IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 2, NO.

3, JULY 2005 329

Spectroradiometric Requirements for the Reflective

Module of the Airborne Spectrometer ARES
Andreas Müller, Rudolf Richter, Martin Habermeyer, Stefan Dech, Karl Segl, and Hermann Kaufmann

Abstract—The Airborne Reflective/Emissive Spectrometer TABLE I

is specified as a whisk-broom imaging spectrometer for re- BASIC PARAMETERS FOR THE ARES REFLECTIVE MODULE
mote sensing of land surfaces covering the wavelength regions
0.47–2.45 m and 8–12 m with 160 spectral bands. The instru-
ment is being built by Integrated Spectronics, financed by the
German Aerospace Agency (DLR) and the GeoResearch Centre
Potsdam (GFZ) and will be available to the scientific community
from end 2005 on. The spectroradiometric design is based on scien-
tific requirements derived from three main application scenarios
comprising vegetation, soil, and mineral sciences. Two of these are
described in this letter. Measured or modeled reflectance spectra
are input to a simulation model that calculates at-sensor radiance
spectra, resamples them with the channel-specific response func-
tions, adds different amounts of noise in the radiance domain,
and performs a retrieval to get the corresponding noisy surface
reflectance spectra. The retrieval results as a function of the sensor
noise level are compared with the accuracy requirements imposed (0.45–2.45 m) is described. A second paper will discuss the
by the different application fields taking into account the technical requirements for the channels in the thermal region (8–12 m).
boundary conditions. The final specifications account for the most
demanding requirements of the three application fields: a spectral
sampling distance of 13–14 nm in the 470–1800 nm region, and II. MAIN APPLICATION SCENARIOS
12 nm in the 2000–2450-nm region. The required noise-equivalent
radiances are 5, 3, and 2 nW cm 2 sr 1 nm 1 for the spectral ARES will be used mainly for environmental applications in
regions 470–1000, 1000–1800, and 2000–2450 nm, respectively. terrestrial ecosystems. The thematic focus is on soil sciences,
Index Terms—Airborne Reflective/Emissive Spectrometer
geology, agriculture and forestry. Limnologic applications
(ARES), environmental remote sensing, imaging spectrometer, should be possible but will not play a key role in the thematic
surface reflectance. applications. For all above-mentioned key application scenarios
the spectral response of soils, rocks, and vegetation as well as
their mixtures contain the valuable information to be extracted
I. INTRODUCTION and quantified. Information extraction from imaging spectrom-

I MAGING spectrometers cover a spectral region with a large

number of narrow contiguous bands. Thus, they are able
to retrieve the spectral reflectance signature of the Earth’s sur-
eter data can be subdivided into three major groups:
• classification of surface materials according to their
spectral response by determining the best fit between var-
face allowing tasks such as mineral identification and abundance ious endmembers and the target spectrum using spectral
mapping, monitoring of vegetation properties, and assessment features;
of water constituents [1], [2]. The Airborne Reflective/Emis- • determination of relative abundances of surface materials
sive Spectrometer (ARES), a new airborne imaging spectrom- using partial or full unmixing approaches;
eter, being built by an Australian consortium [3], will consist • quantitative analysis of spectra using empirical/statistical
of five spectrometer units covering the reflective and thermal methods or model-based approaches.
part of the spectrum. Table I contains the basic sensor parame- In order to obtain the performance of a new imaging spectrom-
ters. The instrument is attached to a standard aerial platform. It eter, the above application scenarios have to be simulated and
can be operated onboard of aircraft with unpressurized cabins. analyzed based on different spectra, observation conditions,
In this letter, the process of defining the spectroradiometric re- and spectroradiometric sensor properties. Based on experience
quirements for the ARES channels in the solar reflective region gained in previous and ongoing research projects, two examples
on vegetation and mineral mapping are presented.
Once the application-dependent sets of input reflectance
Manuscript received February 21, 2005; revised March 31, 2005. spectra have been selected, the spectroradiometric simulation
A. Müller, R. Richter, M. Habermeyer, and S. Dech are with the Remote
Sensing Data Center, German Aerospace Center (DLR), D-82234 Wessling, is carried out and the results are assessed with the following
Germany. three criteria.
K. Segl and H. Kaufmann are with the Remote Sensing Section, Telegraphen-
berg, Geo Forschungs Zentrum Potsdam (GFZ), D-14473 Potsdam, Germany. • Dark target detection: The specified minimum signal-to-
Digital Object Identifier 10.1109/LGRS.2005.848513 noise ratio (SNR) for a low reflectance target
1545-598X/$20.00 © 2005 IEEE
330 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 2, NO. 3, JULY 2005

Fig. 2. (Top) SNR for surface reflectances of  = 0:30 and (bottom curve)
 = 0:05.

noise radiance [6]. Next, the surface reflectance spectrum is re-

trieved using the atmospheric parameters of the forward sim-
Fig. 1. Block diagram of simulation tool. ulation. Then the input reflectance spectrum resampled to the
ARES filter functions can be compared with the retrieved noisy
is specified as 50, 30, and 20 for the atmospheric win- spectrum and the application-dependent differentiation criterion
dows in the spectral regions 470–1000, 1000–1800, and is applied. The simulation process is iterated for various noise
2000–2450 nm, respectively, at a solar zenith angle of 30 . levels and spectral resolutions. This approach is similar to the
• Vegetation parameter retrieval: Determination of leaf one presented in [7] but includes a radiative transfer calculation
water content to an accuracy of 0.005 cm and chlorophyll through the Earth’s atmosphere.
AB to an accuracy of 5 g cm is targeted with ARES
based on the inversion of leaf/canopy radiative transfer III. RESULTS AND DISCUSSION
models. The corresponding noise-equivalent radiance A series of simulations has been carried out varying the spec-
(NER) values were derived assuming that a threshold of tral resolution and sampling distance (SD) of the instrument
20 above the noise level is needed for a stable retrieval of for different application scenarios. The channel bandwidth [full
these parameters. width at half maximum (FWHM)] is always specified as 1.5
• Spectral feature identification: Mineral surface spectra are times the sampling distance, providing enough channel overlap
input to a sensor simulation with different noise levels with the Gaussian response curves to avoid aliasing effects,
and a subsequent reflectance retrieval. The noisy retrieved compare [7]. In this short contribution, only a few selected
spectra are passed to the spectral feature fit (SFF) algo- examples of the large number of simulation cases can be dis-
rithm [4] to determine the separability of targets within a cussed. For the atmospheric radiative transfer simulation with
certain set of materials. Finally, the SFF scores are con- MODTRAN a midlatitude summer atmosphere was used with
verted into a percent classification accuracy. The require- a rural aerosol at 23-km visibility, and a water vapor column
ment is an 80% accuracy. of 2.0 cm (sea-level-to-space). Solar zenith angles were varied
between 20 and 60 , and the flight altitude is 4 km above sea
level.
A. Simulation of Sensor Signal, Surface Reflectance Retrieval
A. Dark Target Detection
The sensor simulation tool employs the application-depen-
dent surface reflectance input spectra. The first step is a radia- The simulation approach established above was iteratively
tive transfer calculation with the MODTRAN code [5] yielding applied to artifical spectra of constant low and
a high spectral resolution at-sensor radiance ; see Fig. 1. medium surface reflectance. Fig. 2 shows SNR re-
The next step is the resampling of the radiance spectrum with the sults for a 4-km flight altitude and a 30 solar zenith angle for the
simulated spectral response functions assumed to be Gaussian. NER levels of 5, 3, 2 nW cm sr nm for the spectral re-
The third step is the generation of noisy ARES radiance spectra gions 470–1000, 1000–1800, and 2000–2450 nm, respectively
for different sensor noise levels typically achievable for whisk- (configuration 1/2/3/4b of Table II). In the atmospheric window
broom imaging spectrometers to study the influence of sensor regions, the minimum SNR requirements (SNR )
noise on the retrieval of surface reflectance spectra. Other noise for a low reflectance surface are achieved. The requirements
sources, e.g., stray light or platform jitter, and aspects of the cannot be met in atmospheric absorption regions because of the
geometric resolution that also influence the SNR requirements inherent low signal.
are not considered. The simulations are performed for a spa-
tial resolution of 8 m corresponding to a flight altitude of 4 km B. Vegetation Parameter Retrieval
above ground level. The appropriate amount of signal-depen- As an example for vegetation studies the retrieval of two
dent photon noise is added to the sensor NER to obtain the total biochemical variables—plant water content and chlorophyll
MÜLLER et al.: SPECTRORADIOMETRIC REQUIREMENTS FOR THE REFLECTIVE MODULE OF ARES 331

TABLE II TABLE III

SPECTRAL CONFIGURATIONS OF THE REFLECTIVE SPECTROMETER NER REQUIREMENTS DERIVED FROM CHLOROPHYLL/PLANT WATER
ARES (SD=SAMPLING DISTANCE) +
RETRIEVALS. 3: CHLOROPHYLL, : PLANT LIQUID WATER

content—from green vegetation with a closed canopy (alfalfa)

has been chosen. The vegetation spectra used as input in the
modeling were simulated with the models PROSPECT [8] and
SAILH. SAILH is based on the SAIL model [9] including a
modeling of the hot spot effect [10]. Modeled spectra have
been chosen as the relationship between the variations of the
foliage chemistry and the spectral response is predefined.
The input parameters are considered free of radiometric and
spectral noise components. The basic vegetation parameters
to constrain the models were taken from the DAISEX ground
measurements [11]. In this letter, the observational parameters Fig. 3. Selected mineral spectra used for simulation. (Plus sign) Calcite.
in SAILH are restricted to 60 sun zenith angle, 0 relative (Diamond) Dickite. (Ttriangle) Kaolinite. (Square) Pyrophyllite.
azimuth, and 90 view zenith angle. Leaf area index and leaf
angle distribution values were kept constant at 5.5 m m and
50 , respectively. The diffuse sky radiance was derived from spectrometers 1 to 3 are complient with the SNR figures esti-
the same MODTRAN [5] simulations as used for the forward mated for the ARES instrument [3].
modeling. The soil reflectance background was given by field
spectral measurements. C. Spectral Feature Identification
Input parameters for PROSPECT such as the struc- The identification and mapping of individual rock and soil
tural parameter , the dry matter content units based on the spectral absoption features of their specific
CM g/cm , and the brown pigment content indicator minerals is a key application in imaging spectroscopy.
BPC were kept constant whereas the chlorophyll and Spectral absorption features of minerals originate from elec-
plant water contents have been varied in realistic margins to tronic and vibrational transitions [12]. The input spectra
study the influence on the vegetation spectrum and thus the selected for the geological applications simulation comprise
detectability of these reflectance differences at different sensor different carbonates (calcite, dolomite, siderite), phyllosili-
noise levels. The chlorophyll AB (Cab) was varied in steps of cates (kaolinite, montmorillonite, dickite, pyrophyllite, illit,
5 g cm at a level of 40 g cm and leaf water content chlorite, vermiculite) and iron oxides/hydroxides (haematite,
(Cw) in steps of 0.005 cm at a reference level of 0.04 cm. The goethite) taken from the U.S. Geolocial Survey spectral library
reflectance difference is defined as C1 C2 where [13]. A set of 826 spectra is employed in this study. Fig. 3
C1 and C2 represent two successive steps of Cab or Cw con- shows the reflectance spectra of four selected minerals in the
centration. For different noise equivalent radiance levels (NER) 2000–2500-nm region [shortwave infrared (SWIR)], taken
the corresponding noise-equivalent reflectance is defined as from the set of 826 spectra. The original laboratory reflectance
NE C1 C1 NER where C1 indicates the values were reduced by a factor of 2 to approximate a brightness
noise-free spectrum, and C1 NER represents the retrieved reduction caused by environmental effects, e.g., aging, coating.
spectrum for a sensor noise level of NER. The threshold with In addition, the absorption depth of spectrally pure materials
respect to the NE level is defined as NE . was varied between 10% and 100% to include a realistic range
Table III summarizes the NER requirements corresponding to of feature variation.
a threshold for the reflective spectrometers of ARES. In The spectral feature fit (SFF) algorithm is applied to the re-
the fourth spectrometer, the high NER requirement of NER trieved ARES spectra; compare Fig. 1. Results are summarized
nW cm sr nm for the liquid water retrieval is diffi- in the confusion matrix [14], where the diagonal matrix ele-
cult to achieve, but due to the broad-band character of the liquid ment of each line contains the number of correct classifications,
water absorption feature (1000—2500 nm) the ARES spectrom- and the sum of the off-diagonal scores represent the misclas-
eters 2 and 3 can be used for this purpose. Spectrometer 1 is sifications. Fig. 4 presents the corresponding classification re-
employed for the chlorophyll AB retrieval leading to the NER sults for different noise levels as a function of the spectral sam-
requirement of 5 nW cm sr nm . The NER values for pling distance. The top curve (dashed) presents the case free
332 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 2, NO. 3, JULY 2005

Key application scenarios were defined and the corresponding

surface reflectance spectra were employed as input to the
simulation to calculate the spectroradiometric requirements.
Results of the spectroradiometric requirements were presented
for a carefully selected number of cases. The required noise
equivalent radiances are 5, 3, and 2 nW cm sr nm
for the spectral regions 470–1000, 1000–1800, and 2000–
2450 nm, respectively.

REFERENCES
[1] A. F. H. Goetz, G. Vane, J. E. Solomon, and B. N. Rock, “Imaging spec-
trometry for earth remote sensing,” Science, vol. 228, pp. 1147–1153,
Fig. 4. Classification accuracy as a function of spectral sampling distance 1985.
(SD). SNR = 50; 100; 200, bottom to top solid curve, respectively. The [2] M. K. Hamilton, C. O. Davis, W. J. Rhea, S. H. Pilorz, and K. L. Carder,
bandwidth is 1.5 times SD. The top dashed curve represents the noise-free case. “Estimating chlorophyll content and bathymetry of Lake Tahoe using
AVIRIS,” Remote Sens. Environ., vol. 44, pp. 217–230, 1993.
[3] I. J. Wilson and T. D. Cocks, “Development of the Airborne Reflec-
of sensor noise, the solid curves represent the classification ac- tive Emissive Spectrometer (ARES)—A progress report,” in Proc.
curacy for three sensor noise levels using SNRs of SNR 3rd EARSeL Workshop on Imaging Spectroscopy, 13–16 May 2003,
Herrsching, Germany, 2003, pp. 50–55.
(bottom to top curves, respectively). For the speci- [4] R. N. Clark, A. J. Gallagher, and A. Swayze, “Material absorption band
fied large set of about 800 minerals, the classification accuracy depth mapping of imaging spectrometer data using a complete band
increases strongly with SNR. shape least-squares fit with library reference spectra,” in Proc. 2nd
A sampling distance (SD) of 10–15 nm yields best results, Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop.
Pasadena, CA, 1990, JPL-Pub. 90-54, pp. 176–186.
with a local classification maximum for SD nm at low [5] A. Berk, G. P. Anderson, P. K. Acharya, M. L. Hoke, J. H. Chetwynd,
SNR values SNR . However, this figure demonstrates L. S. Bernstein, E. P. Shettle, M. W. Matthew, and S. M. Adler-Golden,
that a high SNR value 100 is more important than a small sam- MODTRAN4 Version 3 Revision 1 User’s Manual. Hanscom Air Force
Base, MA: Air Force Res. Lab., 2003.
pling distance and bandwidth. Therefore, if a sensor design with [6] C. Schuler and L. Woody, “Digital electro-optical imaging sensors,” Int.
narrow spectral bandwidths fails to achieve a sufficient SNR, it J. Imaging Syst. Technol., vol. 4, pp. 170–200, 1992.
[7] G. A. Swayze, R. N. Clark, A. F. H. Goetz, T. G. Chrien, and N. S.
is better to sacrifice spectral resolution and increase the band-
Gorelick, “Effects of spectrometer bandpass, sampling, and signal-to-
width to obtain a larger SNR and better classification results. noise ratio on spectral identification using the Tetracorder algorithm,” J.
Most minerals show typical reflectance levels of 30% to 60% Geophys. Res., vol. 108, pp. 5105–5134, 2003.
[8] S. Jacquemoud and F. Baret, “PROSPECT: A model of leaf optical prop-
in the SWIR region. For an ARES design with SD nm
erties spectra,” Remote Sens. Environ., vol. 34, pp. 75–91, 1990.
and FWHM nm and the 30% to 60% surface reflectance [9] W. Verhoef, “Light scattering by leaf layers with applications to canopy
range we will obtain SNR values of 100–200 yielding classifi- reflectance: The SAIL model,” Remote Sens. Environ., vol. 16, pp.
cation accuracies of 84% to 92%. According to Fig. 2 for a 30% 125–141, 1984.
[10] A. Kuusk, “The hot spot effect on a uniform vegetative cover,” Sov. J.
reflectance target an NER of 2 nW cm sr nm leads to Remote Sens., vol. 3, pp. 645–658, 1985.
SNR values above 100 in the 2100–2350-nm region. Therefore, [11] M. Berger, J. Moreno, A. Müller, M. Schaepman, P. Wursteisen, M. Rast,
this NER level is chosen as a design requirement in the SWIR and E. Attema, “The Digital Airborne Imaging Spectrometer Experi-
ment—DAISEX’99,” in Proc. IGARSS, vol. 7, 2000, pp. 3039–3041.
region. [12] G. R. Hunt, “Spectral signatures of particulate minerals in the visible
and near-infrared,” Geophysics, vol. 42, pp. 501–513, 1977.
[13] C. I. Grove, J. J. Hook, and E. D. Paylor, “Laboratory Reflectance
IV. CONCLUSION
Spectra of 160 Minerals, 0.4 to 2.5 Micrometers,” JPL, Pasadena, CA,
A simulation tool for the spectroradiometric design of JPL Pub. 92-2, 1992.
[14] R. G. Congalton, “A review of assessing the accuracy of classifications
multi/hyperspectral earth observing instruments has been of remotely sensed data,” Remote Sens. Environ., vol. 37, pp. 35–46,
developed and applied to the airborne ARES spectrometer. 1991.