Metrics

Views
219

In This Chapter

Advances in Hyperspectral Remote Sensing of Vegetation and Agricultural Crops

Authored by: Prasad S. Thenkabail , John G. Lyon , Alfredo Huete

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Print publication date:  December  2018
Online publication date:  December  2018

Print ISBN: 9781138058545
eBook ISBN: 9781315164151
Adobe ISBN:

10.1201/9781315164151-1

 Download Chapter

 

Abstract

Hyperspectral data (Table 1) is acquired as continuous narrowbands (e.g., each band with 1 to 10 nanometer or nm bandwidths) over a range of electromagnetic spectrum (e.g., 400–2500 nm). A user will use either continuous spectrum (referred to as whole spectral analysis or WSA) or selected optimal hyperspectral narrowbands (OHNBs) in the analysis. The WSA provides subtle differences, throughout the discrete hyperspectral narrowbands (HNBs) of the electromagnetic spectrum, that occur in quantitative measurements or qualitative observations of an object (e.g., crops, soil, water). Thereby, the WSA acts as a spectral signature of a feature or object or a combination of features or objects. For example, plants with erectophile structure like wheat or barley will have a significant slope along the near-infrared (NIR) shoulder (from 760 to 900 nm) whereas plants with planophile structure like soybeans have flat NIR shoulder. This will be completely missing in OHNBs. However, the OHNBs, help select HNBs that provide optimal information, avoiding data redundancy. An overwhelming proportion of the hyperspectral narrow-bands are often redundant for a given application. So, in order to maximize resource use, it will be extremely important to identify and remove the redundant bands from further analysis to ensure best use for resources and to overcome the curse of high-dimensionality of the hyperspectral data or Hughes phenomenon. When number of wavebands increase, the number of samples required to train, classify, and validate a product (e.g., crop types, biomass levels) also increases exponentially. This is called Hughes phenomenon. As a result, OHNBs become attractive to overcome Hughes’ phenomenon. Nevertheless, for full understanding and appreciation of HNBs, many users want to capture every subtle difference that occurs throughout the spectrum such as sensitivity of: 1. NIR shoulder to planophile versus erectophile plant structure, 2. red-edge (700 to 740 nm) to crop stress, 3. Moisture trough in NIR (930 to 1000 nm), 4. Visible bands (e.g., HNBs in 400 to 700 nm) to plant pigment, and many others. In an age of cloud computing, machine learning, and artificial intelligence using WSA with 100s or 1000s of HNBs is becoming increasingly realistic. The OHNBs in studies pertaining to agriculture and vegetation led us to identify 28 optimal hyperspectral narrow-bands (Table 2) that are best suited in the study of vegetation and agricultural crops. The wavebands were identified based on their ability to: (a) best model biophysical and biochemical properties, (b) distinctly separate vegetation and crops based on their species type, structure, and composition, and (c) accurately classify vegetation and crop classes. The study highlighted computation and use of various hyperspectral vegetation indices (HVIs) in studying and determining the most valuable HNBs (Table 2). The specific importance of each of the HNBs to biophysical and biochemical properties of vegetation and agricultural crops are also highlighted in Table 2. Typically, there is more than one waveband for each crop variable. For example, biophysical properties and yield are best modeled using wavebands centered at 682, 845, and 1100 nm. Similarly, moisture sensitivity in leaf\plant by several wavebands centered at 915, 975, 1215, 1518, 2035, and 2260 nm. The 28 wavebands (Table 2) will also allow us to compute 378 unique hyperspectral two-band vegetation indices (HTBVIs). Thereby, there is clear evidence that the combination of the 28 identified HNBs (Table 1.2) and various HVIs computed from them will suffice to best characterize, classify, model, and map a wide array of vegetation and agricultural crop types and their biophysical and biochemical properties.

 Cite
Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.