (4th ed.). Pearson Education.
Lillesand, T., Kiefer, R. W., & Chipman, J. (2015). Remote Sensing and Image Interpretation (7th ed.). Wiley.
NV5 Geospatial Software. (2023). ENVI User’s Guide: Spectral Feature Fitting and Continuum Removal Methods. NV5 Geospatial.