SR-933a A Tunable Fabry-Perot Etalon-Based William J ... · PDF file1 A Tunable Fabry-Perot...
Transcript of SR-933a A Tunable Fabry-Perot Etalon-Based William J ... · PDF file1 A Tunable Fabry-Perot...
SR-933a
A Tunable Fabry-Perot Etalon-BasedLong-Wavelength Infrared Imaging Spectroradiometer
William J. Marinelli, Christopher M. Gittins, Alan H. Gelb, and B. David GreenPhysical Sciences Inc.
20 New England Business CenterAndover, MA 01810-1077
Applied Optics 38(16), 2594-2604 (2000)
Copyright © 2000, Optical Society of America. All rights reserved.
Reprinted by permission of the Optical Society of America.
1
A Tunable Fabry-Perot Etalon-Based
Long-Wavelength Infrared Imaging Spectroradiometer
William J. Marinelli, Christopher M. Gittins, Alan H. Gelb, and B. David Green
Physical Sciences Inc.
20 New England Business Center
Andover, MA 01810-1077
ABSTRACT
Imaging spectrometry enables passive, stand-off detection and analysis of the chemical
composition of gas plumes and surfaces over wide geographic areas. This paper describes the
use of a long-wavelength infrared imaging spectroradiometer, comprised of a low-order tunable
Fabry-Pérot étalon coupled to an HgCdTe detector array, to perform multispectral detection of
chemical vapor plumes. The tunable Fabry-Perot etalon used in this work provides coverage of
the 9.5 to 14 µm spectral region with a resolution of 7 to 9 cm-1. The etalon-based imaging
system provides the opportunity to image a scene at only those wavelengths needed for chemical
species identification and quantification and thereby minimize the data volume necessary for
selective species detection. We present initial results using a brassboard imaging system for
stand-off detection and quantification of chemical vapor plumes against near ambient
temperature backgrounds. These data shows detections limits of 22 ppmv × m and 0.6 ppmv × m
for DMMP and SF6 respectively for a gas/background ûT of 6 K. The system noise-equivalent
spectral radiance is approximately 2 µW cm-2 sr-1 µm-1. Model calculations are presented
comparing the measured sensitivity of the sensor to the anticipated signal levels for two chemical
release scenarios.
Keywords: remote sensing, hyperspectral, imaging, chemical agent, Fabry-Perot, interferometer
2
I. INTRODUCTION
Imaging spectrometry enables passive, stand-off detection and analysis of chemical vapor
plumes spread over wide geographic areas. Remote monitoring of chemical vapor releases is a
logical extension of existing visible and near-infrared imaging spectrometer technology;
however, the absorption and emission features of most organic chemical vapors are weak and
poorly structured in the visible and near-infrared. Sensitive, selective detection of chemical
plumes requires spectroscopic measurements in the 8 to 14 µm (long-wavelength infrared,
LWIR) atmospheric transmission window where most organic vapors have their strongest
identifying signatures. Multispectral imaging enables chemical plume visualization by
exploiting the radiant flux difference between the gas plume and the background to reveal
characteristic absorption and emission spectra of target chemicals. Quantitative analysis of
multispectral imagery requires knowledge of both the spectral signatures of the analyte vapors as
well as the effective temperature difference between the vapor plume and the background scene.
The integration of a spectral filter into an IR focal plane array (FPA) based sensor
provides the spectral resolution requisite to identify the spectral signatures of specific chemical
compounds and the radiometric accuracy necessary to obtain quantitatively accurate
measurements of chemical plume column density. Previous efforts to provide imaging capability
for chemical vapors focussed largely on the use of filter wheels in conjunction with forward-
looking IR (FLIR) sensors1-4. In contrast to filter wheel-based imagers, the transmission
wavelength of a tunable etalon is continously variable. Additionally, the number of spectral
resolution elements available to a tunable etalon-based system is defined by the etalon’s finesse,
typically between 25 and 50, rather than by the number of slots available in a filter wheel. In its
3
simplest configuration, the low-order tunable etalon replaces a filter wheel assembly with a
frequency agile system offering considerable improvement in data acquisition capability.
In contrast to imaging FTIR spectrometers such as the Livermore Imaging Fourier
Transform Infrared Spectrometer (LIFTIRS)5,6 , which covers the 8 to 13 µm atmospheric
transmission window at up to 0.5 cm-1 resolution, the adaptive sampling capability of the tunable
etalon-based imager provides reduced data volume and computational simplicity. In contrast to
dispersive imagers such as the Spectrally Enhanced Broadband Array Spectrograph System
(SEBASS)7, a liquid helium-cooled device which covers the 7.8 to 13.5 µm region at 0.05 µm
resolution (5 cm-1 at 10 µm), the imaging spectroradiometer described in this paper provides
common pixel registry for all wavelengths, ambient temperature operation of the tunable filter,
and higher optical throughput.
II. FABRY-PEROT IMAGING SPECTRORADIOMETER CONCEPT
The LWIR multispectral imaging spectroradiometer system described here is a variation
on a mid-wavelength infrared (MWIR) multispectral imaging spectroradiometer based on the
same tunable etalon technology.8 The optical system concept for the MWIR imager is depicted
in Figure 1: a 128 x 128 pixel InSb IR focal plane array (FPA) views the far field through a
tunable, piezoelectric-actuated Fabry-Pérot interferometer placed in an afocal region of the
optical train. The tunable etalon is in operated in low order (mirror spacing comparable to
wavelength of limited transmitted) and functions as an interference filter which selects the
wavelength viewed by the FPA. This configuration of the interferometer affords both wide a
field-of-view (FOV) and broad wavelength coverage at high spectral resolution.
4
Optical Configuration
Because no suitable large format FPA (�64 x 64 pixels) was available at the time the
LWIR system was constructed, the optical layout of the brassboard system is somewhat different
than that depicted in Figure 1. The layout of the LWIR system is illustrated in Figure 2. Its four-
key components are: 1) an eight- element (2 x 4 format) photoconductive HgCdTe detector array
housed in a liquid nitrogen-cooled dewar; 2) the computer controlled tunable etalon; 3) a rotating
wheel mechanical chopper used to modulate the far field radiation; and 4) a computer controlled
galvo-driven two-axis scanning mirror system to provide a wide Field-of-Regard. The IR
emission from the far field is sampled with the scan mirrors, brought to an intermediate focus by
an off-axis parabolic reflector (OAP), modulated by the mechanical chopper (2870 Hz),
recollimated by a matched OAP, and directed through the tunable etalon. The radiation
transmitted by the etalon is directed into the detector array through a 10.0 to 11.5 µm
transmitting cold filter chosen to closely match the etalon’s free spectral range.
Each 0.5 mm square detector array element is biased and amplified using an eight-
channel, custom-built amplifier board. The use of the mechanical chopper in conjunction with an
eight-channel lock-in amplifier (LIA) board improves the system signal-to-noise performance by
a factor of ~20 over direct sampling of the DC amplifier output. The system noise equivalent
spectral radiance (NESR) achieved using the LIA is similar to that anticipated using a large
format LWIR FPA to provide imaging capability. Conversion of raw data into radiometrically
calibrated output is accomplished using wavelength dependent gain and offset corrections for
each detector element. The gain and offset coefficients are established from a two temperature
radiance calibration of the system at 25 oC and 40 oC. The measured NESR of the system with
the LIA output time constant set at 10 ms is 2 µW cm-2 sr-1 µm-1 (NEûT� 0.1 K) in the middle of
5
the spectral tuning range and ~3 µW cm-2 sr-1 µm-1 at the 10.0 and 11.5 µm operating limits of the
system.
A three-dimensional “datacube”, 2D-spatial by 1-D spectral, is produced by positioning
the interferometer to transmit a chosen wavelength, scanning the galvo mirrors through 288
discrete positions to produce a 48 x 48 pixel image of the field, and repeating the process for all
desired wavelengths. Each 48 x 48 pixel spatial element in the datacube comprises a 40 deg by
40 deg FOV (14 mrad/pixel). The tunable etalon is typically scanned from 10.1 µm to 11.4 µm
in increments of 0.1 µm during chemical vapor plume imaging experiments so datacubes consist
of 14 spectral elements. The galvo mirror scan is the rate-limiting step of the data acquisition
and takes ~9 seconds per wavelength to complete. The use of the eight element array in
conjunction with the scanning mirrors is not an ideal arrangement for rapid acquisition of
multispectral imagery. The current brassboard system demonstrates of the utility of the tunable
Fabry-Perot as the basis for a compact, field deployable LWIR multispectral imaging
spectroradiometer incorporating a large format FPA.
Interferometer
The tunable etalon module used in this work is described in detail in a previous
publication.9 A digital capacitance micrometry system is used to measure mirror spacing and
alignment. A closed loop control system is used to position and align the interferometer to the
desired wavelength in times as short as 1.3 ms. The theoretical basis for the development of the
tunable etalon based imaging spectroradiometer, as well as the consequences and advantages of
low-order operation, is evident from the mathematical description of etalon transmission.
Transmitted wavelengths fulfill the resonance condition:
6
25 × cos� m �0(�)�
× � (1)
I(�)I0(�)
T(�)2
[1R(�)]2× 1 �
2F(�)�
2
sin2 2�5cos��
0(�)1
(2)
where 5 is the mirror spacing, � is the angle of incidence of the light with respect to the optical
axis of the etalon, m is the order of interference, and 0(�) is the phase change upon reflection
from the mirror coating.10
The wavelength dependent filter transmission is described by an Airy function:
where T(�) and R(�) are the wavelength dependent mirror coating transmission and reflectivity
and F(�) is the interferometer finesse. The interferometer is equipped with nominal 94%
reflectivity mirrors spanning the range 9.5 to 14.5 µm. Figure 3 provides an expanded view of
the third order transmission band of the etalon near 10.55 µm. Peak transmission is >60% and
includes slight attenuation by the two anti-reflection coated Ge windows which provide optical
access to the etalon in its hermetically sealed housing. A detailed examination of the lineshape
shows a good fit to the expected Airy function with a FWHM of 6.9 cm-1. The etalon’s spectral
resolution (finesse) is governed not only by mirror reflectivity, but also by surface roughness,
collimation of light incident on the filter, and degree of mirror parallelism.10 For practical
operation in the infrared, the substrate surface roughness is the limiting factor in determining
total finesse. The limiting spectral resolution is seldom greater than û�FSR/40, finesse = 40, due
to this limitation. The observed resolution corresponds to a finesse ~25 and is better than 1% of
� over the entire tuning range of the device.
7
û�FSR �m �m � 1 (3)
û�FSR
�max
m � 1 (4)
The cos � dependence in Eqs. (1) and (2), see also Figure 1, means that a detector array
with an extremely wide FOV acquires a non-monochromatic image. Detector array elements
view shorter wavelengths the further they lie off axis. The maximum useful system FOV is
defined by the user’s requirement for wavelength uniformity in the image, e.g., a maximum
allowable off-axis wavelength shift of FWHM/10, ~0.7 cm-1, would restrict the full angle FOV to
~9 degrees; if the criterion were FWHM/2, then a 20 degree full angle FOV would be allowable.
The maximum incidence angle associated with off-axis rays in the brassboard LWIR imager is
~1.9 degrees. The use of the scanning galvo mirrors provides a wide field-of-regard while
preserving a monochromatic image.
The free spectral range (FSR) of the etalon is defined as the wavelength spacing between
adjacent transmission orders:
where m is the operating order of the interferometer. For an ideal Fabry-Perot etalon, this
expression simplifies to:
where �max is the longest transmitted wavelength when operating in order m. A wide bandpass
filter, matched to the etalon FSR, is placed in front of the detector array to select a single trans-
mission order. Reflected phase dispersion in the dielectric mirror coating, 0(�) in Eqs. (1) and
(2), contracts the observed FSR relative to that predicted by Eq. (4). The observed FSR of our
LWIR etalon in its typical operating mode (m = 3) is ~2 µm. The effective FSR of the system is
8
demonstrated in Figure 4 where we show a transmission spectrum of the interferometer recorded
with an open path FTIR spectrometer.
In contrast to an absorbing or scattering filter, e.g., a LWIR AOTF,11 the tunable etalon
has extremely low self-radiance at when operated at ambient temperature. Self-radiance of
components in the optical train not only reduces the effective detector dynamic range but also
increases photon statistical noise, further reducing detection sensitivity. The etalon is reflective
where it is not transmitting and, in the configuration depicted in Figures 1 and 2, detector
emission at wavelengths outside of the filter passband is refocussed on the array. This self-
radiance is insignificant for a liquid nitrogen-cooled detector. The small but finite absorption of
the mirror coating, ~0.5% per mirror surface, is the primary source of system self-radiance.
Mirror absorption causes the etalon to have a finite emissivity. We calculate 0�0.12 at the
transmission wavelength and 0�0.005 outside its passband. Although we have not made precise
measurements of etalon self-radiance, our experience suggests that the system performance is
consistent with our radiometric model for the Fabry-Perot. The LWIR interferometer has an
apparent field temperature of ~200 K when operated at 298 K.
III. CHEMICAL DETECTION EXPERIMENTS
A chemical vapor plume may appear in either emission or absorption depending upon the
temperature and emissivity differences between the plume and the background. We calculate the
plume spectral signature using a simplified three layer radiative transfer model, as shown in
Figure 5. The model has been used previously by Flanigan12 and is an excellent approximation
when the temperature variation within the plume is small in comparison to the temperature
difference between the plume and the background. The total infrared radiance incident upon the
9
ûN(�) tplume N(�,Tbb) � [1tplume] × N(�,Tplume) N(�,Tchopper) (5)
tplume(�) exp[�ki(�)Ci5] (6)
sensor at a given wavelength (W cm-2 sr-1 µm-1) is the sum of the contributions from each layer.
The brassboard LWIR spectroradiometer measures the difference between the scene radiance and
chopper wheel radiance:
where N(�,T) is the Planck radiance, Tbb is the effective temperature of the background, Tplume is
the plume temperature, Tchopper is the mechanical chopper temperature. The quantity tplume is the
optical transmission of the vapor cloud calculated using Beers’ Law:
where Ci is the average concentration of the chemical compound over the plume length 5 and
ki(�) is its wavelength-dependent absorption coefficient. The sum over index i in Eq. (6) is over
all spectrally relevant chemical species. In Eq. (5) we have made the simplifying assumption that
the atmospheric transmission over our relatively short viewing paths is equal to unity. It is
essential to note that if there is no temperature difference between the background and the plume
then no spectral signature of the chemical plume is observed. The dependence of the minimum
detectable chemical column density, Ci5 in Eq. (6), on temperature difference and molecular
absorption coefficient is discussed below.
Work performed under the auspices of the Army ERDEC has compiled spectral data for a
variety of chemical agents.13,14 We chose dimethyl methyphosphonate (DMMP) as one simulant
in our experimental measurements because of its prior use in this field and because the band
10
spectrum of DMMP is quite similar to the other G-agents. We have also tested the system using
SF6 because it possess a strong absorption band in the spectral region of interest and has low
toxicity. The chemical imaging system was characterized using both direct absorption and
passive IR emission measurements. Initial experiments were conducted utilizing the system to
view an extended blackbody source through a 10 cm path absorption cell. Gas mixtures could be
produced and flowed through the cell using a manifold of mass flow controllers coupled to vapor
generators as well as pure target and diluent gases. An example of an absorption spectrum taken
with the imaging system is shown in Figure 6 for a mixture of 7 ppmv × m SF6 and 76 ppmv × m
DMMP. The SF6 and DMMP absorption coefficients were obtained from laboratory spectra
recorded at 0.5 cm-1 resolution with an FTIR spectrometer. The SF6 absorption spectra were
recorded in an atmospheric pressure flow cell (10 ppmv SF6 in dry N2). The calculated SF6
absorption coefficient was scaled by ~10% to match the published value of the integrated
absorption coefficent of the �3-fundamental band15. This scaling correction is consistent with the
calibration uncertainty in the SF6 mass flow system. The DMMP absorption spectra were
recorded in a sealed cell containing the saturation vapor pressure of the liquid.
The individual components of the calculated transmission spectrum are shown in
Figure 6. Transmission spectra are simulated by convolving high resolution spectra with the
observed Fabry-Perot resolution function. The data agrees quite well with the calculations,
indicating that the spectral resolution of the system as well as the absorption coefficients of the
target gases are both well understood. It should be noted that the mirror coating used in the
brassboard system was initially developed to detect chlorinated hydrocarbon gases at short range.
Hence, the coating was designed for longer wavelengths than are optimal for the detection of
chemical agents and their simulants, such as DMMP. As a result of this constraint, we are forced
11
d[ûN]d!
k(�)×[N(�,Tplume)N(�,Tbb)] (7)
/! �/[ûN]
k(�)× 0N0T
×[TplumeTbb](8)
to use a weaker band of DMMP (10.85 µm peak) than is used in most other studies of chemical
agent detection.
Combining Eq. (6) and with the derivative of Eq. (5) with respect to species column
density allows us to calculate detection limits for analyte species from the measured noise
characteristics of the LWIR imaging system. For a single species present in an optically thin
plume, the instrument’s responsivity for that compound is:
where ! is equal to Ci5. The Planck radiance is very nearly a linear function of temperature near
300 K in the LWIR, so the minimum detectable change in column density is well approximated
as inversely proportional to the absorption coefficient of the species and the temperature
difference between the plume and the background:
The quantity /[ûN] is the measured NESR of the system. Using Eq. (8), the calculated detection
limits against a ûT of 6 K is 0.6 ppmv × m for SF6 using the 10.55 µm absorption band and
22 ppmv × m for DMMP using the 10.9 µm absorption band. Use of the stronger DMMP band
at 9.4 µm would improve our detection sensitivity by approximately a factor of three.
Initial imaging experiments involved viewing the absorption cell containing 20 ppmv × m
SF6 held at 298 K against a 308 K reference blackbody. The location of the SF6 cell in the scene
12
was independently determined by the two-band correlation analysis shown in Figure 7. Figure 7
depicts the measured radiance at 10.55 µm, the peak of the SF6 absorption band, plotted versus
the measured radiance at 10.85 µm, which is outside the SF6 absorption band. This approach
captures differences in apparent emissivity of the scene from pixel to pixel independent of
variations in absolute radiance across the scene. The presence of a structured gas absorption
feature, which is quite different from the more slowly varying background spectrum, is observed
as pixels lying outside the primary vector defined by the majority of the scene pixels. Points
falling above the primary vector would indicate SF6 emission, whereas points lying below the
primary vector indicate absorption. Because the cell is colder than the blackbody background,
only absorption is evident. The pixels corresponding to those points falling >31 below the
primary vector locate the absorption cell in the scene. Figure 8 shows the pixels identified by
two-band correlation analysis highlighted in black. We use the three layer radiative transfer
model to calculate the SF6 column density at the highlighted pixels and find good agreement with
the concentration injected into the cell. Figure 9 depicts calculated column density as a function
of position in the image. The calculated peak column density from the optical measurement,
18 ± 3 ppmv × m, is in excellent agreement with the 20 ppmv × m determined from calibrated
mass flow measurements. Spectral data acquired from these measurements were subsequently
used in spectrally matched filter analysis of SF6 plumes released in an outdoor setting.
We have also tested the system sensitivity against controlled, calibrated plume releases of
DMMP. The DMMP plume releases were performed at the Analytical Chemistry Laboratory at
Argonne National Laboratory (ANL). The laboratory at ANL possesses an apparatus to vaporize
and release low concentration plumes of CW agent simulants.16 The simulant release
concentrations ranged from ~30 to ~220 ppmv × m and a NIST traceable blackbody was placed
13
in back of the plume at its release point so that it is viewed against a calibrated background. The
plume generating apparatus was contained in a laboratory hood and the LWIR imager was
positioned outside the hood at ~1.5 m standoff distance during data acquisition. In this
configuration the calibrated background occupies only a few pixels in the scene and the imagery
is of little interest.
The spectra of the DMMP releases is analyzed by a least-squares fit of experimental data
to Eq. (5). The background blackbody temperature and plume release temperature were
continuously monitored and controlled during the measurements and therefore could be fixed
during the fit to the data. (The plume release temperature was 307 K and the calibrated back-
ground temperature was 313 K.) The column density of DMMP and the mechanical chopper
wheel temperature were treated as independent variables. The effect of treating chopper wheel
temperature as an independent variable was to provide a modest baseline adjustment. The
chopper wheel temperature was measured periodically between data collections. Derived values
of the chopper temperature differ from measured values by <0.5 K. The 11 uncertainty in the
best derived temperature was ~0.1 K for all spectra analyzed.
Figure 10 depicts the best fit of Eq. (5) to representative spectroradiometer data. The
scene radiance in Figure 10 is plotted in absolute units rather than differential units by adding the
appropriate radiance corresponding to the derived chopper temperature. The data is taken from a
point in the image corresponding to the center of the reference blackbody. The released DMMP
column density is 141 ppmv × m based on liquid flow rate into plume generating apparatus and
the value determined from the optical data is 163 ± 29 ppmv × m. Figure 11 shows the DMMP
column density calculated from spectroradiometer data plotted versus release concentration
calculated from the liquid flow rate for the full range of release concentrations covered in the test
14
series. If the correlation between the calculated and released column density were perfect, all
points would fall on the dashed line. The error bars shown correspond to ± 11 for a fit of
experimental data between 10.4 and 11.2 µm to Eq. (5). Statistical analysis of the data in
Figure 11 yields a correlation of 1.01 ppmv × m measured per ppmv × m DMMP released (11
uncertainty: ±0.19) and a detection limit of ±22 ppmv × m. The latter figure is in very good
agreement with the detection limit anticipated from the measured system NESR. When the
region of the data spectral comparison is extended to cover the full 10.1 to 11.4 µm spectral
region stored in the data cube, the 11 uncertainties in correlation and detection limits increase,
but neither the slope nor intercept of the data plot changes appreciably. This increase occurs
because the system NESR is greater near the red and blue limits of the operating range than in the
center of the range.
Finally, the system was tested against an outdoor plume release where a known flow rate
of SF6 was diluted in N2 and released from a pipe of known cross section. This arrangement is
depicted in Figure 12. Gas is released from the short section of black pipe slightly below the
center of the scene. The location of the SF6 plume in the scene was identified using both two-
band correlation analysis and a spectrally matched filter (SMF) algorithm17-19 targeted on the SF6
absorption feature at 10.55 µm. The results of the SMF analysis is shown in Figure 13. The
spatial extent of Figure 13 covers approximately the central horizontal two-thirds of Figure 12.
The SMF analysis locates a larger plume than does the two-band correlation analysis and those
pixels are indicted by lighter tones. The SMF analysis appears to yield fewer false alarms, albeit
through the use of data in the entire spectral data cube. The SF6 column density in the plume at
the pixels identified by two-band correlation (>31 below the primary vector as for the absorption
cell image analysis) was calculated using the three-layer radiative transfer model described
15
previously and is depicted in Figure 14. Unfortunately, the outdoor plume images were recorded
on a windy afternoon and the SF6 plume was rapidly dispersed in the atmosphere. Because of the
rapid dispersion and fact that the opening of the release pipe was less than one pixel wide in the
LWIR image, the effective column density of SF6 at the pipe exit was not well-known. The
building wall, which served as the background, was considerably warmer than ambient
temperature, 313 to 315 K as opposed to 298 K, and the plume is observed most clearly there.
The calculated column density using the radiative transfer model serves as a good cross check for
both the two-band correlation and the SMF analyses. False alarms occurring in each analysis
correspond to column densities very near or below the anticipated detection limits for the species.
The radiative transfer model indicates an SF6 column density of 5 ppmv × m at the apparent ûT
of 15 K. The SNR of the data is approximately 5:1 and is consistent with detection limits
determined using Eq. (8) with the SF6 absorption coefficient appropriate for the system spectral
resolution.
IV. CHEMICAL RELEASE DETECTION
Two scenarios were modeled for comparison with the capabilities of the chemical
imaging system described in this paper: 1) a look down view of a chemical weapons release
from a SCUD-B missile, and 2) observation of dilute emissions from a chemical production
facility stack. In the SCUD-B release scenario, DMMP was used to simulate a volatile G-type
nerve agent such as Sarin. In the stack emissions viewing scenario, DMMP was assumed to
simulate tributyl phosphate, a compound used in the PUREX process for plutonium
reprocessing.20 Two different models were employed to translate chemical releases into infrared
signatures: 1) a Gaussian model of plume dispersion in the atmosphere, and 2) a radiative
16
c(x,y,z) q
�u1y1z
expy 2
212y
exp(zh)2
212z
� exp(z�h)2
212z
(9)
transfer model to describe transport of radiation from the background which traverses the
chemical plume and the intervening atmosphere to reach the sensor. The radiative transfer model
was discussed earlier in this paper, although we no longer assume the intervening atmosphere to
be transparent. The plume dispersion model is discussed below prior to conducting a sensitivity
assessment.
The dispersion of a chemical species in the atmosphere from the release site is the major
source of uncertainty in modeling the ability of a sensor to locate the chemical cloud later in
time. Diffusion and gross scale motions of the atmosphere will dilute and transport the species
of interest along the prevailing wind direction. In this effort a classical Gaussian dispersion
model was used to estimate the concentration of the chemical of interest downstream of the
release site. The Gaussian model is applicable to a highly idealized set of atmospheric
conditions, but nonetheless is used extensively by the EPA and others as a basis for more
extensive models.21 We have used the model to predict the downstream concentrations for
steady-state (chemical stack) and single puff (chemical agent attack) transient releases. These
expressions are given for steady-state and transient releases below.
Steady-State
where
q = mass release rate
u = local wind velocity
17
c(x,y,z) S
(2�)3/21x1y1z
exp(xx1u(tt1)2
212x
(yy1)2
212y
x
exp (zz1)2
212z
� exp (z�z1)2
212z
(10)
1y Ryxry (11)
h = release height
x = wind direction
z = distance above ground
y = distance transverse to wind direction and parallel to ground
and
1x, 1y, 1z = dispersion parameters along x, y, z axis.
Transient
where
S = mass released
and
x1, y1, z1 = release points.
The dispersion parameters, 1x, 1y, and 1z, are empirically derived from atmospheric dispersion
measurements under a range of atmospheric conditions. The parameters 1y and 1z are defined for
distances downwind of the release point from the expressions:
and
18
1z Rzxrz (12)
where the parameters Ry, ry, Rz, and rz are established based on atmospheric conditions referred to
as Pasquill Stability Classes. In the transient emission case 1x = 1y. In our model we have used
values of these parameters defined for a neutral atmosphere and a wind velocity of 10 mph
(4.5 m/s). In the agent dispersion case, the model indicated that the agent cloud grows to
approximately 350 meters in diameter approximately 2 km downstream of the release point,
which is in good agreement with published data on other dispersion models.22
In both modeling scenarios the choice of source strength is critical to the overall
sensitivity estimate. In the chemical agent dispersion case a single release of 100 kg was
assumed to occur. This source strength was thought to be typical of a theater ballistic missile
warhead (SCUD) or of a multiple shell artillery attack. We believe the assumption of a 100 kg
release is conservative. In a chemical delivery mode, the SCUD-B is capable of carrying a
warhead containing 555 kg23 of the persistent agent VX and could carry an equivalent amount of
a G-type agent. Based on data available at the Chemical Weapons Convention web site,24 a
100 kg release would require approximately forty 155 mm howitzer shells, or a similar number of
115 mm rockets. In the stack monitoring case data on anticipated chemical concentrations are
less readily available due to the sensitive nature of the facilities which could be monitored. In
our model, we took an extremely conservative approach to this problem by assuming that the
chemical concentrations at the stack exit were 10 ppmv. From literature available on chemical
process facility design,25 we determined that properly designed exhaust vents have air velocities
which range from 10 to 30 meters/second. This combination of chemical concentration, stack
dimension, and flow velocity sets the mass flow rate of the compound entering the atmosphere.
19
The detection of emissions from chemical stacks represents the more difficult of the two
detection tasks. Initial calculations, based on a model which assumed emissions occurred from a
chemical hood facility (20 m/s flow at 298K) with a 1 meter diameter stack at a concentration of
10 ppmv × m, showed signal levels well below system detection limits. An example of the
model results are shown in Figures 15 and 16. The column density shown in Figure 15 is quite
small, leading to an optical depth of 1.25 x 10-3 for the strongest band in DMMP. The
corresponding change in scene radiance, shown in Figure 16, reveals a net change in radiance of
only 1.5 x 10-8 W cm-2 sr-1 µm-1. This radiance level is well below the detection limits for any
operational or projected multispectral imaging system known to the authors.
It was determined subsequently that emissions from combustion driven stacks
(Texit�150 oC) were most appropriate for the monitoring scenario. Calculations were performed
to simulate this method of chemical release. The monitoring scenario and a plume model was
described by Barden and Kroutil.26 By increasing the plume temperature, the net radiances from
the scene increased approximately 100 fold. For a system with a spatial resolution comparable to
the plume dimensions, the plume radiance would exceed 4 x 10-4 W cm-2 sr-1 µm, which is well in
excess of the calculated NESR for the systems discussed in this paper. Though the increase in
plume temperature increases the signal dramatically, the increase also has implications for the
spectral signature of the background. This change is primarily due to the presence of
combustion-produced species (primarily H2O and CO2) in the stack effluent which tends to
obscure spectral features in the plume
The results of the chemical agent release model were quite promising in relation to the
anticipated performance of the imaging system. The extent of the chemical cloud is indicated in
Figure 17, where the C × 5 product (column density) is shown for the 45 degree look-down view
20
at a range of 5 km. The peak column density was approximately 0.5 cm g liter-1 (1000 ppmv ×
m) DMMP. for the modeled release. (Although these calculations were carried out in cm g liter-1,
units common to the stand-off detection community, ppmv × m is in general more useful because
it does not require knowledge of analyte’s molecular weight.) The DMMP absorption cross-
section at 9.4 µm results in a plume optical depth of approximately 3, implying that it attenuates
95% of the radiation originating from the far field, which in this case is the ground below. The
radiative transfer model utilized the MODTRAN mid-latitude summer atmospheric profile27 to
determine atmospheric transmission from the release point to the sensor. From this model the air
temperature is given as 294 K, the effective blackbody temperature of the ground is 290 K, and
the chemical plume appears in emission. The data presented in Figures 18 and 19 show total and
differential radiance for the DMMP feature at 9.4 µm, the strongest feature in the spectrum. The
primary difference between these data and a similar calculation performed for the weaker
10.85 µm DMMP feature used in the experimental measurements is the flattening of the spatial
radiance distribution near the region of maximum radiance due to the plume being optically
thick. The differential radiance reaches a maximum of 2.5 x 10-5 W cm-2 sr-1 µm-1 in this viewing
scenario, approximately a factor of 12 greater than the NESR of the brassboard imager.
V. CONCLUSIONS AND FUTURE WORK
The objective of this effort was the development of a LWIR hyperspectral imaging system
specifically designed for imaging of gaseous chemical species in air. We have demonstrated that
the tunable Fabry-Perot etalon provides sufficient spectral resolution, tuning range, and tuning
rate to detect and identify vapor plumes of the two tested hazardous vapor simulants, SF6 and
DMMP. The current mirror set and capacitance micrometry system may be improved to extend
21
the spectral tuning range and reduce the tuning time between spectral resolution elements to
nearer the limit imposed by the mechanical tuning limit of the mirror actuators.
The NESR of the LWIR spectroradiometer is sufficient to allow detection of optical thin
chemical vapor plumes against near ambient temperature backgrounds. Our experimentally
determined detection limit of ±22 ppmv × m for DMMP against a temperature drop of 6 K is in
excellent agreement with the value predicted using a simple radiative transfer model and the
measured noise characteristics of the system detection electronics. In future efforts we will be
replacing the scanned 2 x 4 element HgCdTe array with a liquid nitrogen-cooled 64 x 64 pixel
format, 61 µm square pixel staring array. This improvement will increase spatial resolution by
approximately a factor of 10. Furthermore, system models indicate that the sensor NESR can be
improved to approximately 0.5 µW cm-2 sr-1 µm-1. For comparison, the LIFTIRS and SEBASS
imagers report NESR values ~1 µW cm-2 sr-1 µm-1 using liquid helium-cooled large format FPAs
(Refs. 6 and 7).
The brassboard Fabry-Perot imaging spectroradiometer has provided hyperspectral
imagery of sufficient quality to allow quantitative analysis of chemical vapor plume releases.
Data is currently processed off-line, however we continue to upgrade system hardware and data
processing algorithms with the goal of providing a stand-off imaging sensor with real-time
analysis of the scene chemical composition.
ACKNOWLEDGMENTS
Recent development efforts on the LWIR imaging spectroradiometer were funded by the
U.S. Army Night Vision and Electronic Sensors Directorate under Phase I Small Business
Innovation Research contract No. DAAB07-97-C-G006 and by the U.S. Army Edgewood
22
Research, Development, and Engineering Center (ERDEC) under contracts No. DAAD05-97-P-
2618 and No. DAAD05-98-P-0625. The authors thank Dr. Jack Demirgian for conducting
controlled plume releases at ANL, Dr. James Jensen and Dr. Alan Samuels of U.S. Army
ERDEC for helpful discussions, Will Lawrence for assisting with the outdoor SF6 plume imaging
experiments, and George Dippel for development of the system electronics.
REFERENCES
1. H. Walter, Jr. and D. Flanigan, “Detection of atmospheric pollutants: a correlation
technique,” Appl. Opt. 14, 1423-1428 (1975).
2. M.L.G. Althouse, “Spectral filtering of thermal imagers for gas detection,” in Proc. 3rd
Int. Symp. Protection Against Chemical Agents, p. 143-148, Umea, Sweden (1989).
3. L.G. Carr, L. Fletcher, P.L. Holland, J. Leonelli, D. McPherrin, and M.L. Althouse,
“Characterization of filtered FLIR systems designed for chemical vapor detection and
mapping,” in Infrared Imaging Systems: Design, Analysis, Modeling, and Testing, G.C.
Holst, Ed., Proc. SPIE Vol. 1309, p. 90-103 (1990).
4. M.L.G. Althouse and C.I. Chang, “Chemical vapor detection with a multispectral thermal
imager,” Opt. Eng. 30, 1725-1733 (1991).
5. C. L. Bennett, M. R. Carter, and D. J. Fields, "Infrared hyperspectral imaging results from
vapor plume experiments," in Imaging Spectrometry, M.R. Descour, J.M. Mooney, D.L.
Perry, and L.R. Illing, Eds., Proc. of SPIE Vol. 2480, p. 435-444 (1995).
6. C. L. Bennett, M. R. Carter, and D. J. Fields, "Hyperspectral imaging in the infrared using
LIFTIRS," in Infrared Technology XXI, B.F. Andresen and M.S. Scholl, Eds., Proc. of
SPIE Vol. 2552, p. 274-283 (1995).
23
7. J.A. Hackwell, D.W. Warren, R.P. Bongiovi, S.J. Hansel, T.L. Hayhurst, D.J.Marby,
M.G. Sivjee, and J.W. Skinner, "LWIR/MWIR imaging hyperspectral sensor for airborne
and ground-based remote sensing," in Imaging Spectrometry II, M.R. Descour and J.M.
Mooney, Eds., Proc. of SPIE Vol. 2819, p. 102-107 (1996).
8. C.M. Gittins, W.J. Marinelli, and A.J. Ratkowski, “AIRIS Hyperspectral Imaging
Technology,” presented at the 6th Annual AIAA/BMDO Technology Readiness
Symposium, Report No. N97-31890, American Institute of Aeronautics and Astronautics,
Reston, VA, 1997.
9. C.M. Gittins, W.G. Lawrence, and W.J. Marinelli, “Frequency Agile Bandpass Filter for
Direct Detection Lidar Receivers,” Appl. Opt. 37, 0000-0000 (1998).
10. P.D. Atherton, N.K. Reay, J. Ring, and T.R. Hicks, “Tunable Fabry-Perot Filters,” Opt.
Eng. 20, p. 806-814 (1981)
11. D.R. Suhre and E. Villa, “Imaging Spectroradiometer for the 8-12 µm region with a
3 cm-1 Passband Acousto-Optic Tunable Filter,” Appl. Opt. 37, 2340-2345 (1998).
12. D.F. Flanigan, “Prediction of the limits of detection of hazardous vapors by passive
infrared with the use of MODTRAN,” Appl. Opt. 35, 6090-6098 (1996).
13. D.F. Flanigan, “The Spectral Signatures of Chemical Agent Vapors and Aerosols,”
CRDC-TR-85002, US Army Armament, Munitions and Chemical Command, Aberdeen
Proving Ground, MD, April 1985 (unclassified).
14. L.D. Hoffland, R.J. Piffath, and J.B. Bouck, “Spectral signatures of chemical agents and
simulants,” Opt. Eng. 24, 982-984, 1985.
24
15. P. Varanasi, Z. Li, V. Nemtchinov, and A. Cherukuri, “Spectral Absorption-Coefficient
Data on HCFC-22 and SF6 for Remote-Sensing Applications,” J. Quant. Spectrosc.
Radiat. Transfer 52, 323-332 (1994).
16. J. Demirgian, S. Macha, and S.Darby, “The Challenge: Collection of Quantitative
Chemical Release Data,” in Proc. Third Workshop on Stand-Off Detection for Chemical
and Biological Defense, Science and Technology Corp., Hampton VA, p. 299-306 (1994).
17. R.N. Clark, A.J. Gallagher, and G.A. Swayze, “Material absorption band depth mapping
of imaging spectrometer data using the complete band shape least-squares algorithm with
library reference spectra,” in Proc. Third Airborne Visible/Infrared Imaging Spectrometer
(AVIRIS) Workshop, JPL Publication 90-54, p.176-186, Jet Propulsion Laboratory,
Pasadena CA, 1990.
18. R.N. Clark, G.A. Swayze, A. Gallagher, N. Gorelick, and F.A. Kruse, “Mapping with
imaging spectrometer data using the complete band shape least-squares algorithm
simulaneously fit to multiple spectral features from multiple materials,” in Proc. Third
Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop, JPL Publication 91-
28, p.2-3, Jet Propulsion Laboratory, Pasadena CA, 1991.
19. R.N. Clark, G.A. Swayze, and A. Gallagher, “Mapping the minerology and lithology of
Canyonlands, Utah with imaging spectrometer data and the multiple materials,” in
Summaries of the Third Annual JPL Airborne Geoscience Workshop, JPL Publication
92-14, v. 1, p. 11-12, Jet Propulsion Laboratory, Pasadena CA, 1992.
20. U.S. Department of Energy, Richland Operations Web Site,
http://www.hanford.gov/twrs/char.pub/docs/tcr/t107/t107s2.htm
25
21. Atmospheric Chemistry and Physics of Air Pollution, J.H. Seinfeld, John Wiley & Sons,
New York, 1986.
22. E.E. Uthe, N.B. Nielson, and R.D. Kaiser, “Airborne LIDAR and Radiometric Detection
and Analysis of Effluent Plumes,”in Proc. Third Workshop on Stand-Off Detection for
Chemical and Biological Defense, Science and Technology Corp., Hampton VA, p.211-
212 (1994).
23. Fact sheet, Chemical and Biological Defense Information Center,
http://www.cbiac.apgea.army.mil/faq.htm.
24. J.E. O’Pray, “Regional Power Ballistic Missiles: An Emerging Threat to Deployed US
Forces?” Analytical study, Air War College, Air University, Maxwell Air Force Base,
AL, May 1990.
25. Design, Construction and Refurbishment of Laboratories, R. Lees and A.F. Smith, Eds.,
Ellis Horwood Limited, Chichester, England, 1984.
26. J.D. Barden and R. Kroutil, “Development and Preliminary Field Evaluation of a Field-
of-View Near Real-Time 3-D Stack Plume Model Developed for the Measurement
Attributes of Remote Optical Sensors,” in Proc. Third Workshop on Stand-Off Detection
for Chemical and Biological Defense, Science and Technology Corp., Hampton VA, p.
457-467 (1994).
27. MODTRAN2 (Ontar Corp., North Andover MA, 1992).
26
FIGURE CAPTIONS
Figure 1. Basic configuration for the tunable Fabry-Perot etalon based imaging
spectroradiometer. FPA, focal plane array.
Figure 2. LWIR imager optical configuration employing the off-axis parabolic reflectors
(OAPs) as focussing elements
Figure 3. Expanded view of LWIR tunable filter transmission spectrum.
Figure 4. Representative Transmission Spectrum of the LWIR tunable filter module
Figure 5. Schematic diagram of three-layer model for quantitative analysis of scene radiance
levels and plume concentration determination.
Figure 6. Observed and calculated transmission through an absorption cell containing
8 ppmv × m SF6 and 80 ppmv × m DMMP in dry N2. The reference blackbody is
at 308 K and the gas temperature is 298 K. The data represent a six scan co-
average.
Figure 7. Two-band correlation plot of narrowband scene images at 10.55 and 10.85 µm.
See text for details.
Figure 8. IR image of absorption cell at 10.55 µm with the location of SF6 absorption
identified by two-band correlation analysis highlighted in black.
Figure 9. Calculated SF6 column density in the absorption cell containing scene as
determined using the three-layer radiative transfer model.
Figure 10. Best fit LWIR imager data to Eq. (7) for a 141 ppmv × m DMMP plume release.
Figure 11. Measured versus calculated DMMP plume column densities.
Figure 12. Visible image of the outdoor plume imaging configuration. Gas is released from
the black pipe slightly below the center of the scene.
27
Figure 13. SF6 plume from outdoor release experiment as identified using the spectrally
matched filter algorithm. Light pixels indicate a high degree of correlation with
the SF6 absorption spectrum.
Figure 14. SF6 column density in the outdoor plume release calculated using LWIR imager
data and the three-layer radiative transfer model.
Figure 15. Calculated column density (concentration × path length) for stack emission over
the filed of view of the system.
Figure 16. Estimated differential radiance for chemical species in the stack plume for a non-
combustion scenario.
Figure 17. Column density of agent for a simulated release from a SCUD-B.
Figure 18. Total radiance at 9.4 µm for a simulated chemical agent release using DMMP.
Figure 19. Net radiance at 9.4 µm for a simulated chemical agent release. Flattening of the
profile at the highest radiance levels indicates that the cloud is optically thick in
the center.
Figure 1Marinelli
Applied Optics28
λ
R
θ
F-P
Lens
FPA
IncidenceAngle
MirrorSpacing D-3994z
Figure 1. Basic configuration for the tunable Fabry-Perot etalon-based
imaging spectroradiometer. FPA, focal plane array.
Figure 2Marinelli
Applied Optics29
OAP
Detector
Interferometer9.5-11.5 µm
Chopper3 kHz
Horizontal
OAP
OAP
Vertical
Galvo-Scanner
± 20 degHorizontaland Vertical
D-5263
Figure 2. LWIR imager optical configuration employing the off-axis parabolic
reflectors (OAPs) as focussing elements
Figure 3Marinelli
Applied Optics30
Figure 3. Expanded view of LWIR tunable filter transmission spectrum.
Figure 4Marinelli
Applied Optics31
Figure 4. Representative Transmission Spectrum of the LWIR tunable filter module
Figure 5Marinelli
Applied Optics32
Layer 3
Atmosphere
Layer 2
Sensor Plume
Layer 1
D-7659
BlackbodyBackground
Figure 5. Schematic diagram of three-layer model for quantitative analysis of scene
radiance levels and plume concentration determination.
Figure 6Marinelli
Applied Optics33
Figure 6. Observed and calculated transmission through an absorption cell
containing 8 ppmv × m SF6 and 80 ppmv × m DMMP in dry N2. The
reference blackbody is at 308 K and the gas temperature is 298 K. The data
represent a six scan co-average.
Figure 7Marinelli
Applied Optics34
Figure 7. Two-band correlation plot of narrowband scene images at
10.55 and 10.85 µm. See texts for details.
Figure 8Marinelli
Applied Optics35
Figure 8. IR image of absorption cell at 10.55 µm with the location of SF6 absorption identified
by two-band correlation analysis highlighted in black.
Figure 9Marinelli
Applied Optics36
Figure 9. Calculated SF6 column density in the absorption cell containing scene as
determined using the three-layer radiative transfer model.
Figure 10Marinelli
Applied Optics37
Figure 10. Best fit LWIR imager data to Eq. (7) for a 141 ppmv × m
DMMP plume release.
Figure 11Marinelli
Applied Optics38
Figure 11. Measured versus calculated DMMP
plume column densities.
Figure 12Marinelli
Applied Optics39
Figure 12. Visible image of the outdoor plume imaging configuration. Gas is released
from the black pipe slightly below the center of the scene.
Figure 13Marinelli
Applied Optics40
Figure 13. SF6 plume from outdoor release experiment as identified using
the spectrally matched filter algorithm. Light pixels indicate a high degree
of correlation with the SF6 absorption spectrum.
Figure 14Marinelli
Applied Optics41
Figure 14. SF6 column density in the outdoor plume release calculated
using LWIR imager data and the three layer radiative transfer model.
Figure 15Marinelli
Applied Optics42
Figure 15. Calculated column density (concentration × path length) for stack
emission over the field of view of the system.
Figure 16Marinelli
Applied Optics43
Figure 16. Estimated differential radiance for chemical species in the
stack plume for a non-combustion scenario.
Figure 17Marinelli
Applied Optics44
Figure 17. Column density of agent for a simulated relese from a SCUD-B.
Figure 18Marinelli
Applied Optics45
Figure 18. Total radiance at 9.4 µm for a simulated chemical agent
release using DMMP.
Figure 19Marinelli
Applied Optics46
Figure 19. Net radiance at 9.4 µm for a simulated chemical agent release.
Flattening of the profile at the highest radiance levels indicates that the
cloud is optically thick in the center.