Global Hyperspectral Imaging Spectroscopy of Agricultural-Crops & Vegetation (GHISA)

Science Center Objects

This webpage showcases the key research advances made in hyperspectral remote sensing of agricultural crops and vegetation over the last 50 years. There are three focus areas:

Ongoing research in Global Hyperspectral Imaging Spectral-library of Agricultural-Crops (GHISA)

Hyperspectral Imaging Spectroscopy (HIS) provides the most advanced next-generation remote sensing data in the spectral domain in the form of spectral signatures for every pixel observed as opposed to data from a few spectral data points of the current generation of sensors (Figure A). This will help us advance our understanding of Planet Earth through more accurate and precise models, maps, and monitoring capabilities. Given, this importance, Dr. Itiya Aneece and Dr. Prasad Thenkabail of the United States Geological Survey (USGS) conduct a detailed study of five leading world agricultural crops (corn, soybean, winter wheat, rice, and cotton) that occupy 75% and 54% of principal crop areas in the United States and the world respectively using HIS data from Earth Observing-1 (EO-1) Hyperion data by utilizing 99 images spread across various agroecological zones (AEZs) of USA (Figure B). The study established optimal hyperspectral narrowbands (HNBs) to study agricultural crops, conducted dimension reduction studies through data mining approaches, and established accuracies and errors in classifying Hyperion images for crop type mapping using machine learning algorithms (e.g., Support Vector Machines) on the Google Earth Engine (GEE) cloud computing platform.  The study will make a significant contribution to upcoming hyperspectral sensors such as the recently launched HysIS (Hyperspectral Imaging Satellite of India with 55 bands in 400-950 nm VNIR and 165 bands in 900-2500 nm SWIR) by India, and NASA’s Surface Biology and Geology (SBG) mission. This study built a novel, first-of-its-kind, global hyperspectral imaging spectral-library of agricultural crops (GHISA) (Figure A).  For detailed descriptions of Hyperion image processing and compilation of GHISA, please refer to Chapter 9 on Hyperion preprocessing in this book, Google Earth Engine code, and the manuscript “Accuracies achieved in classifying five leading world crop types and their growth stages using optimal Earth Observing-1 Hyperion hyperspectral narrowbands on Google Earth Engine” published in the Remote Sensing open-access journal providing details of Global Hyperspectral Imaging Spectral-library of Agricultural-crops (GHISA).  The GHISA derived hyperspectral library (example Figure B below) will be of immense value to understand, model, map, and monitor agricultural crops within and across agroecological zones (Figure B).

GHISA - Figure A

Click Figure A above for description


GHISA Illustration - Figure B

Click Figure B above for description


Research conducted over the last 20 years (by Dr. Prasad S. Thenkabail’s group)

The primary focus is on the seminal research conducted over the last 20 years by this group using hyperspectral data of agricultural crops and vegetation, gathered from around the world, from various platforms: ground-based, platform-mounted, drone or unmanned aerial vehicle (UAV)-based, airborne, and spaceborne. The leading world crops studied include wheat, rice, corn, soybeans, barley, cotton, and alfalfa (Table 1). Natural vegetation in the African savannas and rainforests were also studied using hyperspectral (or imaging spectroscopy) data. Characteristics of these sensors can be found in Table 2. For example, Figure 1 illustrates Earth Observing-1 (EO-1) Hyperion hyperspectral Data Cube from which agricultural crop spectral libraries are derived for some of the leading world crops. Figure 2 illustrates EO-1 Hyperion hyperspectral data cube for data acquired over African rainforests.

Earth Observing-1 (EO-1) Hyperion hyperspectral data over the Ferghana Valley in Uzbekistan, Central Asia.

Click Figure 1 above for description

Earth Observing-1 (EO-1) Hyperion hyperspectral data cube of study areas in the Congo rainforests in Cameroun

Click Figure 2 above for description

Research conducted over the last 50 years (by the global remote sensing community)

The secondary focus is on the compendium of the hyperspectral remote sensing of vegetation and agricultural crops research conducted by a large community of global researchers over the last 50 years. This is summarized in the book “Hyperspectral Remote Sensing of Vegetation” (Publisher: Taylor and Francis, Inc., see Figure 3).

Hyperspectral Remote Sensing of Vegetation

Click Figure 3 above for description

However, a more comprehensive 4-volume “Hyperspectral Remote Sensing of Vegetation and Agricultural Crops” (Publisher: Taylor and Francis, Inc.) has been published by the end of 2018 or early 2019.

Global hyperspectral imaging spectral-library of agricultural-crops (GHISA)

The tertiary focus is in establishing a global hyperspectral imaging spectral-library of agricultural-crops (GHISA) for some of the major crops of the world (Table 1) for various locations of the world and at various growth stages. The crops include wheat, rice, corn, soybeans, barley, and cotton (Table 1). Hyperspectral data for GHISA are gathered from ground-based (Figure 4) and spaceborne sensors (see Table 2 for data characteristics of these sensors). Data gathered from field spectroscopy using spectroradimeters are illustrated in Figure 5 for some of the leading crops of the world. Such signatures constitute GHISA. Similar GHISA spectral signatures are gathered from various platforms (e.g., Figure 1 and Figure 6 from spaceborne EO-1 Hyperion).

Field spectroscopy to gather hyperspectral data in 100s or 1000s of narrowbands

Click Figure 4 above for description

Hyperspectral signatures of some of the leading world agricultural crops gathered at various crop growing phases.

Click Figure 5 above for description

Distribution of Earth Observing-1 (EO-1) Hyperion hyperspectral images acquired by NASA and USGS.

Click Figure 6 above for description

Hyperspectral Remote Sensing or imaging spectroscopy, originally used for detecting and mapping minerals, is increasingly needed to characterize, model, classify, and map agricultural crops and natural vegetation, specifically in the study of:

  1. Species composition (e.g., Chromolenea odorata vs. Imperata cylindrica),
  2. Vegetation or crop type (e.g., soybeans vs. corn),
  3. Biophysical properties (e.g., LAI, biomass, yield, density),
  4. Biochemical properties (e.g, Anthrocyanins, Carotenoids, Chlorophyll),
  5. Disease and stress (e.g., insect infestation, drought),
  6. Nutrients (e.g., Nitrogen),
  7. Moisture (e.g., leaf moisture),
  8. Light use efficiency,
  9. Net primary productivity and so on.

The goal of using hyperspectral narrowband (HNB) data, as opposed to multispectral broadband (MBB) data, is to increase accuracies and reduce uncertainties in the study of various vegetation and agricultural crop classification outcomes, and in modeling and mapping these quantities. The MBBs have limited bands (e.g., Landsat Operational Land Imager has 11 bands) whereas HNBs are available in 1000s of bands (e.g., EO-1 Hyperion has 242 bands in the 400-2500 nanometer range). A simple example is the depiction of various false color composites (FCCs) from EO-1 Hyperion HNBs relative to Landsat Enhanced Thematic Mapper Plus (ETM+) FCC of 4,3,2 as shown in Figure 7.

Earth Observing-1 (EO-1) Hyperion hyperspectral data depicted in various false color composites (FCCs)

Click Figure 7 above for description

Advances made in using hyperspectral narrowband (HNB) data (e.g., Figure 8, Figure 9), relative to multispectral broadband (MBB) data, include:

  1. significantly improved characterization and modeling of a wide array of biophysical and biochemical properties of vegetation,
  2. ability to discriminate plant species and vegetation types with high degree of accuracy,
  3. reducing uncertainties in determining net primary productivity or carbon assessments from terrestrial vegetation,
  4. improved crop productivity and water productivity models, 
  5. ability to assess stress resulting from causes such as management practices, pests and disease, water deficit or water excess, and
  6. establishing more sensitive wavebands and indices to study vegetation characteristics.
Earth Observing-1 (EO-1) Hyperion hyperspectral signatures of some of the vegetation in the African rainforest.

Click Figure 8 above for description

Earth Observing-1 (EO-1) Hyperion hyperspectral signatures of some of the vegetation in the African rainforest.

Click Figure 9 above for description

This Figure description is the same as that of Figure 8, except that the data are plotted here (Figure 9) in 400-940 nm as opposed to 400-2400 nm in Figure 8.

A wide range of hyperspectral research topics are covered here, such as:

  1. hyperspectral sensors and their characteristics,
  2. methods of overcoming the Hughes phenomenon,
  3. characterizing biophysical and biochemical properties,
  4. advances made in using hyperspectral data in modeling evapotranspiration or actual water use by plants,
  5. study of phenology, light use efficiency, and gross primary productivity,
  6. improved accuracies in species identification and land cover classifications, and
  7. applications in precision agriculture.

One of the first goal of any hyperspectral data analysis is in dimensionality reduction to overcome data redundancy and to optimize the most valuable data for a given application. This requires data mining. Feature selection is necessary in any data mining effort. Feature selection reduces the dimensionality of data by selecting only a subset of measured features (predictor variables). Feature selection methods recommendation is based on:

  1. Information Content (e.g., Selection based on Theoretical Knowledge, Band Variance, Information Entropy),
  2. Projection-Based methods (e.g., Principal Component Analysis or PCA),
  3. Independent Component Analysis (ICA),
  4. Divergence Measures (e.g., Distance-based measures),
  5. Similarity Measures (e.g., Correlation Coefficient, Spectral Derivative Analysis), and
  6. Other Methods (e.g., Wavelet Decomposition Method).

Hyperspectral narrowband (HNB) data are known to provide significant advances in modeling, mapping, and monitoring agricultural crop and vegetation biophysical and biochemical quantities. Biophysical characteristics that are typically studied are (Figure 10):

  1. Biomass: wet and dry (kg\m2),
  2. Leaf area index (LAI),
  3. Green LAI (m2\m2),
  4. Plant height (mm),
  5. Vegetation fraction (%),
  6. Fraction of PAR absorbed by photosynthetically active vegetation (fAPAR) (MJ\m2),
  7. Total crop chlorophyll content (g\m2), and
  8. Gross primary production (g C\m2\yr).

Biochemistry (e.g., plant pigments, water, and structural carbohydrates): Leaf reflectance in the visible spectrum is dominated by absorption features created by plant pigments, such as:

  1. chlorophyll a (chl-a): absorbs in 410-430 nm and 600-690 nm;
  2. chlorophyll b (chl-b): absorbs in 450-470 nm;
  3. carotenoids (e.g., β-carotene and lutein): peak absorption in wavebands <500 nm;
  4. anthocyanins,
  5. lignin, cellulose, protein, and
  6. Nitrogen: relatively low reflectance and strong absorption in SWIR bands by water that masks other absorption features. However, dry leaves do not have strong water absorption and reveal overlapping absorptions by carbon compounds, such as lignin and cellulose, and other plant biochemicals, including protein, nitrogen, starch, and sugars.
Hyperspectral signatures of some of the leading agricultural crops

Click Figure 10 above for description

Hyperspectral vegetation indices (HVIs) (e.g., Table 3) provide an opportunity to study specific biophysical and biochemical quantities more precisely than multispectral broadband derived indices. HVIs are either two-band or multi-band types. The novelties of HVIs include:

  1. physically meaningful indices (e.g., Photochemical reflective index (PRI) as proxy for light use efficiency (LUE)),
  2. significant improvement in sensor saturation relative to broadbands,
  3. providing greater sensitivity (e.g., an index involving NIR reflective maxima @ 900 nm and red absorption maxima @ 680 nm);
  4. new indices not sampled by broadbands such as water-based indices (e.g., involving 970 nm or 1240 nm along with a non-absorption band, and
  5. multi-linear indices involving more than 2 bands.

The comprehensive research also recommends optimal hyperspectral narrowbands (see Table 4) and the best incremental bands (see Table 5) in the study of agriculture and vegetation.

Global Hyperspectral Imaging Spectral-library of Agricultural Crops (GHISA)

Remote sensing is increasingly getting sophisticated with data gathered from increasing spatial, spectral, radiometric, and temporal resolutions. This data are also gathered from multiple platforms such as ground-based, truck-mounted, airborne (e.g., drones, aircrafts), and spaceborne sensors. This allows us to study the Planet with greater precision and accuracy. In this context, hyperspectral remote sensing is considered one of the most advanced tools in the study of the Planet. Hyperspectral data are gathered continuously or near-continuously in narrowbands (< 10 nm) along the entire spectrum (e.g., 400-2500 nm) (e.g., Figure below). The number of bands could be in 10s, or 100s, or 1000s depending on the bandwidth chosen. For example, Earth Observing-1 (EO-1) Hyperion had 242 bands, each of 10 nm bandwidth in the 400-2500 nm range. The data are called hyperspectral not by the number of bands, but by ensuring that a number of narrowbands are chosen throughout the spectral range (e.g., 400-1100 nm or 400 to 2500 nm or 400-12500 nm).

Hyperspectral signatures of some of the leading agricultural crops

Click Figure above for description

Global Hyperspectral Imaging Spectral-library of Agricultural Crops (GHISA) is a comprehensive collection and collation of a wide range of agricultural crop and vegetation hyperspectral data of agricultural crops, vegetation type, species, their growth stages, and their growing conditions. GHISA is developed by linking the hyperspectral signatures to: 1. precise geographic locations, providing latitude and longitude, from where the data are gathered, and 2. Platform (e.g., ground-based, airborne, spaceborne) from which the data are gathered. GHISA will help us better understand, model, map, and monitor biophysical, biochemical, and structural quantities of agricultural crops and vegetation.


For more details, click here.


United States of America:

Study areas throughout the US in various agroecological zones




README, click here.

Dataset, click here.







Central Asia:

Study areas throughout Central Asia



README, click here.

Dataset, click here.









Study areas throughout Africa




README, click here.

Dataset - grasses - shrubs - weeds, click here.

Dataset, click here.