Remote Sensing of Environment 113 (2009) S5S16

Contents lists available at ScienceDirect

Remote Sensing of Environment

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e

Three decades of hyperspectral remote sensing of the Earth: A personal view

Alexander F. H. Goetz
Center for the Study of Earth from Space/CIRES, University of Colorado, Boulder 80309, USA
Analytical Spectral Devices Inc., Boulder, Colorado 80301, USA

a r t i c l e

i n f o

Article history:
Received 8 October 2006
Received in revised form 23 November 2007
Accepted 9 December 2007
Keywords:
Imaging spectrometry
Hyperspectral imaging
Spectroscopy
Earth observations
Remote sensing applications
Sensor development
Historical perspective

a b s t r a c t
Imaging spectrometry, or hyperspectral imaging as it is now called, has had a long history of development
and measured acceptance by the scientic community. The impetus for the development of imaging
spectrometry came in the 1970's from eld spectral measurements in support of Landsat-1 data analysis.
Progress required developments in electronics, computing and software throughout the 1980's and into the
1990's before a larger segment of the Earth observation community would embrace the technique. The
hardware development took place at NASA/JPL beginning with the Airborne Imaging Spectrometer (AIS) in
1983. The airborne visible/infrared imaging spectrometer (AVIRIS) followed in 1987 and has proved to this
day to be the prime provider of high-quality hyperspectral data for the scientic community. Other critical
elements for the exploitation of this data source have been software, primarily ENVI, and eld spectrometers
such as those produced by Analytical Spectral Devices Inc. In addition, atmospheric correction algorithms
have made it possible to reduce sensor radiance to spectral reectance, the quantity required in all remote
sensing applications. The applications cover the gambit of disciplines in Earth observations of the land and
water. The further exploitation of hyperspectral imaging on a global basis awaits the launch of a high
performance imaging spectrometer and more researchers with sufcient resources to take advantage of the
vast information content inherent in the data.
2009 Elsevier Inc. All rights reserved.

1. Introduction
The term hyperspectral imaging was rst coined in a paper
discussing the early results of the technique of imaging spectrometry
(Goetz et al., 1985). The term, it appears, made its way into the
scientic vernacular in the late 1980's by way of the Department of
Defense and intelligence communities as they became interested in
this civilian sector-developed technique and required a catch phrase.
The modier hyper has a negative connotation, meaning too much,
such as in hyperination or hyper-kinetic. However, it is, in fact, an apt
description of the size of the spectral data set collected by the sensors,
which makes solutions possible to problems over-determined in the
mathematical sense. In other words, no single material requires
hundreds of spectral bands spread over several octaves of the
electromagnetic spectrum to be identied uniquely. However, when
mixed with many other materials on the Earth's surface and viewed
through a hostile and changing atmosphere, there is security in
numbers. In order to be able to apply statistical techniques to the data,
an overabundance of spectral bands is needed to tease out the various
material components that make up each ground-instantaneous-eldof-view (GIFOV) or pixel observed with the sensor.
This paper is written to document the origins and major
developments that make hyperspectral imaging of the Earth the
E-mail address: goetz@asdi.com.
0034-4257/$ see front matter 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2007.12.014

exciting remote sensing technique it is today. When the rst eld

spectral measurements were conducted in the early 70's and the
promise of imaging spectrometry became apparent, the technology
was not advanced enough for it to be implemented. In spite of the fact
that humans had walked on the moon, essentially all image processing
was carried out in large centralized computer centers and processing
jobs were loaded using punch cards. Microprocessors didn't yet exist
and data storage for eld instruments was discussed in terms of
kilobytes. The story of hyperspectral imaging is closely tied to
advances in digital electronics and computing capabilities.
While this paper is aimed at remote sensing of the Earth it must be
recognized that an equally exciting application of imaging spectrometry was developed in parallel for planetary exploration and in fact
has provided the majority of the information concerning the nature of
the surfaces of extraterrestrial bodies. The near-infrared mapping
spectrometer (NIMS) on Galileo and the visible and infrared mapping
spectrometer (VIMS) among others are examples. Only rarely have the
two groups of researchers interacted and when, primarily through
Roger Clark of the USGS with a foot in both camps.
The original denition for imaging spectrometry proposed by the
author and others (Goetz et al., 1985) was given as the acquisition of
images in hundreds of contiguous, registered, spectral bands such that
for each pixel a radiance spectrum can be derived. That is still a precise
way of dening the technique (Fig. 1). In particular, it sets this type of
spectral remote sensing apart from multispectral imaging by requiring

S6

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

that of a simple absorption band position comparison. Further

advantages lie in the ability to apply statistical least-squares
techniques to the problem of unmixing pixel spectra to derive the
relative abundances of material components on the surface. These
techniques were pioneered by researchers at the University of
Washington (Smith et al., 1985, 1990) applied to reectance spectra
and Landsat data and Gamon et al. (1993) showed that in AVIRIS data
analysis unmixed plant components were linearly related to their
abundance and Boardman (1993) described the automation of the
unmixing process. Jacquemoud et al. (1995) showed the advantages of
hyperspectral imaging for the extraction of vegetation biophysical
parameters from canopies through radiative transfer modeling.
3. Historical development
Fig. 1. Images collected in many contiguous spectral bands create a continuous radiance
spectrum.

the bands to be contiguous, in actuality overlapping, so that no gaps

occur through which precious information might slip undetected.
The requirement for hundreds of spectral bands is not critical
and detracts from the much more important contiguous in
describing the spectral bands. In reality, if the spectrum is sampled
at least twice for the narrowest spectral feature of interest, then all the
information available in the signal from a surface will be captured.
This concept is demonstrated in Fig. 2.
2. The hyperspectral advantage
The acquisition of the contiguous spectrum for each image pixel
over a selected wavelength interval makes it possible not only to
identify surface materials by their characteristic reectance or
emittance spectrum but also allows the intervening atmosphere to
be characterized to the level required for removal from the measured
radiance signal. This capability is not available from multispectral
imagers and limits these to acquiring data in atmospheric windows
exclusively. The shape of the reectance or emittance spectrum yields
information about grain size, abundance and composition, including
ion substitution in minerals, as well as chlorophyll concentration and
biochemical makeup in vegetation.
Another important advantage is that contiguous sampling makes it
possible to use correlation techniques to compare pixel spectra with
spectral data bases, thereby improving the apparent SNR of the data
by the square root of the number of bands used in the analysis over

Fig. 2. Multispectral (below) vs. hyperspectral sampling (above).

The beginnings of imaging spectrometry of the Earth are rooted in

the launch of Landsat-1, at the time called ERTS-1, in 1972. The author
aided by a co-investigator team including the late Gene Shoemaker of
lunar and cometary fame (Levy, 2000), were chosen by NASA to analyze
Landsat-1 multispectral scanner (MSS) data for geologic interpretation
of the Coconino Plateau south of the Grand Canyon in Arizona (Goetz
et al., 1975). The full resources of the NASA Jet Propulsion Laboratory
(JPL) Image Processing Laboratory (IPL), then arguably the leading
facility of its type in the civilian world, were brought to bear on the
analysis of the rst Earth orbiting satellite, multispectral images. During
eld studies comparing enhanced MSS images with ground observations it became clear that subtle color variations on the images were
difcult to identify in the eld and spectral reectance measurements of
undisturbed, in situ surface samples would be necessary to interpret the
image colors properly. The MSS was in effect a low resolution reectance
spectrometer, albeit with only 4 spectral bands, but providing information beyond the range of human vision.
4. Field spectrometry
The MSS image interpretation effort precipitated the development
of the rst truly portable eld reectance spectrometer (PFRS) that

Fig. 3. The PFRS on the Coconino Plateau, AZ. The instrument consisted of a circular
variable lter and a PbS detector in the optical head and control and recording
electronics in the backpack. A spectrum was acquired in 30 s. The umbrella was
necessary to shield the instrument from the sun.

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

S7

Fig. 4. PFRS spectra from Goldeld, NV. 1) basalt; 2) hematite coating and montmorillonite; 3) hematite coating and alunite; 4) kaolinite; 5) dolomite; 6) vegetation. (Courtesy
of NASA/JPL).

could cover the spectral range of solar reected radiance, 0.42.5 m

(Fig. 3) (Goetz, 1975).
Although 30 s were required to acquire a spectrum and another
30 s to acquire a white reectance standard, PFRS produced spectra
that resulted in the addition of band 7 to the Landsat Thematic Mapper
(TM), the development of the Shuttle Multispectral Infrared Radiometer (SMIRR), own on the second ight of Shuttle in November
1981, and justied the beginning of imaging spectrometer development (Fig. 4).
In order to decrease the acquisition time for eld measurements,
we at JPL developed the Portable Instant Display and Analysis
Spectrometer (PIDAS) (Fig. 5a), which took 4 s to acquire a spectrum
but weighed over 30 kg (Goetz, 1987).
As the demand for eld instruments developed in the research
community, companies such as Geophysical Environmental Research
(GER) of Millbrook, NY and Analytical Spectral Devices (ASD) of
Boulder, CO produced more technically advanced and easier to carry
eld spectrometers (Fig. 5b). The ASD eld instrument acquires a full
spectrum in 100 ms.
Laboratory spectra of minerals (Fig. 6) supported the notion that
reectance spectroscopy in the visible and shortwave infrared (SWIR)
region would be a source of rich information about the composition of
the Earth's surface (Hunt, 1977).
The sharp pyrophyllite absorption was used to dene the spectral
resolution of 10 nm required for imaging spectrometers to identify,
conservatively, 95% of the materials likely to be encountered.
The PFRS was used to verify the spectra of iron minerals in studies
of hydrothermal alteration in Goldeld, NV (Rowan et al., 1974) and
Cuprite, NV (Abrams et al., 1977). The latter area has become a

Fig. 6. Laboratory spectra of common minerals in soils. The width of Thematic Mapper
band 7 is shown for comparison.

standard site for ground truth measurements, and testing and

calibrating airborne and spaceborne instruments for the last 30 years.
5. Shuttle multispectral infrared radiometer
SMIRR was developed at JPL to y on the second ight of Shuttle
(STS-2) in 1979 but was delayed until 1981. The purpose of the
experiment was to test the feasibility of direct mineral identication
from Earth orbit. The instrument was designed as a proler to acquire
radiance values in 10 spectral bands covering the VNIR and SWIR
regions along a 100 m wide track beneath the spacecraft. The
instrument (Fig. 7) consisted of a spare Mariner VenusMercury
mission telescope with a spinning lter wheel in front of an HgCdTe
detector in the focal plane. Two 16 mm ghter-aircraft gun cameras
recorded the ground track.
The pre-ight calibration consisted of a series of outdoor
measurements of white standards and mineral powders of materials
expected to be observed in ight. SMIRR was installed in the payload
bay of STS-2 alongside the Shuttle Imaging Radar A (SIR-A) (Fig. 8).
The experiment was successful and resulted in the rst direct
identication of soil minerals from orbit through the use of the ve
closely-spaced spectral channels in the 2.22.5 m region (Goetz et al.,
1982). The minerals kaolinite and calcite, in the form of limestone,

Fig. 5. (a) PIDAS on the left that utilized lead sulde detector arrays. The PFRS is shown on the right; (b) The ASD FieldSpec3 spectrometer. (Courtesy of ASD Inc.).

S8

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

Fig. 7. Artists' rendition of the SMIRR instrument. (Courtesy of NASA/JPL).

were identied in an Egyptian dry lake and the surrounding hills and
conrmed based on returned samples (Fig. 9).
6. Airborne imaging spectrometer (AIS)

Fig. 8. STS-2 payload bay showing SMIRR with the open telescope cover. The SIR-A
antenna is to the right. (Courtesy of NASA).

In 1979 the rst hybrid focal plane array became commercially

available. This two-dimensional detector array consisted of a matrix of
HgCdTe detectors bonded to a matched silicon charge-coupled device
(CCD) readout array. Although the rst detectors contained only
32 32 elements, they enabled the construction of an imaging
spectrometer that covered the spectral region beyond the 1.1 m
cutoff of silicon arrays. Gregg Vane and the author proposed
successfully to use internal JPL funds for the development of the
Airborne Imaging Spectrometer (AIS) shown in Fig. 10 (Vane et al.,
1984).
The entrance slit to the spectrograph is aligned across-track to the
aircraft motion and the wavelength dispersion is perpendicular to the
slit. The two-dimensional array is read out completely for each image
line. The advantage of a 2-d array system is that no scan mirror is
necessary and the residence time of the signal on the detector is
increased by at least the number of detectors aligned across-track
covered in the one-way scan time of the mirror in an optomechanical
system. The residence time is even greater if the scanner efciency is
less than 100%, which is normally the case. The increase in signal-tonoise ratio is nominally proportional to the square-root of the number

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

S9

Fig. 9. SMIRR 10-channel reectance spectra: a) outdoor calibration measurements; b) orbital measurements; c) Laboratory spectra of eld samples.

of cross-track detectors. The major disadvantage of a pushbroom

imaging spectrometer is that there are tens of thousands of detector
elements that must be calibrated and that may have different drift
characteristics. The inaccuracies in calibration reduce the potential
signal-to-noise ratio gain over an optomechanical instrument, which
typically has a single detector for each spectral band (Goetz et al.,
2003).
The AIS spectral sampling interval was 9.6 nm per pixel and
because the detector had 32 32 pixels, only 307 nm could be covered
with a given grating position. In order to expand the range, the grating
was stepped through 4 positions within the time it took the aircraft to
move forward one cross-track line width (Fig. 11). This technique was
only partially successful because of the time required for the grating to
settle in a new position.
The best data were acquired in the xed grating position covering
the 2.032.32 m wavelength region, approximately the range
covered by the single Landsat Thematic Mapper Band 7.
Fig. 12 shows the rst successful results from an over ight of
Cuprite NV in August 1983. The individual band images were arranged
side by side since each image strip is only 32 pixels wide. Some
spectral variation is seen, but not until the images were processed by
ratioing each pixel to the scene average to bring out band-to-band
differences do spectral reectance variations become apparent (Goetz
et al., 1985).
In Fig. 13(a), the image is from the raw data showing a major
problem associated with area-array detectors that is as vexing today
as it was 25 years ago. The vertical stripes are caused by dead or poorly

calibrated pixels in the detector array. Each spectral-band image has a

different set of stripes depending on the location of the bad pixels.
Fig. 13(b) shows the corrected image in which the bad pixels were
replaced with the average of each of its neighbors in the spectral
direction. Fig. 13(c) image has been normalized to set the sum of all
pixels over all spectral bands to be a constant according to
32
X

Xk i; j = C:

k=1

In Eq. (1), Xk is the image in band k, i and j are the column and row
values. C is a constant. The image clearly shows spectral reectance
variations among regions and they are identied in Fig. 14.
Fig. 14 is historic in that it is the rst image acquired with an
imaging spectrometer that shows unambiguous evidence of mineral
identication, in particular the minerals alunite and kaolinite. The
spectral reectance curves overlay those taken from laboratory
spectra of samples taken from the same eld locations. The area
shown in magenta corresponds to a location containing ash ow tuffs
in which the feldspars have been chemically altered to the clay
kaolinite. The location was nicknamed Kaolinite Hill, which has
endured to the present.
The rst success in identifying surface mineralogy was heartening.
However, in order to satisfy potential critics a second ight over the
Cuprite Mining District was undertaken. Fig. 15 shows an airphoto of
Cuprite with Stonewall Playa in the lower right. The ight lines are

Fig. 10. Exploded view of the AIS instrument that measured approximately 30 30 20 cm. (Courtesy of NASA/JPL).

S10

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

Fig. 11. Optical diagram of the AIS instrument. (Courtesy of NASA/JPL).

shown in white. The ight path with a dogleg resulted from the pilot's
decision to compensate for wind induced drift by raising the right
wing of the C-130 so that the xed-mounted AIS would cover Kaolinite
Hill. After reaching the target, the plane resumed level ight. The
result of the maneuver was fortuitous in that on the resulting image
(Fig. 16) an unknown spectrum was observed.

In Fig. 16, the red spectrum comes from the image; the black
spectrum is from a laboratory measurement of a eld sample. This
material turned out to be the mineral buddingtonite, an ammonium
feldspar (Goetz and Srivastava, 1985).
Field identication was complicated by the fact that the image
could not be georectied since today's sophisticated navigational

Fig. 12. The top image is an air photo of the Cuprite Mining District, NV taken in the visible region with a 35 mm camera aboard the NASA C-130. Below are 32 individual AIS spectral
band images taken on 9.6 nm intervals between 2.03 and 2.32 m.

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

S11

Fig. 13. AIS images of Cuprite, NV. a) Original data with all the striping artifacts showing; b) corrected images, striping removed; c) equal energy normalization applied to enhance the
band-to-band spectral reectance differences.

systems, including GPS, were not available. In addition, the image

tones displayed on the 2 m images did not necessarily correlate with
human perception in the visible spectrum. A number of samples were
returned to the laboratory and among those were several that had
reectance spectra that matched the AIS image spectra with an
absorption minimum at 2.12 m. All the known spectral libraries were
consulted to no avail. The samples were analyzed by X-ray diffraction
(XRD) and shown to contain sanidine, a potassium feldspar. Feldspars
do not exhibit absorption in the 2 m spectral region and, therefore,
some other contaminant was suspected. Larry Rowan of the USGS in
Reston, Virginia reanalyzed the samples with XRD and found
buddingtonite, an ammonium feldspar. The XRD library at JPL had

not been updated for buddingtonite, which has a very similar XRD
spectrum to sanidine. Buddingtonite was rst identied in 1964 in
samples from a site in California and cannot be recognized in hand
specimen (Erd et al., 1964). While buddingtonite occurrences may be
widespread, the nd in Cuprite was alleged to be the fourth or fth
known location in volcanic terrains. At the same time buddingtonite
was discovered in the Carlin gold mining district in Nevada and it was
thought that buddingtonite might become a pathnder mineral for
gold. Buddingtonite appears to form in hot spring systems containing
organic materials in which the ammonium ion substitutes for
potassium in feldspar and in Carlin it was not co-genetic with the
older, micron gold deposits (Felzer et al., 1994).
The mineral identication success, and in particular the buddingtonite discovery which became part of a Public Broadcasting System
(PBS) documentary while the gold pathnder mineral question was
still open, led to a greater interest within NASA to pursue further
sensor development. At the time there was competing pressure by the
Goddard Spaceight Center (GSFC) to spend the R&D dollars on
development of pushbroom linear detector arrays for a future, allelectronic Landsat Thematic Mapper. This competition between GSFC
and JPL for earth sensor development colored the entire history of
imaging spectrometry into the late 1990's.

7. NASA imaging spectrometer development program

Fig. 14. Unsupervised classication image from the spectral data in Fig. 13(c). In yellow
are the laboratory spectra from samples collected in the eld sites.

In 1984, we at JPL proposed an imaging spectrometer program

(Fig. 17) that would encompass an advanced airborne sensor, the
airborne visible/infrared imaging spectrometer (AVIRIS), and two
orbiting sensors, the shuttle imaging spectrometer experiment
(SISEX) and a free-yer, the high resolution imaging spectrometer
(HIRIS).
AVIRIS development was begun in 1984 and the imager rst ew
aboard a NASA ER-2 aircraft at 20 km altitude in 1987. Since then it has
gone through major upgrades as technology changed in detectors,
electronics and computing. AVIRIS is arguably the nest, best
calibrated, airborne imaging sensor ever own, which is the result
of the dedication of the instrument leaders and their teams at JPL,
Gregg Vane (Vane et al., 1993) and Robert O. Green (Green et al., 1998)
and the continuing support of Diane Wickland of NASA Headquarters.

S12

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

Fig. 15. Airphoto of Cuprite, NV overlain with ight tracks in which AIS data were collected in 1983. (Courtesy of NASA/JPL).

Fig. 16. Classied image with occurrences of a material with a 2.12 m absorption feature
shown in red. The red spectrum comes from the image; the black spectrum is from a
laboratory measurement of a eld sample. This material was found to be the mineral
buddingtonite, an ammonium feldspar.

SISEX never moved beyond the design stage because of the

Challenger disaster in 1986 when the Shuttle was destroyed by a
booster rocket malfunction shortly after liftoff.
The HIRIS concept was developed further in response to the Earth
Observing System (EOS) ight proposal opportunity in 1988. Fourteen
individuals from all aspects of Earth observation science made up the
HIRIS Science Team. The goal was to place a pointable imaging
spectrometer covering the 0.42.5 m region with 10 nm contiguous
bands and spanning a 30 km swath from the 705 km orbital altitude of
the EOS platform (Goetz & Herring, 1989). The HIRIS proposal was
accepted for EOS, however, as the overall program was scaled back to
approximately one-fourth of its original size, HIRIS was one of the
casualties. It became quite clear at that time that the Earth science
community had not been sufciently exposed to the advantages of
imaging spectrometry and they had not yet been able to work with
high quality datasets in order to become more familiar with the
concept. In addition, it wasn't yet clear what scientic questions could
be addressed uniquely with imaging spectrometry and, therefore, the
community was not in a position to become an advocate for the
sensor. As one well known EOS scientist put it, as he gesticulated
toward the heavens, why do you need thousands and thousands of
spectral bands in obvious reference to a popular television cosmology
series of the time featuring Carl Sagan. To this day, hyperspectral

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

S13

Fig. 17. The imaging spectrometer program plan in 1984. (Courtesy of NASA/JPL).

imaging suffers from some of the same problems because there are
still many believers in the one spectral band, one biogeophysical
variable way of thinking.
In retrospect, at the time of the demise of HIRIS only a handful of
investigators had actually worked with hyperspectral data mainly
because there were only a few image data sets and they were of poor
quality compared to what is available today. Additionally, there were
not any readily available software tools to manipulate the data sets
that were orders of magnitude larger than those from Landsat, and
desktop computing systems were not yet up to the task. Therefore, it is
not surprising that HIRIS was not included on EOS.
Ultimately, NASA launched Hyperion, an imaging spectrometer
built by TRW, aboard the EO-1 mission in November 2000 (Ungar
et al., 2003). Today, it remains the only continuous source of full-range
hyperspectral data from Earth orbit. The European PROBA/CHRIS
mission launched in 2001 provides 18 spectral channels covering the
VNIR region (Barnsley et al., 2004).
8. Airborne hyperspectral instruments
In the late 1980's several commercial hyperspectral imagers
entered the market. The rst was DAIS from Geophysical Environmental Research of Millbrook, NY (Richter, 1996). In 1989 ITRES
Corporation, Alberta, Canada introduced CASI, an imaging spectrometer covering the visible and near-infrared region to 900 nm
utilizing a 2-d silicon CCD array (Dekker et al., 1992). In 1994 the Naval
Research Lab sensor HYDICE was completed (Basedow & Zalewski,
1995). This sensor was designed around a prism dispersion concept
and a single, hybrid HgCdTe 2-d array to cover the 4002500 nm
spectral region. Although HYDICE began as a dual-use program, it
soon reverted to an all DOD program. Other commercial sensors in this
wavelength region are available. The one most like AVIRIS is the
Australian sensor HyMap from the HyVista Corporation, which
markets a full-range (4002500 nm) hyperspectral imaging service
on a global basis.

spectral bands. Additionally, software written on one platform often

could not be ported to another without writing specic image display
code. During the mid 1980's, the Interactive Data Language (IDL) a
high level array oriented language came into use (Gumley, 2002). It
had the advantage that it functioned seamlessly across all the popular
computing platforms of the time. During this time, Joe Boardman and
Kathryn Kierein-Young, both Ph.D. students at the University of
Colorado, constructed the rst image cube, a representation of
hyperspectral data that caught the imagination of researchers in the
eld and is the icon of hyperspectral imaging to this day (Fig. 18).
The SIPS (Spectral Image Processing System) project at the Center
for the Study of Earth from Space (CSES) at the University of Colorado,
Boulder created hyperspectral image processing software based on
IDL that could be ported to all the platforms supported by IDL (Kruse
et al., 1993). SIPS was the precursor to ENVI (Environment for
Visualizing Images) that was developed by Joe Boardman, Fred Kruse
and others in the private sector and is now owned and distributed by
ITT Visual Information Solutions of Boulder Colorado (ENVI, 1998).
The introduction of ENVI in 1994 made it possible for the rst time for
the broader scientic community to process hyperspectral images and
it provided a strong stimulus as evidenced by the increased number of
peer-reviewed papers being published since then. The concept of pixel
unmixing came into the mainstream of hyperspectral data analysis
with the continuing development of ENVI making it possible for
relative beginners to process images.
Another approach to hyperspectral processing has been advocated
by Roger Clark and is based on library spectra comparison on a pixelby-pixel basis (Clark et al., 2003). This technique results in image
maps identifying the most abundant material in each pixel and has
been applied successfully to both Earth and planetary hyperspectral
data sets. Another current spectral library has been developed by
Simon Hook at JPL, http://speclib.jpl.nasa.gov which include spectra
from the Johns Hopkins University spectral library developed by John
Salisbury.
9. Atmospheric correction

8.1. Development of tools

In 1989 there were no image processing software packages
commercially available that were capable of handling more than 10

As mentioned above, multispectral images are acquired in atmospheric windows so that minimal corrections are required to convert
spectral radiance to spectral reectance. The primary concern is with

S14

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

Fig. 18. Imaging spectrometer data hypercube. The colors represent the reectance of the edge pixels in this image of Boulder, CO. Low values are blue, high values red. (Courtesy of
Joe Boardman).

path radiance, which can be corrected to rst order using dark-object

subtraction (Chavez, 1988). Hyperspectral imagers, on the other hand,
acquire spectral radiance throughout the solar reective region (0.4
2.5 m) in which there are numerous atmospheric absorption features
as shown in Fig. 19.
Correction for the spectrally variable transmission is essential in
order to produce a continuous reectance spectrum. Where the
atmospheric transmission is zero, or nearly so, such as in the strong
water vapor absorption features around 1.4 and 1.9 m, it is not
possible to correct for the atmosphere because no irradiance is
reaching the surface. This fact makes it possible to isolate cirrus clouds
from other clouds at lower elevations (Gao et al., 1993a) and as a result
a spectral band at 1.38 m was added to the Moderate Resolution
Imaging Spectrometer (MODIS) to allow for world-wide cirrus cloud
mapping during daylight (Gao and Kaufman, 2003).
The hyperspectral nature of the data acquisition makes it possible
to create atmospheric correction factors from the data themselves
and, importantly, for each pixel because the atmospheric transmission
changes across the image as a result of elevation and water vapor
differences. Initial work on an atmospheric correction for water vapor
was carried out by Conel et al. (1988) and Gao and Goetz (1990).
Shortly thereafter the program ATREM (ATmospheric REMoval) was
released and licensed by the University of Colorado at no cost to over
300 research groups world-wide (Gao et al., 1993b).
The stage was nally set for advancement in the exploitation of
hyperspectral image data by the general scientic community;
15 years after the rst modern instruments were conceived. The
AVIRIS instrument had been upgraded several times since its rst
ight in 1987 to provide high signal-to-noise ratio data, quality eld
spectrometers were commercially available, commercial software for

displaying and analyzing hyperspectral image data was on the market,

atmospheric correction algorithms to convert radiance to reectance
were available and, last but not least, desktop computing with
sufcient speed, memory and disk space to handle the large datasets
was becoming commonplace.
The long and arduous development of hyperspectral imaging and
its applications is documented in the series of workshops sponsored
by JPL, beginning in 1984 and continuing to the present. The AVIRIS

Fig. 19. Spectral atmospheric transmission in the 0.42.5 m wavelength region for a
visibility of 23 km and total precipitable water vapor of 5 cm. The transmission values
were calculated using MODTRAN (Berk et al., 1998).

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

Geoscience Workshop proceedings are available in PDF format at

http://aviris.jpl.nasa.gov/html/aviris.documents.html.
10. Applications
Hyperspectral imaging has enabled applications in a wide variety of
Earth studies. Some of the pioneering research papers are mentioned
here but it is by no means a complete list. As mentioned before, the
prime motivation for the development of imaging spectrometry was
mineralogical mapping of surface soils and outcrops (Abrams et al.,
1977; Goetz et al., 1985). The reectance spectra of minerals are rich in
electronic as well as overtone and combination vibrational features
that characterize surfaces that are relatively vegetation-free (Clark
et al., 1990). Only approximately 30% of the land surface is relatively
devoid of vegetation and the remaining 70% is covered by vegetation to
the extent that the substrate is rendered inaccessible to remote sensing
identication (Siegal and Goetz, 1977). However, the vegetation cover,
its type, health, vigor and expression of environmental conditions
including the substrate are the subject of many ongoing studies.
Wessman et al. (1988) were the rst to attempt the identication of
tree species based on nitrogen and lignin content in the foliage. They used
statistical regression techniques also known to spectroscopists as chemometrics (Mark, 1989) and built a prediction model based on known
occurrences of broadleaf and evergreen species on Blackhawk Island, WI.
As a follow-on to the HIRIS project, NASA funded the Accelerated Canopy
Chemistry Program in which chemometrics techniques were used
successfully on AVIRIS data acquired over the Harvard Forest, MA (Aber
& Martin, 1995; Martin and Aber, 1997). Other diverse studies of species
and canopy health, water content as well as relative abundances of
photosynthetic (PV) and non-photosynthetic (NPV) vegetation in a pixel
can be found in papers by Gamon et al. (1992, 1993), Ustin et al. (1992,
1998), Roberts et al. (1993, 1998), and Asner and Lobell (2000).
While open oceans can be studied with multispectral imagers such
as SeaWiFS, studies of the costal zone are better served by
hyperspectral imaging, which makes it possible to unmix the bottom
and several in-column constituents (Carder et al., 1993; Lee et al.,
1994). Hyperspectral imaging is equally applicable to the solid water
phase which makes it possible to study the properties of ice and snow,
in particular grain size (Nolin & Dozier, 1993; Painter et al., 1998).
Environmental studies using hyperspectral imaging are yielding
results that would be impossible to obtain or would be prohibitive in
cost or time spent with standard techniques. One example that has
been documented to have saved millions of dollars is in the clean up of
the Leadville, CO Superfund Site in which AVIRIS images combined
with eld spectral measurements identied the waste piles with the
greatest potential for leaching heavy metals into streams and
groundwater (Swayze et al., 2000). Asbestiform minerals have also
been identied in situ from AVIRIS data (Swayze et al., 2005). Maps of
expansive soils, important in construction engineering, can also be
identied in AVIRIS images (Chabrillat et al., 2002).
11. Summary and projections
The development of hyperspectral imaging, the sensors and the
data analysis techniques has been long and arduous. The rate of
understanding and acceptance of the hyperspectral techniques was
tied closely to technology developments, in particular digital electronics and software. It took almost 15 years from the beginning of
funded development to bring all the components together necessary
for use by the general scientic community and another 5 years to
celebrate a sensor in orbit. The 30 years of ups and downs discussed
here are the majority of a professional lifetime, but the continuing
growing interest in this technique for study of the Earth and its
environment has made it all worthwhile.
Speculation about the future is burdened with the knowledge that it
isn't possible to get the details right. However, several trends, actually

S15

needs, can be anticipated. The needs fall into four categories. The rst
involves the trend toward acquiring ever more accurate measurements,
in space and time, of the state of the environment and its processes. The
key question for proponents of hyperspectral imaging is whether
important information can be acquired above and beyond that collected
by global coverage, multispectral instruments such as MODIS. For
instance, published results using AVIRIS data show that valuable
information about vegetation cover can be derived that is not obtainable
with a MODIS type instrument. However, this increased knowledge
comes with a cost in resources, human and monetary, as well as political
will. In the context of all the other demands placed on NASA's resources
is this potential knowledge worth the price? I think it is, but a strong
advocate must come forward to assure that a high enough visibility is
obtained in the struggle for priority in the allocation of resources.
The second need is for the education of students to develop
information and knowledge from the data. Currently, there are only a
handful of university programs that teach remote sensing using
advanced techniques such as hyperspectral imaging. World-wide,
over 800 different universities and research institutions own eld
spectrometers (ASD, personal communication), some multiple units,
but very few have direct access to imaging spectrometers on aircraft.
The potential exists but the eld will not expand signicantly until a
much larger number of students and faculty have ready access to
hyperspectral imagers under their control.
The third trend is a positive one and is derived from the relentless
advance of computer and sensor technology that is providing the ever
increasing ability to acquire, store, access and process large image
datasets. The personal computer revolution in the 1990's that opened up
the eld of hyperspectral imaging to the general scientic community
will be relived as we advance from analyzing 100 km2 aircraft
hyperspectral images to global datasets containing petabytes of data.
There are some practical limitations, however. Imaging spectrometers
based on silicon CCD arrays can be built relatively inexpensively because
the arrays and supporting electronics have other commercial uses and the
prices are concomitantly low. On the other hand, hyperspectral imagers
operating beyond 1 m require much more expensive detectors, which
are manufactured in low volume. This reality combined with the fact that
full-range imaging spectrometer systems will continue to be priced well
over one million USD because of the low volumes. Another factor, export
restrictions, will limit the growth of hyperspectral imaging applications
dependent on systems manufactured in the United States. Ultimately,
when the demand for the increased information becomes acute, there
will be solutions to these problems found.
The nal need is for a hyperspectral imager in orbit that can
produce images of the quality and resolution of AVIRIS that is
radiometrically stable and pointable. For instance, with the high
demand for AVIRIS data, it isn't logistically feasible to conduct
seasonal time-series studies for more than a few sites in North
America. Given the dynamic character of vegetation cover, a snapshot
in time isn't nearly as revealing as a time sequence. The technology
exists but it is a matter of money and priority to implement such a
mission. A strong advocate must step forward to fulll this need.
Acknowledgements
For these 30 years of science and engineering development I am
indebted to many colleagues who have helped move this eld of
endeavor forward. A few stand out especially. Gregg Vane of JPL was a
key partner in getting the rst funding support for AIS and led the
early engineering efforts to build and operate AIS and AVIRIS. Rob
Green has been critical to the success of hyperspectral imaging
research by managing the AVIRIS program in his relentless, 20+ year
long pursuit of creating the highest signal-to-noise ratio and best
calibrated airborne imaging system in history. Joe Boardman, my rst
Ph.D. student at the University of Colorado, introduced me to the true
meaning of n-space and came up with myriads of ingenious solutions

S16

A.F.H. Goetz / Remote Sensing of Environment 113 (2009) S5S16

to image analysis problems, many of which ended up in ENVI. Diane

Wickland of NASA Headquarters deserves special recognition for her
continued support of the hyperspectral imaging research community
and AVIRIS in an often skeptical and hostile environment, particularly
after HIRIS was dropped from the EOS mission. She also provided the
impetus for the founding of ASD Inc. by cutting off funding for eld
spectrometer development at the University of Colorado.
References
Abrams, M. J., Ashley, R., Rowan, L. C., Goetz, A. F. H., & Kahle, A. B. (1977). Mapping of
hydrothermal alteration in the Cuprite mining district, Nevada, using aircraft
scanner imagery for the 0.462.36 m spectral region. Geology, 5, 713718.
Aber, J. D., & Martin, M. E. (1995). High spectral resolution remote sensing of canopy
chemistry. AVIRIS Proceedings, Vol. 951. (pp. 14): JPL Publication.
Asner, G. P., & Lobell, D. B. (2000). A biogeophysical approach for automated SWIR
unmixing of soils and vegetation. Remote Sensing of Environment, 74, 99112.
Basedow, R. W., & Zalewski, E. (1995). Characteristics of the HYDICE sensor, AVIRIS
Proceedings, Vol. 951. (pp. 9): JPL Publication.
Barnsley, M. J., Settle, J. J., Cutter, M. A., & Teston, F. (2004). The PROBA/CHRIS MISSION: A
low-cost smallsat for hyperspectral multiangle observations of the earth surface and
atmosphere. IEEE Transactions On Geoscience and Remote Sensing, 42, 15121520.
Berk, A., Bernstein, L. S., Anderson, G. P., Acharya, P. K., Robertson, D. C., Chetwynd, J. H.,
& Adler-Golden, S. M. (1998). MODTRAN cloud and multiple scattering upgrades
with applications to AVIRIS. Remote Sensing of Environment, 65, 367375.
Boardman, J. W. (1993). Automating spectral unmixing of AVIRIS data using convex
geometry concepts. AVIRIS Proceedings, Vol. 9326. (pp. 1114): JPL Publication.
Carder, K. L., Reinersman, P., Chen, R. F., Mullerkarger, F., Dans, C. O., & Hamilton, M. K.
(1993). AVIRIS calibration and application in coastal oceanic environments. Remote
Sensing of Environment, 44, 205216.
Chabrillat, S., Goetz, A. F. H., Krosley, L., & Olson, H. W. (2002). Use of hyperspectral
images in the identication and mapping of expansive clay soils and the role of
spatial resolution. Remote Sensing of Environment, 82, 431445.
Chavez, P. S. (1988). An improved dark-object subtraction technique for atmospheric
scattering correction of multispectral data. Remote Sensing of Environment, 24, 459479.
Clark, R. N., King, T. V. V., Klejwa, M., Swayze, G. A., & Verga, N. (1990). High spectral
resolution reectance spectroscopy of minerals. Journal Geophysical Research-Solid
Earth Planets, 95, 1265312680.
Clark, R. N., Swayze, G. A., Livo, K. E., Kokaly, R. F., Sutley, S. J., Dalton, J. B., et al.
(2003). Imaging spectroscopy: Earth and planetary remote sensing with the
USGS Tetracorder and expert systems. Journal Geophysical Research-Planets, 108
Art. No. 5131.
Conel, J. E., Green, R. O., Carrere, V., Margolis, J. S., Alley, R. E., Vane, G., et al. (1988).
Atmospheric water mapping with the airborne visible/infrared imaging spectrometer (AVIRIS), Mountain Pass, California. Proceedings of the AVIRIS Performance
Evaluation Workshop, JPL Technical Report 88-38, Pasadena, California (pp. 2129).
Dekker, A. G., Malthus, T. J., Wijnen, M. M., & Seyhan, E. (1992). The effect of spectral
bandwidth and positioning on the spectral signature analysis of inland waters.
Remote Sensing of Environment, 41, 211225.
ENVI programmer's guide (1998). Research systems inc. 930 pp.
Erd, R. C., White, D. E., Fahey, J. J., & Lee, D. L. (1964). Buddingtonite, an ammonium
feldspar with zeolitic water. The American Mineralogist, 49, 831850.
Felzer, B., Hauff, P., & Goetz, A. F. H. (1994). Quantitative reectance spectroscopy of
buddingtonite from the Cuprite mining district, Nevada. Journal of Geophysical
Research, 99, 28872895.
Gamon, J. A., Field, C. B. & Ustin, S. L. (1992). Evaluation of spatial productivity patterns
in an annual grassland during an AVIRIS overight. In Summaries of the Third
Annual JPL Airborne Geoscience Workshop, Robert O. Green (ed.). NASA, Jet
Propulsion Laboratory 9214, Vol. 1, pp. 1719.
Gamon, J. A., Field, C. B., Roberts, D. A., Ustin, S. L., & Riccardo, V. (1993). Functional
patterns in an annual grassland during an AVIRIS overight. Remote Sensing of
Environment, 44, 239253.
Gao, B. C., & Goetz, A. F. H. (1990). Column atmospheric water vapor and vegetation
liquid water retrievals from airborne imaging spectrometer data. Journal of
Geophysical Research-Atmospheres, 95, 35493564.
Gao, B. C., Goetz, A. F. H., & Wiscomb, W. J. (1993). Cirrus cloud detection from airborne
imaging spectrometer data using the 1.38 m water vapor band. Geophysical
Research Letters, 20, 301304.
Gao, B. C., Heidebrecht, K. B., & Goetz, A. F. H. (1993). Derivation of scaled surface
reectances from AVIRIS data. Remote Sensing of Environment, 44, 165178.
Gao, B. C., & Kaufman, Y. J. (2003). Water vapor retrievals using moderate resolution
imaging spectroradiometer (MODIS) near-infrared channels. Journal of Geophysical
Research-Atmospheres, 108(D13) Art. No. 4389.
Goetz, A. F. H., Billingsley, F. C., Elston, D., Lucchitta, I., Shoemaker, E. M., Abrams, M. J.,
et al. (1975). Applications of ERTS images and image processing to regional
geologic problems and geologic mapping in northern Arizona. JPL Technical
Report (pp. 321597). Pasadena, CA: Jet Propulsion Laboratory.
Goetz, A. F. H. (1975). Portable eld reectance spectrometer, Appendix E in Application
of ERTS images and image processing to regional geologic problems and geologic
mapping in Northern Arizona. JPL Technical Report (pp. 321597).
Goetz, A. F. H., Rowan, L. C., & Kingston, M. J. (1982). Mineral identication from orbit:
Initial results from the shuttle multispectral infrared radiometer. Science, 218,
10201024.

Goetz, A. F. H., & Srivastava, V. (1985). Mineralogical mapping in the Cuprite mining
district, Nevada. Proceedings of the Airborne Imaging Spectrometer Data Analysis
Workshop (pp. 2231). Pasadena, CA: JPL Publication 8541, Jet Propulsion
Laboratory.
Goetz, A. F. H., Vane, G., Solomon, J., & Rock, B. N. (1985). Imaging spectrometry for Earth
remote sensing. Science, 228, 11471153.
Goetz, A. F. H. (1987). The portable instant display and analysis spectrometer (PIDAS).
Proceedings of the Third Airborne Imaging Spectrometer Data Analysis Workshop,
Vol. 8730. (pp. 817): JPL Publication.
Goetz, A. F. H., & Herring, M. (1989). The high resolution imaging spectrometer (HIRIS)
for EOS. IEEE Transactions on Geoscience and Remote Sensing, 27, 136144.
Goetz, A. F. H., Kindel, B. C., Ferri, M., & Qu, Z. (2003). HATCH: Results from simulated
radiances, AVIRIS and Hyperion. IEEE Transactions on Geoscience and Remote
Sensing, 41(number 6), 12151222.
Green, R. O., Eastwood, M. L., Sarture, C. M., Chrien, T. G., Aronsson, M., Chippendale, B. J.,
et al. (1998). Imaging spectroscopy and the airborne visible infrared imaging
spectrometer (AVIRIS). Remote Sensing of Environment, 65, 227248.
Gumley, L. E. (2002). Practical IDL programming. San Diego, CA: Academic Press 508 p.
Hunt, G. R. (1977). Spectral signatures of particulate minerals in the visible and near
infrared. Geophysics, 42, 501513.
Jacquemoud, S., Baret, F., Andrieu, B., Danson, F. M., & Jaggard, K. (1995). Extraction of
vegetation biophysical parameters by inversion of the PROSPECT + SAIL models on
sugar beet canopy reectance data. Application to TM and AVIRIS sensors. Remote
Sensing of Environment, 52, 163172.
Kruse, F. A., Lefkoff, A. B., Boardman, J. W., Heidebrecht, K. B., Shapiro, A. T., Barloon, P. J., et al.
(1993). The spectral image processing system (SIPS) Interactive visualization and
analysis of imaging spectrometer data. Remote Sensing of Environment, 44, 145163.
Lee, Z. P., Carder, K. L., Hawkes, S. K., Steward, R. G., Peacock, T. G., & Davis, C. O. (1994).
Model for the interpretation of hyperspectral remote-sensing reectance. Applied
Optics, 33, 57215732.
Levy, D. H. (2000). Shoemaker by Levy: The man who made an impact. NJ: Princeton.
Mark, H. (1989). Chemometrics in near-infrared spectroscopy. Analytica Chimica Acta,
223, 7593.
Martin, M. E., & Aber, J. D. (1997). High spectral resolution remote sensing of forest canopy
lignin, nitrogen and ecosystem processes. Ecological Applications, 7, 431443.
Nolin, A. W., & Dozier, J. (1993). Estimating snow grain-size using AVIRIS data. Remote
Sensing of Environment, 44, 231238.
Painter, T. H., Roberts, D. A., Green, R. O., & Dozier, J. (1998). The effect of grain size on
spectral mixture analysis of snow-covered area from AVIRIS data. Remote Sensing of
Environment, 65, 320332.
Richter, R. (1996). Atmospheric correction of DAIS hyperspectral image data. Computers
& Geosciences, 22, 785793.
Roberts, D. A., Smith, M. O., & Adams, J. B. (1993). Green vegetation, nonphotosynthetic
vegetation, and soils in AVIRIS data. Remote Sensing of Environment, 44, 255269.
Roberts, D. A., Gardner, M., Church, R., Ustin, S., Silbeer, G., & Green, R. O. (1998).
Mapping chaparral in the Santa Monica Mountains using multiple endmember
spectral mixture models. Remote Sensing of Environment, 65, 267279.
Rowan, L. C., Wetlaufer, P. H., Goetz, A. F. H., Billingsley, F. C., & Stewart, J. H. (1974).
Discrimination of rock types and detection of hydrothermally altered areas in
South-Central Nevada by the use of computer-enhanced ERTS images. USGS
Professional Paper 883.
Siegal, B. S., & Goetz, A. F. H. (1977). Effect of vegetation on rock and soil type
discrimination, Photogramm. Engineering and Remote Sensing, 43, 191196.
Smith, M. O., Johnson, P. E., & Adams, J. B. (1985). Quantitative-determination of mineral
types and abundances from reectance spectra using principal componentsanalysis. Journal of Geophysical Research, 90, C797C804 Suppl. S.
Smith, M. O., Ustin, S. L., Adams, J. B., & Gillespie, A. R. (1990). Vegetation in deserts.1. A
regional measure of abundance from multispecrtral images. Remote Sensing of
Environment, 31, 126.
Swayze, G. A., Smith, K. S., Clark, R. N., Sulley, S. J., Pearson, R. M., Vance, J. S., et al.
(2000). Using imaging spectrometry to map acidic mine waste. Environmental
Science and Technology, 34, 4754.
Swayze, G. A., Higgins, C. T., Kokaly, R. F., Clark, R. N., & Sutley, S. J. (2005). Using imaging
spectroscopy to map ultramac rocks, serpentinites, and tremolite-actinolitebearing rocks in California. Geochimica et Cosmochimica Acta, 69, A191.
Ungar, S. G., Pearlman, J. S., Mendenhall, J. A., & Reuter, D. (2003). Overview of the Earth
Observing One (EO-1) mission. IEEE Transactions on Geoscience and Remote Sensing,
41, 11491153.
Ustin, S. L., Smith, M. O., Roberts, D., Gamon, J. A., & Field, C. B. (1992). Using AVIRIS
images to measure temporal trends in abundance of photosynthetic and
nonphotosynthetic canopy components. In Robert O. Green (Ed.), Summaries of
the Third Annual JPL Airborne Geoscience Workshop, JPL Pub. 9241, pp. 57.
Ustin, S. L., Roberts, D. A., Pinzon, J., Jacquemond, S., Gardner, M., Sheer, G., et al.
(1998). Estimating canopy water content of chaparral shrubs using optical
methods. Remote Sensing of Environment, 65, 280291.
Vane, G., Goetz, A. F. H., & Wellman, J. (1984). Airborne Imaging Spectrometer: A new
tool for remote sensing. IEEE Transactions on International Geoscience and Remote
Sensing, GE-22, 546549.
Vane, G., Green, R. O., Chrien, T. G., Enmark, H. T., Hansen, E. G., & Porter, W. M. (1993).
The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Remote Sensing of
Environment, 44, 117126.
Wessman, C. A., Aber, J. D., Petersen, D. L., & Melillo, J. M. (1988). Remote-sensing of
canopy chemistry and nitrogen cycling in temperate forest ecosystems. Nature, 335,
154156.