Home Articles Hyperspectral imaging: Beyond the niche

Hyperspectral imaging: Beyond the niche

Prof. Ian Dowman

Prof. Ian Dowman
Editor – Europe
Geospatial World
[email protected]

Hyperspectral imaging has occupied an important niche in the technology and applications of Remote Sensing for some time. Has it now reached a tipping point where it can become a viable commercial activity across the board of applications of earth observation and imaging science? Read on to know..

Hyperspectral remote sensing involves the acquisition of image data in many narrow, contiguous spectral bands which generate detailed spectral signatures of the objects imaged. This data can be analysed to determine the nature of an object which can be identified by comparing the signature obtained with a signature in a library. Hyperspectral imagery differs from multispectral imagery in that multispectral bands give information in a few discrete bands, while hyperspectral gives continuous information over many bands which are contiguous. Figure 1 illustrates the processes involved in generating spectral signatures from hyperspectral data. A comparable diagram for multispectral imagery would show far fewer images in the spectral dimension and only a few discrete points in the signature.

Hyperspectral remote sensing is synonymous with imaging spectroscopy. Spectroscopy is defined as the study of the interaction between matter and radiated energy. Initially, the main application of hyperspectral remote sensing was the identification of minerals, but applications are now much broader and include agriculture, forestry and water and air quality. Data is acquired from airborne and spaceborne platforms. K. Staenz (Terrestrial Imaging Spectroscopy – Some Future Perspectives; 2009) has given a review of the development of hyperspectral remote sensing and of the issues facing the subject today. The techniques can be illustrated by the sensor HyMap, produced by HyVista Corp. Table 1 gives the spectral configuration of the HyMap sensor for each of the 4 spectral modules containing 32 bands each, totalling 128 spectral bands. The sampling interval of each channel is in the range of 13 – 17nm, which is typical of most sensors, as shown in Table 2.

Figure 2 shows signatures acquired from HyMap of various minerals. The composite images show the delineation of minerals superimposed on the images.

There are also a number of other airborne sensors and spaceborne systems available.

Airborne sensors
Airborne hyperspectral data has been collected since the early 1980s. Since then, sensors have been developed to provide high spectral resolution, normally with a range of 10-20nm. High spatial resolution is possible, although it is dependent on the operational altitude of the platform. Table 2 lists some of the current systems.

Table 2 indicates the range of sensors available and shows that the core spectral range is from the visible (VIS) to the near infrared (NIR) part of the electro-magnetic spectrum. Some instruments extend to the thermal infrared (TIR) while some companies produce separate TIR sensors. The sampling distance is typically 10 – 16nm. For example, DAIS, developed by DLR in Germany, covers the spectral range from the visible to the thermal infrared wavelengths at spatial resolution varying from 3 to 20 m, depending on the aircraft altitude. The DAIS is used for remote sensing applications such as environmental monitoring of land and marine ecosystems, vegetation status and stress investigations, agriculture and forestry resource mapping, geological mapping, mineral exploration as well as for the supply of data for geographic information systems. Six spectral channels in the 8000 – 12000 nm region could be used for the retrieval of temperature and emissivity of land surface objects. These and 72 narrow band channels in the atmospheric windows between 450 and 2450 nm allow investigation of land surface processes with special emphasis on vegetation / soil interactions.

PROBE-1 can be flown over a range of altitudes to provide pixel sizes ranging from 1 to 10 metres and swath widths ranging from <1 km to 6 km. At 2500 metres, PROBE-1 has a swath width of 3 kilometres, with a ground sampling distance (GSD) of 5 metres. From 5000 metres, the swath width is 6 kilometres with a GSD of 10 metres.

NASA-developed AVIRIS (Airborne Visible InfraRed Imaging Spectrometer) is an important instrument in the realm of Earth remote sensing because of its extensive use in the early days of hyperspectral imaging. It is a unique optical sensor that delivers calibrated images of the upwelling spectral radiance in 224 contiguous spectral channels (also called bands) with wavelengths ranging from 400 to 2500 nanometres (nm). AVIRIS has been flown on four aircraft platforms at altitudes between 4 km above sea level and 20 km. HyMap and CASI are two of the most popular commercial systems.

Spaceborne sensors
Since 2000, hyperspectral sensors have been used on spaceborne platforms. These sensors are listed in Table 3.

The first satellite-based hyperspectral sensor was Hyperion on the NASA EO-1 platform; this was a test bed instrument designed to demonstrate hyperspectral imaging from space. This was followed by CHRIS on the ESA PROBA platform, a hyperspectral instrument whose objective is to collect bidirectional reflectance distribution function (BRDF) data for a better understanding of spectral reflectances. The technology objective is to explore the capabilities of imaging spectrometers on agile small satellite platforms.

The use of the HySI hyperspectral sensor on the Indian Moon mission Chandrayan-1, launched in 2008, was considered essential in determining the mineral composition of the lunar surface. It was flown with the terrain mapping camera (TMC) which used stereoscopic imagery to determine a digital terrain model. This enabled scientists to see the variation of mineral content with elevation. This will help plan landing sites for future lunar missions. Pan sharpening of the HySI and TMC data was an invaluable aid in extraction of additional information. A hyperspectral sensor was also used on the Mars Reconnaissance Orbiter flown by NASA.

The goal of the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) is to find the spectral signatures of minerals that formed in liquid water, which could have provided environments suitable for life. The goal is also to measure the changing amounts of water and other volatiles in the atmosphere and as polar ice. It also maps the geology, composition and layering of surface features.

Future space systems include EnMAP (Environmental Mapping and Analysis Program) as part of the Earth observation programme conducted by the German Space Agency (DLR). Using the five successful SAR-Lupe satellites as a basis, the platform is being modified to accommodate a new kind of hyperspectral instrument. It can be seen that the spectral and spatial resolution of spaceborne systems is increasing with time. Other systems under development are CHRIS-2 from Surrey Satellite Technology Ltd (SSTL) and a sensor from Korea (Staenz 2009).

Because of the large volume of data collected from a hyperspectral sensor, data processing is not straightforward. The raw data has to be georeferenced and radiometric correction has to be applied. Radiometric correction must take account of lighting effects due to sun angle and topography, atmospheric transmission and sensor gain. These corrections are important if spectral signatures libraries are to be used. The USGS spectral library contains over 1300 spectra in the wavelength range of 0.2 to 3.0µm.

It has many more minerals, organic and volatile compounds, vegetation and man-made materials. Another existing spectral library is the Jet Propulsion Laboratory (JPL) which includes 160 spectra of minerals in the wavelength range of 0.4 to 2.5µm. The ASTER spectral library based at NASA JPL is a compilation of over 2400 spectra of natural and manmade materials and is based on the other spectral libraries at USGS and John Hopkins University.

Commercial software exists to process hyperspectral data. ENVI is a popular software and comes with wide variety of functionalities and a spectral library. IDL can be integrated with ENVI, providing additional functionality. In many cases, software is also provided by sensor manufacturers and operators.

Nowadays, there is a general interest in making the optimum use of different types of data. Aircraft frequently carry and operate several sensors. The widespread use of laser scanning has encouraged operators to use this with hyperspectral data to combine topographic data to gain additional information on the spatial distribution of land cover and man made features. This has been demonstrated by Chandrayan-1.


Hyperspectral data has been a fertile area for research since data first became available. Much of the scientific research is based on data from AVIRIS, Hyperion and CHRIS provided by NASA and ESA. More recently, the WHISPERS (The Workshop on Hyperspectral Image and Signal Proccesing – Evolution in Remote Sensing) workshop series run by IEEE-GRSS and attended by researchers and practitioners, has been a major event for those involved in hyperspectral remote sensing. Figure 3 shows the range of topics discussed in 2010.

Recent research has looked at such topics as the removal of shadow effect from hyperspectral imagery and classification of roofs and road extraction using hyperspectral data. This recognises that spectral signatures which are the distinctive elements of image pixels cover a variety of materials ranging from man-made materials like asphalt, roof materials like aluminium as well as natural ones like vegetation types. This type of information is designed to be used with 3D city models and other landscape information systems.

There is also an interest in using close range hyperspectral sensors for geological applications. The images can be combined with 3D models from terrestrial laser scanning to improve interpretation through better visualisation. This is another example of data fusion.

Commercial applications
The mining industry was the initial stimulus for using hyperspectral data as it had been known for many years that the presence of minerals could be detected from the chemical composition of the soil along with associated land cover. Multispectral data such as Landsat was used as soon as it became available. As hyperspectral sensors were developed, the data quickly became a tool for mineral exploration. Now, data is used in many areas including agriculture, forestry and land cover. Hyperspectral data is used to identify individual plant and tree species, noxious gases and aerosols, minerals and soil types. They are also used in analysing air quality.

Several companies which manufacture hyperspectral sensors also operate them. Table 4 lists some of the companies which operate airborne hyperspectral sensors and the application areas which they promote. The use of hyperspectral images for mining exploration is a long standing application and an example is given in Figure 2.

The government of Queensland, Australia sees hyperspectral surveys as a way to 'enable rapid, no impact data collection, and complement satellite imagery, aerial photography, magnetics, radiometrics and gravity surveys.' They commissioned a survey covering 25,000 km2 of data from north Queensland using HyMap with a spatial resolution of 4.5 m. The outcomes of the project were released in 2009 and comprise a report and a series of over 500 GIS-compatible image maps showing surface mineralogy, chemistry, vegetation and landscape textures. The resulting mineral, chemistry, vegetation and surface texture maps are of great assistance to mineral explorers by identifying areas of alteration and enabling mineral emplacement models to be tested.

Another example from Australia comes from the Western Australia Minerals and Energy Research News. 'Following on from recent award-winning research into the use of airborne hyperspectral imaging to assess ironore dust loadings in the Port Hedland area, this project evaluates the technology's further use in environmental monitoring. A team led by CSIRO researcher Cindy Ong and Mark Piggott, BHP Billiton Iron Ore, is applying hyperspectral technology to monitor several objectives, including:

  • Developing case histories of mine site and related operations in monitoring iron ore, bauxite and nickellaterite mining;
  • Establishing standards for hyperspectral technology such as reproducibility and accuracy;
  • Developing an operational monitoring system for ironderived dust in Port Hedland
  • Promoting the writing of guidelines for hyperspectral technology as best practise procedures.

This is in concert with legislative bodies such as Environment Protection Authority and Department of Industry and Resources. Anticipated benefits to industry from this 18-month project are more efficient assessment and management of mining environment activities. There also will be significant benefits for people in dangerous working conditions and for environmentally sensitive zones since the need for ground access may be alleviated.'

An application of growing importance is vegetation and forestry, often combining hyperspectral data with LiDAR. Merrick, for example, was awarded a contract to provide high-resolution hyperspectral, LiDAR, and natural colour imagery along the Missouri River in Nebraska and Missouri. The Kansas City Corps of Engineers used the simultaneously collected datasets for habitat modelling in and along the Missouri River. This was a pilot project undertaken by the Corps of Engineers to determine the feasibility of using the three remote sensing datasets in combination with model habitat adequately versus traditional methods. The identification of cottonwood trees habitable for eagles is the primary focus. Merrick acquired hyperspectral data, LiDAR, and true colour imagery during the peak growing season (July 2007) to improve accuracy in species classification and health. The hyperspectral data was collected at 1 meter resolution with 128 spectral bands. Merrick analysed the hyperspectral data and classified it into vegetation classes and vegetation species. The hyperspectral data will be used to classify and determine the health of the vegetation; the LiDAR data will be used to determine the height and density of the vegetation.

Environmental monitoring is another important application. The New Hampshire Department of Environmental Services (DES), in collaboration with the New Hampshire Estuaries Project (NHEP) in the US commissioned a hyperspectral survey to investigate the increase in nitrogen concentration and declining eelgrass beds in Great Bay Estuary which have been observed in the last few decades. One of the hypotheses put forward regarding eelgrass decline was that a possible eutrophication response to nutrient increases in the Great Bay Estuary has been the proliferation of nuisance macroalgae, which has reduced eelgrass area in Great Bay Estuary. This hypothesis was tested by mapping eelgrass and nuisance macroalgae beds using hyperspectral imagery. The result of the survey and analysis is shown in figure 4.


Commercial drivers
Scientists have demonstrated that hyperspectral data can be very useful in a number of applications. Government research organisations and government geological mapping agencies sponsor research and surveys and often there is strong collaboration between these agencies and industry, as shown earlier. This occurs at the CSIRO in Australia, the Geological Survey in Canada, and Geological Survey of Denmark and Greenland (GEUS). Hyperspectral imaging and analysis is used commercially in the exploration part of the mining industry but only by the very big, major companies who own their own instrument and aircraft, like HyMAP. The example from Western Austalia indicates the interest of BHP Bilton in the use of hyperspectral data. It is still too expensive for most major and all minor companies. The costs are equivalent to other airborne surveys and so there will always be financial competition with them. Traditional airborne geophysical surveys can be argued to be generally useful in a wider variety of mineral deposit environments than hyperspectral.

Barriers to development
Staenz (2009) notes that issues that have slowed acceptance and use of hyperspectral data include: inadequate correction for sensor and atmospheric effects, availability and suitability of specific analysis software, data availability and the relative paucity of well-trained scientists to analyse the data. There are also issues of calibration, banding and low signal-to-noise quality. Spaceborne hyperspectral sensors tend to have a narrow field of view which makes it impossible to achieve global coverage and a rapid revisit. Airborne sensors are expensive to operate, especially in remote areas where mineral exploration is required. The spaceborne systems operating at present were designed as technology demonstrators and lack high resolution and revisit capabilities. Operational products are not available and many end users, or potential end users, find using the data and the software challenging. They also find it expensive and labour intensive.

The future
It has been mentioned earlier that the field of hyperspectral remote sensing is developing: commercial sensors, operators and software exist and satellite systems with better resolution are in the pipeline. Staenz (2009) states: 'The good news is that the area of hyperspectral remote sensing is quickly reaching critical mass. Operational atmospheric correction algorithms are now available and many others are under development. Most modern image processing systems can at least handle the high number of spectral bands and new algorithms are under development. Several software systems are available that are specifically designed to perform hyperspectral analysis.' The number of airborne sensors which are available suggests a growing market and certainly many potential applications. There is user resistance, brought about by the cost of hyperspectral data and the steep learning curve required to use the data. Spaceborne systems, with global coverage and rapid repeat cycles, together with user-friendly software would help overcome this resistance. The HyspIRI mission, under study by NASA for possible launch in the 2013-2016 timeframe, includes two instruments mounted on a satellite in low earth orbit. There is an imaging spectrometer measuring from the visible to short wave infrared (VSWIR) and a multispectral thermal infrared (TIR) imager. The VSWIR and TIR instruments will both have a spatial resolution of 60 m at nadir. The VSWIR will have a temporal revisit of approximately 3 weeks and the TIR will have a temporal revisit of approximately 1 week. This is a step in the right direction for encouraging greater use of hyperspectral data.

Another development which has considerable potential is the of unmanned aerial vehicle (UAV) platforms for hyperspectral imaging. These could include fixed wing or helicopters. UAVs are already being used for geophysical surveys. The ready mobilisation and low operational costs make them a much more viable source of data. Lighter and more compact sensors are also being developed so it will only be a short while before UAV hyperspectral surveys become commercially available.

Another important requirement is rapid processing. Work is on to develop real time processing systems that can be brought about by on-board processing. Staenz argues for a fully commercial mission, but this seems a long way off at the moment. It is possible that the demonstrated success of RapidEye would spur the development of a hyperspectral mission. The mining industry could contribute, should they see future potential in rapid access at an economical cost. The development of the operation and processing of hyperspectral data with other data such as LiDAR is also an important development which is seeing practical application.