Questions: Hyperspectral Imaging

Question 1

A geologist needs to distinguish kaolinite (absorption at 2.16 um) from montmorillonite (absorption at 2.21 um). Why would a hyperspectral sensor succeed where a Landsat SWIR band (2.11-2.29 um) would fail?

• A. Hyperspectral sensors have higher spatial resolution
• B. Landsat SWIR integrates across the entire 2.11-2.29 um range, averaging out both narrow absorption features, while hyperspectral bands resolve each absorption separately
• C. Hyperspectral sensors use active illumination that penetrates rock
• D. Landsat cannot image in the SWIR due to atmospheric absorption

Answer: (B)

The absorptions are separated by only 50 nm. A Landsat band spanning 180 nm averages both into a single value. A hyperspectral sensor with 10 nm bands resolves each separately, allowing mineral-specific identification based on exact position, depth, and shape of the absorption.

Question 2

Hyperspectral imaging is always superior to multispectral imaging for remote sensing applications.

• T. True
• F. False

Answer: False

Hyperspectral data has hundreds of highly correlated bands creating processing challenges (curse of dimensionality), requires sophisticated atmospheric correction, demands large training datasets, and produces enormous data volumes. For many applications, multispectral data with 4-12 well-chosen bands provides sufficient discrimination at lower cost and complexity.

Question 3

What is spectral unmixing and why is it particularly important for hyperspectral data?

Model answer: Spectral unmixing decomposes a pixel's spectrum into fractional contributions of constituent materials (endmembers). Because pixels often contain multiple materials, the recorded spectrum is a mixture. With few multispectral bands, unmixing is underdetermined. Hyperspectral data provides enough spectral samples to solve for multiple endmember fractions reliably, yielding sub-pixel abundance maps critical for mineral mapping, fractional vegetation cover, and detecting small targets that do not fill entire pixels.

Spectral unmixing transforms analysis from 'what class is this pixel' to 'what fraction of this pixel is each material' -- enabled by the high spectral dimensionality.