Hyperspectral sensors — imaging spectrometers — measure hundreds of narrow, contiguous wavelength bands for every pixel, capturing an essentially continuous reflectance spectrum. Where multispectral imagery distinguishes broad material classes, hyperspectral data can identify materials: specific minerals, vegetation species, water constituents. It is the closest remote sensing comes to laboratory spectroscopy from a distance.
From bands to spectra
A multispectral sensor samples perhaps ten broad wavelength windows; a hyperspectral instrument samples two hundred narrow ones. The result per pixel is a spectral curve whose absorption features — dips at diagnostic wavelengths — fingerprint materials the way lines identify elements in astronomy.
Analysis matches these curves against spectral libraries or unmixes each pixel into fractions of pure “endmember” spectra, acknowledging that a 30 m pixel rarely contains just one material.
Signature applications
Mineral exploration reads clay, carbonate and iron-oxide absorption features to map alteration halos around ore systems. Precision ecology separates tree species and detects stress before it is visible. Water-quality monitoring quantifies chlorophyll, sediment and coloured dissolved organics. Agriculture distinguishes nutrient deficiencies that broad bands blur together.
Spaceborne missions such as EnMAP and PRISMA, airborne instruments, and increasingly drone-mounted spectrometers supply the data across scales.
Why it is genuinely difficult
Hundreds of correlated bands mean heavy data volumes, low signal per band, and statistical traps (the “curse of dimensionality”) for naive classifiers. Rigorous atmospheric correction is not optional — atmospheric features overprint the diagnostic ones.
Consequently, hyperspectral value is unlocked by specialist processing: careful calibration, dimensionality reduction, physically informed feature selection and validation against field spectra.