• Home
• Tutorials
  • Data processing
    • Hyperspectral
    • LiDAR
    • Sampling design
  • Applications
    • Mapping
    • Modeling
    • Impact assessment
• News
• About

Hyperspectral data processing

What are hyperspectral images?

Hyperspectral sensors provide optical radiance data in >200 channels from 350-2500 nm with a spectral resolution of < 10 nm, generating images of contiguous spectral channels often spanning 400–1050 nm (visible to near-infrared) or 400–2500 nm (visible to shortwave-infrared) wavelength range (Figure 1), that can be use to characterize the chemistry of the vegetation, and allow a better discrimination between subtle physiological differences among plant species.

Figure 1. Hyperspectral image is composed by a large number of narrow spectral bands (a) that results in a spectral curve (b) for each image pixel.

What types of variables can be extracted from hyperspectral data?

Hyperspectral data allows to trace, via spectral separability, subtle differences in the leaf pigment, nutrient and structural properties of the vegetation at different levels of aggregation. Such group-specific spectral properties are known as reflectance signatures that can be use to map and monitor vegetation changes from above.

Hyperspectral data has multiple applications, from allowing the differentiation between bare ground and ground with vegetation cover, and invasive vs. native species, to the accurate identification and mapping of single invasive and native species.

The dataset

To create the dataset provided in this section of the tutorial, we generated an artificial hyperspectral image (Figure 2) from the original APEX data of Sylt island (Germany) collected and preprocessed by the Flemish Institute for Technological Research VITO. The artificial image was created by taking a buffer area from the center of the calibration, validation and background points, and then rearranging the buffer squares together with the corresponding points, which resulted in an image with the real hyperspectral values (grid cells) at each point but reorganize in a smaller file (Figure 3). You can download the dataset for this tutorial from HERE.