Radiometric characterization of hyperspectral imagers using

multispectral sensors

Joel McCorkela, Kurt Thome

b, Nathan Leisso

a, Nikolaus Anderson

a, and Jeff Czapla-Myers

a

aRemote Sensing Group, College of Optical Sciences, University of Arizona, Tucson AZ 85721

bNASA Goddard Space Flight Center, Greenbelt MD 20771

ABSTRACT

The Remote Sensing Group (RSG) at the University of Arizona has a long history of using ground-based test sites for

the calibration of airborne and satellite based sensors. Often, ground-truth measurements at these tests sites are not

always successful due to weather and funding availability. Therefore, RSG has also employed automated ground

instrument approaches and cross-calibration methods to verify the radiometric calibration of a sensor. The goal in the

cross-calibration method is to transfer the calibration of a well-known sensor to that of a different sensor. This work

studies the feasibility of determining the radiometric calibration of a hyperspectral imager using multispectral imagery.

The work relies on the Moderate Resolution Imaging Spectroradiometer (MODIS) as a reference for the hyperspectral

sensor Hyperion. Test sites used for comparisons are Railroad Valley in Nevada and a portion of the Libyan Desert in

North Africa. Hyperion bands are compared to MODIS by band averaging Hyperion’s high spectral resolution data with

the relative spectral response of MODIS. The results compare cross-calibration scenarios that differ in image acquisition

coincidence, test site used for the calibration, and reference sensor. Cross-calibration results are presented that show

agreement between the use of coincident and non-coincident image pairs within 2% in most bands as well as similar

agreement between results that employ the different MODIS sensors as a reference.

Keywords: cross calibration, hyperspectral, desert sites

1. INTRODUCTION

There are dozens of earth observing satellite sensors currently on orbit for the purpose of land- and ocean-based remote

sensing. Data reported by these sensors provide a basis for a wide range of environmental studies. It is critical that

sensors remain calibrated in order to achieve synergy among coexisting sensors and legacy among generations of

sensors. There are many available methods of radiometric calibration of satellite sensors including prelaunch, on-board

lamp, lunar, or solar illumination, and methods that use terrestrial scenes acquired on-orbit. The RSG has historically

employed the reflectance-based method of vicarious calibration for a large array of sensors. The basic method primarily

relies on manned, in situ measurements of a test site’s surface and atmospheric conditions close in time to when the

sensor of interest views that site1. Recent developments by the RSG allow these measurements to be autonomous

without the requirement for on-site personnel2.

One critical goal of the in-flight radiometric calibration of a sensor is to determine that sensor’s calibration as soon as

possible after launch. The rigors of launch often invalidate the results of the preflight calibration. On-board calibration

hardware has also been shown to change in transition from the laboratory to the space environment, therefore additional

validation is preferred3. Approaches relying on in-situ measurements have been shown to provide this validation but

often prove ineffective for rapid evaluation of sensor calibration. Poor weather conditions, lack of personnel, or limited

financial resources can prevent the number of usable data collections and limit the frequency of the in-situ data. Cross-

calibration methods provide an opportunity for vicarious calibration at a higher temporal frequency than the in-situ

approaches.

The two sensors involved in a cross-calibration each have their own orbit therefore their imagery will be acquired with

unique viewing and solar geometries. The ideal case for cross-calibration is when the two sensors share the same orbit

or even the same spacecraft. For example, before arriving to its destination orbit, Landsat 7 was placed in a tandem orbit

with Landsat 5 allowing for the cross calibration of the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) and Landsat

5 Thematic Mapper (TM) to establish calibration continuity4,5

. Less ideal is the case where the two sensors acquire

imagery at different times. This occurs attempting the comparison between ETM+ and Terra MODIS, which lags ETM+

Earth Observing Systems XIV, edited by James J. Butler, Xiaoxiong Xiong, Xingfa GuProc. of SPIE Vol. 7452, 745210 · © 2009 SPIE · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.828608

Proc. of SPIE Vol. 7452 745210-1

by approximately 40 minutes6. These and other, less ideal, scenarios for cross-calibration of two satellite sensors are

listed below:

- Same geometry and coincident acquisitions

- Nearly same geometry and nearly coincident acquisitions

- Different geometry and nearly coincident acquisitions

- Same geometry and different acquisition times/dates

- Different geometry and different acquisition times/dates

In this paper, geometry refers to the viewing zenith angle and solar zenith angle at the time of image acquisition. As

mentioned, the most accurate case is when two sensors view the same area at the same time with the same geometry.

Any difference in acquisition time or date increases uncertainties due to changes in surface and atmospheric conditions

affecting the radiance incident at the two sensors. The effect of the atmosphere and surface are further exacerbated when

the two sensors view the same area at different view angles.

This work studies cross calibration results for two of the above mentioned scenarios: 1) the sensors have the same

geometry and nearly coincident image acquisitions; and 2) the sensors have the same geometry but different acquisition

dates. Results for both of these scenarios are presented for the comparison between Hyperion and MODIS (Terra and

Aqua). Studying both scenarios with the same sensor pair is possible because Hyperion’s orbit preceded MODIS by 40

minutes early in its mission, but experienced an orbit change in mid-2005, after which only non-coincident comparisons

are possible.

This paper gives a brief overview of the sensors used for the cross-calibration and then discusses the test sites used in the

comparisons. Methodology is presented on how the comparisons were made for the two different cross-calibration

scenarios. The results section discusses the agreement between comparisons made with the nearly coincident datasets

and comparisons made with datasets that have been acquired on different dates. Also shown is the agreement between

results obtained with a Hyperion-Terra MODIS comparison and a Hyperion-Aqua MODIS comparison. To provide a

comparison between Hyperion and MODIS independent of the method described in this work, the cross-calibration

results are compared with results obtained from the reflectance-based approach of vicarious calibration, which takes

advantage of ground-truth measurements.

2. SENSORS

Multispectral sensors tend to have wider swath widths and more frequent acquisitions of any given site relative to a

hyperspectral imager. In addition, hyperspectral sensors are usually tasked by an operator, meaning that even when the

sensor is over a desired area there may be no image acquisition. The outcome is that access to a multispectral, wide

swath image of a desired test site will not mean access to a similar hyperspectral image. On the other hand, there is a

high probability of having an overlapping multispectral scene for any given hyperspectral acquisition.

There are many multispectral sensors in orbit that have trusted absolute radiometric calibrations. This work uses data

from the MODIS sensors on the Terra and Aqua platforms due to their remarkably accurate radiometric calibration7,8

.

Frequent revisit times of every other day at the equator and daily coverage at mid-latitudes is essential for this work and

is provided by the MODIS sensors. Terra was launched in December 1999 with a 10:30 am equatorial crossing time and

Aqua was launched in May 2002 with a 1:30 pm equatorial crossing time. Terra is commonly known as the ‘AM’

satellite and Aqua, the ‘PM’ satellite due to their equatorial crossing times. MODIS has spatial resolutions of 250, 500,

and 1000 m, depending on the spectral band, covering a swath width of 2330 km. The MODIS sensors have a total of 36

bands, 19 of which are within the solar reflective region of the electromagnetic spectrum.

Hyperion is the hyperspectral imager selected for this work. One advantage to using the Hyperion sensor is RSG’s

history of vicarious calibration using the reflectance-based approach providing results that can be compared with other

studies. Hyperion is onboard the EO-1 spacecraft launched in November 2000. It has 198 processed spectral bands in

the 0.4-2.5 μm range with nominal bandwidths of 10-11 nm and about 10 nm between band centers9. The imager has a

ground resolution of 30 m covering a swath width of 7.7 km.

The reflectance-based calibration results of Hyperion uses imagery that is processed to level 1R. Level 1T data are used

for this cross-calibration study. Level 1R could not be obtained for this study. Level 1T data are freely available on the

internet and includes geolocation data. For comparing the results from the reflectance-based approach and the cross-

calibration study in this work, it is important that the radiometric scale between the two different levels be known. The

Proc. of SPIE Vol. 7452 745210-2

4

2

0

24

500 1000 1500 2000 2500Wavelength (nm)

1%

solid line in Figure 1 shows the difference between level 1R and level 1T radiances averaged over an area of 1 km 1

km and five different acquisitions at Railroad Valley. The positive vertical axis represents a higher radiance reported in

level 1R. The dashed lines represent plus and minus one standard deviation about the average of the five scenes. With

the largest difference approaching 1% in the shortwave infrared, radiometric differences between data levels are

assumed to be negligible for comparisons between results of reflectance-based approach and cross-calibration.

Figure 1. The percent difference curve (solid line) shows a comparison of Hyperion L1R and L1T imagery of a 1 km 1km

area of the Railroad Valley test site averaged for five image pairs. The dashed lines represent plus and minus one

standard deviation about the average.

3. TEST SITES

The ideal test site for reflectance-based calibration has many characteristics. These include a relatively high surface

reflectance and surface properties that are spatially uniform, temporally invariant and near-lambertian. It is also desirable

that the site reside at a high altitude; cover a large area; be an arid region to limit water vapor; and must also be

accessible10

. Many of these same properties are needed for accurate cross-calibration as well, and the RSG sites serve as

the starting point for this work. One site evaluated extensively by RSG is Railroad Valley Playa in Nevada that has been

found to satisfy many of the above criteria10

. RSG has a reasonably large archive of ancillary data and imagery from

many sensors at Railroad Valley. Railroad Valley is located at approximately 38.5 N, 115.7 W and is accessible via a

common US highway. It offers a relatively high reflectance across the visible and short wave infrared spectrum, short-

term stability, frequently cloud-free skies, and is easily locatable in imagery. It is about 10 x 15 km2 in area at 1430 m

altitude.

A one-kilometer square area of Railroad Valley that has been used by RSG for the reflectance-based calibration of

MODIS is used in this work11

. A typical reflectance spectrum obtained from in-situ measurements of the area is shown

in Figure 2. This area represents a single pixel of the low-resolution bands of MODIS depicted in Figure 3.

Proc. of SPIE Vol. 7452 745210-3

0.5

0.4

0.3

0.2

0.1

00500 1000 1500 2000 2500

Wavelength (nm)

Figure 2. Reflectance of the Railroad Valley test site according to in situ measurements on 13 July 2002.

One criterion listed above not required for cross-calibration is accessibility. Work has shown that portions of the

Saharan desert are suitable invariant sites and also meet many of the characteristics listed for an ideal site, especially that

of minimal cloud cover and precipitation12

. Previous work that utilizes the Saharan desert includes temporal trending of

sensor calibration13,14

, sensor intercomparisons15,16,17

, flat fielding wide field of view sensors by characterizing several

sites18

, and comparison of aerosol optical depth products over desert sites19

. For this work, a Saharan test site offers the

opportunity to utilize a site much larger than Railroad Valley and evaluate the accuracy of using a site for which ground-

based data are not available.

The selection of test site area within the Saharan desert is based on the availability of high temporal frequency Hyperion

imagery that was readily found in seemingly uniform areas. This search led to scenes in Libya, more specifically in

World Reference System (WRS) path 181, row 40, in which nearly 200 scenes have been acquired from 2004 to present.

In order to compare consistent portions of the Libyan surface, an area is selected that is slightly narrower than

Hyperion’s swath width and long as possible while keeping a large buffer away from less uniform areas. Initial

judgment of selecting this area was done manually by stretching scenes from several dates and selecting an area that had

minimal change spatially and temporally. The resulting area is shown as the highlighted rectangle in Figure 4 and is

defined by these coordinates: [29.1056, 23.8091], [29.0958, 23.8687], [28.8180, 23.7956], and [28.8281, 23.7348]. This

represents an area of approximately 6 km 31 km or about 210,000 Hyperion pixels and almost 200 MODIS pixels.

This area is fully contained within the 50 Libyan Hyperion scenes processed in this work despite random longitudinal

shifts of spatial coverage.

Figure 3. A portion of a typical Hyperion scene of

Railroad Valley in WRS path 40 row 33. The

highlighted square in the middle of the playa

represents the area used for comparisons.

Figure 4. A portion of a typical Hyperion scene of

Libya in WRS path 181 row 40. The large

highlighted rectangle represents the area used for

comparisons.

One way to assess the uniformity of a test site is to find the standard deviation of the radiance or reflectance values

across the site. However, the results of this assessment will also contain the variability in the instrument performing the

measurements such as noise and flat fielding errors. This assessment is shown for the Railroad Valley in Figure 5 where

the solid line shows the standard deviation of the average radiance reported by Hyperion and the dashed line shows the

standard deviation of the average reflectance obtained with in situ measurements. Both the Hyperion imagery and in situ

reflectance data were acquired on 13 July 2002. Since the Railroad Valley site represents only one pixel of MODIS, a

Proc. of SPIE Vol. 7452 745210-4

10

0

500 1000 1500 2000 2500Waveleriqth (nm)

10

0

500 1000 1500 2000 2500Waveleriqth (nm)

standard deviation of MODIS data is not provided. The variability of the Libyan region is shown in Figure 6 as percent

standard deviation of the radiance of the site where the solid line represents Hyperion on 4 June 2008 and the points

represent Terra MODIS on 8 June 2008. The standard deviations shown here are representative of most clear days at

this site. Except for atmospheric absorption bands, the standard deviations of the uniformity of the site and instrument

uniformity are found to be below 3% for both RRV and the Libyan Desert site, making them both good candidates for

this study.

Figure 5. Indication of the RRV site variability.

Percent standard deviations of the average

radiance and reflectance measured by Hyperion

(solid line) and in situ instruments (dashed line),

respectively. Both data are from 13 July 2002.

Figure 6. Indication of the Libyan site variability.

Percent standard deviations of the average

radiance measured by Hyperion (solid line) and

Terra-MODIS (small markers). The dates of the

data are 4 June 2008 and 8 June 2008 for

Hyperion and MODIS, respectively.

4. METHOD

This work finds cross-calibration results for two scenarios that can essentially be labeled as coincident and non-

coincident, referring to the image acquisition time and geometry. The coincident case, as well as results from the

reflectance based approach, are held as a baseline for comparing the non-coincident case. Application of the non-

coincident case is far more general since not nearly as many sensors fall under the coincident case.

The method used for both cases is nearly identical, the only difference being that the non-coincident case requires a

search of image pairs from the two sensors that have matching geometries. Both coincident and non-coincident cases

apply to the comparison between Hyperion and MODIS since they shared the same orbit with a 40-minute difference

until mid-2005 at which time Hyperion deviated and only non-coincident comparisons are possible.

Imagery from both sensors is freely downloaded from the internet and radiance values from the Railroad Valley and

Libyan tests sites are found. Hyperion level 1T data are obtained from the United States Geological Society (USGS) and

Terra and Aqua MODIS calibrated radiance products MOD021KM and MYD021KM, respectively, are obtained from

National Aeronautical and Space Administration (NASA). Processing routines automate the procedure of defining the

pixels that make up the test sites, finding averages and standard deviations of the radiance values, and recording any

ancillary data such as view and solar geometry. The routine also applies a factor to correct the difference in solar zenith

angle.

For the non-coincident case, additional routines are developed that find Hyperion-MODIS image pairs that share similar

view and solar geometry. Inputs to this routine are the library of images from both sensors, a maximum angular

difference of the geometries between the two sensors, and the maximum temporal difference between the acquisition

dates of the image pair. Restricting the geometry difference between an image pair reduces uncertainties induced by the

lack of knowledge about surface bidirectional functions and atmospheric effects. Restricting the temporal difference

between the image pair may reduce the effect of differing atmospheric conditions. However, the more restriction placed

on these parameters results in fewer image pairs satisfying the geometry requirements. The results section will briefly

explore the difference in results when the image pair matching restrictions are varied.

A critical aspect of radiometric cross-calibration is accounting for spectral band differences of the sensors under study.

For example, although TM and ETM+ have very similar spectral filters, it has been shown that their cross-calibration is

Proc. of SPIE Vol. 7452 745210-5

affected by their spectral band differences at the 2% level20

. This work attempts to alleviate spectral effects by

emulating MODIS bands with Hyperion bands. This allows for direct comparison of the two sensors. This is

accomplished by band averaging the high spectral resolution data of Hyperion with the known relative spectral response

functions of Aqua and Terra MODIS.

5. RESULTS

The results presented here consist of percent differences in radiance values of Hyperion and a reference with a functional

form of

100LReference LHyperion

LReference

where LReference and LHyperion are the radiance values from the reference and Hyperion, respectively. The references

employed in this work are at-sensor radiance values provided by the reflectance-based approach of vicarious calibration,

solar-zenith corrected Terra-MODIS, and solar-zenith corrected Aqua-MODIS. When using a MODIS sensor for a

reference, the Hyperion radiance is band averaged according to that sensor’s relative spectral response. The MODIS

bands used in this work are approximately centered at (units of nm): 442 (band 9), 466 (band 3), 553 (band 4), 645 (band

1), 856 (band 2), 904 (band 17), 1243 (band 5), 1632 (band 6), and 2119 (band 7).

The results compare cross-calibration scenarios that differ in image acquisition coincidence, test site used for the

calibration, and reference sensor. Each plot is separated into visible-near-infrared (VNIR) and shortwave-infrared

(SWIR) sections. Some of the scenarios will appear in more than one plot for comparison purposes, and in this case the

plotting symbol for the scenario will remain consistent. All of the results are averages of multiple data points and

therefore have some variance. This is displayed by the one standard deviation error bars in every plot.

The nearly ideal cross-calibration case of similar acquisition times with the same viewing geometry was possible for

Hyperion and Terra MODIS image pairs before EO-1’s orbit change in mid-2005. RSG performed the reflectance-based

approach of vicarious calibration at Railroad Valley for Hyperion during this time. Results from both of these

calibration methods are shown in Figure 7. RSG’s ground-truth method provides hyperspectral radiance which is band

averaged to the narrow spectral bands of Hyperion for comparison. Since there are nearly 200 points, the data is plotted

as a line (solid black). The sharp fluctuations in these results suggest that there is a spectral misregistration between the

in situ measurements and Hyperion. The circles in Figure 7 represent differences between nearly coincident Terra-

MODIS radiance and band averaged Hyperion radiance. The error bars on the symbols throughout this paper represent

plus and minus one standard deviation of the averaged data. The solar models assumed by Hyperion and MODIS differ

in the SWIR spectral region. These differences show up as calibration differences in the results of this work. To show

consistency with the reflectance-approach, the blue triangles in Figure 7 show the original cross-calibration results (red

circles) with a scale applied that takes out the effect of the different solar models.

Proc. of SPIE Vol. 7452 745210-6

20 +

15

10

Band aved cross-calRSG refi-based vicarious

Wavelength (nm)

400 500 600 700 800 900

20

15

10

-101200

Band aved cross-caly Cross-cal w/ equiv solar scale

RSG refi-based vicarious

1400 1600 1800

Wavelength (nm)

2000 2200

20 + I A Libya site - coincident dataRRV site - coincident data

15

10

5

0

-5

-10 + I I I I I

400 500 600 700 800 900

Wavelength (nm) Wavelength (nm)

20

15

10

5

0

-5

-10 I

A Libya site - coincident dataRRV site - coincident data

I I I -I--

1200 1400 1600 1800 2000 2200

Figure 7. The solid line shows the average difference between RSG’s predicted radiance using the Railroad Valley test site

and the radiance reported by Hyperion. The dashed lines show the plus and minus one standard deviation for the five

dates of calibration. The solid circles show the difference in radiances between nearly coincident Terra-MODIS and

Hyperion for 17 dates using Railroad Valley as test site. The solid triangles in the SWIR spectral region show the same

results as the circles but on an equivalent solar scale between Hyperion and MODIS.

Results that employed coincident image pairs using different test sites is shown in Figure 8. As seen in the error bars of

the two scenarios, the calibration dataset using the Libyan test site has much less variability than the one employing

Railroad Valley. A possible reason that contributes to the lower variability in the Libyan results is that it is a much

larger site, which allows for more area to be sampled and reduces effects of misregistration of the test site in the cross-

calibration image pairs. Other than the two blue bands, the two different test sites provide very consistent results.

Figure 8. Percent difference between nearly coincident Hyperion and Terra-MODIS where the triangle are results using the

Libyan test site and circles are results using the Railroad Valley test site.

The ability of cross-calibrating sensors that do not share similar orbits is valuable since this is the case for most sensors,

but confidence in the results is reduced due to possible temporal variations of the atmosphere and surface conditions.

Use of a stable test site to reduce this variability is attempted and the following results are found. Figure 9 shows

calibration differences between Hyperion and Terra-MODIS using coincident image pairs acquired before Hyperion’s

orbit change and non-coincident image pairs from 2008. Pairing of the non-coincident images from Hyperion and Terra-

MODIS in these results was restricted to a maximum of three degrees for differing view and solar geometries, and a

maximum of 18 days between image acquisitions. The dates of the Hyperion and Terra-MODIS image pairs used for the

non-coincident results are listed in Table 1 along with the associated geometries. Results of the coincident and non-

Proc. of SPIE Vol. 7452 745210-7

-5

-10--I-- I I I I I -

ij

4

400 500 600 700 800 900

Wavelength (nm)

20 + A Libya - coincidentLibya - not coincident (2008)

15

10

20

15

10

5

-10

A Libya - coincidentLibya - not coincident (2008)

-I I I

1200 1400 1600 1800 2000 2200

Wavelength (nm)

400 500 600 700 800 900

Wavelength (nm)

20 + I I A Aqua-based results - 2008N Terra-based results - 2008

15

10

5

-H0

-5

-10

Wavelength (nm)

20

15

10

5

0

-5

-10 I

A

NAqua-based results - 2008Terra-based results - 2008

I I I -I--

1200 1400 1600 1800 2000 2200

coincident data sets are in agreement within 2% in most bands. This provides confidence that any two sensors can be

radiometrically compared if they have matching geometries at any point in their orbit over a stable test site such as

Libya.

Figure 9. Percent difference between Hyperion and Terra-MODIS comparing coincident and non-coincident datasets from

the Libyan test site. The triangles represent results from coincident image pairs (pre-2005) and hourglasses represent

image pairs from different dates.

Further consistency in the cross-calibration method is shown with results obtained using independent reference sensors.

Except for the 442 nm band, differences in results between using independent reference sensors are below 1.5% as seen

in Figure 10. Dates of the image pairs and the geometries of the imagery for both Terra- and Aqua-MODIS are listed in

Table 1.

Figure 10. Percent difference between Hyperion and two different MODIS sensors for reference. Both datasets use non-

coincident image pairs and the Libyan Desert as a test site.

Proc. of SPIE Vol. 7452 745210-8

20 + I I A Small temporal constraintF Large temporal constraint

15 T -

10

5

0

-5

-10 + I I I I I

400 500 600 700 800 900

Wavelength (nm) Wavelength (nm)

20

15

10

5

0

-5

-10 I

A Small temporal constraintF Large temporal constraint

I I I -I--

1200 1400 1600 1800 2000 2200

Table 1. List of Hyperion-MODIS image pairs over the Libyan Desert with geometries restricted to a three degree

difference in geometry and 18 day temporal difference.

Date View zenith Solar zenith

Hyperion MODIS Hyperion MODIS Hyperion MODIS

Terra-MODIS 27-Apr-2008 17-Apr-2008 14.25 11.64 26.10 23.98

07-Sep-2008 10-Sep-2008 15.66 13.95 31.64 30.14

17-Sep-2008 03-Oct-2008 1.64 0.84 35.40 36.35

10-Oct-2008 19-Oct-2008 3.57 1.02 41.82 41.61

23-Oct-2008 26-Oct-2008 11.71 11.90 44.77 43.44

Aqua-MODIS 19-Apr-2008 02-Apr-2008 10.59 13.02 30.43 28.72

27-Apr-2008 16-Apr-2008 14.25 12.82 26.1 27

27-Apr-2008 18-Apr-2008 14.25 13.39 26.1 24.55

27-Apr-2008 02-May-2008 14.25 12.74 26.1 23.98

07-May-2008 25-Apr-2008 2.7 0.79 25.89 24.08

07-Sep-2008 09-Sep-2008 15.66 12.93 31.64 29.9

17-Sep-2008 16-Sep-2008 1.64 0.39 35.4 33.33

23-Oct-2008 11-Oct-2008 11.71 13.2 44.77 41.85

23-Oct-2008 27-Oct-2008 11.71 13.12 44.77 47.28

Previous results have compared Hyperion and MODIS non-coincident imagery that has pairing restrictions of a

relatively short time difference of 18 days. This allowed nine matches between Hyperion and Aqua-MODIS within the

imagery of the Libyan test site during the year 2008. Expanding this restriction to 60 days returns 23 image pairs.

Depicted in Figure 11, the differences in results for the two temporal restrictions are 1% or lower for all bands.

Figure 11. Percent difference between Hyperion and Aqua-MODIS using one dataset that allows a maximum of 18 days

between image pairs and one that allows a maximum of 60 days between image pairs.

Proc. of SPIE Vol. 7452 745210-9

6. CONCLUSION

Cross-calibration between a hyperspectral imager and a multispectral sensor is presented. The high spectral resolution

data of Hyperion is band averaged with the relative spectral response of MODIS to allow for direct comparison. A

survey of imagery is performed to find image pairs of the two sensors that have similar geometries so that the need of

characterizing the effect the of atmosphere and directional reflectance of the surface is reduced.

Several parameters of cross-calibration are explored and their results are compared. The most significant comparison is

that of the coincidence of the sensors’ image acquisitions. This comparison is convenient with Hyperion and Terra-

MODIS since they shared a similar orbit for several years early in their missions but have had different orbits since

2005. Results of the coincident and non-coincident data sets are in agreement within 2% in most bands providing

confidence that any two sensors can be radiometrically compared if they have matching geometries over a stable test

site, such as Libya, at any point in their orbit. The temporal period between image pairs used for calibration

comparisons is shown to have minimal affect on the 1% level for small and large temporal gaps in the image pair. Just

as loosening the temporal constraint, the ability to use more than one reference sensor also increases the number of

image pairs available for comparison. The difference in results attained from using Terra- and Aqua-MODIS as

reference sensors is less than 1.5% for all but one spectral region studied here.

Future work with cross-calibration procedures discussed here will extend the spectral range from simple band averaged

comparisons between the hyperspectral and multispectral systems to a full hyperspectral characterization using spectral

models based on in situ data when available, such as for Railroad Valley, and hybrid models from well-calibrated

multispectral data fused with spectral libraries.

REFERENCES

[1] Slater, P.N., S.F. Biggar, R.G. Holm, R.D., Jackson, Y. Mao, M.S. Moran, J.M. Palmer, B. Yuan, “Reflectance- and

radiance-based methods for the in-flight absolute calibration of multispectral sensors,” Remote Sensing of

Environment, 22, 11-37 (1987). [2]

Thome, K.J., J.S. Czapla-Myers, and S.F. Biggar. "Ground-monitor radiometer system for vicarious calibration,"

Proceedings of SPIE, 5546, 223-232 (2004). [3]

Biggar, S.F., K.J. Thome, and W.T. Wisniewski, “In-flight radiometric calibration of the Advanced Land Imager

and Hyperion sensors on the EO-1 platform and comparisons with other earth observing sensors,” Proceedings of

SPIE, 4814, 289 (2002). [4]

Teillet, P.M., J.L Barker, B.L Markham, R.R. Irish, G. Fedosejevs, J.C. Storey, "Radiometric cross-calibration of

the Landsat-7 ETM+ and Landsat-5 TM sensors based on tandem data sets,” Remote Sensing of Environment, 78,

39-54 (2001). [5]

Teillet, P.M., B.L. Markham, and R.R. Irish, “Landsat cross-calibration based on near simultaneous imaging of

common ground targets,” Remote Sensing of Environment, 102, 264-270 (2006). [6]

Thome, K.J., E.E. Whittington, and N. Smith, “Radiometric calibration of MODIS with reference to Landsat-7

ETM+,” Proceedings of SPIE, 4483, 203-210 (2003). [7]

Xiong, X., K. Chiang, J. Sun, N. Che, and W. Barnes, “Aqua MODIS First Year On-orbit Calibration and

Performance,” Proceedings of SPIE, 5234, 391-399 (2004). [8]

Xiong, X. and W. Barnes, “An Overview of MODIS Radiometric Calibration and Characterization,” Advances in

Atmospheric Sciences, 23, 69-79, (2006). [9]

Pearlman, J.S., P.S. Barry, C.C. Segal, J. Shepanski, D. Beiso, and S.L. Carman, “Hyperion, a Space-Based Imaging

Spectometer,” IEEE Trans. On Geoscience and Remote Sensing, 41, 1160-1173 (2003). [10]

Scott, K., K. Thome, and M. Brownlee, “Evaluation of the Railroad Valley playa for use in vicarious calibration,”

Proceedings of SPIE, 2818, 159-166 (1996). [11]

Thome, K., N. Nandy, and K. Scott, “Vicarious calibration of MODIS using Railroad Valley Playa,” IEEE Trans.

On Geoscience and Remote Sensing, 3, 1209-1211 (2001). [12]

Cosnefroy, H., M. Leroy and X. Briottet, “Selection and characterization of Saharan and Arabian desert sites for the

calibration of optical satellite sensors,” Remote Sensing of Environment, 58, 101–114 (1996). [13]

Markham, B.L., J.A., Barsi, D.L. Helder, K.J. Thome, and J.L. Barker, “Evaluation of the Landsat-5 TM

Radiometric Calibration History using Desert Test Sites,” Proceedings of SPIE, 6361, 63610V (2006).

Proc. of SPIE Vol. 7452 745210-10

[14] Heidinger, A.K., J.T. Sullivan, and C.R.N. Rao, “Calibration of visible and near-infrared channels of the NOAA-12

AVHRR using time series of observations over deserts,” International Journal of Remote Sensing, 24, 3635-3649

(2003). [15]

Cabot, F., O. Hagolle, H. Cosnefroy, and X. Briottet, "Inter-calibration using desertic sites as a reference target,"

Geoscience and Remote Sensing Symposium Proceedings, 5, 2713-2715 (1998). [16]

Cabot, F., O. Hagolle, and P. Henry, "Relative and multitemporal calibration of AVHRR, SeaWiFS, and

VEGETATION using POLDER characterization of desert sites," Geoscience and Remote Sensing Symposium

Proceedings, 5, 2188-2190 (2000). [17]

Rao, C.R.N., C. Cao, N. Zhang, “Inter-calibration of the Moderate-Resolution Imaging Spectroradiometer and the

AlongTrack Scanning Radiometer-2,” International Journal of Remote Sensing, 24, 1913-1924 (2003). [18]

Cosnefroy, H., P. Soule, X. Briottet, O. Hagolle, and F. Cabot, “POLDER multiangular calibration using desert

sites: method and performances,” Proceedings of SPIE, 3221, 141-148 (1997). [19]

Martonchik, J.V., D.J. Diner, R. Kahn, B. Gaitley, and B.N. Holben, “Comparison of MISR and AERONET aerosol

optical depths over desert sites,” Geophysical Research Letters, 31, L16102 (2004). [20]

Teillet, P.M., G. Fedosejevs, K.J. Thome, and J.L. Barker, “Impacts of spectral band difference effects on

radiometric cross-calibration between satellite sensors in the solar-reflective spectral domain,” Remote Sensing of

Environment, 110 (3), 393-409 (2007).

Proc. of SPIE Vol. 7452 745210-11