Remote Sensing of Environment 78 (2001) 27 – 38

www.elsevier.com/locate/rse

Absolute radiometric calibration of Landsat 7 ETM+ using the

reflectance-based method
K.J. Thome
Remote Sensing Group, Optical Sciences Center, University of Arizona, 1631 E. University Boulevard, Tucson, AZ 85721, USA
Received 1 June 2000; received in revised form 22 December 2000; accepted 30 April 2001

Abstract

A key to the continuation of quantitative data from the Landsat series of sensors is the radiometric understanding of the sensor. Vicarious
calibration methods are one approach that has been used successfully for the absolute radiometric calibration of Thematic Mapper (TM). One
of these vicarious methods is the reflectance-based approach that is applied here to the radiometric calibration of the Enhanced Thematic
Mapper Plus (ETM+) sensor on the Landsat 7 platform. This method is described for application to ETM+. Results from ground-based
measurements of atmospheric conditions and surface reflectance made at Railroad Valley Playa, Nevada, Roach Lake Playa, Nevada, and
White Sands Missile Range, New Mexico are presented including descriptions of the test sites. The gains derived from four dates using these
sites spanning the period from June 1999 to October 1999 agree to within 5% of each other and to better than 7% with the prelaunch,
laboratory-derived gains. This is within the combined 5% uncertainty of the prelaunch values and the estimated 3 – 5% uncertainty of the
reflectance-based method. The gains determined from the reflectance-based method are all lower than the prelaunch values for bands 1 – 5
and the values for band 7 exceed the prelaunch gains. These biases could be caused by errors in the treatment of atmospheric aerosols at
shorter wavelengths and uncertainties in the assumed solar irradiances used to convert the relative radiances to absolute values at longer
wavelengths. D 2001 Elsevier Science Inc. All rights reserved.

Keywords: Absolute radiometric calibration; Vicarious calibration

1. Introduction From a platform standpoint, the Landsat series of satel-

lites can be viewed in two distinct parts. The first includes
The Landsat series of satellites provides the longest Landsats 1, 2, and 3 that carried the return beam vidicon
running continuous data set of high-spatial resolution (RBV) camera and the Multispectral Scanner System
imagery dating back to the launch of Landsat 1 in 1972 (MSS). The second phase of Landsat includes the sub-
and continuing with the April 15, 1999 launch of Landsat 7. sequent Landsat platforms from Landsats 4 to 7. These
A large part of the success of the Landsat program can be platforms have a lower orbit, and have higher spatial and
attributed to the knowledge of the radiometric properties of spectral resolution sensors, and a faster repeat cycle. These
the Landsat sensors. This knowledge is due to the combina- platforms omitted the RBV cameras but Landsats 4 – 6 still
tion of prelaunch and postlaunch efforts using laboratory, carried the MSS. Landsats 4 and 5 included the Thematic
on-board, and vicarious calibration methods (where vicari- Mapper (TM) sensor. Details on the design of these early
ous calibration refers to any method not relying on on-board Landsat sensors can be found in Engel and Weinstein
calibrators) (Thome, Markham, Barker, Slater, & Biggar, (1983), Lansing and Cline (1975), Markham and Barker
1997). The radiometric calibration of these systems not only (1987), and Slater (1980). A great deal of research was done
helps characterize the operation of the sensors, but, more during the early days of Landsat to understand these
importantly, this calibration allows the full Landsat data set systems, including the extensive Landsat Image Data Qual-
to be used in a quantitative sense for such applications as ity Assessment (LIDQA) program (Markham & Barker,
land use and land-cover change. 1985). The last two Landsat platforms, Landsats 6 and 7,
have included improvements to the TM sensor to give better
spatial resolution of the thermal infrared band and to include
E-mail address: [email protected] (K.J. Thome). a high-spatial resolution panchromatic band. Unfortunately,

0034-4257/01/$ – see front matter D 2001 Elsevier Science Inc. All rights reserved.
PII: S 0 0 3 4 - 4 2 5 7 ( 0 1 ) 0 0 2 4 7 - 4
28 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

Landsat 6 failed to achieve orbit. The Landsat 7 platform direct solar irradiance and solar diffuser calibrators should
carries only one sensor, the Enhanced Thematic Mapper have similar stability as the ICs.
Plus (ETM+) sensor, having dropped the MSS sensor. One thing that must be kept in mind regarding the on-
In order to understand the necessity for vicarious cal- board calibrators is that they cannot provide a calibration
ibration approaches for the calibration of ETM+, it is helpful that is of higher accuracy than the preflight, laboratory
to understand the methods and usefulness of the on-board calibrations. That is, the accuracy of the in-flight, absolute
calibrators of past Landsat sensors. On-board calibration of calibration must be worse than the preflight calibration,
the MSS and TM used solar- and lamp-based approaches for since the preflight calibration source is used to calibrate the
the solar reflective bands. For Landsats 1 – 3, the MSS on-board calibrators. This gives good justification for devel-
included a partial aperture, partial-path solar calibrator. Data oping a calibration approach that is independent of the
from the solar calibrator on Landsat 1 were problematic preflight calibration. In addition, it is possible that degrada-
because of degradation of the optics (Horan, Schwartz, & tion of an on-board calibration system can occur over long
Love, 1974). Problems with the attitude control of Landsat 2 periods of time. This degradation is typically difficult to
made the data difficult to interpret, but the solar calibrators determine without means of an additional independent
on both Landsats 2 and 3 appeared to operate normally calibration source. This is the advantage to having the three
(Lansing, 1986). However, the data from these systems have separate calibration approaches for ETM+, but since two of
not been used for absolute radiometric calibration of the these methods are essentially untested, there is still a
MSS sensors because the on-board lamp systems worked justification for including an additional, well-understood
well (Lansing, 1986). calibration methodology. This additional method is the
The on-board lamp calibrator (also referred to as the subject of this work and relies on the vicarious calibration
internal calibrator or IC) for MSS used a shutter wheel and technique referred to as the reflectance-based approach.
a pair of redundant, tungsten-filament lamps. Within the Many methods have been proposed and used for the in-
shutter wheel is a wedge-shaped neutral density filter that flight radiometric calibration of satellite sensors. Hovis,
provides variable attenuation as it rotates with the shutter Knoll, and Smith (1985) made one of earliest attempts at
wheel. The shutter wheel also serves the purpose of vicarious calibration by measuring radiances above a
preventing light from the entrance aperture of MSS reach- ground target from a high-altitude aircraft to verify the
ing the focal plane. The IC used for TM is somewhat degradation of the response of the Coastal Zone Color
different with three lamps rather than one. The image of Scanner’s shorter wavelength bands. Since then, many types
each lamp filament falls on a different attenuating filter of vicarious calibration have been developed. Kaufman and
such that eight different irradiance levels can be obtained Holben (1993) present a method using large-view angles
by varying the choice of lamps. For ETM+, a similar IC is and molecular scatter to characterize the short-wave visible
used but the design only uses a single lamp level. In channels of the Advanced Very High-Resolution Radio-
addition to the lamp-based IC, ETM+ has two other on- meter. Vermote, Santer, Deschamps, and Herman (1992)
board calibrators. The first is a return to the direct solar used a similar approach to calibrate the short-wavelength
illumination technique with a partial aperture approach channels of Systeme Pour l’Observation de la Terre-1,
using a multifaceted optical device to direct solar energy Haute Resolution Visible (SPOT-1 HRV) sensor where the
onto the focal plane. The third on-board calibrator is one contributions from aerosols and sea-surface reflection were
that is new to Landsat. It is a full-aperture, full-optical-path determined from data at longer wavelengths. These two
solar diffuser (Markham, Barker, Boneyk, Kaita, & Helder, approaches are different from Hovis et al. (1985) in that they
1996) that is the first implementation of a spaceborne do not require in-situ measurements. However, because
diffuser for absolute radiometric calibration of a high- assumptions are made based on climatology, the results of
spatial resolution sensor. these methods can suffer from larger uncertainties than
The primary advantage to the on-board calibrators is that those using in-situ methods, though methods without in-situ
a calibration is performed with high-temporal frequency. For data can be applied with higher-temporal frequency.
MSS, the IC provides data after every other scan, while for In the late-1980s, the Remote Sensing Group (RSG) at
TM and ETM+ it is at the end of each scan. The partial the University of Arizona developed three vicarious tech-
aperture solar calibrator on ETM+ provides data once per niques of absolute calibration that rely on in-situ measure-
orbit and the full-aperture solar calibrator on ETM+ is used ments. These methods are referred to as the reflectance-,
at least once per month. The stability of the lamps that are irradiance-, and radiance-based techniques (Biggar, Santer,
part of the IC is also such that variability over several scan & Slater, 1990; Slater et al., 1987). These three techniques
lines is quite small. This makes the on-board lamps ideal for have been used successfully for the SPOT HRV (Gellman
examining within scene variability of detectors. Experience et al., 1993), Landsat 5 TM (Slater et al., 1987; Thome
with the lamps on the early Landsat systems indicates that et al., 1993), a Daedalus scanner (Balick, Golanics, Shines,
large, abrupt changes in the lamp output do not occur. Thus, Biggar, & Slater, 1991), and the Airborne Visible and
the lamps are also excellent calibration sources over the Infrared Spectrometer (Vane et al., 1993). A variation of
period of weeks to months. Though untested for ETM+, the the radiance-based approach is cross-calibration where the
 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38 29

calibration of one satellite sensor is transferred to other (6) Spectral uniformity of the site is considered important
uncalibrated systems. Teillet et al. (1990) and Che et al. over as wide a spectral region as possible. This simplifies
(1991) employed this approach to calibrate several of band integrations and decreases the effects of spectral
the Advanced Very High-Resolution Radiometers. This mismatch between the ground-based and ETM+ sensors.
method is especially useful for large-footprint sensors for (7) Accessibility of the site and distance from Tucson are
which it is more difficult to apply the reflectance- and also important factors.
irradiance-based approaches. There is no ideal calibration site that satisfies all of the
In this work, the results of the radiometric calibration of above conditions. In the southwestern United States, there
the solar-reflective, multispectral bands of ETM+ using the exist several fairly uniform reflectance sites, which have
reflectance-based approach are presented. Studies on the been used over the course of many years by the RSG for
calibration of the thermal band of ETM+ can be found calibrations of Landsat-TM, SPOT-HRV, and other airborne
elsewhere in this special issue. Results from the panchro- and satellite-borne imaging sensors. Three of these test sites
matic band are not presented here due to the difficulty in were used for the current work and details of these test sites
accounting for the broadband spectral response of this band. are given below.
Future work will evaluate the calibration of this band using
reflectance-based methods. The vicarious calibration results 2.1. White Sands Missile Range
shown here rely on ground-based data collected at the
RSG’s historical test site at White Sands Missile range as The White Sands Missile Range test site in New Mexico
well as data collected at dry lake test sites in California and has been in use for vicarious calibration since the mid-
Nevada. Section 2 describes these test sites followed by a 1980s. It is located in the desert southwest of the United
description of the reflectance-based measurements made for States in a region of low aerosol loading and an elevation of
the current work. The results from this work are then 1.2 km. The test site used here for ETM+ is commonly
presented showing that the ETM+ sensor does not show referred to as Chuck Site and is located in the alkali flats
any significant degradation during the first 6 months of region. The coordinates of the test site are 32.919
N latitude
operation. Section 6 gives areas for future work. and 106.351
W longitude. The site is relatively devoid of
vegetation, though the area near the site includes regions of
greater vegetation and large gypsum dunes. In the VNIR,
2. Test site descriptions the White Sands site has a fairly flat spectral reflectance that
is quite high, however, the reflectance is much lower and
Vicarious calibration, whether it is the reflectance-, spectrally structured in the SWIR. The level of reflectance
irradiance-, and radiance-based techniques of the RSG or varies with season with the lowest reflectance values occur-
one of many other approaches, requires careful selection of ring during the winter months when portions of the missile
the test site used. For the RSG’s work, there are several range are either underwater or wet from the higher water
critical characteristics of an ideal test site and, in brief, these table. Highest reflectance values are typically seen in late
are (Scott, Thome, & Brownlee, 1996) the following. fall after the surface has dried after summer-season rains.
(1) A high-reflectance site reduces the impact of errors in The size of the White Sands area is the largest of the test
determining the radiance due to atmospheric scattering. A sites with an overall size of about 50 km.
site reflectance greater than 0.3 ensures that the radiance due
to reflection of the direct solar irradiance from the surface is 2.2. Railroad Valley Playa
the dominant contributor to the at-sensor radiance.
(2) An elevation of at least 1 km reduces the amount of Railroad Valley Playa is a dry lakebed in Nevada with a
atmospheric aerosols and the errors associated with predict- composition dominated by clay. The coordinates of this test
ing their characteristics and concentrations. site are 38.504
N latitude and 115.692
W longitude and its
(3) High-spatial uniformity over a large area minimizes site is located at 1.3 km above sea level between the cities of
the effects of scaling the reflectance data to the size of the Ely and Tonopah, NV. It is a desert site with no vegetation
full test site. The level of uniformity required is difficult to and aerosol loading is typically low. Railroad Valley Playa is
quantify other than to say, ‘‘the more uniform the better.’’ the largest of the playa test sites used by the RSG, but is still
(4) Changes with season should be minimal. This about one-fourth of the area of White Sands. While the
implies a site that is free of vegetation. Arid regions spectral reflectance of the playa sites is typically lower than
typically improve the probability of a temporally stable that of White Sands, especially in the blue part of the
site, while at the same time increasing the probability of spectrum, the spectral reflectance is reasonably flat through-
cloud-free days. out the spectral range of ETM+. Temporal records for this
(5) The site should be nearly lambertian to decrease site do not exist for as long a period of time as for White
uncertainties due to changing solar and view geometry. A Sands, so it is difficult to state how the reflectance varies as
flat site has the advantage of reducing BRF effects and a function of time of year. Early indications are that it too
eliminating shadow problems. has lowest reflectance in the winter months due to a rising
30 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

water table. The site is also more susceptible to cloudiness tance effects and a longer data collection time increases the
than the White Sands site with peak cloudiness in the winter probability that changes in atmospheric conditions can lead
and late summer months. to changes in the downwelling irradiance on the surface.
Selecting a smaller site would either mean sampling fewer
2.3. Roach Lake Playa pixels in the cross-track direction or fewer detectors. In
addition, this would mean having different sized sites than
The Roach Lake Playa test site is at an elevation of 0.8 km used for ongoing work with TM.
located near the California – Nevada border along Interstate To obtain the reflectance of the test site, a spectroradi-
15, which is the major highway between Los Angeles, CA ometer is transported across the entire site. The primary
and Las Vegas, NV. The coordinates of the test site are instrument for the surface-reflectance collection is a com-
35.651
N latitude and 115.367
W longitude. This playa is mercially available spectrometer that gives 1.4-nm spectral
immediately north of another playa, Ivanpah Playa, that is resolution from 350 to 1000 nm and 10-nm resolution for
also used by the RSG. The size of Roach Lake is approx- the 1000- to 2500-nm spectral range. The output from the
imately 3
3 km. This is somewhat smaller than the Ivanpah spectroradiometer is interpolated within the data collection
Playa, which is approximately 3
7 km, hence, it is a software to report results at a 1-nm spacing across the entire
secondary site for the RSG. Unfortunately, at the time of spectral range. The instrument is transported across the site
the work for ETM+, access to the Ivanpah Playa was using a backpack device that extends the instrument away
restricted by the Bureau of Land Management to prevent from the body of the user and raises the foreoptics to a
damage to the playa surface while studies of its use were height of about 2 m above the ground. An 8
field of view
being made. The spatial uniformity of the Roach Lake test is used for the measurements giving a circular sample on
site is the best of the three sites used here but its size is the the ground of approximately 0.3 m in diameter. A larger
smallest. Its reflectance has a similar spectral shape as that of field of view would give better spatial sampling but would
Railroad Valley, but is significantly brighter than Railroad be more susceptible to surface bidirectional reflectance
Valley while darker than White Sands in the visible and near effects. A smaller field of view forces a longer integration
infrared. Since this marked the first time that Roach Lake time, and thus increases the time needed to measure the test
was used, the temporal stability of the reflectance is not site. The user walks a path parallel to the cross-track
understood at this time. Work at Ivanpah Playa shows that direction of Landsat 7 through the center of the four
this playa’s reflectance is quite stable with time except for the cross-track pixels for all 16 along-track pixels. The spec-
few days following heavy rainfall. troradiometer is configured to average 30 spectra per
sample and 10 samples are collected within a single
‘‘pixel.’’ This gives a total of 19,200 spectra collected over
3. Reflectance-based approach the site and 640 samples. In essence, this means that
approximately 1400 m2 is sampled by the reflectance
This section gives details of the reflectance-based measurements. Considering that the site is 57,600 m2 in
method used in this work. The four sections describe the size, it should be clear that either a homogeneous or
basic parts of this approach. Uncertainties in the results of randomly varying site is needed since measurements of
these measurements are discussed in a later section. < 2.5% of the site area are used to represent the entire area.
It takes 45 –60 min to collect these data when the reference
3.1. Surface reflectance retrieval measurements are included.
Reflectance of the site is determined by ratioing the
The reflectance-based approach relies on ground-based, measurements described above to those of a reference panel
surface reflectance measurements of a selected site. For for which the bidirectional reflectance factor has been
ETM+, this site is a rectangular area that is 480
120 m determined in the laboratory. Measurements of the reference
with the long side of the site oriented approximately in the are made at the start and end of the data collection, as well
along-track direction of Landsat 7. This is the size that was as after every eight pixels (or, equivalently, every 80 test-site
originally selected by the RSG for work with Landsats 4 and samples). This level of sampling reduces the level of
5 TMs and ensured that all 16 detectors of the multispectral uncertainty due to changes in instrument response with time
bands were sampled as well as all of the detectors for the and changing atmospheric conditions, while keeping the
thermal bands. This means that four ground samples for data collection time to a reasonable level. Knowing the
each of the 16 multispectral detectors are collected corres- bidirectional reflectance of the reference allows the reflec-
ponding to the nominal 30-m ground spatial resolution of tance of each sample to be computed taking into account
the multispectral bands. Changing the size of the site to a effects due to sun-angle changes and reflectance panel
larger size would provide more statistical sampling for each bidirectional reflectance over the 60 min of data collection.
detector, but increasing the size also causes the data collec- Global, downwelling irradiance data are also collected near
tion to take longer. This can then create problems due to the test site to determine if there are significant changes
changes in solar zenith angle causing bidirectional reflec- in diffuse skylight illumination during the measurement
 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38 31

period. Once each of the spectral samples of the site is Chappuis absorption band). The Junge parameter described
converted to reflectance, all 640 data points are averaged to in Section 3.2 that is derived from the solar radiometer
give a single spectral reflectance for the entire site. measurements is used to compute Mie scattering phase
It should be clear that a critical part of this reflectance functions used in the code. While the radiative transfer
retrieval is the characterization of the reference panel in the code can include nonlambertian effects of the surface,
laboratory. The calibration of this panel is done with bidirectional reflectance measurements of the surface were
reference to a standard made from pressed polytetrafluoro- not available for this work, so the surfaces are assumed to
ethylene based on a prescribed approach defined by be lambertian.
National Institute of Standards and Technology (NIST) Strong gaseous absorption effects due to water vapor are
(Biggar et al., 1988). The calibration reference is a dir- determined using MODTRAN3 to compute transmittance
ectional-to-hemispheric reflectance standard provided by for the sun-to-surface-to-satellite path for 1-nm intervals
NIST. Polynomial fits are made to the measured data to from 350 to 2500 nm (Berk, Bernstein, & Robertson, 1989).
calculate the reflectance of the field standard for the sun- This sun-to-ground-to-sensor transmittance is multiplied by
view geometry and wavelengths for a given set of field the at-sensor radiance output from the radiative transfer
measurements (Biggar et al., 1988). Past work relied on code to correct the radiances for this strong absorption.
Halon as the standard of diffuse reflectance, but Halon is no While this approach is an approximation that excludes
longer available. Spyak and Lansard (1997) investigated interactions between diffusely scattered radiances and
another polytetrafluoroethylene powder, Algoflon F6, and absorption, it does not cause large uncertainties for applica-
found it to be a suitable replacement and this is currently tion to ETM+ because of the small effect of absorption
being used. within the ETM+ bands and the high surface reflectance of
the test sites used in this work.
3.2. Atmospheric characterization The relative radiances that are the output of the radiative
transfer code are converted to absolute radiances by mul-
Atmospheric characterization data are collected at the tiplying by a supplied solar irradiance curve corrected for
same time as the surface reflectance measurements. This changes in earth – sun distance. The solar irradiance standard
characterization relies on solar extinction measurements used in this work is that from MODTRAN3. These absolute
from a 10-band solar radiometer (Ehsani, Reagan, & Erxle- radiances are then band-averaged over the ETM+ spectral
ben, 1998). The solar radiometer is relatively calibrated responses to obtain band-averaged, at-sensor radiances.
immediately prior to, during, or after each field campaign.
Data are used in a Langley method retrieval scheme to 3.4. Determination of image digital numbers (DNs) and gain
determine spectral atmospheric optical depths (Gellman,
Biggar, Slater, & Bruegge, 1991). The optical depth results The final step needed to determine the sensor gain is to
are used as part of an inversion scheme to determine ozone compare the DN output from ETM+ to the predicted
optical depth and an aerosol size distribution (Biggar, Gell- radiances. The DN output is determined by averaging the
man, & Slater, 1990). In this work, the aerosols are assumed output for the 64 pixels related to our ETM+ test site. Level
to follow a power law distribution, also referred to as a 0R data are used in this work to reduce effects due to
Junge distribution. The advantage to such a distribution is resampling of the image data. Because the Level 0R data do
that it requires only one number to define the aerosol size not have a geometric correction applied, it is necessary to
distribution, the so-called Junge parameter. Optical depths at perform a rudimentary geometric correction of the region
1-nm intervals from 350 to 2500 nm are computed from this near our site. This is done by first locating our test site in the
derived size distribution (Junge parameter) and columnar image by finding blue tarpaulins that were laid out at
ozone. Columnar water vapor is derived from the solar opposing corners of the site. The tarpaulins darken the
extinction data using a modified Langley approach (Thome, pixels that contain them and serve as ground control points
Herman, & Reagan, 1992). in the imagery. Using these tarpaulins and also linear
features in the image, we linearly shift the pixels in the
3.3. Radiative transfer code Level 0R data to find the 64 pixels in the image corres-
ponding to the surface reflectance measurements.
These atmospheric and surface data are inputs to a The DNs from these 64 pixels are then averaged and
radiative transfer code that computes hyperspectral, at- ratioed to the predicted radiance from the radiative transfer
sensor radiances (Thome et al., 1996) based on a Gauss – code to obtain the gain for the selected band using (Eq. (1)):
Seidel iteration approach (Herman & Browning, 1965).
The code assumes a plane-parallel homogeneous atmo- DNl  Ol
Gl ¼ ð1Þ
sphere and divides this atmosphere into layers to account hLil
for the vertical distribution of scatterers and weak absorp-
tion due to ozone in the visible and near infrared (approx- where Gl is the gain for the given spectral band in units of
imately the 400- to 800-nm spectral range know as the DN/[W/(m2 sr mm)], DNl is the average DN for the
32 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

Table 1
Average atmospheric conditions for 10-min period about the given overpass time for each of the four calibration dates
Date (1999) Overpass time (MST) 550-nm aerosol optical depth Junge parameter Column ozone (cm-atm) Column water vapor (cm)
June 1 11:17 0.1024 3.112 0.172 1.139
July 20 11:08 0.0325 2.970 0.250 1.595
October 8 11:09 0.0381 3.093 0.187 1.135
October 20 10:32 0.0207 3.264 0.235 0.758

spectral band based on the 64 pixels, Ol is the offset in for the overpass time shown in Table 1. Further indication
DN from the calibration file, and hLli is the band-averaged of the atmospheric conditions can be seen in Fig. 1
spectral radiance predicted at the sensor from the radiative showing pyranometer data scaled to fit on the same graph
transfer code. as the optical depth data that are also shown. These
pyranometer data give a qualitative measurement of the
downwelling total irradiance and large, abrupt changes in
4. Recent Landsat 7 ETM+ results the output are indicative of clouds covering the sun.
Clouds in other parts of the sky are indicated by smaller
The results presented here are from three separate scale fluctuations in the irradiance.
calibration sites for four separate dates spanning the time Also shown in Fig. 1 are the inferred aerosol optical
period from May to October 1999. In addition to the four depths at 550 nm. These are obtained using the methods
data sets presented here, campaigns to White Sands described in Section 3. The data give both an indication
Missile Range on May 21 and 26, Ivanpah Playa on of the level of turbidity of the atmosphere as well as its
May 22, Roach Lake Playa on September 22, and Rail- temporal stability and these data clearly show the variable
road Valley Playa on May 27 and July 27 were also nature of the sky early in the day, while the period around
attempted. All of the data collections in May were during overpass was clear and temporally stable. Table 1 gives
the initial checkout period of the ETM+ sensor. Thus, the the average optical depth at 550 nm for a 10-min time
orbit of the sensor was not that of the Worldwide period about the overpass time of ETM+. Also given are
Reference System path/row and Landsat 7 was not at its the average Junge parameter, column ozone, and column
final 705-km orbit. While this created some difficulties water vapor derived from the solar extinction measure-
because of the different sensor inclination and oversam- ments. While it is clear that the early morning and late
pling due to the lower orbit, it also allowed multiple afternoon atmospheric conditions were cloudy, the stand-
calibration attempts at test sites without having to wait for ard deviation of the 10-min averages were all less than the
the 16-day repeat cycle. The ground-based data collections expected uncertainties in the derived quantities, implying
from May 21 and 26 during this checkout period were that atmospheric variability is a small contributor to any
successful, however, the reflectance of the White Sands
test site was such that all bands of ETM+ were saturated.
The surface data collection for May 22 was also success-
ful, but, unfortunately, the imagery from the ETM+ sensor
was one of the few data sets lost during the checkout
period. The May 27, July 27, and September 22 data sets
were contaminated by clouds. This is an indication of the
difficulties of vicarious calibration using in-situ measure-
ments. Of 10 total data collection attempts, only four high-
quality data sets were obtained and problems will some-
times be encountered with weather-related difficulties and
the vagaries of the sensor.

4.1. June 1 — Railroad Valley Playa

The first successful data collection by the RSG with

both a combination of clear skies and successful image
acquisition was June 1 at the Railroad Valley Playa. The
weather on this date was mostly cloudy in the morning
Fig. 1. Aerosol optical depth at 550 nm inferred from solar extinction
with clearing skies. There were cumulus clouds visible measurements from Railroad Valley on June 1, 1999. Also shown for
throughout the sky at the time of overpass, but no clouds reference are scaled pyranometer data to indicate downwelling global
were over the site at overpass, nor was the sun obstructed irradiance.
 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38 33

uncertainties. All of the values shown in Table 1 are Table 2

Summary of retrieved surface reflectance results for all seven bands
typical of our test sites, although the optical depth on this
date is higher than normal. However, it is still quite low Date (1999) Band 1 Band 2 Band 3 Band 4 Band 5 Band 7
when compared to values from more typical areas around June 1 0.253 0.332 0.365 0.393 0.402 0.318
the world. July 20 0.274 0.397 0.493 0.550 0.632 0.583
October 8 0.260 0.375 0.468 0.517 0.580 0.528
The other component to the reflectance-based method is
October 20 0.480 0.559 0.602 0.645 0.454 0.170
the retrieved surface reflectance at the time of overpass.
Fig. 2 shows the spectral reflectance derived for the test
site on this date. In addition, the standard deviation of this
average is shown as a percent of the average reflectance. the silicon-based detector of the spectroradiometer has
The average reflectance shown is that of all data collected poorer response at these wavelengths, again leading to
during a 60-min period about the ETM+ overpass time. As poorer signal to noise. The abrupt change in the percent
mentioned previously, the pyranometer and optical depth standard deviation at 1000 nm is due to a change in the
data shown in Fig. 1 indicate quite a bit of atmospheric spectroradiometer to a different detector that has better
variability early in the day and later in the day after response than silicon at these wavelengths. The third cause
overpass. However, the period during which the surface of larger standard deviations is due to changes in the
reflectance data were collected was stable enough that no column water vapor amounts between the measurements.
corrections for atmospheric variability were made in the This is especially important when considering differences
retrieved reflectance. Table 2 presents the band-averaged in water vapor amounts between the reference panel
reflectance for each of the six multispectral bands of measurements and the site measurements. This factor is
ETM+ derived from the spectral reflectance curve in considered to be small in this case due to the temporal
Fig. 2. stability of the retrieved water vapor amounts from the
Gaps seen in the reflectance and standard deviation solar extinction data.
curves are regions of the spectrum strongly affected by Also seen in Fig. 2 is an increase in the percent standard
water vapor absorption. Hence, the spectroradiometer deviation at shorter wavelengths. This is due to the fact that
measurements suffer from poor signal-to-noise effects in the spatial homogeneity of the surface is relatively constant
these spectral regions. This can be seen somewhat in the with wavelength in reflectance units. That is, if the reflec-
graph around the 940-nm water vapor absorption feature tance at 600 nm decreases by 0.02 in reflectance, then there
(it is more noticeable in related figures for the other test is a similar decrease in reflectance at other wavelengths.
sites in Figs. 4 and 5). While the spectral reflectance in Because the reflectance of the surface decreases at shorter
this part of the spectrum looks reasonable, the noisiness of wavelengths, this leads to a larger percent standard devi-
the data is larger than in other parts of the spectrum. This ation. A similar effect is seen at these wavelengths for
is due to a combination of three separate factors. The first measurements at Roach Lake and at longer wavelengths at
is that the absorption reduces the incident solar irradiance White Sands Missile Range.
in this part of the spectrum reducing the measured signals The results of these measurements were used as inputs
and hence the signal-to-noise-ratio of the instrument. The to the radiative transfer code. Table 3 gives the predicted
second factor is similar in nature. It is due to the fact that at-sensor radiances for each of the bands considered here
as well as the average DNs determined from the imagery.
The uncertainty for the DNs given in the table is the
1  s standard deviation of the average of the 64 pixels
corresponding to the test site used. In addition, the table
also contains the minimum and maximum average DNs
determined by shifting the 64-pixel area by one pixel in
all directions. This gives an indication of the uncertainty
due to a misregistration between the ground-based reflec-
tance data and the imagery. The differences caused by
misregistration are all < 1%, while the percent standard
deviation of the 64-pixel average is approximately the
same as that of the percent standard deviation of the
reflectance data.
Finally, combining the data in Table 3 allows for the
sensor gain to be computed. The units used here are in DN
per unit radiance with radiance having the units of W/[m2 sr
mm]. The values for each of the multispectral bands is given
Fig. 2. Spectral reflectance results for Railroad Valley Playa site for June in Table 4. The table also includes the prelaunch, laboratory-
1, 1999. derived gains for reference.
34 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

Table 3
Site digital number results for each band and each date examined. Also shown are the predicted radiances from the reflectance-based method for each date
and band.
Band 1 Band 2 Band 3 Band 4 Band 5 Band 7
June 1 Average site DN 194.4 201.8 242.1 178.1 194.5 156.1
1  s S.D. 2.5 2.7 3.3 2.1 2.5 2.6
Minimum DN 193.6 199.9 239.8 176.6 192.9 154.8
Maximum DN 195.3 202.9 243.6 179.1 195.6 156.8
Radiance (W/[m2 sr mm]) 153.7 168.5 152.7 109.8 24.61 6.038
July 20 Average site DN 203.3 231.2 255.0 234.1 255.0 254.0
1  s S.D. 3.6 3.4 – 2.3 – 1.5
Minimum DN 202.9 230.8 255.0 233.9 255.0 253.1
Maximum DN 203.6 231.6 255.0 234.6 255.0 254.3
Radiance (W/[m2 sr mm]) 161.9 193.7 200.4 150.1 37.84 10.70
October 8 Average site DN 157.7 176.6 235.6 180.0 212.7 190.2
1  s S.D. 3.4 3.9 5.2 3.6 3.5 4.1
Minimum DN 156.9 176.0 234.9 179.4 212.1 189.3
Maximum DN 158.7 177.8 236.4 180.7 213.7 191.0
Radiance (W/[m2 sr mm]) 125.1 148.7 153.7 114.5 28.15 7.82
October 30 Average site DN 240.0 237.2 255.0 210.5 160.7 68.8
1  s S.D. 5.1 5.5 0.0 4.6 3.3 1.8
Minimum DN 239.4 236.5 255.0 210.0 160.7 68.6
Maximum DN 241.0 238.2 255.0 211.5 161.2 69.0
Radiance (W/[m2 sr mm]) 194.1 198.3 181.2 133.8 20.62 2.365

4.2. July 20 — Roach Lake Playa saturation of some pixels in band 7. Thus, it is not possible
to calculate gains for these bands on this date. The retrieved
The weather on this date was clear with no clouds. Fig. 3 gains from this campaign are given in Table 4.
shows the derived 550-nm optical depth as a function of
time for this date, as well as the scaled pyranometer data.
This graph is similar to that of Fig. 1, except the y-axis scale 4.3. October 8 — Roach Lake Playa
is one order of magnitude smaller. It is clear from this figure
that the atmosphere was quite stable during the entire period Conditions on October 8 at Roach Lake were very
of measurement as well as the time around sensor overpass. similar to those on July 20. Table 1 summarizes the
The atmospheric results are summarized in Table 1. Fig. 4 atmospheric results and Table 2 the reflectance data. From
shows the spectral reflectance derived for the test site on this Table 2, it is clear that the surface of the playa is somewhat
date in a similar fashion to Fig. 2. By comparison, the darker with the reflectance in the visible and near-infrared
spectral curves are similar in shape, but Roach Lake is about 5% darker while the short-wave infrared darkened
clearly brighter, as can also be seen in Table 2. Also evident
in the figure is the lower percent variability of the site. Table
3 summarizes the results of this collection for each spectral
band considered here, as well as the DNs derived from the
Level 0R image data. One problem that was encountered for
this data collection was the saturation of bands 3 and 5 and

Table 4
Derived gains for each of the four dates and the six visible and near-infrared
multispectral bands of ETM+ (values for April 15 indicate prelaunch,
laboratory-derived gains)
Gains [DN/(W/(m2 sr mm))]
Date (1999) Band 1 Band 2 Band 3 Band 4 Band 5 Band 7
April 15 1.22 1.18 1.51 1.51 7.59 21.75
June 1 1.167 1.108 1.487 1.486 7.294 23.370
July 20 1.163 1.116 – 1.560 – –
October 8 1.141 1.087 1.435 1.441 7.024 22.410 Fig. 3. Aerosol optical depth at 550 nm inferred from solar extinction
October 30 1.159 1.121 – 1.462 7.067 22.751 measurements from Roach Lake on July 20, 1999. Also shown for reference
Values also determined after removing 15-DN offset for all bands and dates. are scaled pyranometer data to indicate downwelling global irradiance.
 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38 35

discussion for Fig. 2, this decrease in reflectance in the

short-wave infrared then leads to much larger percent
standard deviations at these wavelengths. The band-aver-
aged reflectance for this campaign is given in Table 2. Table
3 summarizes the average DNs for the site and the predicted
at-sensor radiances, and Table 4 gives the retrieved gains
from this campaign.

4.5. Discussion

Examining Table 4 leads to two key conclusions. The

first is that the agreement between different dates is quite
good with the range of gains being < 3.5% for all bands.
This implies two things. The first is that the response of the
ETM+ sensor was stable during this time period. The
Fig. 4. Spectral reflectance results for Roach Lake site for July 20, 1999. second is that the precision of these results is quite good,
giving confidence that trends in instrument response could
slightly more (8 –9%). Much of this was due to recent be determined given enough data points. The second con-
rainfall in the area that had left standing water over clusion is that the vicarious results agree well with the
portions of the playa. The DNs retrieved from imagery prelaunch gains. The estimated 1  s uncertainty in the
are given in Table 3. Here, it is clear that the site was less prelaunch values have been given as 5% absolute (B.
homogeneous than for the previous date. However, the Markham, personal communication), and, as will be dis-
level of inhomogeneity is still < 2% and not a large source cussed in Section 5, the estimated uncertainties in the
of uncertainty. The retrieved gains from this campaign are vicarious calibrations are 3 – 5%. Thus, the vicarious cal-
given in Table 4. ibration results agree to better than the combined uncertain-
It should be noted that the reflectance results from this ties in the two methods. An obvious interpretation of this
date are an excellent example of the need for ground-based result is that the prelaunch calibration for ETM+ is still valid
measurements at the time of sensor overpass to character- for this time period.
ize the surface reflectance of the test site. While the It should be noted that while the results of the prelaunch
surface reflectance of the sites used by the RSG has low calibration and the postlaunch vicarious calibration agree to
variability with time, the 5 – 9% change in reflectance within their combined uncertainties, there appears to be
would have a similar effect on the predicted at-sensor biases between the results. The vicarious results for bands
radiance. Thus, using the retrieved reflectance values from 1 –5 are all less than the prelaunch value (except for the
July 20 would lead to a 5 –9% error in the retrieved gain value obtained for band 4 on July 20), while the vicarious
and this is unacceptable when attempting to absolutely results for band 7 give gain values that are higher than the
calibrate a sensor to better than 5%. This effect has been prelaunch. It is not clear, at this point, what is the cause of
well known from the RSG’s work at White Sands Missile this difference, but work is currently underway to under-
Range where the test site’s reflectance has been known to stand these ‘‘biases.’’
vary by more than 20% depending upon the time of year
(Thome et al., 1993).

4.4. October 30 — White Sands Missile Range

The final data set presented here uses data from White
Sands Missile Range at the RSG’s Chuck Site area. Weather
on this date was similar in nature to that of the Roach Lake
data sets. The retrieved atmospheric conditions are summar-
ized in Table 1 and show that conditions were indeed similar
to those of July 20 and October 8. Fig. 5 shows the spectral
reflectance derived for the test site along with the percent
standard deviation of the average. The large spectral vari-
ability of the reflectance in the short wave is a clear
indication for selecting the dry-lake test sites for vicarious
calibration. In addition, Fig. 5 shows the surface to be less
homogeneous, and the reflectance in the short-wave infrared Fig. 5. Spectral reflectance results for White Sands Missile Range for
is much lower than for the playa sites. As described in the October 30, 1999.
36 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

5. Uncertainties distribution and the aerosol complex index of refraction.

These two parameters lead to top-of-the-atmosphere-radi-
There are two ways in which to view the uncertainties of ance uncertainties of 2.0% and 2.5%, respectively. If it is
the results. One way is to determine the uncertainties of each assumed that the above errors are independent, then a root-
separate field measurement and then determine how these sum square approach leads to an overall uncertainty of 5%.
uncertainties impact the predicted at-sensor radiances. The To help evaluate this estimated accuracy, a joint field
other is to compare independently derived results from campaign to Lunar Lake Playa was organized under support
several groups measuring the same site at the same time from NASA’s EOS project (Thome et al., 1998). This
and produce results for identical spectral bands. Results of international campaign included scientists from three sen-
both approaches have been presented previously and, since sors on the EOS Terra Platform, the ETM+ sensor, and other
this earlier work is still valid for application to ETM+, they non-EOS investigators. This campaign pointed out differ-
are only summarized here (Biggar, Gellman, & Slater, 1994; ences between groups in the order of 5% in the spectral
Thome et al., 1998). region corresponding to the bands of ETM+. Most of the
From the description of the reflectance-based method differences seen were attributed to several groups using
given earlier, it can be seen that there are four basic areas directional – hemispherical reflectances and other groups
of uncertainty in the method: (1) atmospheric characteriza- using directional – directional reflectances for the reflectance
tion, (2) surface characterization, (3) radiative transfer code, of their reflectance panels. When these differences were
and (4) computation of the site-average DNs. The factors taken into account, even better agreement was found. It
leading to uncertainties in determining the site average DN should be pointed out that these differences are more an
are an incorrect determination of the site’s location in the indication of the precision of the vicarious calibration rather
image and the subsequent misregistration of the site’s surface than the accuracy, since many of the groups used very
reflectance to DN. As discussed in Section 3, the use of similar retrieval algorithms (for example, the assumption of
tarpaulins allows us to determine the location of our site to a Junge size distribution and use of similar reflectance
better than one pixel. Even allowing a misregistration of up references). However, the comparisons included a variety
to two pixels leads to less than a 1% uncertainty in the site- of radiative transfer codes, atmospheric instrumentation, and
averaged DNs and thus the computation of the calibration surface reflectance data collection techniques and equip-
coefficient. Uncertainties caused by the radiative transfer ment. Thus, this comparison is a good indication that the
code are its inherent numerical accuracy and assumptions. uncertainties of the vicarious calibrations is lower than 5%.
Biggar et al. (1994) lists the uncertainty due to these as < 1% One last point, however, is that the above cross-comparisons
in the top-the-atmosphere radiance computed by the code for were made on the relative at-sensor radiances. An additional
cases when the radiative transfer codes are used with uncertainty is added when converting these relative radian-
identical atmospheric assumptions and inputs. ces to absolute radiances.
The uncertainties in the retrieved reflectance are harder to
quantify since there are two errors to consider here. The first
is the error in an individual measurement, that is, the 6. Conclusions
accuracy of a single reflectance measurement. The uncer-
tainty in a single measurement of the site reflectance is < 2% Four separate vicarious calibration campaigns were held
for the range of reflectances seen at our field sites (that is a for the purpose of evaluating the responsivity of the solar-
reflectance value of 0.01 at a reflectance of 0.50). The reflective, multispectral bands of the Landsat 7 ETM+
primary uncertainty sources for the surface reflectance sensor. These campaigns took place early in the mission
retrieval are the calibration of our field reference panels, with the first set of results obtained 47 days after launch.
instrument noise of our field radiometers, and diffuse sky- The final data set shown here was collected 198 days after
light corrections. Recall that these individual reflectances are launch. The vicarious results agreed with each other to
averaged to determine an overall site reflectance. The uncer- better than 3.5% indicating both the stability of the ETM+
tainty due to this can be evaluated by resampling the sensor and the quality of the vicarious results.
reflectances that are used to determine the site reflectance. Past work with the reflectance-based method used here
Randomly removing as many as one-half of the samples indicate that the method has absolute uncertainties in the
from the averaging has less than a 1% effect on the retrieved 3 –5% range. Comparisons of the reflectance-based results
reflectance. Thus, while the small-scale variability of the to the prelaunch, laboratory-based gains agree to better
surface reflectance of these sites can be quite large, as than 7% in all cases, which is within the combined
evidenced by the large standard deviations of the site uncertainties of the two approaches. Further work is
averages, this variability is random enough that it does not needed to better understand the possibility of biases
represent the uncertainty of the retrieved reflectance. between the prelaunch and vicarious methods since the
The last area of uncertainty is the atmospheric character- reflectance-based values for bands 1 – 5 are less than those
ization. Much of our uncertainty here is due to an incorrect from prelaunch and the values for band 7 exceed the
determination of the aerosol type, both the aerosol size prelaunch. A possible source of the bias could be incorrect
 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38 37

assumptions about aerosol absorption and this would Che, N., Grant, B. G., Flittner, D. E., Slater, P. N., Biggar, S. F., Jackson,
R. D., & Moran, M. S. (1991). Results of calibrations of the NOAA-11
strongly affect the shorter wavelength bands. Biases at
AVHRR made by reference to calibrated SPOT imagery at White Sands
the longer wavelengths could be due to uncertainties in the N.M.. Proceedings of SPIE, 1493, 182 – 194.
assumed solar irradiances used to convert the relative Ehsani, A. R., Reagan, J. A., & Erxleben, W. H. (1998). Design and
radiances to absolute values. Further work will also performance analysis of an automated 10-channel solar radiometer
include comparisons to the on-board calibrators of ETM+ instrument. Journal of Atmospheric and Oceanic Technology, 15,
to better understand the accuracy of the vicarious results 697 – 707.
Engel, J., & Weinstein, O. (1983). The Thematic Mapper — an over-
and further understand these possible biases. view. IEEE Transactions on Geoscience and Remote Sensing, GE-21,
Future work with the ETM+ sensor related to these 258 – 265.
vicarious approaches includes the implementation of the Gellman, D. I., Biggar, S. F., Dinguirard, M. C., Henry, P. J., Moran, M. S.,
irradiance- and radiance-based approaches. In addition, Thome, K. J., & Slater, P. N. (1993). Review of SPOT-1 and -2 cali-
cross-calibration attempts will be made using sensors on- brations at White Sands from launch to the present. Proceedings of
SPIE, 1938, 118 – 125.
board NASA’s Terra platform launched in December 1999. Gellman, D. I., Biggar, S. F., Slater, P. N., & Bruegge, C. J. (1991). Cali-
This platform is in an orbit that is 30 min behind Landsat 7, brated intercepts for solar radiometers used in remote sensor calibration.
providing comparisons with the sensors on the Terra plat- Proceedings of SPIE, 1493, 175 – 180.
form. Because of the excellent quality of the ETM+ data, it Herman, B. M., & Browning, S. R. (1965). A numerical solution to the
should be possible to improve the vicarious methods that are equation of radiative transfer. Journal of Atmospheric Science, 22,
559 – 566.
being implemented for the Terra platform to create a Horan, J. J., Schwartz, D. S., & Love, J. D. (1974). Partial performance
quantitatively consistent data set. degradation of a remote sensor in a space environment and some pos-
sible causes. Applied Optics, 13, 1230 – 1237.
Hovis, W. A., Knoll, J. S., & Smith, G. R. (1985). Aircraft measurements
for calibration of an orbiting spacecraft sensor. Applied Optics, 24,
Acknowledgments
407 – 410.
Kaufman, Y. J., & Holben, B. N. (1993). Calibration of the AVHRR visible
The author wishes to acknowledge the Atmospheric and near-IR bands by atmospheric scattering, ocean glint, and desert
Sciences Laboratory at WSMR for the use of the site and reflection. International Journal of Remote Sensing, 14, 21 – 52.
data support at White Sands, the Bureau of Land Manage- Lansing, J. C. (1986). Solar calibration performance in Landsat multispec-
tral scanners. Applied Optics, 25, 333 – 334.
ment offices in Tonopah, NV and Needles, CA for their help
Lansing, J. C., & Cline, R. W. (1975). The four- and five-band multispectral
in accessing our test sites in California and Nevada. I would scanners for Landsat. Optical Engineering, 14, 312 – 322.
also like to acknowledge the many students and staff who Markham, B. L., Barker J. L. (Eds.) (1985). Special issue: LIDQA final
helped with the field collections used for this work — symposium, Photogrammetric Engineering and Remote Sensing 51,
Arsalan Ahmad, Nik Anderson, Wayne Barber, Xiancheng 1245 – 1493.
Markham, B. L., & Barker, J. L. (1987). Radiometric properties of US
Ding, Ann Gordon, Hyunki Kim, Rob Kingston, Keith processed Landsat MSS data. Remote Sensing of Environment, 22,
Krause, Michele Kuester, Brian Magi, Marek Mienko, John 39 – 71.
LaMarr, Prabal Nandy, and Emily Whittington. This Markham, B. L., Barker, J. L., Boncyk, W. C., Kaita, E., & Helder, D. L.
research was supported by NASA grant NAG5-2448. (1996). Landsat-7 Enhanced Thematic Mapper Plus in-flight radiomet-
ric calibration. International Geoscience and Remote Sensing Sympo-
sium, 2, 1270 – 1272 (Lincoln, NE).
Scott, K. P., Thome, K. J., & Brownlee, M. R. (1996). Evaluation of the
References Railroad Valley Playa for use in vicarious calibration. Proceedings of
SPIE Conference, 2818, 158 – 166.
Balick, L. K., Golanics, C. J., Shines, J. E., Biggar, S. F., & Slater, P. N. Slater, P. N. (1980). Remote sensing: optics and optical systems ( pp. 465 –
(1991). The in-flight calibration of a helicopter-mounted Daedalus mul- 509). Reading, MA: Addsion-Wesley Publishing.
tispectral scanner. Proceedings of SPIE, 1493, 215 – 223. Slater, P. N., Biggar, S. F., Holm, R. G., Jackson, R. D., Mao, Y.,
Berk, A., Bernstein, L. S., & Robertson, D. C. (1989). MODTRAN: a Moran, M. S., Palmer, J. M., & Yuan, B. (1987). Reflectance- and
moderate resolution model for LOWTRAN7. GL-TR-90-0122. Tech- radiance-based methods for the in-flight absolute calibration of multi-
nical Report. Geophysics Directorate, Phillips Laboratory, Hanscom, spectral sensors. Remote Sensing of Environment, 22, 11 – 37.
AFB, MA. Spyak, P. R., & Lansard, C. (1997). Reflectance properties of pressed
Biggar, S. F., Gellman, D. I., & Slater, P. N. (1990). Improved evaluation of Algoflon F6: a replacement reflectance-standard material for Halon.
optical depth components from Langley plot data. Remote Sensing of Applied Optics, 36, 2963 – 2970.
Environment, 32, 91 – 101. Teillet, P. M., Slater, P. N., Ding, Y., Santer, R. P., Jackson, R. D., & Moran,
Biggar, S. F., Gellman, D. I., & Slater, P. N. (1994). Uncertainties in the M. S. (1990). Three methods for the absolute calibration of the NOAA
in-flight calibration of sensors with reference to measured ground sites AVHRR sensors in-flight. Remote Sensing of Environment, 31, 105 – 120.
in the 0.4 to 1.1 mm range. Remote Sensing of Environment, 48, Thome, K. J., Gellman, D. I., Parada, R. J., Biggar, S. F., Slater, P. N., &
245 – 252. Moran, M. S. (1993). In-flight radiometric calibration of Landsat-5
Biggar, S. F., Labed, J. F., Santer, R. P., Slater, P. N., Jackson, R. D., & Thematic Mapper from 1984 to present. Proceedings of SPIE, 1938,
Moran, M. S. (1988). Laboratory calibration of field reflectance panels. 126 – 131.
Proceedings of SPIE, 924, 232 – 240. Thome, K. J., Gustafson-Bold, C. L., Slater, P. N., & Farrand, W. H. (1996).
Biggar, S. F., Santer, R. P., & Slater, P. N. (1990). Irradiance-based cali- In-flight radiometric calibration of HYDICE using a reflectance-based
bration of imaging sensors. International Geoscience and Remote Sens- approach. Proceedings of SPIE Conference, 2821, 311 – 319.
ing Symposium, 1, 507 – 510 (College Park, MA). Thome, K. J., Herman, B. M., & Reagan, J. A. (1992). Determination of
38 K.J. Thome / Remote Sensing of Environment 78 (2001) 27–38

precipitable water from solar transmission. Journal of Applied Meteor- Vane, G., Green, R. O., Chrien, T. G., Enmark, H. T., Hansen, E. G., &
ology, 31, 157 – 165. Porter, W. M. (1993). The airborne visible/infrared imaging spectrom-
Thome, K., Markham, B., Barker, J., Slater, P., & Biggar, S. (1997). Radio- eter (AVIRIS). Remote Sensing of Environment, 44, 127 – 143.
metric calibration of Landsat. Photogrammetric Engineering and Re- Vermote, E., Santer, R. P., Deschamps, P. Y., & Herman, M. (1992). In-
mote Sensing, 63, 853 – 858. flight calibration of large field of view sensors at short wavelengths
Thome, K., Schiller, S., Conel, J., Arai, K., & Tsuchida, S. (1998). Results using Rayleigh scattering. International Journal of Remote Sensing, 13,
of the 1996 Earth Observing System vicarious calibration campaign at 3409 – 3429.
Lunar Lake Playa, Nevada (USA). Metrologia, 35, 631 – 638.