SCRIPPS INSTITUTION OF OCEANOGRAPHY LIBRARY
OPTICAL SYSTEMS GROWT H
UNIVERSITY OF CALIFORNIA, SAN DIEGO
OPTICAL SYSTEMS GROUP
TECHNICAL NOTE NO. 234
May 1993
3 1822 03387 3324
AUTOMATED WHOLE SKY IMAGERS
FOR DAY AND NIGHT
CLOUD FIELD ASSESSMENT
UNIVERSITY
OF
CALIFORNIA
SAN DIEGO
J. E. Shields
R. W. Johnson
T. L. Koehler
M. E. Karr
CRSI
НЕ
RNIA
TERE
(1868
ESPERO
The material contained in this note is to be considered
proprietary in nature and is not authorized for distribution
without the prior consent of the Marine Physical Laboratory
SCRIPPS
INSTITUTION
MARINE PHYSICAL LAB San Diego, CA 92152-6400
OF
OCEANOGRAPHY
""""""
"
"
蒙蒙
​,
TECHNICAL NOTE NO. 234
AUTOMATED WHOLE SKY IMAGERS
FOR DAY AND NIGHT
CLOUD FIELD ASSESSMENT
J. E. Shields, R. W. Johnson, T. L. Koehler, and M. E. Karr
SUMMARY
A new 24-hour automated Whole Sky Imager (WSI) has been developed at the
Marine Physical Laboratory at Scripps Institution of Oceanography. This system is used
for assessment and documentation of cloud fields and cloud field dynamics. The WSI is
a ground-based system which monitors the upper hemisphere under lighting conditions of
daylight, moonlight, and starlight. It is a visible range, passive, electronic imager
acquiring multi-spectral images under fully automated control. From these images, the
presence of clouds is assessed using automated cloud decisions algorithms for daylight
imagery; night algorithms are in development. This technical note describes the system,
designated the Day/Night WSI, including sample imagery and a discussion of the
algorithms.
TABLE OF CONTENTS
Summary :
-
List of Illustrations ..
........
E:
WINRANXVMA
1.0
Introduction
-
ULIVII ...............
2.0
..................................................................
Overview of the Day/Night WSI System ..
2.1 The WSI Sensor ..
2.2 The WSI Controller...
3.0
WSI Data and Acquisition .....
3.1 Sample Imagery .........
3.2 WSI Data Acquisition ...
Automated Cloud Assessment.
4.0
000 va Aww W
5.0
Conclusion
6.0
Acknowledgements .......
7.0
References..
LIST OF ILLUSTRATIONS
Fig. #
Figure Title
1
WSI Day/Night Sensor Assembly ...........
...
1
2
WSI Day/Night Sensor with Environmental Housing....
Day/Night WSI Hardware Block Diagram ...
WSI Raw Data Image Acquired with Daylight at 650 nm ...
Processed Cloud No Cloud Decision Image.
WSI Raw Data Image Acquired with Moonlight at 650 nm.....
6
WSI Raw Data Image Acquired with Starlight in Open-hole Configuration ......
Clear Starlight Image Enhanced to Emphasize Star Constellation Patterns ........
Natural Illumination Levels ..........
10
Field Data Acquisition Conceptual Flow Chart...
11
12
Cloud Data Processing Conceptual Flow Chart
Cloud Decision Image Series from a 28-Minute Data Acquisition Interval.
...............
1. INTRODUCTION
Over a period of nearly ten years, the Marine Physical
Lab (MPL) at Scripps Institution of Oceanography has
developed a series of imaging systems for assessment of
the atmosphere (Johnson, et al, 1986; Johnson, et al,
1989; Shields, et al, 1990; and Shields, et al, 1992). The
Whole Sky Imagers (WSI) are automated imagers used
for assessment and documentation of cloud fields and
cloud field dynamics. The WSI is a ground-based elec-
tronic imaging system, which monitors the upper hemi-
sphere. It is a passive, i.e. non-emissive system, which
acquires multi-spectral images of the sky dome. From
these images, the presence of clouds is assessed using
automated cloud decision algorithms.
Clouds are such pervasive features of the atmospheric
environment that they have a very significant impact on
applications ranging from military test support to global
warming research. Requirements can range from a
simple need to know the cloud cover fraction at a given
point in time and space, to a need to know the locations
of clouds within the scene or more complex parameters
such as the persistence of cloud free line of sight as a
function of look angle.
Whole Sky Imagers provide a determination, at ap-
proximately 200,000 points in the upper hemisphere, of
the presence of opaque clouds and thin clouds in the line
of sight. WSI's may also be adapted to provide calibrated
radiance at each of these points. Under fully automated
control, the WSI's provide this information with minimal
human intervention. The data may be used for statistical
analysis of relations between cloud free line of sight and
cloud cover, evaluation of persistence and recurrence,
and other high resolution spatial and temporal character-
istics of the cloud field. The systems are used in military
test site support, providing documentation of field condi-
tions during tests such as missile tracking tests. For
Global Warming applications, the WSI's will provide
cloud cover and cloud distribution information at test
sites. These data will be used for studies of cloud
radiative forcing and feedback mechanisms, and in un-
derstanding relationships between the cloud field and
incoming solar radiation at the surface.
Over the last decade, the Optical Systems Group
(OSG) at MPL, and formerly at the Visibility Lab, has
developed a family of WSI systems, beginning with the
fielding of the first two generations, EO System 1 and EO
System 2, in 1984. During the late 80's, OSG developed
and fielded Day-only WSI's (EO System 5), fully auto-
mated systems which acquire image sets every minute
for 12 hours per day. These images undergo a series of
calibrations and processing by automated algorithms to
yield a cloud decision image with 1/3 degree spatial
resolution. Several of these daytime WSI's operated in
the field over a period of 2 to 3 years per site (Shields, et
al, 1991; Johnson, et al, 1991; and Koehler, et al, 1991).
Building on the experience gained with the Day WSI
systems, OSG has more recently developed a Day/Night
WSI (EO System 6) capable of image acquisition under
daylight, moonlight, and starlight conditions (Shields. et
al, 1993). This note discusses the Day/Night WSI sys-
tems, including control of data acquisition and interpre-
tation of the data.
2. OVERVIEW OF THE DAY NIGHT WSI
SYSTEM
The Day/Night WSI is a ground-based electronic
imaging system. The sensor package consists of a solid
state CCD (Charge Coupled Device) camera, solar/lunar
occultor, filter changer, and environmental protection.
The control package consists of an IBM PC-compatible
computer for communications and system control, a
backup archival unit, and an Accessory Control Panel to
enable a manual interactive link with the sensor assem-
bly.
2.1 The WSI Sensor
The WSI sensor is illustrated in Fig. 1. The primary
components in this figure are discussed below.
The

.
......
...
..
.....
..__..._.................._.._
__.__
.
...
.
..........
...
..
.......
.....................
.
.....................
.
..........
.....
FIG. 1
WSI DAY/NIGHT SENSOR ASSEMBLY
(NOT TO SCALE)
SOLAR/LUNAR
ATTENUATOR ASSY
PROTECTIVE
ACRYLIC DOME
VOTO
.
610mm
FISHEYE LENS, 4.8* DIA.
NIKKOR, 8mm, 1 2.8
MODIFIED W/AUX, FILTERS
(A to B: EBFL = 47.7mm)
1
AUDIO
2010
L01
MPL
OCCULTOR
DRIVE
MPL
CUSTOM DUAL
INTERFACE WI
AUX. FILTERS
OLTASZOI
.......
..
.
....
.
.
.........
.
.
.....
WWW
ELECTRO-MECH.
SHUTTER ASSY
*
w
ww
.
.
.
.
DIXO
*
*
*
Utta
#
.
COOLED
CAMERA
CHAMBER
A
MODIFIED
25mm FIBER
OPTIC TAPER ASSY.
.
.
#
m
.
>
MODIAED CH250
LIQUID COOLED
CAMERA HEAD &
HOUSING
100mm
CAMERA
ELECTRONICS
UNIT FOR
PM 512 ARRAY
..
1000000
...
The fisheye lens is a Nikkor 8mm f/2.8 lens. It has a
full 180 degree field of view for viewing the complete sky
dome simultaneously. The lens has equi-distant projec-
tion, i.e. the zenith angle in object space is nearly linear
with respect to the position of the corresponding pixel in
image space.
Like the Day-only WSI, the Day/Night WSI uses an
optical filter changer designed by the OSG. There is a
significant difference between the optical system in the
two WSI's however. The optical system in the Day-only
WSI uses an optical relay to convert the fisheye image
size and location to the appropriate camera chip format.
The lenses which form this optical relay are part of the
filter changer in the Day-only WSI.
The Day/Night WSI uses a different concept for
converting the fisheye image format. The filter wheels
are placed directly between the exit aperture and the back
focal plane of the fisheye. The back focal plane is
coincident with the surface of a fiber optic taper, which
then de-magnifies and transfers the image to the chip.
The fiber optic taper is bonded to the chip, for a proximity
focus system. The losses and distortions in the taper are
minimal, in comparison with relay systems we consid-
ered, so that optical quality and sensitivity are preserved.
The optical filter changer contains two independently
controlled filter wheels, each containing up to four fil-
ters. One wheel is intended for spectral control, the other
for flux level control. The standard set-up is currently as
follows:
This selection of filters is somewhat different from
that used in the Day WSI. In particular, the Day WSI used
two pairs of red and blue filters. This was due to the
limited dynamic range of the CID (Charge Injection
Device) camera which is used in the Day Only WSI.
Since the CID had a useful dynamic range of approxi-
mately 1.2logs, it was necessary to use a second filter pair
offset by.5 log to yield a dynamic range of approximately
1.7logs. The Day/Night WSI uses a CCD with a dynamic
range of approximately 3 logs or more, and the use of a
second filter pair is unnecessary.
The Day/Night WSI's electronic camera is a
Photometrics Slow Scan CCD. In our early development
of nighttime capability, we tested a number of options,
including on-chip integration with a CID camera, and use
of an image intensifier. We found on-chip integration to
be very non-linear, particularly as the sensor chip aged.
Our image intensifiers proved to be noisy and unstable in
our test applications; they also may be damaged by
exposure to excess flux, which made their use in unat-
tended field situations problematic.
Following tests of several CCD cameras, we chose to
go with a Photometrics slow scan camera. This camera
has outstanding noise and sensitivity characteristics as
well as high image quality. Its very low noise and high
sensitivity allow acquisition of night imagery, even un-
der starlight conditions. Its 16 bit digitization, in combi-
nation with the low readout noise, allow for an outstand-
ing combination of large dynamic range with fine radio-
metric resolution. However, in slow scan camera opera-
tions, data readout is far slower than in the earlier RS-170
class video systems such as used in the Day WSI. What
you trade for of course is nighttime capability with high
image quality.
The camera housing shown in Fig. 1 is temperature
stabilized. It is sealed and purged with dry nitrogen at a
slight positive pressure for protection of the sensor ele-
ments from moisture.
The tracking solar/lunar occultor is a dual drive
occultor, with separate control of the azimuth drive and
zenith drive. The computer logic supplies the appropri-
ate occultor gear angle as a function of date and time, and
the occultor is driven (automatically) to the proper posi-
tion. Most of the occultor is opaque, however the central
portion consists of a 4 log neutral density filter, so that the
sun or moon position may be detected. This aids in
validation of computer clock time, WSI leveling, and
lens geometric calibration.
As shown in Fig. 2, the WSI sensor and camera
chamber are further protected by an environmental en-
Filter Position
Spectral Wheel
ND Wheel
AWN
opaque (blocked) open
open
2 log ND
650 nm
3 log ND
450 nm
optional
Most of the above filter positions also include opti-
cally transparent trim filters, to enable pixel registration
(i.e. same size image for all possible filter combinations).
Spectral position 1 is used for acquiring a dark image,
used to correct the raw field image for dark offset and
pixel nonuniformity. Spectral 2 is open-hole, and may be
used for calibration or for acquisition under rural star-
light conditions. Spectral positions 3 and 4 are for
acquiring the red and blue images. The use of this filter
pair has enabled development of cloud algorithms based
on the spectral character of the sky scene. Neutral
Density position 1 is used for night acquisition, 2 is for
twilight, and 3 is for daytime. Filter position 4 on the ND
wheel may be used for optional test configurations.
Fig. 3
closure. The system includes a chiller assembly, which
cools the camera chamber. In addition, the chilled liquid
is provided to the hot side of the TE (thermo-electric)
cooler which cools the sensor chip, thus further enhanc-
ing the low noise characteristics of the CCD. (Nitrogen
cooling is not required.) The environmental enclosure
also houses the camera electronics unit, and coolant flow
monitor.

DAY/NIGHT WSI
HARDWARE BLOCK DIAGRAM
ENVIRONMENTAL ENCLOSURE
HIGH RESOLUTION
VGA
MONITOR
10
PHOTOMETRICS
ATX 200L
SOLID STATE
SLOW SCAN
CAMERA &
ELECTRONICS UNIT

TMI COMPUTER
(IBM PC CLONE)
with
TMI 80486 CPU
254 mm
.
DIGITAL
DOCTO
.
t
Fig. 2
WSI DAY/NIGHT SENSOR WITH
ENVIRONMENTAL HOUSING
(NOT TO SCALE)
:
AT 200
CONTROLLER CARD
-
-
24*
610 mm
AUTOMATIC
SOLARLUNAR
OCCULTOR
ASSY.
OPTIONAL
COMMUNICATIONS
MODULE
-
-
VII.
EXABYTE EXB - 8200
2.2 Gbyte 8 mm Cartridge Tape System
REMOTE
CONTROLLED
OPTICAL FILTER
ASSY.
.itakin
www
wowWWWWWWWWW
*
ANALOG
www
w
w
wwwww..........
.
.....
..www.w
ww.
ACCESSORY
CONTROL PANEL .
COOLING
COILS
-
STOWED KEYBOARD
TEMPERATURE
AND
STATUS
SENSE
COOLED
CAMERA
CHAMBER
ttö
EXTERIOR SENSOR
INSTALLATION
INTERIOR CONTROLLER
INSTALLATION
CAMERA
ELECTRONICS
UNIT
COOLANT FLOW
MONITOR
& NC CUT-OFF
6.5
-
.
14.5
COOLANT
RESERVE
& PUMP
ASSY
CHILLER ASSY
2.2 The WSI Controller
The WSI exterior sensor system is connected to a
controller, consisting primarily of electronics and the
computer package. This controller must be in an office
environment or equivalent protected environment. These
units are normally connected by 100-foot cable. Fig. 3
illustrates the primary components of both the exterior
sensor and the interior controller, and their connection.
In addition to the PC computer and monitor, the
controller includes an Accessory Control Panel (ACP).
This ACP enables control of the filter changer and
occultor either manually or through computer control.
The controller also includes an Exabyte tape backup
system with a data storage capacity of 2.3 Gbytes.
3. WSI DATA AND ACQUISITION
3.1 Sample Imagery
A sample Day/Night WSI image is shown in Fig. 4. In
this illustration, the zenith is in the center, with the
horizon on the edge of the round image. The south is at
the top, and east is to the right. The black square near the
right is the solar/lunar occultor. When not obscured by
clouds, the sun is imaged through the 4 log neutral
density filter in the occultor. The red, blue, and dark
imagery may be processed to yield a cloud decision
image, as shown in Fig. 5. Regions of this image which
have been identified by the automated algorithm as clear
sky are false colored blue. Opaque and thin clouds are
identified with white and yellow respectively. A cloud
decision is made independently at each pixel location.
A moonlight image acquired by the WSI on-site at
Kirtland AFB, NM, is shown in Fig. 6. In this image, the
occultoris nearly overhead, with the moon imaged through
the 4 log neutral density filter. Nearby buildings and
terrain may be seen on much of the image edge, with
clouds to the north (bottom) of the image. A few stars
appear in the image to the east (right). Under moonlight,
the path radiance or moonlight scattered into the path of
site masks most stars.

Image Type: Raw red
Location: WSMR, Helst
Date: 18 Sept. 92
Time: 15172
An image acquired by the WSI on-site at White Sands
Missile Range, NM, is shown in Fig. 7. This image was
acquired under no-moon conditions. A cloud field en-
croaching from the north-east covers approximately half
the sky dome. The lights to the south (near the top of the
image) are from the sky light over El Paso, approximately
40 miles from the site. Note that in the clear regions,
many more stars are visible due to the reduced path
radiance under no-moon conditions.
Spectral 2 (Open), 24 Feb. 1992, 0900 local time
Acquired in New Mexico, no moon
60 sec. exposure, 500-1300 display, Dark corrected
MPIT
Orion's Belt
Fig. 4. WSI Raw Data Image Acquired with
Daylight at 650 nm
Image Type: Cloud Decision
Location: WSMR, Helst
Date: 18 Sept. 92
Time: 1517Z
IMPU
Fig. 7. WSI Raw Data Image Acquired with
Starlight in Open-hole Configuration
MPD
Fiore 36
Fig. 5. Processed Cloud/No Cloud
Decision Image
Red Image, 14 Oct. 92, 08162
Kirtland AFB NM, Moonlight
40 sec. exposure, 0 - 10000 display, Dark corrected
An interesting feature of the camera performance may
be seen in this figure. With 16 bit digitization, the system
has approximately 65,000 gray levels. This gives the
system a useful radiometric range of over three logs,
while retaining very good radiometric resolution. In this
particular image, in order to show the features such as the
stars, the gray level range from 500 counts to 1300 counts
has been displayed. Anything above 1300 is shown in
Fig. 7 as white, and anything below 500 appears black.
The white areas near the top of the image which appear
to be offscale bright in this reproduction are actually
onscale in the digital image, just as the black areas near
the bottom are onscale in the digital image. It is this large
digital range, in combination with the low noise, which
allows the system to acquire features at a large range of
brightnesses, while retaining the digital information which
allows the user to enhance the image and view brightness
features which occupy a narrow portion of this brightness
range. Fig. 8 shows a similar clear sky, enhanced in such
a way as to emphasize the constellations.
3.2 WSI Data Acquisition
One of the important design criteria for the Day/Night
WSI is the large range of flux levels the system must be
able to deal with. Figure 9 shows the naturally occurring
MP3
Fig. 6. WSI Raw Data Image Acquired with
Moonlight at 650 nm

illuminance levels under a variety of lighting conditions.
These data are from the work of Brown 1952, and are
consistent with irradiance measurements acquired by our
group at the Visibility Lab over a period of many years.
In Fig. 9, the daytime illuminance conditions the Day
WSI has had to deal with are shown in the top two curves
on the right side of the plot. These represent clear to dark
storm conditions for sun zenith angles 0 to 90 degrees.
The Day/Night WSI is operational through quarter moon
conditions shown on the bottom right curve, and down to
the starlight conditions shown on the left side of the plot.
This represents approximately a 9 log range of lighting
conditions. The sensor is designed to obtain the neces-
sary sensitivity range by using the approximately 3 to 3.5
log sensitivity of the camera chip, approximately 3 logs
range from exposure control, and approximately 3 logs
range through neutral density filter control.
The normal data acquisition sequence is shown in Fig.
10 (specific details of the program vary with the user
application). In order to apply the daylight cloud algo-
rithm, it is necessary to acquire a red image at 650 nm,
and a blue image at 450 nm. In addition, a “dark image"
is acquired at the same exposure level, with the light
blocked. This dark image provides a measure of the dark
current and electronic bias applied to the chip, and is
subtracted from each of the images to provide the dark
correction.
Start
Initialize camera,
computer, tape drive
Read clock,
determine start time
Spectral 2 (Open), 24 Feb. 1992, 0754 local time
Acquired in New Mexico, no moon
60 sec. exposure, 500-1500 display, Dark corrected
Determine flux control
parameters, set ND and exposure
Orion's Belt
Determine desired occultor
position, move occ
Grab red, blue,
(dark) image
Big Dipper
Embed ident
headers
Save images to
Exabyte tape
Fig. 8. Clear Starlight Image Enhanced to
Emphasize Star Constellation Patterns
i no
yes
Change
Таре
yes/End of \ no V
Week
Next
grab
time?
5.00
4.00-
Unobscured Sun
3.00 -
Black Storm
2.00
Fig. 10. Field Data Acquisition
Conceptual Flow Chart
1.00
Illumination in Footcandles (logs)
0.00
Full Moon
-1.00
-2.00
Quarter Moon
-3.00
Starlight
-4.00
-5.00
-40.00-
-30.00-
-20.00-
- 10.00
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
Normal daylight exposures are 100 msec, with a 3 log
neutral density filter in the optical path. As light levels
decrease, a 2 log filter and slightly longer exposure times
are used. At night, the open-hole position in the neutral
density wheel is used, and exposure times are increased
as necessary to provide the added flux. Under urban
conditions in San Diego, exposures of 30 seconds pro-
vide quality red and blue images. Under rural conditions
encountered to date, these exposures are reasonable for
full moon conditions. Under the darkest rural no-moon
Solar and Lunar Elevation Angles
Fig. 9. Natural Illumination Levels.
These measurements, from Brown 1952, illustrate flux
conditions the Day/Night WSI should Encouner.
conditions, it is necessary currently to use the open hole however. This technique essentially allows one to base
(no spectral filter) acquisition with approximately 1 the cloud decision on the color, or spectral signature as
minute exposures.
defined by this ratio. The red/blue ratio is the basis of the
The flux control algorithm is designed to allow the
current daytime cloud decision algorithm.
system to automatically determine the appropriate in- The cloud decision algorithm is illustrated conceptu-
strument settings to enable acquisition of on-scale data. ally in Fig. 11. In order to compute the ratio images, one
In the Day-only WSI, this flux control was based on must first acquire radiometric calibrations to character-
minute-by-minute assessment of the prevailing lightize sensor performance. Calibrations of the dark image
levels. The Day/Night WSI uses a different scheme, characteristics, temporal and spatial signal variances,
which is essentially predictive, in order to handle the very linearity as a function of exposure, linearity as a function
quickly changing flux levels during the hours near sun of flux level, and relative response as a function of
rise and sunset.
spectral and neutral density selection have been acquired
The data illustrated in Fig. 9 were used as the first
and evaluated for an existing Day/Night WSI. These
results are used in the ratio computation algorithm. This
estimate of relative changes in downwelling irradiance.
ratio computation procedure is simpler than with the
For flux control, we are more interested in the average
sky radiance than in the downwelling irradiance of the
older Day-only WSI, due to the improved linearity and
pixel registration of the current WSI's.
sky. For this reason, the diffuse irradiance, which is more
closely related to the average sky radiance, was estimated
from the downwelling irradiance curves in Fig. 9, using
Kondrat’yev (1965). For characterizing the flux levels
Raw Images
under moonlit conditions, the algorithm also makes cor-
rections for earth-lunar distance and for moon phase
Apply dark correction
angle (Haphe, 1963).
Combining the expected radiant field information
Apply calib (function of
with the operating characteristics of the camera, a deter-
exposure, SP filter, ND filter,
mination was made of the approximate desired minimum
pixel position)
signal for various conditions. From this, tables of neutral
density filter and exposure setting as a function of solar
Ratio red/blue
zenith angle, lunar zenith angle, lunar phase and lunar
distance were generated for use by the system. The
selected exposure/ND filter is changed whenever the flux
Mask occultor
is expected to change by approximately .2 logs.
4. AUTOMATED CLOUD ASSESSMENT
Compute Clear Sky
background
Once quality imagery is acquired, it is desirable to
make an automated, pixel-by-pixel determination of the
Determine haze correction
presence of clouds in the images. This might appear to be
an easy task, since the clouds are so obvious to a human
viewing the properly displayed images. However, to
Compute Hazy Sky
background
generate an algorithm which identifies clouds consis-
tently over many months of data, at a variety of sites, is
not trivial.
Identify opaque clouds
(fixed threshold)
In our early work with the cloud algorithms (1984),
we determined that radiance thresholding is generally not
Identify thin clouds
adequate, because clouds can often be darker than the
(perturbation wrt Hazy Sky)
adjacent sky. Edge detection provided additional infor-
mation, however many of the clouds have quite diffuse
Save images
edges which are not easily identified with the techniques
used in the early tests. Similarly, texture analysis pro-
vided useful but inadequate information. Detection
Fig. 11. Cloud Data Processing
based on the red/blue ratio, proved to be quite effective
Conceptual Flow Chart

Once the calibration-corrected ratio is computed, the
opaque clouds are identified on the basis of this ratio
alone. A simple threshold technique is used in which
pixels that exceed a specified ratio are identified as
opaque cloud. In other words, the opaque cloud discrimi-
nation is based on spectral signature, as defined by the
red/blue ratio.
Thin clouds are not defined with a specific spectral
signature, but rather as a given deviation from the spec-
tral signature of the background sky. In simpler terms,
sky. In simpler terms,
thin clouds are not necessarily white, however they are
whiter than the sky background in a given direction. The
sky ratio varies both directionally (i.e. as a function of
both look angle and solar zenith angle) and as a function
of visibility. The thin cloud decision algorithm selects the
appropriate table, which characterizes the directional
variance in background ratio for the given site and given
solar zenith angle (interpolating to the actual solar zenith
angle). Determination of the appropriate normalization
factor, to adjust for variations in visibility, is based on an
estimated normalization factor computed for each image
using the clear portions of the sky. Thus a sky back-
ground ratio is generated for the given site, solar zenith
angle, and haze condition, as a function of azimuth and
zenith angle. The thin clouds are then identified at each
pixel on the basis of their fractional deviation from the
sky ratio.
The day cloud algorithm has been applied to much of
the Day WSI data base A data base of approximately 900
Gigabytes of raw image data (approximately 4600 data
days) has been generated with Day WSI's. Of this data,
14 months at each of 4 stations have been processed to the
cloud decision image (Johnson, 1991). The results
compare quite well to the standard observer (Shields,
1990, and Koehler, 1991).
The Day/Night WSI systems have not yet been used to
acquire an extensive data base from which the cloud
algorithm accuracy may be assessed. However sample
imagery results are quite encouraging. The sample
shown earlier in Fig. 5 is an example of the current cloud
decision algorithm. Figure 12 shows a series of four

Image Type: Cloud Decision
Location: WSMR, Helst
Date: 5 Feb. 93
Time: 1622Z
Image Type: Cloud Decision
Location: WSMR, Helsti
Date: 5 Feb. 93
Time: 1631Z

Image Type: Cloud Decision
Location: WSMR, Helst
Date: 5 Feb. 93
Time: 164OZ
Image Type: Cloud Decision
Location: WSMR, Helst
Date: 5 Feb. 93
Time: 1650Z
IMPE
Fig. 12. Cloud Decision Image Series from a 28-minute Data Acquisition Interval
evaluation of persistence and recurrence, as well as cloud
development vs. translation studies become feasible.
Finally, if full absolute radiometric calibrations are ob-
tained prior to fielding the system, the upper hemisphere
absolute radiance distribution may be extracted from the
data. As these systems continue to develop in capability,
flexibility, and convenience, they should continue to
have important applications including military test sup-
port and global warming research support.
6. ACKNOWLEDGMENTS
This work was sponsored by Air Force Phillips Lab,
Army Atmospheric Sciences Lab, and Navy SPAWARS,
through the Office of Naval Research. Our sincere
thanks to Maj. Tom Dorsey of Phillips Lab, Robert W.
Endlich of ASL, and Fred Diemer of ONR for their
encouragement and advice. We would also like to
recognize the outstanding efforts of our colleagues at
Marine Physical Laboratory: Harry Sprink and Jack
Varah for hardware development and support; Bill Davy,
George Trekas, and Tim Stoesz for precision machining
shop work; Carole Robb for publications support; and
Pat Jordan and staff for administrative support.
cloud decision images from a set of data acquired during
a 28 minute interval as a stratocumulus band passed
overhead.
There are several features of this cloud identification
technique which should be noted. First, unlike schemes
involving human evaluation of images, the technique is
both fast and consistent. Secondly, through application
of the calibration corrections, most of the bias due to
camera characteristics such as non-linearity is removed.
Third, through use of the ratio technique, as opposed to
an identification based on radiative brightness, one cor-
rectly identifies even clouds which are darker than the
sky background. Finally, through correction of the
background sky ratio for aerosol load and directional
variance, the system avoids much of the directional bias
inherent in human assessment; for example, a cirrus
streak from an aircraft is correctly identified both upsun
and downsun.
For moonlit images, it should be possible to use a
cloud decision algorithm which is conceptually similar to
that used in the Day WSI. A study of night radiance
distributions acquired by our group in 1968 and 1969 has
been made to evaluate the red/blue ratios under a variety
of conditions, (Gordon, 1989). This study indicates that
for full moon to quarter moon lighting conditions, the
red/blue ratio should be a reasonably good indicator of
clouds, but for starlight conditions the spectral character
of the sky is quite different, and a different algorithm will
be required. Development of the moonlight algorithm,
along with characterization of the contamination of the
light field due to terrestrial sources (such as urban lights)
is currently under development.
5. CONCLUSION
Both the Day/Night Whole Sky Imagers and the Day
WSI's, developed by the Marine Physical Lab, acquire
imagery appropriate for automated identification of cloud
fields. Depending on sponsor requirements, these WSI's
may either yield a data archive for post processing, or
yield near-real time results. The latest version of the WSI
units acquires imagery not only during the daytime, but
under moonlight and starlight conditions, for full 24-
hour automated data acquisition.
In use by the military for several years, these systems
yield data for determination of cloud cover, as well as
cloud field spatial characteristics such as angular distri-
butions of cloud free line of site. Statistical information
such as the frequency of cloud obscuration of direct solar
flux as a function of cloud cover and solar zenith angle are
readily extracted (limited only by the mechanics of
dealing with a large data base). Temporal studies such as
7. REFERENCES
Gordon, J. I., 1989: “Assessment of Night Sky Radiance
Data from Project Shedlight", Viz. Ability, Inc., 2941
Ashwood Dr., Corvallis, Oregon, 97330.
Hapke, B. W., 1963: "A theoretical Function for the
Lunar Surface”, Journal of Geophysical Research,
Vol. 68, No. 15, 4571-4585.
Johnson, R. W., W. S. Hering, and J. E. Shields, 1986:
"Imagery Assessment for the Determination of Cloud
Free Intervals”, University of California, San Diego,
Scripps Institution of Oceanography, Visibility Labo-
ratory, Atmospheric Visibility Technical Note No.
200.
Johnson, R. W., W. S. Hering, and J. E. Shields, 1989:
"Automated Visibility and Cloud Cover Measure-
ments with a Solid-State Imaging System”, Univer-
sity of California, San Diego, Scripps Institution of
Oceanography, Marine Physical Laboratory, SIO 89-
7, GL-TR-89-0061, NTIS No. ADA 216906.
Johnson, R. W., J. E: Shields, and T. L. Koehler, 1991:
“Analysis and Interpretation of Simultaneous Multi-
Station Whole Sky Imagery”, University of Califor-
nia, San Diego, Scripps Institution of Oceanography,
Marine Physical Laboratory, SIO 91-33, PL-TR-91- :
2214.
whose
.
.
.
Koehler, T. L., R. W. Johnson, and J. E. Shields, 1991:
“Status of the Whole Sky Imager Database", Proceed-
ings of the Cloud Impacts on DOD Operations and
Systems - 1991 Conference.
Kondrat’yev, K. YA., 1965: “Actinometry", NASA TT
F-9712, Translation of "Aktinometriya",
Gidrometeorologicheskoye Izdatel' stvo, Leningrad,
1965.
Shields, J. E., T. L. Koehler, M. E. Karr, and R. W.
Johnson, 1990: "Automated Cloud Cover and Vis-
ibility Systems for Real Time Applications” Optical
Systems Group Technical Note No. 217, University
of California, San Diego, Scripps Institution of Ocean-
ography, Marine Physical Laboratory.
Shields, J. E., R. W. Johnson, and T. L. Koehler, 1991:
"Imaging Systems for Automated 24-Hour Whole
Sky Cloud Assessment and Visibility Determina-
tion”, Proceedings of the Cloud Impacts on DOD
Operations and Systems - 1991 Conference.
Shields, J. E., R. W. Johnson, and M. E. Karr, 1992: “An
Automated Observing System for Passive Evaluation
of Cloud Cover and Visibility”, University of Califor-
nia, San Diego, Scripps Institution of Oceanography,
Marine Physical Laboratory, SIO 92-22, PL-TR-92-
2202.
Shields, J. E., R. W. Johnson, and T. L. Koehler, 1993:
“Automated Whole Sky Imaging Systems for Cloud
Field Assessment”, Fourth Symposium on Global
Change Studies, 17 - 22 January 1993, Anaheim, CA.
Published by the American Meteorological Society,
Boston, MA.