INFRARED CLOUD IMAGING SYSTEMS CHARACTERIZATION
by
David Walter Riesland
A thesis submitted in partial fulfillment
of the requirements for the degree
of
Master of Science
in
Optics and Photonics
MONTANA STATE UNIVERSITY
Bozeman, Montana
November 2016
c©COPYRIGHT
by
David Walter Riesland
2016
All Rights Reserved
ii
DEDICATION
To my friends and family,
for your loving support.
To my wife,
my steadfast companion over oceans and years.
iii
ACKNOWLEDGMENTS
I would like to thank my advisor, Dr. Joseph Shaw for your guidance
and support, and for providing such valuable opportunities throughout my
undergraduate and graduate career. I would also like to thank graduate
committee members Dr. Wataru Nakagawa for providing such great advice
and feedback, and Dr. Charles Kankelborg for inspiring me to pursue optics
as an undergraduate through the NASA National Student Solar Spectrograph
Competition. A special thanks to Paul Nugent for putting up with me in the
lab through all the years, late nights, and coffee; and for being such a great
sounding board. I also thank the entire Optical Remote Sensor Lab team at
Montana State University for all of your support. A big thank you to Dr. Jed
Hancock and Dr. James Peterson for allowing me to come take measurements
at Space Dynamics Lab, and for such a great summer. I would also like to thank
the team at NASA Glenn Research Center, Dr. James Nessel, Michael Zemba,
and especially Jacquelynne Houts for your advice and collaboration.
Funding Acknowledgments
I would like to acknowledge the generous support of the NASA Montana
Space Grant Consortium, the NSF Arctic Observing Network Program, NASA
Glenn Research Center, NASA EPSCoR, Montana Research and Economic
Development Initiative, Space Dynamics Lab, and NASA JPL for various
aspects of this work.
iv
TABLE OF CONTENTS
1. INTRODUCTION ........................................................................................1
2. ICI METHODOLOGY AND BACKGROUND ...............................................7
Introduction .................................................................................................7
Conclusion ................................................................................................. 10
Acknowledgments ................................................................................ 11
3. INFRARED CLOUD IMAGER HARDWARE OVERVIEW .......................... 12
Introduction ............................................................................................... 12
Camera ...................................................................................................... 13
Microbolometer Focal Plane................................................................. 14
Lenses and Windows ........................................................................... 15
Support Electronics Upgrades...................................................................... 16
Camera Communication Module .......................................................... 17
Microprocessor/Analog to Digital Converter.......................................... 17
Enclosure............................................................................................ 18
Environmental Module Upgrades ................................................................. 20
Proposed Cooling System..................................................................... 24
NASA GRC Deployment...................................................................... 25
Conclusion .......................................................................................... 27
4. RADIOMETRIC CALIBRATION BACKGROUND...................................... 28
Introduction ............................................................................................... 28
Initial Assumptions..................................................................................... 28
Simple Camera Calibration ......................................................................... 29
Temperature-Dependent FPA Calibration .................................................... 30
Conclusion ................................................................................................. 31
5. LWIR CAMERA RELATIVE SPECTRAL RESPONSE................................ 32
Introduction ............................................................................................... 32
Methodology .............................................................................................. 34
Simulated Tau2 Camera RSR ...................................................................... 36
Simulated Photon Camera RSR................................................................... 38
Deviation from Manufacturer’s RSR ............................................................ 39
Conclusion ................................................................................................. 40
vTABLE OF CONTENTS - CONTINUED
6. TAU2 CHARACTERIZATION.................................................................... 43
Introduction ............................................................................................... 43
Low-signal Suppression and Shutter Offset Correction................................... 44
External Flat-Field Correction ............................................................. 45
External FFC of a False-Temperature Internal Shutter .......................... 47
Tau2 Calibration Results............................................................................. 48
Noise-equivalent Radiance ........................................................................... 50
Conclusion ................................................................................................. 51
7. TAMARISK CHARACTERIZATION .......................................................... 53
Introduction ............................................................................................... 53
Calibration Algorithm................................................................................. 54
Calibration Results ..................................................................................... 56
Conclusion ................................................................................................. 57
8. SYSTEM COMPARISONS.......................................................................... 59
Introduction ............................................................................................... 59
Photon vs Tau2 .......................................................................................... 59
Photon vs Tamarisk.................................................................................... 61
Image Comparison............................................................................... 62
Radiometric Comparison...................................................................... 62
Conclusion ................................................................................................. 65
9. CLOUD PHASE DETECTION ................................................................... 67
Introduction ............................................................................................... 67
Algorithm Development .............................................................................. 70
Algorithm Comparisons .............................................................................. 73
Conclusion ................................................................................................. 76
10. CONCLUSIONS ......................................................................................... 77
REFERENCES CITED.................................................................................... 80
vi
LIST OF FIGURES
Figure Page
1.1 Emission from clouds can be seen with an infrared imager at
Montana State University, 23 January 2016, 1:44 am MST. ...................3
1.2 Size progression of microbolometer cameras, from left to
right: 1999 (Amber Sentinel), 2009 (FLIR Photon 320), 2013
(FLIR Tau2), and 2016 (FLIR Lepton). Photo courtesy of
Joseph Shaw. ......................................................................................6
2.1 Cloud data taken by the Svalbard ICI during a validation
period at Bozeman, MT. Total atmospheric-emitted radiance
is seen in (a); the modeled atmosphere is subtracted and (b)
is the residual cloud radiance, while (c) shows the calculated
optical depth and (d) shows the expected cloud attenuation
at 550 nm...........................................................................................9
2.2 MODTRAN simulations showing radiance contributions and
clouds for zenith measurement in Bozeman, MT. ................................ 10
3.1 CAD Drawing of GRC ICI System, Spring 2015.................................. 12
3.2 Flir Tau2 Camera. Photo courtesy of Paul Nugent. ............................. 13
3.3 Microbolometer topology ................................................................... 14
3.4 Air mass follows a predicted trend with zenith angle, here
simplified as a secant......................................................................... 16
3.5 JPL ICI legacy support electronics from 2014. Photo
courtesy of Joseph Shaw.................................................................... 16
3.6 The GigE communication module, shown here attached to
the back of a Tau2 camera, provides more camera control
while reducing overall component footprint ......................................... 17
3.7 JPL tube enclosure mounted to cooling box. Paint can
be seen peeling from windowed flange due to previous
deployment. Tube enclosure has been refinished with powder
coat for longevity. ............................................................................. 19
3.8 Svalbard and White Sands ICI systems deployed at Montana
State University. Photo courtesy of Joseph Shaw. ............................... 20
3.9 Heater prototype tests....................................................................... 22
vii
LIST OF FIGURES - CONTINUED
Figure Page
3.10 Cooling option prototype CAD drawing, here shown in blue.
Systems were designed for modularity................................................. 22
3.11 Prototype tube ICI cooling tests ........................................................ 23
3.12 Operational field deployment at Glenn Research Center in
Cleveland, OH. ................................................................................. 26
4.1 Simple two-point radiometric camera calibration ................................. 30
5.1 FLIR Camera RSRs measured at SDL................................................ 33
5.2 RSR simulation flowchart .................................................................. 34
5.3 Tau2 camera measured and simulated RSRs ....................................... 37
5.4 Tau2 calibration error due to RSR uncertainty. ................................... 38
5.5 Photon camera measured and simulated RSRs .................................... 39
5.6 Photon calibration error due to RSR uncertainty ................................ 40
5.7 RSR measurements vs manufacturer-specified spectra ......................... 41
5.8 Scatter plots of cloud scene radiances calculated with the
SDL-measured and manufacturer-specified RSR spectra for
(a) Photon and (b) Tau2 cameras. ..................................................... 42
5.9 Uncertainty in percentage of scene radiance due to generic
RSR from manufacturer .................................................................... 42
6.1 Manual FFC vs External FFC at different FPA temperatures .............. 46
6.2 Proposed external FFC data collection routine.................................... 47
6.3 Corrected radiance for a Tau2 Camera using MSU standard
algorithm over time in comparison with uncorrected and
source radiance. Images were collected every 20 seconds. ..................... 49
6.4 Distribution of error in calibrated radiance relative to the
source radiance ................................................................................. 50
7.1 Tamarisk camera first sky image, with tripod and operator
in view. The colorbar shows digital number output of the FPA............ 53
viii
LIST OF FIGURES - CONTINUED
Figure Page
7.2 Tamarisk FPA temperature sensor output has a fairly linear
response with external camera temperature ........................................ 54
7.3 Tamarisk digital FPA temperature, digital number response,
and scene radiance can be solved with a linear regression..................... 55
7.4 Radiance correction standard deviation for Tamarisk camera
over FPA.......................................................................................... 57
8.1 Svalbard ICI (G02) and ICI3 reported dB compared over time............. 59
8.2 Statistical analysis of instrument comparisons. .................................... 60
8.3 Photon (left) and Tamarisk (right) ICI systems deployed side
by side ............................................................................................. 61
8.4 Radiometric image comparison of cloudy sky ...................................... 63
8.5 Thin clouds can be seen above the noise when doing scene-
to-scene subtraction .......................................................................... 64
8.6 Distribution of scene comparisons during deployment period................ 65
8.7 Scatter plot showing correlation of 0.996............................................. 66
9.1 The imaginary part of the index of refraction is plotted
against wavelength. Arrows show the differences at 1.64 µm
and 1.7 µm. ...................................................................................... 68
9.2 Observation of clouds simulated in MODTRAN .................................. 69
9.3 Sunlight scattered from simulated simulated ice clouds (cir-
rus) have a much different spectral shape than liquid water
clouds .............................................................................................. 70
9.4 The Knap (2002) algorithm allows ice clouds to be distin-
guished from liquid clouds in MODTRAN simulations......................... 71
9.5 PWV has little affect on the algorithm for OD > 1. ............................ 72
9.6 The spectral ratios between ice and water clouds become less
discrete after gaussian 150 nm bandwidth filters are used. ................... 73
ix
LIST OF FIGURES - CONTINUED
Figure Page
9.7 Adding a third integrated radiance channel allows for greater
separation of data vs optical depth..................................................... 74
9.8 Optical Depth can be replaced by the RMS addition of the
radiance from each channel ................................................................ 75
9.9 Two-channel algorithm using calibrated radiance values ...................... 76
xABSTRACT
Infrared cloud imaging (ICI) is a useful tool for characterizing cloud cover for a
variety of fields. Clouds play an important role in free-space high frequency (optical
and mm-wave) terrestrial communications. Ground-based infrared imagers are used
to provide long-term, high resolution (spatial and temporal) cloud data without the
need for sunlight. This thesis describes the development and characterization of two
ICI systems for deployment at remote field sites in support of Earth-to-space mm-
wave and optical communication experiments. The hardware upgrades, calibration
process, sensitivity analysis, system validation, and algorithm developments are all
discussed for these systems. Relative spectral response sensitivity analysis is discussed
in detail, showing as much as 35% calibrated scene radiance uncertainties when
using generic manufacturer data in comparison with measured spectral responses.
Cloud discrimination algorithms, as well as cloud phase (ice or water discrimination)
algorithms are also discussed.
INTRODUCTION
NASA has a bandwidth problem. Don Cornwall, the director of the Space 
Communications and Navigation Program at NASA HQ, stated in 2015 that ”We 
are leaving 90% or more of our data on the surface of Mars” [1]. Modern space-borne 
science instruments are generating much more data than what can be sent back to 
earth through legacy communication systems. This communications bottleneck is 
due to the reliance on RF technology for deep-space communication. At the time of 
this paper, almost all NASA missions are using RF technology for earth-to-space and 
space-to-space communication. However, as space-borne missions are becoming more 
complex, there is a push within NASA to move to higher-frequency communication 
systems to allow for more data to be moved in a smaller amount of time. There are 
several missions underway in both NASA [1–3] and ESA [4] to increase communication 
bands from RF frequencies to mm-wave and optical frequencies. However, as these 
communication bands increase in frequency they become more susceptible to cloud 
attenuation in terrestrial satellite links. Therefore, it is becoming more important 
to model and understand cloud conditions and their relation to signal attenuation 
and distortion at operational and potential high-frequency earth-to-space ground 
communication sites [5–7].
A popular method for cloud detection is satellite observation. Many weather 
satellites are being used by forecasters to predict weather conditions across wide 
areas. However, these satellites have either poor temporal or poor spatial resolution 
at any given ground site. Most geostationary weather satellites such as GOES-15, 
MTSAT-2, and Meteosat-8 are stationed in equatorial orbit. This geometry makes it
1
2difficult or impossible to characterize clouds for high-latitude sites, which are most
often of interest for both climate studies and earth-space communications. Likewise,
satellites that scan the surface of the earth from a polar orbit, such as NOAA-17 and
Metop-A, have higher spatial resolution due to their lower altitude, but only visit the
same site twice a day, thus having poor temporal resolution.
A ground-based system at the observational site can be used to increase both
temporal and spatial resolution. Such cloud detection instruments include LIDAR,
visible cameras, and IR cameras. Visible cameras give the most intuitive information
since their responsivity is similar to the human observer. Visible imagers only reliably
detect clouds during daytime and either do not work at night or do not provide
consistent results throughout day and night [8,9]. This suggested that visible cameras
may have degraded performance during nighttime periods.
Active lidar or mm-wave radar sensing also provides excellent cloud data [10,11],
but with larger and more costly instruments that require scanning to cover a large
fraction of the sky. Dual-channel microwave radiometers are frequently deployed at
propagation study sites for measuring precipitable water vapor [12], but they do not
allow detection of ice clouds and have only weak sensitivity to thin liquid clouds.
Furthermore, these other instruments typically measure at only a single point and
require complex scanning systems to obtain spatial information.
Many of these disadvantages can be solved through the use of long-wave infrared
(LWIR) imaging systems. LWIR cameras measure emitted cloud radiance, which
provides a consistent signal during both day and night. Multiple elevation angles
can also be detected for simultaneous radiometric analysis with an imaging system.
This allows for both high spatial and high temporal cloud statistics at the site of
interest [9, 13].
3Figure 1.1: Emission from clouds can be seen with an infrared imager at Montana
State University, 23 January 2016, 1:44 am MST.
As shown in Figure 1.1, thermal emission from clouds enable observation with
an infrared system. A clear sky has relatively low absorption, while clouds have
relatively high absorption at the IR wavelengths used by ICI systems. Invoking
Kirchoff’s law of thermal radiation, a perfect blackbody will emit as much radiation
as it absorbs. Although clouds are in thermal equilibrium with the surrounding
air, there is a substantial radiometric contrast between cloud emission and clear-air
emission because the clouds have high emissivity (due to high absorption) while the
clear air has much lower emissivity (due to lower absorption). Therefore, clouds will
emit more thermal radiation than the surrounding clear sky [14]. With a carefully
characterized infrared camera, the emission from clouds lead to the classification of
the cloud, both in terms of optical depth and optical attenuation.
4Infrared imaging with low-cost, uncooled imaging systems was enabled for
civilian use when microbolometer technology was declassified in 1992. Many details of
the development of the microbolometer technology are cited in these references [15,16].
Uncooled microbolometer focal planes allowed for relatively low-cost systems to be
used for research and commercial applications [17].
The earliest infrared cloud imager (ICI) system was developed for atmospheric
research by Joseph Shaw at NOAA starting in 1997 in collaboration with the Japanese
Communication Research Laboratory or CRL (now National Institute of Information
and Communication Technology, NICT) [18]. The project then moved with Dr.
Shaw to Montana State University (MSU) in 2001 and the ICI system was moved
to MSU in January 2002. This system was later deployed to Barrow, AK [9, 13]
while other ICI systems were also deployed to Barrow, AK, and Poker Flat, AK,
for cloud characterization studies in the Arctic. Later systems have been used to
support cloud cover characterization campaigns at JPLs Table Mountain Facility
(TMF) [19] and Goldstone Deep Space Communications Complex (GDSCC), both
operated by the NASA Jet Propulsion Laboratory (JPL) in support of free-space
optical communication studies.
Researchers found through experimentation at sites such as NASA Glenn
Research Center’s (GRC) Near-Earth Network (NEN) that current atmospheric
models do not correctly account for total atmospheric attenuation at low elevation
angles (propagation near the horizon) for Ka-band propagation. They believed that
thin clouds may be undetected or poorly accounted for in the models, which pointed
to the need for a cloud-detection instrument to characterize clouds and their effects
on high-frequency communication [6]. Since ICI systems are perfectly suited for this
role, a narrow-field-of view ICI was developed to generate cloud statistics both along
the propagation beam and at zenith.
5The development of this new system revealed some limitations of miniaturized
infrared cameras. As infrared imaging technology progressed, focal plane arrays
became optimized for detection of higher temperatures at the cost of low temperature
detection. The primary reason for this was that to increase the resolution of the focal
plane array, pixels needed to get smaller and smaller. This progression can be seen in
Figure 1.2 which shows cameras produced from 1999 on the left to a camera produced
in 2016 on the right. If the f/# of the camera stays the same, a smaller pixel will
generate a smaller electrical signal due to smaller throughput for an extended source,
resulting in a decreased signal-to-noise ratio. Since the infrared signal coming from a
cloud is so much smaller than the signal coming from something like a jet plume, and
many more customers measure jet plumes than clouds, cameras that were useful for
cloud imaging were being phased out of production. Optimization in manufacturing
for small pixels caused significant problems for cloud detection. A specific problem
arose when the manufacturer of the microbolometer cameras we used in 2013 changed
their firmware so that it subtracted our low cloud emission signal from the scene by
proprietary noise reduction algorithms before the data became accessible to the user.
The firmware and internal processing algorithm updates were the largest contribution
to challenges in cloud detection not seen in legacy ICI systems.
This thesis describes two ICI systems that were built for long-term deployment
in hostile environments using primarily OEM components. The design, analysis,
characterization and calibration challenges of those systems will be addressed in this
thesis, as well as meaningful lessons learned for future systems.
The second chapter of this thesis gives more detailed background on ICI
methodology. Chapter three gives an overview of the hardware used in the NASA
Glenn ICI Systems. Chapter four provides a brief background of radiometric
calibration and shows calibration algorithms previously developed at Montana State
6Figure 1.2: Size progression of microbolometer cameras, from left to right: 1999
(Amber Sentinel), 2009 (FLIR Photon 320), 2013 (FLIR Tau2), and 2016 (FLIR
Lepton). Photo courtesy of Joseph Shaw.
University. Chapter five addresses one of the primary sources of uncertainty for an
ICI system, the relative spectral response of the camera. Chapter six talks about
some of the calibration challenges faced when upgrading the system with a Tau2
camera with updated firmware. Chapter seven shows a preliminary look at another
infrared camera (the Tamarisk by DRS) that could be used in future cloud imaging
systems. Chapter eight shows comparisons between the Tamarisk, Tau2, and legacy
Photon camera systems. Chapter nine is a discussion of possible radiometer units to
use in the future to improve cloud optical depth retrievals of ICI systems through the
retrieval of cloud thermodynamic phase.
ICI METHODOLOGY AND BACKGROUND
Introduction
This chapter discusses the methodology behind infrared cloud imaging and some
of the science behind the process. As stated previously, cloud imaging is enabled
through observing cloud thermal emission. Since we assume that radiance from the
cloud is the difference between the scene radiance and the estimated atmospheric-
emitted radiance, cloud radiance Lcloud is calculated through Equation (2.1)
Lcloud = Lmeasured − Lclearsky. (2.1)
Lclearsky is either determined through imaging algorithms or radiative transfer
modeling. One component of the clear-sky radiance estimation involves comparing
the measured angular radiance pattern with the theoretically expected variation of
radiance with elevation angle.
This effect is modeled with a radiative transfer code for water vapor with a
uniform horizontal distribution. Since the imager is able to detect multiple elevation
angles simultaneously, any deviations from this previously established trend are
assumed to be radiance from clouds.
Once the clear-sky radiance is estimated and removed, the residual radiance is
used to estimate the cloud optical depth (path-integrated cloud extinction), which is
defined in Eq. (2.2) in terms of the extinction cross section σ and the particle number
density N . Another component of the clear-sky radiance identification involves
estimating the emission of the atmospheric column from the path-integrated water
vapor content, called ”precipitable water vapor” (PWV), which is often measured
7
8with a microwave radiometer [12] or solar radiometer [20].
τ =
∫ ∞
0
σ(z)N(z) dz (2.2)
An empirical relationship between cloud optical depth and radiance is derived
using methods established in [14,21]. Radiance is what is measured with an infrared
imaging system and since radiance depends on both emissivity and temperature,
knowing the temperature of the cloud (dependent on cloud height) reduces uncertainty
of the emissivity estimate. For this reason, it is helpful to run an ICI system next to
an atmospheric lidar.
A useful reference wavelength near the middle of the visible spectrum is 550
nm, so optical depth is modeled here to determine optical depth for the entire
visible spectrum. Of particular interest are cirrus clouds, because even though they
are difficult to detect, they will still attenuate a free-space communication laser
beam. Attenuation is expected to remain spectrally flat for the optical communcation
channels of 532 nm, 860 nm, 1064 nm, and 1550 nm since ice particles are much larger
than these wavelengths [19]. Figure 2.1 shows examples of emitted radiance measured
with an ICI system, the residual radiance that remained after the simulated clear-sky
emission was subtracted, and the corresponding cloud optical depth and attenuation.
Infrared cloud imaging is done primarily in the 8-14 µm atmospheric window.
These wavelengths were chosen because of the relatively low absorption, and thus
emission, from atmospheric gases and the higher emission from clouds in this spectral
region. H2O, CO2, and O3 are all contributors to gaseous atmospheric emission in
this band. However, the largest contributor is H2O. Fig. 2.2 shows a MODTRAN [22]
simulation of atmospheric emission for the 1976 U.S. standard atmosphere, both with
and without additional water vapor. The figure also shows the radiance expected from
9Figure 2.1: Cloud data taken by the Svalbard ICI during a validation period at
Bozeman, MT. Total atmospheric-emitted radiance is seen in (a); the modeled
atmosphere is subtracted and (b) is the residual cloud radiance, while (c) shows
the calculated optical depth and (d) shows the expected cloud attenuation at 550 nm
a low altitude cumulus cloud at 0.066-km with an optical depth of 2, and a cirrus
cloud with an optical depth of 1 at 8-km altitude. These are all with the observer
altitude at 1524 m. Column-integrated water vapor content (called precipitable water
vapor or PWV) is shown in the figure, scaled by 1.5 and 2 times that of the standard
atmosphere. Atmospheric emission also increases with water vapor, and this shows the
affect of humidity on radiance. Atmospheric emission is modeled down to 80◦ zenith
angle (degrees from zenith). The model requires a well-characterized atmosphere
and Montana State University has been researching the use of image processing
10
Figure 2.2: MODTRAN simulations showing radiance contributions and clouds for
zenith measurement in Bozeman, MT.
Source: Reference [7]
algorithms to improve cloud detection at low elevation angles. Once the emission
from the atmosphere is removed, the residual radiance (cloud emission) needs to
be characterized. The characterization of this residual radiance requires a careful
radiometric calibration of the camera, methods for which have been developed at
Montana State University [23–25]. These methods will be covered in greater detail in
later chapters.
Conclusion
This chapter discussed some of the methodology behind the ICI data collection
mechanisms. First, atmospheric radiance is measured by imaging the sky with a
11
thermal camera. The radiance contribution of clear sky is estimated through an
empirical model and subtracted from the scene. The residual radiance is then assumed
to be radiance from clouds. This radiance is then converted to optical depth, and
optical depth is converted to attenuation for a visible laser. This is all enabled through
an atmospheric window in the 8-14 µm spectral region. This process requires both
knowledge of the atmosphere and a well-characterized camera.
Acknowledgments
This section is derived from previous work by the author [7].
INFRARED CLOUD IMAGER HARDWARE OVERVIEW
Introduction
The following chapter describes the hardware systems developed for NASA
Glenn Research Center in support of RF satellite propagation studies in two separate
locations. The original design was constrained to be identical to legacy systems
deployed at JPL (that I had built as an undergraduate) with the exception of
environmental control. However, as problems were seen in the field at JPL, necessary
design improvements were implemented. A CAD model of the newest ICI tube
enclosure system is seen in Figure 3.1.
Figure 3.1: CAD Drawing of GRC ICI System, Spring 2015
12
13
An ICI system consists of 3 main components: the camera, support electronics,
and the environmental control module. The camera system is housed in a weatherized
enclosure designed for long-term deployment at a variety of sites.
Camera
The system used long-wave infrared cameras mounted behind germanium
windows, each with an uncooled microbolometer focal plane. The cameras were
chosen to provide a narrow field of view that had high resolution across the ground-
to-satellite atmospheric path. The uncooled microbolometer allowed for the camera
to have lower upfront cost while maintaining reasonable cloud detection capabilities.
The system used FLIR Tau2 cameras (Figure 3.2). The camera core had an FPGA
interface that performed data preprocessing to reduce noise and focal plane non-
uniformity. Image data were output via LVDS protocol, and camera-specific data
were output by a RS232 interface.
Figure 3.2: Flir Tau2 Camera. Photo courtesy of Paul Nugent.
14
Microbolometer Focal Plane
A microbolometer focal plane can be thought of as an array of small resistors
that are thermally coupled with the scene. As the resistive elements heat up or cool
down because of a change in incident thermal radiation, their resistance changes. If a
known voltage is injected across this resistive element, its current can be monitored,
and a change in current is inversely proportional to a change in resistance via Ohm’s
law. This allows for a measurement of thermal radiation.
Figure 3.3: Microbolometer topology
In practice, the incident radiation is absorbed by a cavity that is matched to
the optical bandwidth of interest [17]. One side of the cavity is a highly reflective
substrate and the other side is a resistive element. The resistive element is mounted
with resistively conductive arms approximately one fourth of the wavelength above
the reflective substrate, usually 2.5 µm. Owing to physical manufacturing constraints,
this distance can vary slightly from camera to camera. This physical limitation
causes some uncertainty in detector responsivity, since the distance determines the
15
depth of the cavity and the cavity determines the wavelength selectivity. Since the
wavelength selectivity is particularly important for atmospheric remote sensing, this
uncertainty was improved with a relative spectral response measurement. Relative
spectral response will be discussed in more detail in Chapter 5.
Lenses and Windows
Because of the long wavelengths involved in infrared imaging, LWIR cameras use
specialty windows that transmit energy in this spectral band. Lenses consist primarily
of germanium, and the external window for the camera is made of carbon-coated
germanium. Although the window emits radiation that depends on its temperature,
a correction for window emission can be applied when using IR windows in a LWIR
system [26].
The lens chosen for the GRC systems had a focal length selected to provide a
narrow field of view to give high spatial resolution in the expected front radiation
lobe of the RF antenna. The chosen lens was a 25-mm, f/1.1 lens with a 13◦ × 10◦
field of view, and 0.680 millirad instantaneous field of view.
Heritage ICI systems had a much larger field of view, which assisted in identifying
clear sky. From the point of view of a stationary observer looking on the ground
towards zenith, radiance from a clear sky increases with zenith angle because the
atmospheric path length increases with angle. This effect is illustrated in Figure 3.4
for a simple plane-parallel atmosphere.
Since a narrow-field-of-view system will observe a narrower range of zenith angles
than a wide-angle imager, this effect is minimized and the ability to classify clear sky
is also reduced. However, near the horizon the higher angular resolution of a narrow
field of view may provide a small advantage over a wider field (such as a fisheye lens
field of view) because the airmass increases so rapidly there with zenith angle.
16
Figure 3.4: Air mass follows a predicted trend with zenith angle, here simplified as a
secant.
Support Electronics Upgrades
Support electronics for the ICI system required significant upgrades in order to
achieve both the level of communication needed with the camera and the required
environmental control of the unit. A legacy support electronics rail is seen in Figure
3.5. Most of the electronics seen here were phased out with instrumentation upgrades.
Figure 3.5: JPL ICI legacy support electronics from 2014. Photo courtesy of Joseph
Shaw
17
Camera Communication Module
The legacy camera communication module consisted of a Pleora R© Ethernet
control module that was modified by FLIR R© to limit control of the LWIR camera
to a few specific functions. However, these functions were not adequate to perform
the specialized functions needed for modified shutter control, so the Ethernet control
module was replaced with a Pleora R©GigE module that interfaced to the FLIR camera
through a NWB Sensors R© interface board. The GigE module had a smaller footprint
than the Ethernet control module and was needed to provide full control over the
FLIR camera. Figure 3.6 shows the modified communication module upgrade.
Figure 3.6: The GigE communication module, shown here attached to the back
of a Tau2 camera, provides more camera control while reducing overall component
footprint
Microprocessor/Analog to Digital Converter
The legacy system used a Beaglebone R© onboard microcontroller for temperature
and humidity sensor data acquisition, which then communicated over Ethernet to the
instrument computer. From long-term deployment at JPL’s Table Mountain Facility,
18
it was learned that power cycling would sometimes corrupt the microcontroller
memory, resulting in a loss of environmental sensor data. To prevent microcontroller
memory corruption, the microcontroller system was replaced by an ADAM R©
industrial analog to digital converter. Moving to an industrial component increased
system reliability and reduced complexity. The Ethernet physical layer remained in
place, but Ethernet communication protocol was replaced by RS485.
Enclosure
Since the camera had a narrow field of view, a window could be used without
limiting the view of the camera. This was advantageous for several reasons. Since the
instrument would be subject to both blowing sand and corrosive sea water conditions,
a window would protect the camera from the elements. A hatch system could be used,
such as the hatch used in the ICI3 system deployed to Barrow, AK, but it would not
provide protection from the blowing desert sand whenever it opened.
The legacy system deployed at JPL had a corrosion-resistant stainless steel tube
enclosure with a carbon-coated germanium window. This enclosure was chosen for
the GRC systems as well, specifically to provide protection both from the corrosive
environment in the high arctic and the sandy conditions in the desert. The enclosure
was treated with a powder coating finish to eliminate peeling and increase container
survivability.
Although the enclosure originally seemed ideal for long-term deployment, it had
several disadvantages. The enclosure was made from a Stainless Steel tube, alloy 636.
This specific alloy is difficult to machine and the cylindrical shape was not conducive
to modification. The tube was also notoriously difficult to work with in the field. If
the system needed to be checked for any reason, the tube would need to be taken
inside and dismantled. This became increasingly difficult as additional components
19
Figure 3.7: JPL tube enclosure mounted to cooling box. Paint can be seen peeling
from windowed flange due to previous deployment. Tube enclosure has been refinished
with powder coat for longevity.
were packed into the tube. Temperature sensors on the front plate prevented quick
access to key components, and the optical window would need to be handled every
time the system was taken apart. This resulted in long maintenance times, even for
relatively simple tasks.
The tube also constrained the maximum efficiency of the thermal control system,
due to constricted air flow and thermal sink separation. The stainless steel tube seen
in Figure 3.8 does not provide enough thermal isolation from the ambient environment
to allow for adequate cooling. Due to the small size of the chamber, additional
insulation inside the tube would constrict airflow from the cooling module further.
20
Figure 3.8: Svalbard and White Sands ICI systems deployed at Montana State
University. Photo courtesy of Joseph Shaw.
Environmental Module Upgrades
Two instruments were developed: one for deployment in the high arctic, the other
for a desert site. Each of these sites posed significant challenges for environmental
control. Svalbard, Norway was the chosen Arctic site with temperatures ranging from
20 ◦C to -20 ◦C. Since Svalbard is an island in the high Arctic, the instrument was
expected to experience periodic corrosive salt water spray as well as periods of low
temperature. Due to multiple Barrow deployments, the effects of salt water spray
on systems are well known. The two-year ICI deployment in Barrow showed major
corrosion issues, as well as problems with peeling paint. If not planned for, these
problems could have detrimental effects mid-deployment.
21
White Sands, NM was the chosen desert site, with temperatures ranging from
45 ◦C to -7 ◦C. Blowing desert sand was expected at the site, as well as periods of
high temperature. Due to the drastically different environmental scenarios for each
system, modularity was implemented for the environmental control. The core system
would consist of the camera and support electronics. Either a heating or cooling
module could then be implemented based upon need.
Environmental Control Environmental control of the enclosure originally con-
sisted of an on/off temperature-controlled relay driving a small resistive heater. In
order to control both heating and cooling from the same controller, a more advanced
controller needed to be used. An ADAM R© PID controller replaced the legacy control
system, primarily since it could share the same RS485 communication interface as
the analog to digital controller.
Heating Module Due to the cold temperatures involved at Svalbard, a high-
capacity heater was chosen for this system. The heater was made by Hi-Heat
Industries R© from flexible silicon. The heating element can be seen in initial testing
in Figure 3.9b. Initial tests showed the heater was able to compensate for ambient
temperature fluctuations down to -18◦ C in a snowstorm, even without insulation.
Cooling Module Due to the remote instrument deployment site of the instru-
ment, the cooling module needed to minimize as much support equipment as possible.
This ruled out the use of freon, other cooling liquids, or compressed air. Instead,
thermoelectric Peltier devices were used to provide cooling and heating using only
electrical power. Heat sinks were used on either side of the device to dissipate heat
and transfer energy from one side to the other. The peltier device is rated to give a
specific temperature difference between the two plates with the application of power.
22
(a) 200 W silicon heater tested with enclo-
sure in -18◦ C weather.
(b) Front end of enclosure with camera
inside insulation and copper heating shield.
Figure 3.9: Heater prototype tests
Figure 3.10: Cooling option prototype CAD drawing, here shown in blue. Systems
were designed for modularity.
23
(a) Visible image of stainless steel tube
with attached box cooling system. Interior
temperatures were monitored.
(b) LWIR image of cooling system showing
relative temperature. Black regions are the
coldest in the scene.
Figure 3.11: Prototype tube ICI cooling tests
Because of the cost and difficulty of machining 636 stainless steel, it was decided
to use the end of the tube for heat exchange. An air-to-air cooler was purchased
and ruggedized for outdoor use. The calculated temperature difference between the
outside and inside of the chamber was 20 ◦C with a 40 ◦C ambient temperature.
Original cooling tests showed that the cooling system was drawing heat out of
the tube and was able to overcome the power dissipated by the electronics. In Figure
3.11 the lens at the end of the tube is black, showing that the window is the coldest
temperature (brightness temperature) in the scene.
The cooling system and tube enclosure were tested with a solar simulator to
simulate thermal loading at an ambient temperature of 40 ◦C. Without a sun shield,
the camera temperature stabilized at 37.2 ◦C, which meant the TEC was able to
dissipate the heat generated in the enclosure by electronics and any thermal loading
from the sun. The thermal loading was expected to be reduced with painting the
enclosure white and adding a sun shield.
24
Since the power from the electronics was being dissipated by the TEC, the TEC
would have been adequate if insulation were added to the exterior of the tube and
box assembly. Any future systems should have isolated thermal layers between the
camera and the external housing.
As long as we were constrained to use the stainless steel enclosure, there existed
multiple cooling efficiency losses between the cooling module and the camera. The
main efficiency loss was due to restricted airflow between the heat exchanger and
camera. A cooling manifold was designed to optimize airflow across the heat sink
and electrical components were taken out of the tube and inserted into the cooling
compartment. Packing the electronics into the cooling compartment provided better
airflow to the camera, but made the electronics almost impossible to maintain. Any
troubleshooting of the system required many man hours just to take the system apart.
Although these modifications slightly improved performance, a better solution would
have been to move to a different enclosure and abandon the tube system altogether.
Proposed Cooling System
When concerns were raised with NASA GRC engineers about the suitability
of the tube system to perform adequate cooling performance in desert conditions, a
legacy enclosure developed by NASA Glenn RF engineers was disclosed. The legacy
enclosure had been deployed in White Sands to provide temperature stability for RF
electronics at 50 ◦C. The enclosure was an insulated fiberglass 14”×16”×10” box with
the back side replaced by a large heat sink. Peltier devices were linked between the
heat sink and an interior bottom plate. Components then could be attached to the
bottom plate and cooled/heated conductively. The camera could then be conductively
coupled to the cooling plate, reducing the efficiency loss introduced when transferring
25
conduction to convection. The White Sands system is currently being retrofitted at
Glenn Research Center with this enclosure.
NASA GRC Deployment
The Svalbard ICI tube system was deployed at NASA Glenn Research Center
in January 2016 in a heating configuration. The software was written in MATLAB,
which can be seen in Figure 3.12a. The software reports sky radiance, real-time
processed attenuation, and cloud classification. Figure 3.12b shows a visible image of
the sky near the time seen in the GUI.
26
(a) Svalbard ICI GUI showing cloud attenuation over time
(b) Visible image looking along enclosure during same time period
Figure 3.12: Operational field deployment at Glenn Research Center in Cleveland,
OH.
27
Conclusion
This chapter discussed the hardware improvements for the GRC ICI systems in
comparison to the JPL ICI systems deployed a year prior. Most of the improvements
to the system were necessitated by problems associated with firmware upgrades to
the Tau2 cameras, which will be discussed in Chapter 6.
The tube enclosure was not optimal for thermal control. Although the TEC was
able to compensate for the thermal dissipation of the electronics, space did not allow
for thermal isolation with the ambient environment. Future systems should always be
well insulated when thermal control is used. When space and air flow is constricted
in an enclosure, thermal control should be enabled through conduction instead of
convection. This will allow for a more efficient transfer of energy from the cooler to
the camera system. If a conductive thermal control system is used, the entire camera
should be enclosed to prevent thermal gradients across the FPA. If adequate thermal
control is used, thermal drift of the FPA should be minimized. This should simplify
the calibration significantly and reduce many problems seen when the camera core
gets too hot. NASA GRC is in the process of fabricating a box enclosure that should
help the temperature stability of the camera core significantly, while at the same time
allowing for easier maintenance with a larger enclosure.
RADIOMETRIC CALIBRATION BACKGROUND
Introduction
The calibration process for microbolometer cameras has been well documented
in previous work at the Optical Remote Sensor Laboratory (ORSL) at Montana State
University [23–25]. However, a cursory review will be placed here to provide context
for the rest of this document. This chapter covers the building blocks and basic
formulas for calibration of thermal imagers.
Initial Assumptions
A perfect blackbody will emit spectral radiance according to the Planck radiation
equation, equation (4.1),
LBBλ =
2hc2
λ5
1
e
 hc
λkT

− 1
, (4.1)
where LBBλ is spectral radiance [W/(m
2sr), c is speed of light in vacuum 2.998x108
[m/s], h is Planck’s constant 6.6626x10−34 [J s], k is Boltzmann’s constant 1.381x10−23
[J/K], and T is the absolute temperature in Kelvin of the blackbody [K]. The radiant
flux incident on a detector in a radiometer viewing a blackbody source that overfills
the radiometer’s field of view can be calculated from Equation (4.2),
Pd = AdΩep,fp
∫ λ2
λ1
LBBλ T
l
λ T
a
λ R
f
λ dλ , (4.2)
where Ad is the active area of the detector, Ωep,fp is the projected solid angle of
the entrance pupil as seen from the focal plane, T lλ is the spectrally dependent
28
29
transmittance of the lens, T aλ is the spectrally dependent transmittance of the
atmosphere between the radiometer and the blackbody, and Rfλ is the relative spectral
response of the focal plane. The signal from the analog-to-digital (A2D) converter
inside the camera is then shown in Equation (4.3)
D# = GiPd +Do, (4.3)
where D# is the digital number output of the A2D Gi is the temperature-independent
gain, and Do is the digital number offset.
Simple Camera Calibration
Since Equation (4.3) shows a linear relationship between detected flux and digital
number, a linear calibration can be derived. The simplest camera correction uses two
blackbody source temperatures, one hot and one cold. A camera correction is then
calculated using Equations (4.4) through (4.6), visualized by Figure 4.1.
Lsi = gD# + d (4.4)
where
g =
LHOT − LCOLD
DHOT −DCOLD (4.5)
d = LCOLD − gDCOLD. (4.6)
30
Figure 4.1: Simple two-point radiometric camera calibration
Temperature-Dependent FPA Calibration
Since a microbolometer focal plane array (FPA) experiences resistance change
caused by both incident scene radiation and changes in its own temperature, it should
not be surprising that this detector is extremely sensitive to its own temperature. In
the case of cloud imaging, the radiation from the camera body is usually larger than
the radiation from the scene. The basic temperature model for a microbolometer
FPA is a resistive divider, where the resistance of the detector changes with signal
from the blackbody target, and the reference resistor changes resistance with respect
to the camera body temperature. A more thorough model of systematic contributors
to this temperature dependence can be found in References [17,27].
Dr. Shaw’s group at Montana State University has developed methods to
compensate for a microbolometer camera’s FPA-temperature response, thereby
31
stabilizing the camera so that the signal variations can be related only to changes
in the scene radiation. In the primary method, the FPA-temperature response is
compensated through a regression based on one linear term and 3 non-linear terms.
This correction is applied to the camera’s digital numbers before the stabilized digital
numbers are used in the linear regression shown in Equation (4.4). The correction
takes the form
DNC =
DN − b(∆T )
1 +m(∆T )
, (4.7)
where
b(∆T ) = b1∆T + b2∆T
2 + b3∆T
3 − σ1. (4.8)
where DNC is the corrected digital number, DN is the uncorrected digital number
from the FPA, b(∆T ) is the temperature-dependent bias correction, m(∆T ) is the
temperature-dependent gain correction, ∆T is the difference in FPA temperature
from 25◦ C, b1, b2, b3 are constants derived from the temperature-difference matrix
inversion, and σ1 is a constant bias term. Further detail of the derivation of Equations
(4.7) and (4.8) may be found in Reference [27].
Conclusion
The Optical Remote Sensor Group at Montana State University has developed
algorithms that thoroughly correct for microbolometer temperature drift. Once the
greatest contributor of system error, this is now one of the smallest contributors to
error if adequate information is collected from the camera system. A more thorough
discussion of system error contributions is found in Chapters 5 and 6.
LWIR CAMERA RELATIVE SPECTRAL RESPONSE
Introduction
This chapter discusses how the relative spectral response (RSR) of the camera
can cause uncertainty in cloud characterization with infrared cloud imagers. The
spectral response of the camera is a product of the detector’s spectral response
function and the optics’ and filter’s spectral transmittance. For the purposes of this
discussion, it is assumed that the spectral response of the camera does not change with
temperature. Uncertainty in the RSR can cause error when calculating integrated
scene radiance.
In summer 2016 I worked with colleagues at the Space Dynamics Laboratory
(SDL) in Logan, Utah, to measure the spectral response of two different LWIR
cameras. SDL has performed spectral response measurements for infrared space
systems such as Wide-field Infrared Survey Explorer (WISE), Michelson Interfer-
ometer for Global High-resolution Thermospheric Imaging (MIGHTI), and Radiation
Budget Instrument (RBI) telescopes, and has developed a technique to measure the
spectral response of detectors using a Fourier Transform Spectrometer [28, 29]. The
measurements taken at SDL can be seen in Figure 5.1.
The uncertainty for the Tau2 and Photon camera spectral response measure-
ments were estimated to be ±10%. Using the measured spectral response as a known
RSR for a simulated camera, the effect of a shift of ±10% was then analyzed to
determine the corresponding calibration error in radiance.
32
33
Figure 5.1: FLIR Camera RSRs measured at SDL
34
Methodology
The following is a sensitivity analysis of the error in calculated radiance that are
caused by uncertainties in the relative spectral response function (RSR), including
RSR-induced errors in the calibration coefficients. The calibration coefficients are
determined from (in this simple case) a two-point calibration using blackbody targets.
The error in the derived coefficients are due to determining integrated radiance for
the blackbody targets when using an uncertain relative spectral response. A flowchart
of the overall simulation can be seen in Figure 5.2.
Figure 5.2: RSR simulation flowchart
This process includes assumptions that the emissivity of the blackbody is 1,
atmospheric transmittance is 1, and therefore the radiance from the blackbody can
be characterized perfectly by the Planck Formula. The reference spectral responsivity
of the simulated cameras are the relative spectral responses of the Tau2 and Photon
cameras collected at SDL. The camera is assumed to have no noise, and that the
camera has no temperature dependence for this simulation.
Relative spectral responses within the ±10% bounds were simulated to give 1000
statistically relevant RSRs. These RSRs were then used to simulate 1000 different
35
calibrations. Once the simulated calibrations were completed, measured cloud spectra
radiance scenes were used to test the error between known scene integrated radiance
and calculated scene integrated radiance within the bounds of error of the relative
spectral response measurement.
The RSR-induced calibration errors arise from incorrect calculations of how the
camera responds to wavelength. For a simple two-point calibration, one blackbody
target is used at two different temperatures. This incident radiation is expected to
follow the Planck Radiation formula for spectral radiance, already shown in Equation
(4.1). The uncertainty in the spectral response causes an error when integrated
with the expected spectral scene radiance to produce integrated scene radiance.
This integrated scene radiance is then used to determine the linear temperature-
independent gain of the camera.
To determine how this uncertainty in calibration coefficients affects measure-
ments, a simulated camera was used. This perfect camera was known to have the
response g and d, which were taken from an actual Tau2 calibration to give realistic
digital numbers. G and B were then calculated according to Equations (5.1) and (5.2).
The known digital number of the simulated camera were then calculated according
to Equation (5.3)
G =
1
g
(5.1)
D =
−d
g
(5.2)
D# = G
∫ λ2
λ1
Ls(λ)R(λ) dλ+D, (5.3)
36
where Ls(λ) is the spectral scene radiance, and R(λ) is the true relative spectral
response of the camera. This gave simulated digital numbers in response to known
scene spectral radiances. These simulated digital numbers were then used to calculate
integrated radiance using erroneous calibrations. The error between the known
integrated radiance and the simulated integrated radiance then gave an estimate
of the radiance uncertainty arising from the relative spectral response uncertainty.
The simulated-error RSRs were used to calculate the integrated radiance on the
focal plane array when viewing a perfect blackbody by multiplying the blackbody
spectral radiance by the simulated RSR. The spectral radiance was then integrated,
and these resulting integrated radiances were then used as data points in a two-point
calibration to derive gn and dn for each calibration. The values gn and dn were then
used to calculate the integrated radiance Lsim for each scene, as shown in Equation
(5.4). The simulated scenes where cloud spectra collected at the Atmospheric
Radiation Measurements (ARM) program site in Barrow, AK from July 2012 to
July 2014.
Lsim = gnD# + dn. (5.4)
The radiance difference between the known integrated radiance of the scene obtained
by integrating spectral radiance with the known RSR and the calculated integrated
radiance simulated with the perturbed RSR can then be shown as
∆L = Ls − Lsim. (5.5)
Simulated Tau2 Camera RSR
The spectral uncertainty was calculated using 1000 perturbed RSRs, using the
SDL RSR as a reference. The perturbed RSRs can be seen in Figure 5.3. The radiance
37
uncertainty for these simulated data is shown in Figure 5.4b to be 3.5% of the scene
radiance.
Figure 5.3: Tau2 camera measured and simulated RSRs
The simulated scenes were built from 2.4 million spectra measured of the Arctic
atmosphere. The scenes varied from 5 to 45 W/(m2 sr) integrated scene radiance.
The data shown in Figure 5.4a show the strongest bias between known and simulated
data to occur at the highest scene radiance values. This is due to a small change in
gain having the largest correction error for large digital numbers for a linear correction
term.
In order to equally compare uncertainty over the span of scenes, error in terms
of percent of scene radiance is shown in Figure 5.4b, where the uncertainty for the
coldest scene was calculated to be 3.5% for the Tau2, or 0.175 W/(m2 sr) uncertainty
in a 5 W/(m2 sr) scene.
38
(a) Tau2 known vs calculated integrated
scene radiance.
(b) Tau2 camera uncertainty in terms of
percent scene radiance.
Figure 5.4: Tau2 calibration error due to RSR uncertainty.
Simulated Photon Camera RSR
I also worked with colleagues at SDL to measure the RSR of a FLIR Photon
camera. Its spectral response is different from the Tau2, having a slower falloff for
higher wavelengths and less structure within the 10-12 µm region. The reference and
simulated RSRs can be seen in Figure 5.5.
Similar to the Tau2, the error in terms of radiance is largest at the hottest scenes,
shown in Figure 5.6a. Since the scenes of most interest are for clear sky and thin
clouds, the uncertainty terms were presented in percentage of scene radiance. Due to
the photon RSR having significantly less structure within the 10-12 µm region, the
calibrated radiance uncertainty arising from the spectral response uncertainty for a
cold scene is 4.7% uncertainty in a 5 W/(m2 sr) scene, seen in Figure 5.6b.
This uncertainty is 14% larger than the Tau2 for a cold scene. This could be due
to the Tau2 having a wider response than the Photon, where having more area under
the curve increased the calculated integrated radiance more than the Photon. This
39
Figure 5.5: Photon camera measured and simulated RSRs
wider response may have helped calibration uncertainty, but may also mean more
error in atmospheric retrieval due to temperature of CO2 for the Tau2.
Deviation from Manufacturer’s RSR
Results so far have focused on the error due to the uncertainty within the
RSR measurement. To see how well previous systems were calibrated, the error
in radiance was calculated between the SDL RSR and the typical RSR from the
manufacturer. Figure 5.7 shows the RSR measured at SDL and the RSR provided by
the manufacturer as a typical curve for both cameras.
The same calibration error analysis was done using the supplied FLIR RSRs for
the Photon and Tau2 cameras. As expected, the error seen due to the incorrect RSR
is larger than it was when perturbing the measured spectra within its uncertainty
bounds. The difference in integrated radiance can be seen in Figure 5.8. Error for the
40
(a) Photon known vs calculated integrated
scene radiance
(b) Photon camera uncertainty in terms of
percent scene radiance.
Figure 5.6: Photon calibration error due to RSR uncertainty
Photon camera was > 25% scene radiance at 5 W/(m2 sr), while error for the Tau2
was 10% scene radiance at 5 W/(m2 sr).
When scene radiances were plotted against each other in Figure 5.8, one can
hardly distinguish between the two for the Photon, while a definite bias can be seen
in the Tau2 data. Note that the Tau2 gain crosses the 1:1 line in Figure 5.8b. This
causes a zero percent error at 10 W/(m2 sr) in Figure 5.9b, which quickly goes to a
large error either side of this null point.
Conclusion
Knowing the RSR of the infrared cameras used in cloud imaging allowed for an
analysis of how much of the calibration uncertainty arises because of this parameter.
The Tau2 uncertainty due to RSR for cold scenes was found to be 0.175 W/(m2 sr),
and the Photon RSR uncertainty was found to be 0.235 W/(m2 sr) for cold scenes
assuming a ±10% RSR measurement uncertainty. In other words, a ±10% RSR
41
(a) Photon measured vs generic manufac-
turer RSR
(b) Tau2 measured vs generic manufacturer
RSR
Figure 5.7: RSR measurements vs manufacturer-specified spectra
measurement uncertainty caused a 4.7% calibration uncertainty for the Photon, and
a 3.5% calibration uncertainty for the Tau2 for cold scenes.
Both the generic spectral responses from FLIR were found to be out of the
uncertainty bounds of the SDL measurement. The differences between the SDL and
FLIR RSR showed a high sensitivity toward knowing the RSR for very cold scenes.
In other words, not measuring the RSR of the Photon camera would give a 33% error
for clear sky scenes, and a 15% error for the Tau2 camera. This points towards a
need to understand the RSR better than what is given for a generic production run
from the manufacturer if the characterization of thin clouds are to be prioritized.
42
(a) Photon RSR scene radiance compar-
isons
(b) Tau2 RSR scene radiance compar-
isons
Figure 5.8: Scatter plots of cloud scene radiances calculated with the SDL-measured
and manufacturer-specified RSR spectra for (a) Photon and (b) Tau2 cameras.
(a) Photon RSR calibration error (b) Tau2 RSR calibration uncertainty
Figure 5.9: Uncertainty in percentage of scene radiance due to generic RSR from
manufacturer
TAU2 CHARACTERIZATION
Introduction
This chapter is a discussion of the characterization of a FLIR Tau2 Camera.
Multiple issues were found in the calibration of this camera, even with respect to
previous cameras of the same type. This chapter will discuss these issues and show
the final calibration and characterization results.
In 2012 when the first Tau2 cameras were purchased, the Tau2 was the next-
generation microbolometer LWIR camera from FLIR. Since the MSU ICI sytems in
the mid-2000s had been using FLIR Photon cameras, it was natural to progress to
the Tau2. The Tau2 boasted a smaller pixel size with 17 µm pitch pixels, and was
a smaller, more lightweight camera. The largest driving factor, however, was that
FLIR Photon cameras were no longer being produced. This necessitated changing to
a different camera core. Even though the spatial resolution of the Tau2 was higher,
it was not necessarily a better camera for cloud imaging. The smaller pixel pitch and
larger focal length meant a smaller instantaneous field of view, decreasing radiometric
throughput and therefore the signal on the detector. This led to an increase in noise
with respect to the signal, or a lower signal-to-noise ratio (SNR).
In 2013, two ICI systems with Tau2 cameras were deployed to NASA JPL. A
year later, two more Tau2 cameras were purchased for NASA GRC. Between the
two purchases, FLIR upgraded the firmware of the Tau2 cameras. The following are
findings from calibrations and deployments completed after the firmware upgrade.
43
44
Low-signal Suppression and Shutter Offset Correction
When the GRC system was deployed at MSU during the initial testing period,
the camera would get quite hot without the assistance of a cooling system. Once
the focal plane temperature reached approximately 34 ◦C, the camera would report
digital values of zero. The firmware of the Tau2.7 (the upgraded firmware version of
the Tau2) had optimized the preprocessing algorithms internal to the camera for hot
scenes. Since the temperature of the scene being analyzed was significantly below the
temperature of the camera, the data were being filtered out with the noise, resulting in
blank sky images. The imager was also giving noticeable jumps in calculated radiance
with respect to small changes for certain camera temperatures. After discussion with
FLIR engineers, it was discovered that using the Tau2.7 in manual flat-field mode
(as we were) would automatically correct the digital number output in response to
camera temperature according to factory-set tables. This complicated the calibration
significantly.
A flat-field correction (FFC) is a non-uniformity correction that FLIR systems
applies using proprietary software. It fixes pixel-to-pixel deviations when viewing a
perfectly flat scene. The uncooled microbolometers will both drift and reduce contrast
over time, so FFCs need to be performed approximately every five minutes to fix these
fluctuations.
With the software update, internal correction tables were applied to the FFC
algorithm. This linked the shutter temperature to the FFC and could only be disabled
by applying an external FFC instead of a manual FFC. Previous systems used a
manual FFC that allowed us to have control over when the FFC of the camera
happened. The early versions did not preprocess gain and offset settings according to
internal tables for manual FFCs, allowing us to rely entirely on our custom calibration.
45
To resolve this problem an external FFC was used. This meant that either an external
shutter was used, or the internal camera shutter was told to to close so a FFC could
be performed as if the shutter were an external to the camera. With using an external
FFC, the internal tables were not used in the non-uniformity correction algorithm, so
FFC behavior was similar to previous systems.
External Flat-Field Correction
It was discovered through experimentation that clouds could be seen with high
focal plane temperatures by doing an external FFC on the sky. This very cold scene
caused the camera to increase its internal offset settings to the point where clouds
could be seen above the noise. This effect can be seen in Figure 6.1. The images
on the left show what happened when a manual FFC was used. Directly after the
image on the left was taken, the camera was switched to external FFC on the sky.
The images on the right were recorded after the external FFC was applied. These
images show fixed-pattern noise, but the skilled observer can see cloud patterns above
the noise. In this case, the clouds produced a very small signal. The external FFC
algorithm effectively subtracts the atmospheric signal (since the algorithm believes
the camera is looking at a flat scene), and differences from this flat scene are seen in
subsequent images. Using this method is very similar to scene-to-scene subtraction
used in some of our cloud detection algorithms.
This technique may be beneficial in the future for use in detecting clouds with
low optical depth. By doing a flat field correction on the sky, a scene subtraction is
accomplished. Assuming clear sky can be observed in the frame, anything different
in the next frame should be radiance from clouds as they move through the field
of view. Cloud determination would then need to be accomplished through image-
to-image subtraction. An advantage to this technique is that self-emission from the
46
(a) Manual flat field correction 35◦ C (b) External flat field correction 34◦ C
(c) Manual flat field correction 54◦ C (d) External flat field correction 51◦ C
Figure 6.1: Manual FFC vs External FFC at different FPA temperatures
camera body, camera optics, enclosure window, and atmospheric emission are all
being removed from the scene, so only changes in signal due to the changing scene
are being observed.
The main disadvantage of technique is that normal calibration is lost when the
gain and offset values change to unknown values. However, an in-situ calibration
may still be accomplished by taking an image of the shutter and doing a throughput
calculation according to the MSU-developed method discussed in Reference [23]. This
idea should be pursued in future studies, especially if the systems are meant to be
optimized for sub-visual cirrus cloud detection.
47
Another disadvantage of this technique is that every time an image is taken, an
external FFC must be done the frame before. The reason this is a problem is that
our previous experience showed that the shutter will wear out rapidly if used more
often than approximately once every 5 minutes. If the scene does contain clouds, the
scene will be changing over time as the clouds move across the scene. This scene
is obviously not flat, so the cloud will show up as a negative radiance as scene-to-
scene subtraction is performed. If the cloud moves and is replaced by another cloud
of similar optical depth, the scene will not seem to change. A compromise to this
technique may be to use the FFC from the shutter for the first 2.5 minutes, then flat
field on the sky and take data for 2 minutes more. When dropping the shutter at the
end of the 2 minutes of the sky external FFC data set, an image of the shutter can
be taken before the next FFC is performed to establish throughput calibration. This
proposed collection routine is visualized in Figure 6.2.
Figure 6.2: Proposed external FFC data collection routine
External FFC of a False-Temperature Internal Shutter
Since it is desirable to have an absolute calibration, another method was pursued.
The internal gain and offset of the camera were observed to change dependent on the
scene observed. If the camera did an external FFC on the sky, which is much colder
than the actual shutter, the gain and offset were automatically changed to allow
48
for small cloud signals. Since this change depended on the algorithm knowing the
temperature of the scene, changing this temperature input should also change the
internal offset and gain.
NWB Sensors R© had developed a technique of doing an external flat-field
correction on the shutter, and fixing the shutter temperature to a set value so the
internal gain and offset tables of the camera would not change. I found out about
this technique through discussion with NWB engineers [30], and decided to try the
technique for cloud imaging applications. Instead of fixing the shutter temperature
to near the camera temperature, I told the camera that the shutter temperature was
actually much higher than it really was. I first tried telling the camera that the
shutter temperature was 80 ◦C. This moved the offset of the camera up enough to
allow the cloud data to be above the threshold of the noise-compensation algorithm,
while keeping the internal offset and gain constant throughout the measurement.
Data could then be calibrated through MSU’s standard algorithm.
Tau2 Calibration Results
Using MSU’s standard algorithm for FPA temperature correction and digital-
number-to-radiance calibration, the data were processed from the experiment de-
scribed next. A large-area blackbody source was set to six different temperatures
while the camera was set inside a thermal chamber set to ”soak” the camera at a
steady temperature. This was repeated for five camera temperatures. The camera
was then put in a ramp environment, where the environmental chamber temperature
(hence the camera temperature) was ramped up and down while the camera viewed
the source.
The results of this process are shown in Figure 6.3. The blue line shows
the blackbody source radiance, which was calculated by integrating the Planck
49
function over the camera’s relative spectral response function for the blackbody source
temperature. The yellow-orange line shows the camera’s output converted to radiance
with a standard two-point radiometric calibration, but without any correction for
the drifting FPA temperature. The red line shows the radiance values obtained
from the two-point calibration after the FPA-temperature variation compensation
was applied. The blackbody radiance was calculated, and then the compensation
mechanisms corrected for radiance error. The uncertainty of this correction can be
seen in Figure 6.4.
Figure 6.3: Corrected radiance for a Tau2 Camera using MSU standard algorithm
over time in comparison with uncorrected and source radiance. Images were collected
every 20 seconds.
Source: Reference [7]
50
Figure 6.4: Distribution of error in calibrated radiance relative to the source radiance
Source: Reference [7]
Noise-equivalent Radiance
Sensitivity of the ICI system depends on the Noise Equivalent Radiance (NER)
at the focal plane array. Previous ICI systems have been able to discriminate sub-
visual cirrus clouds above pixel noise. A subvisual cloud can be defined as ”any
cloud that cannot be detected” by a lidar system or otherwise and has been roughly
approximated to a cloud optical depth of 0.03 [31].
NER is calculated by determining the pixel-to-pixel variation over time. This is
done while the camera is viewing a flat scene, and the images are taken as quickly as
possible to minimize any thermal changes within the camera over the image collection
period. For this test, the camera was at room temperature and 200 images were taken
as quickly as possible (60-Hz). Each pixel in the image was then evaluated over 200
images. The standard deviation in digital number was then calculated for each image
set to determine pixel-to-pixel noise. Image sets were taken every 5 minutes for
51
one hour. The pixel noise was then converted to radiance after the calibration was
completed, and the mean of this noise in radiance was reported as NER.
The NER for the Svalbard ICI system at room temperature has been calculated
to be 0.043 W/(m2sr) using an extended source. This NER is equivalent to a cirrus
cloud of 0.004 optical depth, located at 8-km altitude. This is significantly below
subvisual cirrus as defined previously. Such a cloud would have a transmission loss
of 0.3% at 550 nm. A cloud this thin would mostly likely be undetectable due to
atmospheric transmittance from ground to 8 km.
It should also be noted that the system uncertainty is more than one order of
magnitude higher than the NER. This means that although we can see thin clouds
above pixel noise, we can only classify that their radiance is below the radiometric
system uncertainty.
The radiometric system uncertainty for an ICI system can be calculated using the
method described in Reference [27]. The uncertainty of the radiometric calibration
can be calculated to be the Root Sum Square (RSS) of the uncertainty of the
radiometric camera correction (how well we can correct for temperature drift of the
FPA), the uncertainty of the RSR, and the uncertainty of the source. This is then
RSSed with the uncertainty of the atmospheric model for the deployment site to give
the total system uncertainty. Because of this system uncertainty, cloud statistics are
binned into 7 discrete cloud thicknesses [19]. The discussion of NER analysis follows
from previous work by the author [7].
Conclusion
The Tau2 camera had significant challenges associated with its characterization
for cloud imaging. However, once the internal compensation algorithms were
52
overcome, a standard FPA temperature-compensation algorithm could be applied
to the camera.
Since the uncertainty of this calibration is the RSS error of the RSR uncertainty
and the radiometric correction uncertainty (source uncertainty is negligible), the total
calibration uncertainty can be calculated. For the Tau2 camera, RSR uncertainty
error was 0.175 W/(m2 sr) from Chapter 5 and radiometric correction uncertainty
was calculated to be 0.50 W/(m2 sr) in this chapter, so total calibration uncertainty
was calculated to be 0.53 W/(m2 sr). When calculating the calibration uncertainty
for the Photon, the radiometric correction uncertainty was 0.29 W/(m2 sr) [27] and
the RSR uncertainty was 0.24 W/(m2 sr) from Chapter 5. This resulted in a total
calibration uncertainty of 0.38 W/(m2 sr) for the ICI3 Photon. This shows that while
the Photon calibration was still better than the Tau2, the error from the Photon was
only 16% better than the error of the Tau2.
If the RSR was not measured, but instead the manufacturer RSR was used,
this would cause an uncertainty of 33% for the Photon and 15% error for the Tau2.
This would cause an error in integrated radiance for a cold scene of 1.65 W/(m2 sr)
for a Photon camera, and an error of 0.75 W/(m2 sr) for the Tau2. This shows
that if not properly measured, RSR uncertainty can dominate radiometric calibration
uncertainty. This is an important result, since before this work the RSR had not been
measured and only an assumed RSR was available [27]. The measured RSR for these
cameras should be considered camera specific. Different lens configurations, lens
manufactures, and camera core production runs will cause error in the RSR.
Even though the calibration accuracy might be similar for these two systems,
the Photon still will have better radiometric throughput and less noise than a Tau2,
making it a better camera core for cloud imaging.
TAMARISK CHARACTERIZATION
Introduction
This chapter discusses the preliminary investigation of a DRS R© thermal camera
for use in cloud imaging. This was an uncooled microbolometer camera with a 17-
micron-pitch focal plane array. The goal was to investigate this camera as a potential
replacement for the similarly configured Tau2 discussed in Chapter 6. A camera
was loaned to Montana State University for evaluation with regard to infrared cloud
imaging. The first sky image taken with this camera can be seen in Figure 7.1.
Figure 7.1: Tamarisk camera first sky image, with tripod and operator in view. The
colorbar shows digital number output of the FPA
The evaluation camera was a DRS R© Tamarisk camera core packaged by Sierra
Olympic R© to include a framegrabber interface. Sierra Olympic R© referred to the
53
54
camera as a Viento. The Viento 640 was a 640×480 30-Hz camera with a f/1.4 lens.
The field of view was 90◦ × 67◦ with a 2.45 milliradian iFOV.
Calibration Algorithm
The Tamarisk core does not have a thermal sensor on the FPA, but uses the
digital number from a masked pixel to determine the calibration tables for thermal
compensation. Due to the time constraint of the evaluation period, we were only
able to calibrate with soak temperatures (with no temperature ramping). Therefore,
the legacy focal plane temperature correction could not be used in determining focal
plane array temperature response. To ensure the digital number response was linear
with camera temperature in this range, output digital numbers were plotted against
camera temperatures measured with a thermocouple placed on the camera housing.
Figure 7.2: Tamarisk FPA temperature sensor output has a fairly linear response
with external camera temperature
55
To compensate for FPA temperature response, a system matrix was developed
to determine the response of the camera with respect to scene radiance and focal
plane array temperature. The ability to linearize the data in this temperature range
was proven by plotting the data on a three-axis scatter plot, as shown in Figure 7.3.
Figure 7.3: Tamarisk digital FPA temperature, digital number response, and scene
radiance can be solved with a linear regression
The plane in Figure 7.3 was then solved for using a linear regression based
upon Equation 7.1, where A is camera-temperature-independent gain, B is camera
temperature dependent gain, C is a mixing term of A and B, and D is an offset
term. In order to reduce calibration error, B held both FPA temperature and case
temperature terms. Equation (7.1) was then solved through a matrix inversion.
56
[
Lbb
]
=
[
A B C D
]
×

M1
M2
M3
M4

(7.1)
It should be noted that this inversion may be unstable for temperatures and
radiances outside of the bounds of the collected data in the calibration. For instance,
low radiance corrections may not be accurate due to extrapolation beyond the bounds
measured with our laboratory blackbodies. This problem would be minimized by
collecting ramp data and applying temperature difference algorithms discussed in
previous chapters. The time it would take to collect these data, however, was beyond
the evaluation time given for the camera.
Calibration Results
The image shown in 7.4 is the standard deviation of the radiance error in each
pixel when comparing the calibration results to the blackbody. The high uncertainty
in the corners of the image are due to clipping of the window of the blackbody. Since
the Tamarisk has a 90◦ × 67◦ FOV for x and y directions, the diagonal FOV of the
image is actually 112◦. These large angles are difficult to calibrate with a limited-
area blackbody source. The camera is required to be fairly close to the blackbody,
which can both cause heating of the lens and reflection of lens emission from the
blackbody surface (since the blackbody emissivity is not 1). Also, it has been shown
that emissivity of a blackbody source will change with large field angles [32]. However,
these effects were ignored for this calibration since their effects are small compared
to the uncertainty in the FPA correction.
57
Figure 7.4: Radiance correction standard deviation for Tamarisk camera over FPA
Conclusion
The initial calibration for the Tamarisk camera proved successful and shows a
standard deviation of 1.44 W/(m2sr) with a distribution over the scene shown in
Figure 7.4. With more temperature data, this would most likely be reduced due to a
more accurate FPA correction, but it shows that calibration of the Tamarisk camera
is possible and will give reasonable results.
The initial Tamarisk calibration has shown that the camera core performs very
similarly to the Tau2. The advantages to the system are that the internal calibration
tables and settings are more accessible than the Tau2 system. For instance, the
Tamarisk core will report back which internal tables it is using in its non-uniformity
corrections. However, the Tamarisk was not tested for temperatures above 32 ◦C.
As shown in Figure 7.2, when camera temperature compensation is disabled, the
58
temperature sensor only reported temperatures between 13 ◦C and 32 ◦C. This may be
solvable through more serial commands, but it was not investigated for this purpose.
Another solution might be to enable the temperature compensation and do
separate calibrations for each table. Both the Tau2 and Tamarisk cores have shown
that microbolometer arrays are less sensitive to cold scenes (like clear sky) when FPA
temperature is above 32 ◦C.
A disadvantage to the Tamarisk core is that it only has a temperature sensor on
the FPA and not on the case as well. This may mean that even with a more thorough
calibration, external sensors will be needed to added to perform the same tasks as
the Tau2. However, if future systems will continue to strive for thin cloud detection,
external temperature sensors should be used anyway on the case as well as the lens
to further characterize self-emission from the camera.
This chapter has shown that calibration of the Tamarisk is possible, and the
system seems to perform very similarly to the Tau2. However, the Tamarisk cores are
currently restricted to have a 17 µm pixel pitch. Since the Tamarisk and Tau2 cores
seem to be so close in performance, I would advocate for choosing a Tau2 324 over a
Tamarisk 640 in future systems. The Tau2 324 has 25 µm pitch with a 324×256 array,
with a 63◦ × 50◦ FOV and a 3.333 iFOV. The larger iFOV will increase throughput,
which should mean the Tau2 324 is currently a better camera for cloud detection than
either the Tau2 640 or Tamarisk 640 cores.
SYSTEM COMPARISONS
Introduction
This chapter discusses the performance parameters of different ICI systems as
the unit under test when compared with a legacy system, the ICI3 cloud imager. The
ICI3 system has been validated against visible cloud imaging instruments and other
remote sensing equipment at the ARM Research Site in Barrow, AK [27], so it is
well understood. This system provides a reliable real-time data set for side-by-side
comparisons with newer systems.
Photon vs Tau2
The Tau2 Svalbard system was placed beside the Photon ICI3 system for
validation in January 2016. The Svalbard system was to deploy in Cleveland, OH,
and the calibration needed to be validated against a known system. This comparison
also was reported in previous work by the author [7].
Figure 8.1: Svalbard ICI (G02) and ICI3 reported dB compared over time.
Source: Reference [7]
59
60
(a) dB Scatter Plot (b) dB difference distribution
Figure 8.2: Statistical analysis of instrument comparisons.
Source: Reference [7]
The Tau2 and Photon cameras were deployed with drastically different fields
of view. In order to adequately compare collected data, both fields of view were
decreased to 10◦ through software. Pointing error required a time shift in the data
and a low pass filter applied to the data. The ICI3 uses a hatch system that closes
whenever it rains, so data taken while the hatch was closed were disregarded. Both
imagers were compared over a two-day period, as seen in Figure 8.1.
Data show good agreement during the comparison period. Figure 8.2 shows that
dB values from each instrument had a correlation of 0.978. The standard deviation
of error between the two instruments is 1.13 dB. The Tau2 reported less attenuation
during transient periods than the ICI3 using a full-field average. This is most likely
due to error in completely matching the fields of view.
The distribution in Figure 8.2(b) shows a slight bias in small dB measurements,
which correlates to radiances of relatively low optical depth. This bias may be due
to uncertainty in the spectral responsivity of the two cameras.
61
Photon vs Tamarisk
The Photon and Tamarisk cameras were compared side-by-side for half of a day
in May 2016. The Tamarisk was at Montana State University under evaluation for
a short time, so a short calibration was performed, followed by a short comparison
period. The purpose of the evaluation was to determine if the Tamarisk camera could
be calibrated and if it had a low enough NER to observe thin clouds.
Figure 8.3: Photon (left) and Tamarisk (right) ICI systems deployed side by side
In Figure 8.3 the ICI3 enclosure can be seen on the left. An astute observer
can see that the sliding hatch is open and the ICI3 is taking data at the time this
photograph was taken. The Tamarisk camera was smaller in size to the Photon, but
a large legacy container (from the first ICI deployment to Barrow) was used to house
the electronics. The Tamarisk enclosure can be seen on the right of Figure 8.3. The
Tamarisk lens protruded from a PVC enclosure. A defocus was inadvertently put on
62
the camera as a result of the tight fit between the camera lens and the enclosure.
While image quality was degraded during this test period, radiometry still remained
within acceptable parameters.
Image Comparison
Both the Photon and Tamarisk cameras had 90◦ degree fields of view, so the
images matched much better than the previous test with the Photon and Tau2. Side-
by-side images of the same cloud can be seen in Figure 8.4. There is a bit of an
orientation error between the two cameras, but the cloud structure in the images
clearly shows the same cloud. While Figure 8.4 shows a fairly thick cloud, Figure
8.5 shows a relatively clear sky. Using scene-to-scene subtraction, clouds can be seen
traversing the clear sky.
Uncertainty of the radiometric correction of the Tamarisk camera was 1.44
W/(m2 sr), and the uncertainty of the RSR is unknown. Even if the uncertainty of the
calibration was 1.44 W/(m2 sr), one can see from the scene-to-scene subtracted image
in Figure 8.5d that thin clouds less than this uncertainty can be seen. These clouds
would be classified as simply being lower than the uncertainty of the calibration.
Radiometric Comparison
Scene radiance was compared over time using averaged scene radiance of the
images. The images were masked to remove the effect of buildings in the field of
view.
The standard deviation of average scene difference between the two cameras is
shown in Figure 8.6 as 0.71 W/(m2 sr), which is impressive considering the accuracy
of the radiometric correction for the Tamarisk camera was 1.44 W/(m2 sr). In
other words, the reported standard deviation between the two cameras is half of the
63
(a) ICI3 Photon
(b) Viento Tamarisk
Figure 8.4: Radiometric image comparison of cloudy sky
64
(a) ICI3 Photon (b) Viento Tamarisk
(c) ICI3 Photon scene subtraction (d) Viento Tamarisk scene subtraction
Figure 8.5: Thin clouds can be seen above the noise when doing scene-to-scene
subtraction
65
Figure 8.6: Distribution of scene comparisons during deployment period
uncertainty of the Tamarisk, which means they compare very well. The correlation
between the two systems is shown in Figure 8.7 to be 0.996.
Conclusion
The Tau2 and ICI3 systems compared fairly well over the course of the three-
day period in terms of optical depth. Since the FOV was dramatically smaller than
the FOV of the ICI3 system, both systems had their FOV reduced to 10◦. The two
systems had a correlation of 0.978 for reported dB values. Although the standard
deviation of the reported differences was 1.13 dB between the ICI3 and Svalbard ICI,
this may have been due to the different RSR values between the Photon and Tau2
imagers. Pointing errors may have also contributed to this error, since both analysis
field of views were so narrow. The Svalbard ICI system should be calibrated well
enough to give the needed cloud statistics for the Svalbard site.
66
Figure 8.7: Scatter plot showing correlation of 0.996
The Viento and ICI3 data compared better than the Tau2 and the ICI3 in terms
of radiance. This may have been due to the large FOV contributing to a better-
matching average scene. Since the FOV was so large for both systems, small changes
in the scene would not affect the reported radiance as much as it would in comparing
two narrow-field-of-view instruments. Even with a 17 µm pitch, the Tamarisk camera
had a shorter focal length, contributing to a larger iFOV than the Tau2 system; this
produced a more closely matched throughput. These contributing factors may make
the Tamarisk appear to be a better system than the Tau2, even with a worse FPA
temperature-correction. Another consideration is that the Tamarisk and ICI3 systems
were only compared over a few hours, so the data may be biased for certain cloud
conditions. Overall, both Tau2 and Tamarisk cameras compared well to the ICI3 and
should be considered to be viable options for infrared cloud imaging.
CLOUD PHASE DETECTION
Introduction
A thermal infrared image contains information about cloud optical depth up to
a value of approximately 4. Optically thicker clouds simply behave as a blackbody,
but optically thinner clouds emit thermal radiation that varies predictably with cloud
optical depth [31]. However, recent comparisons of cloud optical depth derived from
ICI images and lidar data show that the data tend to accumulate into two clumps.
Since the ICI cloud optical depth algorithms were developed for cirrus clouds, it is
possible that there is a systematic difference between the results for clouds made of
ice or liquid water.
In addition to the question of how cloud phase affects the cloud optical depth
retrievals from an ICI system, thermodynamic phase is used in atmospheric physics to
determine cloud formation mechanics [33,34]. Cloud phase also has a large impact on
attenuation and beam polarization. For instance, for KA band propagation, ice clouds
have an almost negligible attenuation coefficient and only affect polarization in this
band, while suspended water particles (clouds) have a large attenuation coefficient
approximated through Rayleigh scattering [35]. This could lead to an instrument
that would support KA propagation studies even further at NASA GRC’s NEN site.
From our perspective as remote sensing instrument designers and users, this presents
an opportunity to explore some possible ways of remotely sensing cloud phase with
passive imagers.
Therefore, this chapter describes a study that was undertaken to assess the
possibility of using passive short-wave infrared (SWIR) imaging to determine cloud
phase. The hope would be that this could eventually lead to a new instrument to
operate alongside a long-wave infrared ICI to obtain enhanced spatial and optical
67
68
information about clouds. The satellite remote sensing community has developed
methods to estimate cloud phase from SWIR radiance measurements [36–41]. There
have also been investigations into using the LWIR for cloud phase determination [42].
These methods are based on differences in the refractive index of liquid water and
ice, as indicated in Figure 9.1.
Figure 9.1: The imaginary part of the index of refraction is plotted against
wavelength. Arrows show the differences at 1.64 µm and 1.7 µm.
Source: Data are from [43] for liquid water and [44] for ice.
In order to classify cloud phase, spectral shape can be used to determine if the
cloud consists of ice or water [37]. The imaginary part of the index of refraction is
spectrally dependent, as seen in Figure 9.1 [37]. Since attenuation depends on the
imaginary part of the index of refraction, two clouds with different phases illuminated
by the same source will attenuate the illumination differently. There is a distinct slope
difference between ice and water refractive indices. To simulate this effect for different
cloud types, a radiative transfer code was used to simulate clouds according to the
69
observation constraints seen in Figure 9.2. The sun is behind the observer at 180◦
azimuth, 45◦ elevation. The illuminated cloud is viewed in front of the observer at
45◦ elevation.
Figure 9.2: Observation of clouds simulated in MODTRAN
The simulated clouds varied in cloud height and droplet distribution according
to six standard cloud types. Extinction was then varied to simulate different
concentrations of water or ice in the cloud. Cirrus was simulated as a pure ice
cloud. All other clouds contain only liquid water. The clouds shown use MODTRAN
simulated cloud profiles for cumulus, altostratus, stratus, stratus&stratocumulus,
nimbostratus and cirrus. The optical depth of each cloud type was varied by
keeping cloud altitude and thickness the same as a standard cloud and modifying
the extinction. The cirrus (ice) clouds show a distinct slope difference from 1500 to
1700 nm, which leads to discrimination of these clouds vs all others simulated. The
spectral shape difference vs cloud type for the same optical depth can be seen in
Figure 9.3. This slope can be characterized by taking apectral at narrow bandwidths
70
for different regions, integrating them, and taking the ratio between regions for best
classification algorithm.
Figure 9.3: Sunlight scattered from simulated simulated ice clouds (cirrus) have a
much different spectral shape than liquid water clouds
There are small spectral features seen in Figure 9.3 near 1600 nm that are due to
CO2 absorption. The assumption with the following algorithms are that CO2 content
will be constant between measurements of ice clouds and water clouds.
Algorithm Development
Several different cloud-phase calculation algorithms have been tested using
simulated MODTRAN data. Most of these algorithms broke down when clouds
reached an optical depth less than 1. The Knap et al (2002) method seemed to be the
71
Figure 9.4: The Knap (2002) algorithm allows ice clouds to be distinguished from
liquid clouds in MODTRAN simulations
most separable algorithm for phase determination using 150-nm bandwidths for the
1700 and 1640-nm center wavelengths. The ratio of integrated radiances for different
cloud types is plotted against optical depth in Figure 9.4 to show the effectiveness of
this algorithm with simulated data.
Precipitable water vapor (PWV) can cause absorption in the spectral region of
interest, so the standard 1976 atmosphere PWV was scaled by 50% to 150% in 10%
increments to see the effect of PWV on the algorithm. Figure 9.5 shows the algorithm
applied to these different PWV levels. The trend of the ratio does not change with
PWV, but some smearing starts to occur at very small optical depths. This is due
to both curves converging to clear sky, and the differences in radiance becomes very
small.
72
Figure 9.5: PWV has little affect on the algorithm for OD > 1.
73
Figure 9.6: The spectral ratios between ice and water clouds become less discrete
after gaussian 150 nm bandwidth filters are used.
Algorithm Comparisons
The algorithm was tested using simulated data and applying gaussian band-
widths from 10 to 500 nm for 1640 and 1700 nm center wavelengths. Figure 9.6
shows how the ratios degrade as filter bandwidth is increased. The data in red are
cirrus (ice) clouds. All other clouds are water or mixed phase. The different lines for
each color annotate different optical depths.
In order to improve separability between ice cloud data and water cloud data,
3 vectors were used; the (1700-1640)/1640 weighted integrated radiance difference,
the (1550-1640)/1550 weighted integrated radiance difference, and optical depth.
74
Figure 9.7: Adding a third integrated radiance channel allows for greater separation
of data vs optical depth.
The data from this algorithm can be visualized in Figure 9.7 This shows improved
separability for precipitable water vapor, and also shows a convergence for all cloud
types to clear sky (the circles on the far left of the plot). This algorithm shows that
by adding a third channel, greater separability between ice and water data can be
achieved. This algorithm requires optical depth to be known.
Since cloud optical depth is not known without a lidar support instrument or
other instrument to measure cloud phase, another numeric needed to be used for
a stand-alone radiometer system. Since three integrated bands are being used to
determine cloud phase, the integrated value of each band can be evaluated as a vector,
and these vectors can be summed in 3D space such as an RGB image. The magnitude
of this vector is therefore sqrt(A2 +B2 +C2), where A is integrated radiance of 1550
nm center wavelength, B is integrated radiance of 1640 nm center wavelength, and
75
Figure 9.8: Optical Depth can be replaced by the RMS addition of the radiance from
each channel
C is integrated radiance of 1700 nm center wavelength. The magnitude of this vector
is then used instead of optical depth for determining cloud phase. This data can be
visualized using Figure 9.8.
If only the magnitude of the 1640 nm channel is known, and the relative
magnitude of the 1700 nm channel is known when compared to the 1640 channel,
the following algorithm can be used. This is by far the simplest method explored.
Since the radiance from a cirrus cloud stays much lower than the radiance from any
other cloud type at the same optical depth, this parameter can be used to separate
cloud phase for clouds with an optical depth of 0.1 or more. Using simulated data,
this has higher separability using magnitude instead of cloud optical depth and does
76
Figure 9.9: Two-channel algorithm using calibrated radiance values
not require the use of a support instrument. This algorithm could be used when a
third channel is not available. This algorithm is visualized in Figure 9.9.
Conclusion
Using MODTRAN simulations of different cloud types, it has been shown that
cloud phase can be determined using the ratios between two different wavelength
channels. Specifically, the center wavelengths that show the most promise are 1550,
1640, and 1700 nm using 150 nm bandwidths.
CONCLUSIONS
This thesis described how five major challenges were overcome relating to
infrared cloud imaging systems. The first challenge involved modifying an existing
enclosure to perform in multiple environments. The second challenge was to overcome
the loss of data collection imposed by OEM manufacturer firmware. The third
challenge was to measure and show the impact of a relative spectral response function
for a LWIR camera. The fourth challenge was to calibrate and compare different
LWIR cameras for infrared cloud imaging studies. The fifth and final challenge was
to analyze how cloud phase could be determined to update cloud optical depth models
derived for ICI systems.
Chapter two discussed some of the methodology behind infrared cloud imaging,
and some of the physical mechanisms that enable the technology. Chapter three de-
scribed two ICI systems that integrated OEM components into a research instrument.
The constraint of the enclosure used by NASA JPL caused significant problems in
terms of routine maintenance, instrument accessibility, and air flow. The JPL tube
system was ultimately abandoned by NASA GRC in favor of a legacy environmental
enclosure. Chapter four provided a brief introduction to radiometric calibration and
the calibration algorithms previously developed at Montana State University. This
enabled the reader to have a thorough understanding of later chapters.
The effect of relative spectral response uncertainty had mostly been ignored in
previous ICI systems due to the belief that the FPA temperature correction was
the largest source of error in the calibration. Measurement of instrument RSR
and analysis of measurement error produced surprising results. If left unmeasured,
calibration error due to uncertainty in manufacturer published spectra is by far
77
78
the largest contributor to error in an ICI system (approximately 35% error). The
measurement process and uncertainty analysis was discussed in chapter five.
Since research experiments are not typically the normal customer for OEM
developers, sometimes optimization for certain customers results in the OEM product
failing to perform well for research applications. This was the case for the Tau2
camera discussed in chapter six, which had a firmware upgrade on the preprocessing
unit inside the camera core. This upgrade caused the preprocessor to reject the
signal from clouds as noise. Through perseverance and collaboration, a method was
discovered to not only favorably bias the camera, but possibly enhance the cloud
detection capabilities of the Tau2.
Due to the difficulties and intricacies involved in calibrating the Tau2, another
microbolometer camera was evaluated to see how it compared. The new camera core
was called the Tamarisk, and it performed much like the Tau2 in the initial evaluation
period.
Chapter eight compared three different cloud imaging systems, the ICI3 photon,
the Svalbard ICI Tau2, and the Viento Tamarisk. The ICI3 system was used as a
benchmark instrument due to its long characterization history and other comparison
tests. The Svalbard ICI was deemed ready for deployment; and the Tamarisk camera
showed that it not only matched ICI3 data very well, it was also able to detect thin
clouds.
Chapter nine is an analysis for future radiometer units to use alongside ICI
systems in order to improve cloud optical depth retrievals. Three different center
wavelengths were chosen as a basis for future instrumentation. Cloud phase will be
an important part of cloud classification algorithms, both for optical depth retrieval
and expected attenuation estimates.
79
Cloud classification is an important part of propagation studies in the mm
and optical wavelengths as it becomes a larger focus for NASA earth-to-satellite
communication links. The work in this thesis showed hardware development, system
characterization, and algorithm development for cloud classification systems. This
work can be used as a baseline for relative spectral response analysis for analyzing
ICI atmospheric model sensitivity, as well as future cloud phase radiometer systems.
Future ICI systems should investigate the use of filters to reduce the affect of O3 and
CO2 emission in integrated atmospheric radiance.
Even though MSU has been able to model and correct for FPA temperature
fluctuations in microbolometer cameras, future systems should also employ the use
of environmental enclosures optimized for temperature stability because this will
allow the temperature-stabilization routines to work better over a smaller range of
camera temperatures. This will decrease the time involved in the calibration process,
and possibly allow for detection of thinner clouds. Environmental enclosures should
be large enough to allow access for periodic maintenance and troubleshooting. An
optimal configuration would consist of a large panel that allowed access to test points.
As systems are miniaturized, care must be taken to ensure that systems remain
accessible to minimize maintenance time.
80
REFERENCES CITED
[1] D. M. Cornwell, “NASA’s optical communications program for 2015 and
beyond,” in SPIE LASE, pp. 93540E–93540E, International Society for Optics
and Photonics, 2015.
[2] S. Shambayati, D. Morabito, J. S. Border, F. Davarian, D. Lee, R. Mendoza,
M. Britcliffe, and E. Weinreb, “Mars Reconnaissance Orbiter Ka-band (32 ghz)
demonstration: Cruise phase operations, paper presented at AIAA SpaceOps
Conference, Am. Inst. of Aeronaut. and Astronaut,” 2006.
[3] D. E. Raible and A. G. Hylton, “Integrated RF/optical interplanetary networking
preliminary explorations and empirical results,” in 30th AIAA International
Communications Satellite Systems Conference, Ottawa, Canada, 2012.
[4] Z. Sodnik, B. Furch, and H. Lutz, “Free-space laser communication activities in
Europe: SILEX and beyond,” in LEOS 2006-19th Annual Meeting of the IEEE
Lasers and Electro-Optics Society, pp. 78–79, IEEE, 2006.
[5] H. E. Green, “Propagation impairment on Ka-band SATCOM links in tropical
and equatorial regions,” IEEE Antennas and Propagation Magazine, vol. 46,
no. 2, pp. 31–45, 2004.
[6] J. Nessel, J. Morse, and M. Zemba, “Results from two years of Ka-band propa-
gation characterization at Svalbard, Norway,” in The 8th European Conference
on Antennas and Propagation (EuCAP 2014), pp. 3511–3515, IEEE, 2014.
[7] D. W. Riesland, P. W. Nugent, J. A. Shaw, M. J. Zemba, and J. Houts, “Infrared
cloud imaging in support of Ka-Band progagation studies,” in AIAA Proc.
International Communications Satellite Systems Conference, (Cleveland, OH),
2016.
[8] C. N. Long, J. M. Sabburg, J. Calbo´, and D. Page`s, “Retrieving cloud
characteristics from ground-based daytime color all-sky images,” Journal of
Atmospheric and Oceanic Technology, vol. 23, no. 5, pp. 633–652, 2006.
[9] B. Thurairajah and J. A. Shaw, “Cloud statistics measured with the infrared
cloud imager (ICI),” IEEE Transactions on Geoscience and Remote Sensing,
vol. 43, no. 9, pp. 2000–2007, 2005.
[10] J. R. Campbell, D. L. Hlavka, E. J. Welton, C. J. Flynn, D. D. Turner,
J. D. Spinhirne, V. S. Scott III, and I. Hwang, “Full-time, eye-safe cloud
and aerosol lidar observation at atmospheric radiation measurement program
sites: Instruments and data processing,” Journal of Atmospheric and Oceanic
Technology, vol. 19, no. 4, pp. 431–442, 2002.
81
[11] E. E. Clothiaux, K. P. Moran, B. E. Martner, T. P. Ackerman, G. G. Mace,
T. Uttal, J. H. Mather, K. B. Widener, M. A. Miller, and D. J. Rodriguez, “The
atmospheric radiation measurement program cloud radars: operational modes,”
Journal of Atmospheric and Oceanic Technology, vol. 16, no. 7, pp. 819–827,
1999.
[12] V. Mattioli, P. Basili, S. Bonafoni, P. Ciotti, and E. Westwater, “Analysis and
improvements of cloud models for propagation studies,” Radio Science, vol. 44,
no. 2, 2009.
[13] J. A. Shaw, P. W. Nugent, N. J. Pust, B. Thurairajah, and K. Mizutani,
“Radiometric cloud imaging with an uncooled microbolometer thermal infrared
camera,” Opt. Express, vol. 13, pp. 5807–5817, Jul 2005.
[14] J. A. Shaw and P. W. Nugent, “Physics principles in radiometric infrared imaging
of clouds in the atmosphere,” European Journal of Physics, vol. 34, no. 6, p. S111,
2013.
[15] P. W. Kruse, Uncooled thermal imaging: arrays, systems, and applications,
vol. 2003. SPIE press Bellingham, WA, 2001.
[16] P. W. Kruse and D. Skatrud, Uncooled Infrared Imaging Arrays and Systems.
No. v. 47 in Semiconductors and semimetals, Academic Press, 1997.
[17] M. Vollmer and K.-P. Mo¨llmann, Infrared thermal imaging: fundamentals,
research and applications. John Wiley & Sons, 2010.
[18] Personal Communication. Dr. J.A. Shaw.
[19] P. W. Nugent, J. A. Shaw, and S. Piazzolla, “Infrared cloud imager development
for atmospheric optical communication characterization, and measurements at
the JPL Table Mountain Facility,” InterPlanetary Network Progress Report,
vol. 42, no. 192, pp. 1–31, 2013.
[20] B. N. Holben, T. Eck, I. Slutsker, D. Tanre, J. Buis, A. Setzer, E. Vermote,
J. A. Reagan, Y. Kaufman, T. Nakajima, et al., “AERONET-A federated
instrument network and data archive for aerosol characterization,” Remote
sensing of environment, vol. 66, no. 1, pp. 1–16, 1998.
[21] J. A. Shaw, P. W. Nugent, N. J. Pust, B. J. Redman, and S. Piazzolla, “Cloud
optical depth measured with ground-based uncooled infrared imagers,” in Proc.
SPIE, vol. 8523, p. 85231D, 2012.
[22] A. Berk, G. P. Anderson, P. K. Acharya, L. S. Bernstein, L. Muratov,
J. Lee, M. Fox, S. M. Adler-Golden, J. H. Chetwynd Jr, M. L. Hoke, et al.,
“MODTRAN5: 2006 update,” in Defense and Security Symposium, pp. 62331F–
62331F, International Society for Optics and Photonics, 2006.
82
[23] P. W. Nugent, J. A. Shaw, and N. J. Pust, “Radiometric calibration of infrared
imagers using an internal shutter as an equivalent external blackbody,” Optical
Engineering, vol. 53, no. 12, pp. 123106–123106, 2014.
[24] P. W. Nugent and J. A. Shaw, “Calibration of uncooled LWIR microbolometer
imagers to enable long-term field deployment,” in SPIE Defense+ Security,
pp. 90710V–90710V, International Society for Optics and Photonics, 2014.
[25] P. W. Nugent, J. A. Shaw, and N. J. Pust, “Correcting for focal-plane-array
temperature dependence in microbolometer infrared cameras lacking thermal
stabilization,” Optical Engineering, vol. 52, no. 6, pp. 061304–061304, 2013.
[26] P. W. Nugent, J. A. Shaw, N. J. Pust, and S. Piazzolla, “Correcting calibrated in-
frared sky imagery for the effect of an infrared window,” Journal of Atmospheric
and Oceanic Technology, vol. 26, no. 11, pp. 2403–2412, 2009.
[27] P. W. Nugent, Deployment of the third-generation infrared cloud imager, a
two year study of arctic clouds at Barrow Alaska. PhD thesis, Montana State
University, 4 2016. Not yet published.
[28] J. Peterson, J. Cardon, P. Sevilla, J. Hancock, C. Englert, C. Brown, K. Marr,
and J. Harlander, “MIGHTI spectral calibration,” in Proc. Conference on
Characterization and Radiometric Calibration for Remote Sensing, (Logan, UT),
2016.
[29] J. Peterson, H. Latvakoski, G. Cantwell, J. Champagne, and J. Cardon, “200 nm
to 100 um, with extremely low uncertainty requirements: Challenges of the RBI
spectral calibration,” in Proc. Conference on Characterization and Radiometric
Calibration for Remote Sensing, (Logan, UT), 2016.
[30] Personal Communication. P.W. Nugent.
[31] K. Sassen and G. G. Mace, Ground based remote sensing of cirrus clouds. Oxford,
New York, NY, 2002.
[32] P. W. Nugent and J. A. Shaw, “Large-area blackbody emissivity variation with
observation angle,” in SPIE Defense, Security, and Sensing, pp. 73000Y–73000Y,
International Society for Optics and Photonics, 2009.
[33] P. Pilewskie and S. Twomey, “Cloud phase discrimination by reflectance
measurements near 1.6 and 2.2 µm,” Journal of the atmospheric sciences, vol. 44,
no. 22, pp. 3419–3420, 1987.
[34] P. Pilewskie and S. Twomey, “Discrimination of ice from water in clouds by
optical remote sensing,” Atmospheric research, vol. 21, no. 2, pp. 113–122, 1987.
83
[35] “Attenuation due to clouds and fog.” ITU Recommendation ITU-R P.840-6, 09
2013.
[36] B. A. Baum, D. P. Kratz, P. Yang, S. Ou, Y. Hu, P. F. Soulen, and S.-C.
Tsay, “Remote sensing of cloud properties using MODIS airborne simulator
imagery during success: 1. data and models,” Journal of Geophysical Research:
Atmospheres, vol. 105, no. D9, pp. 11767–11780, 2000.
[37] W. H. Knap, P. Stammes, and R. B. Koelemeijer, “Cloud thermodynamic-phase
determination from near-infrared spectra of reflected sunlight,” Journal of the
atmospheric sciences, vol. 59, no. 1, pp. 83–96, 2002.
[38] S. L. Nasiri and B. H. Kahn, “Limitations of bispectral infrared cloud phase
determination and potential for improvement,” Journal of Applied Meteorology
and Climatology, vol. 47, no. 11, pp. 2895–2910, 2008.
[39] S. Platnick, M. D. King, S. A. Ackerman, W. P. Menzel, B. A. Baum, J. C. Rie´di,
and R. A. Frey, “The MODIS cloud products: Algorithms and examples from
terra,” IEEE Transactions on Geoscience and Remote Sensing, vol. 41, no. 2,
pp. 459–473, 2003.
[40] P. Chylek, S. Robinson, M. Dubey, M. King, Q. Fu, and W. Clodius,
“Comparison of near-infrared and thermal infrared cloud phase detections,”
Journal of Geophysical Research: Atmospheres, vol. 111, no. D20, 2006.
[41] A. Kokhanovsky, O. Jourdan, and J. Burrows, “The cloud phase discrimination
from a satellite,” IEEE Geoscience and Remote Sensing Letters, vol. 3, no. 1,
pp. 103–106, 2006.
[42] D. D. Turner, S. Ackerman, B. Baum, H. E. Revercomb, and P. Yang, “Cloud
phase determination using ground-based AERI observations at SHEBA,” Journal
of Applied Meteorology, vol. 42, no. 6, pp. 701–715, 2003.
[43] K. F. Palmer and D. Williams, “Optical properties of water in the near
infrared∗,” J. Opt. Soc. Am., vol. 64, pp. 1107–1110, Aug 1974.
[44] S. G. Warren, “Optical constants of ice from the ultraviolet to the microwave,”
Appl. Opt., vol. 23, pp. 1206–1225, Apr 1984.