UNIVERSITY OF CALIFORNIA, SAN DIEGO
3 1822 04429 7166
ATMOSPHERIC VISIBILITY
TECHNICAL NOTE NO. 200
MAY 1986
IMAGERY ASSESSMENT FOR THE DETERMINATION
OF CLOUD FREE INTERVALS
Offsite
(Annex-Jo
rnals)
QC
974.5
.T45
no. 200
R.W. Johnson
W.S. Hering
J.E. Shields
UNIVERSITY
The material contained in this note is to be considered
proprietary in nature and is not authorized for distribution
without the prior written consent of the Visibility Laboratory
and Air force Cambridge Research Laboratories.
OF
CALIFORNIA
SAN DIEGO
Contract Monitor, Donald D. Grantham
Atmospheric Physics Division
SITY
VERSI
00
sino
ORNT
CHE
ONY
@ee
Prepared for
Air Force Geophysics Laboratory, Air Force Systems Command
United States Air Force, Hanscom AFB, Massachusetts 01731
SCRIPPS
INSTITUTION
OF
OCEANOGRAPHY
VISIBILITY LABORATORY La Jolla, California 92093
UC SAN DIEGO LIBRARY
UNIVERSITY OF CALIFORNIA, SAN DIEGO
.
3 1822 04429 7166
TECHNICAL NOTE NO.200
IMAGERY ASSESSMENT FOR THE DETERMINATION
OF CLOUD FREE INTERVALS
by
R.W. Johnson
W.S. Hering
J.E. Shields
COPY__
OF_
COPIES
TABLE OF CONTENTS
1.0 INTRODUCTION ....
.....1
2.0 INSTRUMENTATION
2.1 Electro-optic Camera Systems
2.1 Photographic Cameras ..............
2.3 Hardware Organization ..................
2.4 Summary ............
.............
3.0 EXPERIMENTAL PROCEDURES
3.1 Data Collection
3.2 Data Base Summary
0000 irin AA Awin www
.....
4.0 ANALYSIS OF ACQUIRED IMAGES ..............
4.1 Description of Sample EO Camera Images .
4.2 Comparison of Camera System Images ......
4.3 Development of Cloud Identification Techniques ................
5.0 CLOUD STATISTICS: EXTRACTION AND RESULTS ........
5.1 Automated Extraction of Cloud Statistics from Images
5.2 Evaluation of Automated Extraction Accuracy ..........
5.3 Evaluation of Extracted Statistics ...
.........
8
10 .....
..... 11
6.0 SUMMARY AND ONGOING DEVELOPMENTS
7.0 REFERENCES .
8.0 ACKNOWLEDGEMENTS .....
APPENDIX A ..............
Ñ
APPENDIX B
LIST OF TABLES AND ILLUSTRATIONS
Table No.
2.1
Filter assembly assignments .........
Data base summary
3.1
........
......... 17
0.000
Fig. No.
2-1
2-2
........14
2-3
EO Camera optical layout
Equipment organization schematic
External equipment layout ........
Instrumentation hut on van ...........
Interior equipment layout ..........
...... 15
3-1
3-2
Experimental procedure - whole sky camera system .........
Field data log ....
.
4-2
..........
4-3
4-4
4-5
4-6
4-7
SO
Imagery from EO Camera II for a variety of sky conditions .........
Imagery from EO Camera I and Automax I for comparison with EO Camera II images
Line plot of cumulus cloud image for EO Camera II, EO Camera I and Automax I .....
Measured signal (from radiance at 650nm) as measured by the EO system ................
Test of radiance thresholding for cloud detection ......
Gradient information in cloud detection test image ......
Test of cloud discrimination using ratio of test image to clear image data ...
...................
Ratio of images taken with 0.7 log filter and 1.39 log filter ....
Difference image, shows / signal (x) - signal (x+1) | ..........
Cloud discrimination for 1 row with hybrid algorithm .......................
DO
4-8
4-9
4-10
5-1
5-2
10
Conceptualized chart of EO Camera data processing program ................
Sample imagery utilized for geometric calibration.
Cloud free line of signt, frequency of occurrence vs zenith angle
Cloud free arc length, cumulative frequency of occurrence vs arc length ................
Relative frequency of cloud cover at C Station, White Sands Missile Range .............
5-5
:
.
.
.
.
.
.
.
.
.
.
IMAGERY ASSESSMENT FOR THE DETERMINATION
OF CLOUD FREE INTERVALS
R.W. Johnson, W.S. Hering, & J. E. Shields
1.0 INTRODUCTION
This report describes the activities of the
Visibility Laboratory in support of a multi-team
system performance and site characterization
exercise which occurred at the White Sands Missile
Range, New Mexico, during the months of August
and September, 1984. The goal of this
demonstration was to illustrate the functional utility
of an electro-optic camera system as a reliable data
acquisition device for operations and research
requiring automatic specification of cloud cover: its
amount and distribution, both temporal and spatial.
Samples of the acquired imagery are given in
Section 4. The quality of the digital images is
compared with that of the photographic images.
This is followed by a discussion of preliminary
techniques for the interpretation of these images
using a mainframe computer with an image
processing system. For this prototype system, in-
house processing followed the field deployment,
however the eventual goal is the derivation of
algorithms which could be applied in real time.
The techniques developed on the image
processing system were applied to the full New
Mexico data base of one of the electro-optical
cameras, to yield initial statistical descriptions of
the cloud cover. These results are given in Section
5. Finally, in Section 6, the current directions in
hardware, algorithm, and software development
are briefly discussed.
More specifically the initial goal, for the
Visibility Laboratory effort, was to deploy a
prototype hardware system for the timely
assessment of local cloud cover. This system was
designed to acquire visible spectrum, whole-sky
images in specific wavelength bands, in a digital
format suitable for automated cloud analysis. A
related and in the long term a more demanding
task, was the initial development of the algorithms
necessary to use this class of data for the real-time
specification of cloud-free interval statistics and
their related probabilities.
2.0 INSTRUMENTATION
Since the test exercise schedule was relatively
firm, and hardware deliveries were somewhat
uncertain, several combinations of on-hand and on-
loan composite instrument systems were prepared
for deployment to New Mexico. Two electro-
optical cameras and two 35mm photographic
cameras with their associated control and recording
electronics were provided. These systems are
discussed in Section 2.
A number of systems were deployed to New
Mexico by the Visibility Laboratory, in order both
to obtain the desired sky images and to fully
evaluate the character of these images in relation to
the cloud characteristics. The most important of
the systems were two EO (electro-optic) camera
systems in somewhat different stages of
development. In addition, two photographic
cameras were deployed, one color and one black
and white. These two cameras provided a more
traditional data source for comparison with the
digital imagery acquired by the EO systems.
2.1 Electro-optic Camera Systems
A two-person field team and the equipment were
initially on-site Monday 27 August 1984, and
began data collection on Wednesday, 29 August.
Imagery was collected daily through Friday, 21
September, for a total of 24 days. A discussion of
the collection procedures and a summary of the
acquired data is given in Section 3.
Both EO camera systems consisted primarily of
a standard GE solid state camera, with Visibility
Laboratory optical and filter assemblies attached.
In each case, the camera contained a CID (charge
injection device) sensor. The external optics were
designed to allow the use of a fisheye lens, for
nearly full hemisphere coverage, as well as smaller
field of view lenses. The filter assembly, which
physically contained part of the optics mentioned
above, allowed the spectral filtering of the light, as
well as the control of light levels through use of
neutral density filters. Finally, each system
included control electronics as well as data archival
electronics and systems. The components of these
cameras are listed in Appendix A.
d) Since the range of the CID cameras was rather
limited in comparison with observed sky and cloud
radiances (just over 1 log of radiometric sen-
sitivity), a number of neutral density filters were
included in the filter changer in order to allow the
input flux level to be changed in a controlled way.
The use of the f-stops on the fisheye lens offered
additional ranging capability.
There were several important design consid-
erations affecting the development of these
instruments. Some of these are given below.
Figure 2.1(a) illustrates the optical layout of the
EO Camera I. The filters mounted in the filter
assembly are listed in Table 2.1. EO Camera I
could be run with either the 425nm filter or the 605
nm filter, either with or without a .3 log neutral
density filter. Additional control of flux level was
obtained through the use of the various f-stops on
the fisheye lens. Also, a 555nm filter was included
in this system for acquiring pseudo-photopic test
data.
a) In the analysis of the images, it may be
necessary to know the absolute radiance of each
point in the image, in order to compare this
radiance with a model or standard value. Even if
knowledge of absolute radiance is not required,
many of the potential techniques for cloud
identification require knowledge of relative
radiance within the image. In either case, this
implies the use of a fixed gain system, that is one
which does not change the electronic amplification
as a function of the image itself. This was a major
reason for choice of a CID imager with fixed gain.
The system block diagram for EO Camera I is
illustrated in Fig. 2.2(a). The camera was a GE
solid state CID camera, Model 2200. Camera
control and data logging were accomplished
through an HP mini-computer in conjunction with
a Chieftan frame grabber. The data were recorded
digitally on magnetic tape, and viewed simul-
taneously on a monitor Digital tapes were
returned to the laboratory for analysis. In this
system, the iris assembly was manually controlled,
as was the occultor assembly. The occultor
assembly consisted of a shading device placed
external to the fisheye lens to shade it from
sunlight, thus reducing stray light within the
system.
b) It is convenient to obtain data for the whole sky
in one image. In this particular application, it was
deemed adequate to obtain all but near-horizon look
angles. Consequently an upward looking fish eye
lens was used. A related consideration, however,
is that optical resolution must be adequate to image
sufficiently small clouds. Whereas the backup
system, EO Camera I, has a 128 x 128 picture
element (pixel) array, with a resulting resolution of
approximately 1.5° (zenith angle) per pixel, the
prototype system, EO Camera II, has approx-
imately 256 x 256 pixels, with a resolution of
approximately 0.7° per pixel.
The optical layout for EO Camera II is illustrated
in Figure 2.1(b), with the filters listed in Table 2.1.
This system is similar to EO Camera I, with a
somewhat more compact optical assembly and a
higher resolution sensor array. The sensor was a
GE Model 2505 CID Camera. Data could be
obtained either with a 650nm filter or spectrally
unfiltered, and with any of four neutral density
filters (or none) superimposed in the path.
c) In order to obtain maximum radiometric
resc olution between clouds and sky, a filter changer
was utilized which would allow data to be collected
through a red filter. In this wavelength regime, the
sky is relatively dark in comparison with the
clouds. In addition, as backup, it is desirable to
obtain data through a blue filter, so that clouds may
be potentially identified through analysis of the
measured spectral radiance ratios. For this
deployment, the backup EO Camera I was fitted
with a blue filter in addition to the red, whereas the
prototype EO Camera II had only a red filter.
Figure 2.2(b) illustrates the system block
diagram for EO Camera II. Figure 2.2(c),
showing the "Proposed" system, illustrates the
Zenith-driven system which was originally
intended for this deployment, but not deployed.
Third party delivery delays resulted in the
deployment of the "As-built" system in Fig.
2.2(b). In this system, the video signal was routed -
through a time/date generator to a video recorder
Table 2-1. Filter assembly assignments
camera support electronics, as well as the elevated
platform for the instrumentation hut containing the
actual cameras.

Camera
Ident
Filter
Position
Filter
EO CAMI
2-450 nm
6-605 nm
3-555 nm
open hole
open hole
0.6 log ND
The cameras are shown on the instrumentation
hut in Fig. 2.4. In this illustration, the fisheye
lenses are partially visible, as are the occultor
assemblies shading each of the four cameras. The
filter-control and camera assemblies are inside the
hut, and thus not visible in this illustration.
NM to como
EO CAM II
1.3 log ND
0.9 log ND
0.7 log ND
0.4 log ND
open hole
5-650 nm
The support eletronics are shown in Fig. 2.5.
The taller rack to the left contains the computer and
control panels for the EO Camera I, while the
darker blue rack to the right, and the shorter middle
rack contain the equivalent controls for the
prototype EO Camera II. The tape drive beneath
the table and the plotter to the left of the scene
support the archieval and real-time display of the
EO Camera I system.
.
.
.
2.4 Summary
and a monitor. The primary disadvantage of the
prototype system was the variable gain inherent in
the VCR. The video tapes of data from this system
were transported to the laboratory where they could
be digitized and read into the mainframe computer
for analysis. With this system, the iris assembly
and occultor assembly were controlled remotely.
In summary, there were four data acquisition
cameras deployed to the test site.
I. The back-up system, EO Cam I, 128 x 128 array
II. The prototype system, EO Cam II, 256 x 256 array
III. The primary 35mm camera, black & white film
IV. The secondary 35mm camera, color film
2.2 Photographic Cameras
In addition to the two electo-optic camera
systems, two 35mm photographic camera systems
were deployed. Each of these systems utilized a
fisheye lens, as well as an occultor assembly
similar to that utilized with the EO cameras. Both
were Automax Model G1, 35mm systems.
The back-up system, EO Camera I, included all
the peripherals required to obtain fixed-gain digital
data. The prototype EO Camera II, with its higher
pixel resolution, yielded potentially superior data to
that of EO Camera I, but was for this deployment
limited to recorded VCR data. The two 35mm
cameras were included for site documentation and
to provide data quality standards.
3.0 EXPERIMENTAL PROCEDURES
The primary photographic camera system
utilized black and white film, with a red filter in the
optical path. The data from this camera may be
used for comparison with the EO camera data. The
back-up photographic camera system utilized color
film. The primary purpose of this camera was to
generally document the scene. These photographs
are also useful as visual back-up by the analyst
working with the EO camera images. The
components of these photographic cameras are
listed in Appendix A.
3.1 Data Collection
2.3 Hardware Organization
The data collection plan involved the capture of
a pre-selected sequence of images from each of the
four camera systems in close temporal proximity.
The data collection procedure was designed to
support the experimental procedure illustrated in
Fig. 3.1. This figure shows the proposed
application of the data, as briefly discussed in the
previous section. In order to support this scenario,
the minimum data required were as follows:
The four camera systems were mounted, for this
deployment, on an expandable van as illustrated in
Figure 2.3. The van was provided as GFE by the
ASL host organization through their DELAS-AT-O
group as coordinated by Mr. R. W. Endlich. The
van provided both the protective enclosure for the
A. On an hourly basis, during daylight, coordin-
ated with Met observations:
3.2 Data Base Summary
1. One 128x128 image in each of two spectral
bands, 450nm, and 605nm.
2. One 256x256 image at 650nm.
3. One black and white photograph.
4. One color photograph.
A summary of the data base obtained during the
twenty-four day deployment interval is listed in
Table 1. On a typical ten hour data day, one would
complete events 5 through 15 on the data log (Fig
3.2) 10 times, and events 1 through 4 somewhat
less often.
B. During periods of dynamic cloud changes.
(Optional)
1. Sequence A above, at 5-minute intervals for
30-minute period.
2. Special sequences at the discretion of the
scientist in the field, for example shorter
intervals for a shorter period, substitute
small FOV for short samples, etc.
Appendix B contains a listing of those data from
EO Camera II which were subsequently digitized
and extracted for analysis. This digitization and the
subsequent analysis are discussed in the next
sections.
4.0 ANALYSIS OF ACQUIRED IMAGES
. A sample data log is shown in Figure 3.2.
Items A 1 through 4 in the above list are included
as events 5 through 9 in this data log. In addition,
the data log includes space for several optional
events: measurements of downwelling irradiance;
measurements both before and after the EO Camera
II with the other systems, and measurements with
more than one neutral density filter setting if
desired.
Following the field deployment, the camera data
were shipped back to the Visibility Laboratory for
analysis. The EO Camera I data consisted of tapes
of computer compatible 128 x 128 images. The
EO Camera II data were stored on video tape, in 15
second blocks of images recorded at roughly thirty
256 x 256 images per second. After these analog
video data arrived in-house, one image for each 15
second block of fisheye data was digitized,
yielding a computer compatible image which was
then stored on tape.
Thus the normal procedure was the collection of
the following data:
1. EO Camera I, Downwelling Irradiance:
450nm, 605nm, 555nm.
2. EO Camera II, Downwelling Irradiance:
650nm.
3. Automax 1, black and white (Pre EO II)
4. Automax 2, color (Pre EO II)
5. EO Camera I, fisheye: 450nm, 605nm
(Pre EO II)
6. EO Camera II, fisheye: 650nm
7. EO Camera I, fisheye: 450nm, 605nm
(Post EO II)
8. Automax 1, black and white (Post EO II)
9. Automax 2, color (Post EO II)
In this section, sample images from EO Camera
II are illustrated and described. They are then
compared with images from EO Camera I and from
the photographic camera, Automax I. Finally, the
work done with the image processing system on
these images to develop a preliminary cloud
detection algorithm is described.
4.1 Description of Sample EO Camera
Images
Sample images for EO Camera II are shown in
Fig's. 4.1 a-d. These figures illustrate a clear sky,
scattered cumulus, broken altocumulus, and high
broken cirrus conditions respectively.
In the above sequence, an Automax mea-
surement consists of generally 2-3 photographic
frames. A measurement by the EO Camera I
consists of one data file, which contains four
images taken one second apart. These four images
are essentially duplicates, giving a means to check
system noise. An EO Camera II measurement
consists of 15 seconds of data collection on the
VCR, and thus approximately 450 images. Only
one image for each of these measurements would
generally be extracted from each camera during
post-deployment processing.
Although the sensor array was rectangular, the
image placed on this array by the optics was round.
In these fisheye images, the center of the image is
the zenith, and the edges represent the horizon.
(The actual field of view extends to about 7 degrees
above the horizon.) Various structures may be
seen near the horizon, such as the small line near
the top of the image, which is a telephone pole to
the south of the van. The black rectangular object
is the sun occultor, used to shade the lens. The
occultor was an opaque frame supporting a 3-log
neutral density filter. The attenuated solar radiance
appears as a bright spot. A small grey scale
attached to the underside of the occultor may also
be seen. The time and date inserted by a time-date
generator on the recorded image appears near the
bottom of the scene
significantly inferior to the other images. The
clouds may be seen, however cloud edges are not
clearly defined, and cloud identification is difficult,
even for subjective viewer analysis. The EO
Camera II images are sufficiently superior to offset
the disadvantage of video recording, and for this
reason the remaining analysis in this report is based
on the EO Camera II data.
Although EO Camera II is referred to as having
256 x 256 resolution, the actual situation was
somewhat more complicated. The original CID
array had a 388 x 248 array, which was partially
filled by the circular image. The image was
recorded in video format, at which point the signal
was analog. When the signal was redigitized, it
was digitized into a 512 x 512 array. Thus the
actual images manipulated on the computer were
512 x 512, however the original resolution was
close to 256 x 256.
The Automax images shown in Fig. 4.2 are
slightly superior to those of the EO Camera II.
This is primarily due to the limited dynamic range
of the electro-optic camera. Use of the Automax
for the accumulation of real time results or
extended data bases is not practical, for two
reasons. One is the difficulty associated with
converting the photographic images to digital
images which could be utilized with a computer.
The second is the difficulty in maintaining
radiometric control with the photographic images,
thus making absolute calibration and comparison
with model clear day skies very difficult. The use
of the electro-optic camera overcomes these
shortcomings. From comparison with the
Automax I it is apparent that it does so with only
limited loss of digitized image clarity.
Figure 4.3 illustrates a line grab from the images
shown in Fig's. 4.1b, 4.2a and 4.2b. The signal
level for a row just below the center of the image is
shown for all three systems. In this figure, one
can see that the cloud edges are reasonably sharp
for both EO Camera II and the Automax I, and
somewhat more diffuse for EO Camera I.
The images in Fig. 4.1 qualitatively look quite
reasonable for our purposes. In the clear image,
one can see the higher radiance near the horizon
and in the upsun direction, especially near the
aureole. The cumulus cloud case shown in Fig.
4.1b shows a number of clouds of various sizes
and brightnesses. Both the angular resolution and
radiometric resolution are adequate for these clouds
to be quite obvious. The somewhat more difficult
cases of smaller clouds and thinner clouds are
shown in Fig. 4.1 c&d. In each of these scenes
the clouds are apparent to the eye. It should be
noted that the quality of the images as stored in the
computer and viewed on the image processing
display is somewhat higher than viewed in these
reproductions due to the degradations inherent in
converting the images to hard copy.
43. Development of Cloud Identification
Techniques
4.2 Comparison of Camera System
Images
For the images illustrated in the beginning of
this section, it is apparent that the information
necessary for cloud discrimination is inherent in the
image. That is, a subjective viewer of the image
can readily identify which sky sections are clear,
and which are obscured by clouds. At this point
then, the problem is to determine a quick and
reliable procedure by which a computer may make
objective identification of the cloud cover.
An image from EO Camera II is compared with
the images from EO Camera I and from the black
and white Automax I in Fig. 4.2. The EO Camera
I image is 128 x 128. The Automax image which
was originally recorded through a red filter, was
created by digitizing the film image. The digitizing
was done with a 100 um spot size, in order to
duplicate the pixel size of the EO Camera II as
closely as possible. With the spot size, the
resulting image was 256 x 256 pixels.
In Fig. 4.2, one can see that for reasons
discussed in Section II the EO Camera I image is
An immediately obvious feature of these
measured images is that the clouds tend to be
brighter than the background sky. An illustration
of this is shown in Fig. 4.4, which is a composite
plot of relative radiance values extracted along
identical sky arcs from electro-optical images
recorded on two successive days at approximately
identified are the overhead sky between the two
bright clouds, which is shown in grey and thus
erroneously identified as cloud, and a small patch
of cloud near the west horizon (the left edge of the
image) which is identified as sky. Thus one
trouble spot is near the sun between clouds, where
the sky becomes quite bright. Another is in clouds
near the horizon, where the radiance may become
less than the sky radiance over most of the image.
Measured Gradient
the same local time on each day. The first day, 3
September, was cloudy and the second day, 4
September, was clear. The surface weather obser-
vation for 3 September at 1258 LST shows 6/10
cumulus clouds with bases of 6500 ft. and a
prevailing surface visibility of 55 miles. For the
corresponding time on the following day, the sky
was clear except for a few isolated cumulus clouds
on the NE horizon, and the estimated visibility was
50 miles. The prescribed arc is a straight-line path
on the recorded image, extending from west to east
and passing near the zenith. In this case, a simple
instantaneous comparison of the relative signal
profile at 650nm on the cloudy day versus the
expected clear sky radiance profile at 650nm yields
good discrimination of the 3 September cloud field
along the designated arc segment. In the future we
plan to have images which are radiometrically
calibrated, and make a direct comparison with
modeled partly cloudy skies and spectral radiance
ratios. An algorithm could then be based on the
comparison of measured data to model calculations
derived from inherent image information.
Another source of information is the positional
rate of change of the radiance. In general, the
cloud edges in these images represent large abrupt
changes in radiance. Figure 4.6 shows an
application of this information. Figure 4.6a shows
a pseudo gradient derived from the cloud field
image. This image is the difference between
adjacent x pixel values, rectified and added to the
rectified difference between adjacenty pixel values.
Cloud edges may be clearly seen, even on the small
clouds in the lower half of the image. Near the
horizon the edges may be seen, however definition
of cloud vs hole becomes more difficult. A mask
based on the gradient image is superimposed on the
threshold image in Fig. 4.6b. Here one can see
that where the threshold detection fails, overhead,
the gradient provides the additional needed
information. Development of a simple algorithm to
make use of the gradient information is not trivial
however, due to the complexity of the gradient
field.
For this deployment, the lack of absolute
calibration precluded the use of algorithms based
on model vs measured radiances. It should also be
noted that algorithms based on the use of the
blue/red radiance ratio could not be applied, since
EO Camera II data were acquired only with a red
filter. However, for purposes of preliminary
analysis, the development of alternate algorithms
has been approached. This development helps give
the analyst experience with the quality of the
images, and the results may be useful in the later
development of algorithms for application to
calibrated imagery.
Identification using Image Ratios
Identification by Threshold
One problem with the radiance threshold
detection scheme is that the radiance of a clear sky
varies with look angle. Thus a threshold low
enough to correctly identify downsun clouds near
the horizon will incorrectly identify the upsun clear
sky, particularly near the aureole. One way around
this problem is to ratio the image with a clear day
image with comparable sun elevation and turbidity
conditions, thus normalizing out the clear sky
radiance changes.
The application of several test algorithms to one
image is illustrated in the next several figures.
Figure 4.5a illustrates the sample test image. The
simplest test was the definition of any point with a
radiance over a given threshold as cloud. Figure
4.5b illustrates this test. In this case, the threshold
chosen was 50 (where the recorded digitized signal
may vary from 0 to 255, and is monotonically
related to radiance). For this illustration, all points
with signal less than 50 are coded blue, and those
with signal greater than 50 retain their original
black and white color from the display screen.
This threshold worked reasonably well, as a first
approximation. Most of the clouds are correctly
identified. The two main areas not correctly
Figure 4.7 illustrates the ratio of the test image
signal with the clear sky image signal taken the
following day near the same time. In this image,
any ratio less than or equal to 1 has been color
coded blue, while any ratio between 1 and 1.4 has
been coded pink. Ratios higher that 1.4 have been
left black and white. In this image, the holes
between clouds have ratios very close to 1 for most
of the sky, as expected. The cloud edges
themselves have higher ratios, as does the region
directly overhead where there is a hole between
two bright clouds close to the solar aureole. These
pink-coded regions most probably represent the
enhanced path radiance which is expected near
cloud edges.
differed by 0.69 logs. Thus the radiance reaching
the sensor changed by a factor of 4.9 (or 10.69).
These cameras are reasonably linear with radiance
(not with log of radiance), so the signal should
have changed by an approximately constant factor
of about 4.9. Taking the ratio of the actual images,
it was found that the signal ratio varied from about
1.1 over much of the downsun sky to about 2.4 in
the upsun clouds. Figure 4.8 illustrates the ratio of
the two images. Checking other similar cases
revealed that the signal level change caused by
changing f-stops or neutral density filters was in
general not close to the expected value, nor similar
from one case to the next, and as illustrated in Fig.
4.8, not even constant within one image.
From Fig. 4.7a, it is apparent that for this test
case, identification of the clouds based on a signal
ratio threshold of 1.4 would be quite accurate.
That is, if the ratio of test image to the clear image
taken at the same time is greater than or equal to
1.4, that point would be characterized as cloud.
This procedure was applied to the image, with the
binary result given in Fig. 4.7b. In this figure,
points identified as clear or cloud are assigned
values of 0 or 1 respectively. Comparison with the
original image indicates that this result is
reasonably accurate.
One problem with applying this particular
identification algorithm to the general case would
be that the clear sky radiance itself depends on the
aerosol load in the atmosphere. One advantage of
comparison with a model rather than a measured
clear day image is that this variation could
potentially be allowed for.
The result of this unfortunate variance caused
presumably by the variable gain, is that two images
recorded with different set-ups cannot be compared
accurately, for this deployment. Even if one image
were normalized to the other at some point, the
within-image variance caused by this problem is
too large. Thus, for this data set it is not possible
to use the ratio of "test" to "clear" image algorithm
except in those limited cases where the cloudy
image is taken with the same setup as the clear
image.
Difference Image
Variable Gain Effects
For the example given in the previous
paragraphs, the cloudy and clear day images were
taken with the same f-stop and neutral density on
the camera system, and comparison of the
uncalibrated radiances yielded reasonable results.
For most of this deployment however, the images
were recorded with a variety of f-stop settings and
neutral density filters interposed in the path.
Unfortunately, although a change of neutral density
filter or f-stop causes a fixed ratio change of input
radiance, it does not cause a fixed change in signal
level, partly because the VCR which was used to
record the data had a variable gain which could not
be controlled.
Since comparison of signal levels between
images was a problem for this data set, it was
decided to try to develop a technique which would
identify the clouds within an image based on the
signals in that image alone. For this reason,
another look at the gradient field seemed in order.
There are various techniques for looking at closed
contours, however we needed an algorithm which
could be applied to one line from an image after it
had been extracted. For this reason, the difference
between the signal in adjacent x pixels was
investigated. Figure 4.9 shows this difference
image, color coded by magnitude. Points which
appear blue show little change between adjacent x
pixels. Points which are yellow, red, and black
show correspondingly greater differences between
x pixels.
To analyze the magnitude of the problem caused
by the variable gain in conjunction with the chang-
ing f-stops and neutral density filters, two images
were extracted from 11 September. These two
images were recorded approximately one minute
apart, first with a 0.7 log neutral density filter in
place, and then with the 1.3 log neutral density.
These values differ by 0.6 logs. At the actual wave-
length of 650nm, the measured filter responses
Several features may be noted in this
illustration. First, not all of the clouds have
particularly large differences at the edges. In fact,
some of the within-cloud edges show more
difference than the cloud edges. The line to the left
of point "a" in the illustration is one such within-
cloud variance. Point "b" points out one of the
5.0 CLOUD STATISTICS:
EXTRACTION AND RESULTS
actual cloud edges which shows less difference
than observed at point "a". Thus the simple
technique of looking for a large positive change
followed eventually by a large negative change,
and identifying the points between as cloud, would
be ineffective. Second, the clouds in general are
characterized by higher x differences than seen in
the clear sky sections of the image. However, the
centers of large clouds in several cases (such as
point "a") show low differences; thus identifying a
cloud by a threshold in the x-difference would be
ineffective.
One of the goals of this deployment was the
demonstration of a system which could be used to
automatically determine a variety of cloud statistics
for a given region. As discussed in previous
sections, the EO Camera system was successfully
deployed, yielding sky images in which the clouds
could be readily seen. This prototype system was
limited in that data were acquired using a video
recorder with variable gain. As shown in the
previous section, this limits the type of cloud
identification algorithms which can be applied.
Hybrid Difference and Threshold Algorithm
Although neither clouds nor cloud edges are
characterized by a given gradient in Fig. 4.9 (not
all cloud pixels have high gradient), it is noticeable
that all pixels with high gradient are cloud. This
was utilized in the following test.
The data were however adequate to generate a
reasonably accurate algorithm, which may be
applied to the data for demonstration purposes.
Accordingly, a computer program was developed
which could extract the desired data and apply the
prototype cloud algorithm, and extract illustrative
statistical data.
First, all points with a difference in x of more
than 3 were identified as cloud. Next, all those
points which passed this test (and were within a
center section of the row) were averaged, to yield
an approximate value for cloud radiance within this
image. Last, any additional points on that row
which had signals greater than or equal to .85 times
this average were identified as cloud also. Thus an
average cloud signal was determined based on
those points with a large x difference; then, any
points with a signal close to or higher than this
average were also identified as clouds.
In addition, cloud/no-cloud results were
manually extracted from the acquired imagery, and
resulting statistics were determined. This section
includes a brief description of the automated
extraction. This is followed by an analysis of the
resulting statistics, and a comparison with the
manually extracted data.
5.1 Automated Extraction of Cloud
Statistics from Images
This hybrid difference and threshold algorithm
worked very well for the test images. Two sample
images are shown in Fig. 4.10, in which the row
to be extracted is color coded blue if the algorithm
identifies the point as clear, and red if it is
identified as cloud. The technique was tested on
one row extracted for each image in the data base
(as discussed in the next section), and found to be
reasonably accurate except in a few cases. The
algorithm has the advantage that it is not highly
affected by radiance level or signal gain and thus
can be easily extended from a test case to all the
images. The main source of error is in the case
when the extracted row covers a range from upsun
to downsun (near dawn or sunset), as will be
discussed in the next section.
The automated extraction procedure makes use
of EO Camera II data extracted from the video
tapes and converted to digital format. These data
are processed using a mainframe computer. A
conceptualized flowchart of the processing
program is illustrated in Fig. 5.1. The program
extracts one row or column from each desired
image, and associates it with the appropriate
time/date information. A system such as illustrated
in Fig. 2.2c could automatically record time, date,
and filter information. With this prototype system
utilizing the video recorder, the time and date were
extracted from the manual field logs for input to the
program, and verified using the time/date
superimposed on each image by the time/date
generator.
The procedures for extracting one row for the
entire data base and applying the algorithm are
discussed in the next section, along with the results
of this processing
An algorithm is then applied to the line of data
extracted for each image, to convert it to a line of O
or 1 values, corresponding to a no cloud or cloud
decision respectively at each pixel along the line.
the longest cloud-free interval along the line is
determined for each image (here longest is defined
as the largest zenith angle change, not necessarily
the largest pixel count along the line). These
values are then combined to yield the cumulative
frequency of cloud free arc length. These results
will be discussed later in this section.
This decision algorithm takes the form described in
Section 4. That is, those points with high variance
along the line are assigned a value of 1, or
identified as cloud. An average of the signal values
for certain of these points is computed. Then other
points with signal greater than .85 times this
average are also assigned 1 values. The program
allows one to determine one average based on the
central portion of the line, or determine separate
averages for several segments of the line. Test
runs have been made which determine the
sensitivity of the results to the various constants
used in the algorithm. The constants derived on
the basis of the figures shown in Section 4 also
yield the best cumulative results in comparison
with the manual data.
5.2 Evaluation of Automated Extraction
Accuracy
Next, the zenith angle associated with each pixel
is computed. This computation takes the form
[(x-xo)2 + (y-y.)2]
* [a2(x-x.)2 + b2(y-y.)2] 1/2
Sample output from the cloud extraction
program are shown in Fig's. 5.3 and 5.4. Fig. 5.3
gives the frequency of cloud free line of sight as a
function of zenith angle for the extracted row. That
is, at each angle, which in turn corresponds to a
pixel location in the image, the percentage of cases
in the data base which were cloud free at that angle
is given. Figure 5.4 gives the cumulative
frequency for cloud free arc length. For each
image, the longest segment which is cloud free is
identified, and these results for all of the images
were combined to yield the cumulative frequency.
The statistical summary includes data extracted
from 233 images at 1-hour intervals for the hours
of 0700-1600 LST inclusive for the 23-day period
ending 20 September 1984. The prescribed arc
extends west-east and passes near the zenith.
where Omax = max 0 at edge of lens
Xo, Yo = center of elliptical image
a,b = axes of elliptical image
The geometric calibration was derived using
images such as illustrated in Fig. 5.2. This image
was acquired in a large hemispherical room, with
the camera lens mounted level with the base of the
hemisphere. The angles associated with the edges
of the various lit panels are known. The pixel
positions of these panel edges were extracted and
used to derive the above geometric calibration
equation. The results were found to be linear
within 2 degrees in zenith angle, and within less
than one pixel uncertaintly in azimuth angle.
In each plot two curves are given: the program
output, created using the program discussed in
Section 5.1, and the manual output, created from
the data pulled manually from the video images.
The third curve in Fig. 5.3 will be discussed in the
next section. Given the expected uncertainties
inherent in the current algorithm, (resulting from
the use of the uncalibrated data), the computed and
manual results are surprisingly close. In Fig. 5.3,
both curves indicate clear skies overhead in 80-
90% of the cases, with a lesser probability of cloud
free sky at the horizon. This drop near the horizon
is to be expected partly because a layer of clouds
overhead with even linear spacing in the horizontal
direction will have decreasing angular spacing as
the path of sight becomes more slanted away from

After the program applies the geometric
calibration to the data, the line for each image is
printed, yielding a listing of O's and i's as a
function of zenith angle along the line. When the
azimuthal angle exceeds 180 degrees (left half of
image), the value of -O is printed as a code in place
of 0.
The program then combines the results from all
images to yield certain statistics. The first extracted
statistic is the percentage of cases which were
cloudy at each angle, as a function of angle. Next,
In Fig. 5.3, although the computed and manual
results compare quite well for zenith angles near
-80 through +50 (west horizon through overhead
partway to east horizon), the computed results drop
lower than manual for zenith angles near +80.
Thus near the east horizon the algorithm tends to
identify too many cases as cloud.
9
Although the cumulative statistics in Fig's. 5.3
and 5.4 look quite good, this interim algorithm is
subject to uncertainty. The algorithm essentially
identifies cloud signal on the basis of edges and
any other points with considerable variance, and
then finds the rest of the clouds by comparison
with this signal. It is to be expected that problems
will occur if the cloud signal itself varies in certain
ways, or if the sky signal is too close to the cloud
signal.
frequency of cloud cover by category as recorded
at Ĉ Station for (1) the specific time period of EO
system operation which included the daylight hours
from 0700 LST to 1600 LST from 29 August to 20
September 1984, and (2) the long term average for
all hours in September for the period 1951 to 1973.
Making some allowance for an expected enhanced
frequency of small scale convective clouds during
the daylight hours (1/10-5/10 category), the cloud
conditions during the experimental period appear to
be about normal for late August and September at
White Sands, New Mexico.
Several individual cases were examined to
determine the extent of these problems. In most
cases, the identification of the cloud locations for
the extracted line compared quite well with the
manual identification. Probably the most common
inaccuracy occured near sunrise or sunset, when
the extracted line represents look angles quite near
the sun. In these cases, the upsun sky signals
were sometimes identified as cloud.
Systematic extraction of cloud/no-cloud data
from the digital EO camera imagery along
prescribed arcs yields the information required for
the objective determination of the observed
frequency of cloud-free intervals of varying lengths
along the track over selected periods of time. An
example is the observed cumulative frequency of
arc length for a pre-selected arc over the period of
EO observations at White Sands Missile Range, as
shown in Fig. 5.4. Note, for example, that the
frequency of cloud-free arc length greater than or
equal to 90 degrees was 75 percent for the White
Sands location for the specified period of time.
Another example of a case in which the
algorithm did less well was a case in which the
cloud edges were very bright relative to the rest of
the cloud. In this case the center of the cloud was
identified as sky, and only the proverbial silver
lining was identified as cloud.
The current algorithm is good enough to identify
many cases correctly, and can be used to
demonstrate the types of computations which may
be made with a system of this type. However, the
statistical results must be taken as approximate, and
the results for individual images cannot be trusted
in every case. The future availability of EO
Camera data with fixed gain and radiometric
calibration will give us the ability to apply
potentially much more accurate algorithms based
on comparison with clear day images and models,
and comparison of blue/red ratios with model
ratios.
The cumulative frequency of the longest arc
length should, of course, reflect the observed
frequency of clear skies for this period, which as
calculated from Station C surface weather
observations was 37.8 percent. The manually
extracted results fall close to this value, however,
the automatically extracted results are somewhat
lower. The 38% cumulative frequency occurs at an
arc length of 139 degrees, rather than at the
maximum, 148 degrees. Thus, the current
algorithm yields clear day cloud free arcs which are
often too short, indicating incorrect identification of
near-horizon pixels.
5.3 Evaluation of Extracted Statistics
The field site for the electro-optical (EO) system
experiments was established adjacent to C Station,
White Sands Missile Range, where a complete
program of surface weather observations is
maintained in accordance with prescribed standard
procedures. Sky condition observations from C
Station provide a basis for general assessment of
cloud cover during the course of the experiments
relative to climatological expectancy. Shown in
Fig. 5.5, is a comparison between the relative
Another stochastic feature of the observed sky
conditions easily derived from the archived digital
imagery is frequency distribution of cloud-free-line-
of-sight, CFLOS, as shown previously in Fig.
5.3. For the prescribed arc, the observed
frequency of CFLOS for all zenith viewing angles
less than 40 degrees ranged between 85 and 92 %.
Proceeding with the knowledge that observed
average sky conditions during the experimental
program did not differ markedly from normal, it is
of interest to compare the observed CFLOS
frequency along the single designated path with
model calculations. With the assumption that the
average cloud amount for all sky condition
-10-
observations designated clear on the Surface
Observation Form MFI-10 is 0.25, the calculated
average percent sky cover over the specific period
of 233 observations was 0.248. The probability of
CFLOS corresponding to this average cloud cover
as calculated by the geometric model of Allen and
Malic (1983) is given by the dashed line in Fig.
5.3. Their model, which fits the Lund and
Shanklin (1973) cloud data base reasonably well,
represents the probability of CFLOS by the
relationship
Data were acquired hourly for 24 days, and
subsequently digitized in-house. The resulting
images show the clouds clearly, in a variety of
meteorological conditions. The primary system
evaluated in this report, designated EO Camera II,
is superior to the prototype EO Camera I, which
has lower resolution. Comparison with
photographic camera images demonstrates the
excellent resolution and enhanced utility of the EO
Camera II images.
PCFLOS = P, (1 + btano )
where 0 is zenith viewing angle,
As a result of delivery problems, the primary
system was deployed uncalibrated. This limits the
types of cloud decision algorithms which may be
applied to this preliminary data base. However, an
algorithm has been developed which gives
reasonable representation for most images. Using
this algorithm as a demonstration, sample statistics
characterizing the observed cloud fields have been
extracted.
Po=1-n ( 1+3n )/4,
and
b = 0.55 - n/2
where n is sky cover in tenths.
The close correspondence in the simplified
model calculations and the observed CFLOS
frequency for this very limited example is
undoubtedly fortuitous. Furthermore, application
of the modeling approximation to the "average"
cloud cover is not necessarily commensurate with
the approach used in its development. However,
the important point is that the EO camera system
can be used to help establish a reliable and
comprehensive cloud data base in a form that can
be used readily to validate existing models and to
modify, adapt and extend these techniques. A
radiometrically calibrated EO system offers the
advantages of economical and reliable performance
for systematic field use and, with appropriate
algorithm development, the resultant data can be
tailored to provide specific and objective
determinations of sky condition and its statistical
behavior as required for a variety of applications.
Following the deployment of the camera system
in the field, a number of modifications have taken
place or are being developed at this time. These
may be briefly summarized here. Perhaps the most
important modification from the analyst's point of
view is the change from video recording to straight
digital output, with the frame grab performed
electronically and controlled by computer in the
field. This allows the recording of fixed gain, well
controlled data. As a result radiometric calibration
of the instrument becomes feasible. This in turn
makes comparison with radiative transfer model
calculations more practical.
In the realm of recent advances, the electronics
control function has been considerably enhanced
with the addition of computer control of the filter
mechanism. This filter mechanism includes both a
spectral wheel and a neutral density wheel,
allowing independent variation of the two filter
types. As a result, both blue filtered and red
filtered images may be acquired routinely at a
variety of neutral density settings. Additional
microcomputer control functions include writing of
identifying header information to tape preceeding
the image data, and writing of embedded time,
date, and filter information in each image.
6.0 SUMMARY AND ONGOING
DEVELOPMENTS
A preliminary data base has been obtained,
which demonstrates the feasibility of acquiring sky
images suitable for evaluation of cloud field
characteristics. The imaging system consists of a
solid state camera with various optics for control of
radiance levels and spectral content, with
associated electronics for controlling data
acquisition and storage.
Simultaneously with these mechanical
developments, software and analysis developments
have occurred. A version of the FASCAT
atmospheric model (Hering 1985) has been
implemented on the microcomputer. As a result of
parametric studies using this model, it has been
determined that, given fixed gain and calibrated
pow erment
imagery from the EO camera system, the blue-red
spectral ratio technique is potentially the most
promising candidate method of cloud discrim-
ination. Radiative model calculations clearly
indicate that a ratio of simultaneous images made
with narrow-band red and blue filters will provide
good cloud recognition over a broad range of
meteorological conditions. . Problems center
primarily on cloud analysis in the upsun direction,
near the horizon at sunrise and sunset, and of
course the identification of thin cirrus. For the
latter problem, analysis of the spatial variations in
the measured radiance field in the solar aureole
region appears to be the most promising approach.
7.0 REFERENCES
Allen H. and J.D. Malick (1983). "The
Frequency of Cloud-Free Viewing Intervals",
AIAA 21st Aerospace Sciences Meeting Paper,
AIAA-83-0441, 5, Reno, Nevada.
Hering, W.S. and R.W. Johnson (1985), "The
FASCAT Model Performance Under Fractional
Cloud Conditions and Related Studies", Univer-
sity of California, San Diego, Scripps Institu-
tion of Oceanography, Visibility Laboratory,
SIO Ref. 85-7, AFGL-TR-84-0168.
Lund, I.A. and M.D. Shanklin (1973), "Universal
Methods for Estimating Probabilites of Cloud-
Free Lines-of-Sight Through the Atmosphere",
J. Appl.Met., 12, pp 28-35.
In summary, as the continuing mechanical
improvements proceed, evaluation of images and
development of algorithms is continuing, utilizing
both the mainframe computer (primarily for
evaluation), and the minicomputer. The goal is, as
before, the development of accurate techniques
which may be implemented on a minicomputer and
deployed in the field for real time assessment of the
cloud information inherent in the acquired imagery.
Using the results of the White Sands field
demonstration of the prototype system, and
utilizing the improved capabilities of the current
hardware and software systems, we feel this goal
to be fully realizeable.
8.0 ACKNOWLEDGEMENTS
This work is performed under contract to Air
Force Geophysics Laboratory, Contract No.
F19628-84-K-0047. The authors are grateful to
Mr. D. Grantham of AFGL for his valuable
suggestions and guidance. We wish to thank Mr.
R.W. Endlich and the host for the field test,
Atmospheric Sciences Laboratory, for their
coordination and assistance. Finally we wish to
thank Mr. J.C. Brown, Mr. J. Rodriquez and Mr.
J.S. Fox of the Visibility Laboratory for their
support.
-12-

19"
mum 3.25"
3.25?
5.21"
--
-
M
- 6.50"
M
illi
willantolatum
1
.
"
.
, 128 x 128 ARRAY
.NEWS
.A
m.
RELAYED
IMAGE
LOCATION
In
Www
L21h 13
.
.
.
www.
.
with
WIB
SOLIGOR
FISHEYE
ADAPTER
Mill
:
GE TN2200
SOLID STATE CAMERA
(Front Housing Removed)
310-107-2
PRESSURE
WINDOW
HOUSING
310-107-6
SIX FLAG
FILTER
ASSEMBLY
310-106-55
TE TEMP CONT HOUSING
(Main Housing Only)
(a) Multi-spectral imaging fisheye scanner (EO Camera I)

common 11.03"
-
3.25"
-
3.25" -
seventeensandalen
3.75"
-----
Www
...........
242 X 377
ARRAY
L2 L3
L
LS
Nu
III|||||
SOLIGOR
FISITEYE
ADAPTER
GE 2505
SOLID STATE CAMERA
310-107-2
PRESSURE
WINDOW
HOUSING
310-107-6
SIX FLAG
FILTER
ASSEMBLY
ADAPTER
PLATE
AND
LENS CELL
(b) Multi-spectral whole sky imager (EO Camera II)
FIG. 2-1. EO Camera optical layout.
-13-

E/O CAMI
IN 2200
CAMERA ASS'Y
PN 2110A
CAMERA
CONTROLLER
CHIEFTAN
6809uP
FRAME GRABBER
PANASONIC
SOLID STATE
CCTV MONITOR
TR-920M
IRIS
ASSY
OCCULTOR
ASS'Y
MANUAL
CONTROL
HEWLETT/PACKARD
9825B
SYSTEM CONTROL
COMPUTER
DYLON 1015B
TAPE DRIVE
CONTROLLER
(a) Back-up system as-built for AUG'84 deployment.
(EO Camera I)
CIPHER
F 880 640
TAPE DRIVE
(b) Prototype system as-built for AUG '84 deployment. (EO Camera II)

PANASONIC
SOLID STATE CCTV
MONITOR
TR-920M
..
..
E/O CAMII
TN 2505
CAMERA ASSY
FOR. 9
VTG-22
TIME/DATE
GENERATOR
.
.
IRIS
OCCULTOR
ASSY
ASSY
SONY VO 5600
3/4" VCR
REMOTE PANEL
IRIS OCCULTOR
CONTROL CONTROL
(c) Composite system proposed for CFLOS algorithm development.

..
.......
....
....
......
MONITOR
VIDEO
E/O CAMII
TN 2505
CAMERA ASSY
VIDEO
TIME/DATE
GENERATOR
VIDEO
IRIS
ASSY
OCCULTOR
ASSY
VCR
REMOTE
VIDEO
DIGITAL
POYNTING 505
FRAME GRAB
ITS-100
CONTROLLER
CIPHER 9 TRACK
DIGITAL
RECORDER
IRIS
| OCCULTOR
ASSY
ASSY
DIGITAL
ZENITH
Z-100
COMPUTER
FIG. 2-2. Equipment organization schematic.
-14-

Fig. 2.-3. External Equipment Layout
Fig. 2-4. Instrumentation Hut on Van
-15-

SER
00
III
OOO
333333333333333333333
o
0
0
@CO
0
0
0
0
1 1 1
• TIMU II
Fig. 2-5. Internal Equipment Layout
- 16-
Table 3.1. Data Base Summary

EJO CAMERA I
Number of 4-Image Files
Fisheye
Filter 2 Filter 3 Irrad.
www
.
E/O CAMERA II ..
Number of 15-sec
Cassette Image Sets
| Number Fisheye irrad.
AUTOMAX 1 & 2
Number of
Film Sets
B&W Color
Data
.
Total
.
www
.
.......
.
.
.
..
...
....
........
29 AUG
30 AUG
31 AUG
1 SEP
2 SEP
3 SEP
4 SEP
5 SEP
6 SEP
7 SEP
8 SEP
9 SEP
10 SEP
11 SEP
12 SEP
13 SEP
14 SEP
15 SEP
16 SEP
17 SEP
.
..... W
....
on ooo oo w nan Bütöv Et ñ ñ ã no
WWV
W
WW
Wwwwwwwwwwwwwwwwwww ..
..
SEP
Tape
Ident
Number
8/9
10
11/12
vvvvoooo on orů AAWw wWNNNNN-
OOOOOOOOOOONNO W W OOO OO
....
...
...
19 SEP
20 SEP
21 SEP
-17-
mm


BACK-UP SYS
E/O CAMI
TN 2200
PROTOTYPE SYS
EO CAM II
TN 2505
35 mm AUTOMAX
BLACK & WHITE
35 mm AUTOMAX
COLOR
WHOLE SKY IMAGE
128 X 128 ARRAY
DIGITAL DATA
9 TRACK TAPE
WHOLE SKY IMAGE
242 X 377 ARRAY
COMPOSITE VIDEO
3/4" VCR
WHOLE SKY IMAGE
KODAK PLUS-X
PHOTOGRAPHIC RECORD
100 FT. REEL
WHOLE SKY IMAGE
KODAK VPS-3
PHOTOGRAPHIC RECORD
100 FT. REEL
.
...
...
..
.......
.
.
CHEMICAL
PROCESSING
CHEMICAL
PROCESSING
....................
..
.
www.
COMPUTERIZED
DIGITIZATION
E/O SCANNER
DIGITIZATION
RADIOMETRIC
RADIOMETRIC
GEOMETRIC
GEOMETRIC
RESOLUTION
COMPARISONS
RESOLUTION
COMPARISONS
VISUAL
RESOLUTION
ASSESSMENT
WWW
.
SELECTED ARC
EXTRACTION
PROCEDURE
....
APPLICATION OF
CLOUD DISCERNMENT
ALGORITHMS
WN
CALCULATION OF
CLOUD DISTRIBUTION
STATISTICS
FIG. 3-1. Experimental procedure - whole sky camera system.
-18-
E/O CAM FIELD DATA LOG
Standard Hourly Set

START
DATA WCMR An 15 sep 84
DATA
SITE
DATE
CHANGES SINCE LOG #1 OF THIS DATE (SET-UP):
LOG# FOR
THIS DATE
TIME
1141
.
..
........
..
....."
CLOUD CONDITIONS:
TONS: 60%
cirrus-
looks
like
a
, موه/ ورک
W
-
Event
No.
Filter
Flag
DIN
Comments
.
IDENT:
System
Frame, File,
(Hdr.)
Time
| or Count: | f/stop
E/O CAM-100_1141 file 371 ll
E/O CAM-1 (0) 11142 | file 38
E/O CAM-1 (1) 11142 file 39,40 1
Beg. 114245
E/O CAM-2 End 114300
Irrad.
Irrad.
4 1o Irrad. sun came out
w
ww
.
4
in
Irrad
ht
mm
Sky onscale
..
16
Automax 1 11145 11957-8
6 Automax 2 1145 11959-10 | - | - | -
E/O CAM-1 (G) 11148 file 41
8 E/O CAM-1 (G) 11148 | file 42 | ll 3 lo
Beg. 11475
9 I E/O CAM-2 Endivyo
10 E/O CAM-1 (G) 11149 file 4]
E/O CAM-1 (G) 11149 file 44
E/O CAM-1 (G) 1151 file 45
13 |E/O CAM-1 (G) 1751 | file 46
14 Automax 1 1or52 1961
15 Automax 2 1 1/521 1962
Clds onscale
.
NUM
w
ww
.
......
.
.
FIG. 3-2. Field Data Log.
-19-

mg-HALASAN
a) Clear sky
b) Cumulus clouds
09-
L
il:54: IL
c) Alto cumulus
d) Cirrus clouds
Fig. 4-1. Imagery from EO Camera Il for a variety of sky conditions. Images were digitized from video, piped
through image processing, displayed into matrix camera.
-20-

a) EO camera I expanded to 256 x 256
b) Automax I digitized at approximately 256 x 256
Fig. 4-2. Imagery from EO Camera I and Automax I for comparison with EO Camera II images. Images
acquired on 3 SEP 84 near 1401 hrs.

.
.
.
.
.
..
...
.
.
....
.......
.
.
......
.............
..
.
..
........
..
.
.
...
.
.
..
..
...
....
(a) EO Camera II. (see Fig. 4-1b)
(b) EO Camera I. (see Fig. 4-2a)
150 f
.
....
..
m,
..
.....w.
.
7.
...
.
....
......
.........
..
.
...
.
........
.
.
.
.....
so se escon
........
.
.
.
.
.
.
.
. ..
256
........
.
....
...
...
.
.
..
..
::...
. v
I'"
we'
*
.
;.
X
w., wir
,,
.
.......
......
......
........
..
.
..
.
SIGNAL
SIGNAL
.
: ..............
...
...
.
.....
..
..:
0...
512
64
128
256
PICTURE ELEMENT (Pixel) NUMBER
PICTURE ELEMENT (Pixel) NUMBER
(c) Automax I. (see Fig. 4-2b)
1001
SIGNAL
...............
.....
256
128
PICTURE ELEMENT (Pixel) NUMBER
Fig. 4-3. Line plot of cumulus cloud image for EO Camera II, EO Camera I and Automax I. The data were
acquired 3 SEP 84 near 1401 hrs. Row 282 from EO Camera II and corresponding rows of EO Camera I and
Automax | were extracted.


.
..
...
..
.
....
..
......w.w
Y
.....
.
......
.
....
....
..
..
..
..
.
......
150
- ..
.
..
.
....
... wwwwwah
.
..
8..w...
.
..
..........
.
.
.
...
.
.
...
..
.
..22.
..
.
..
.
.
.
w
3 SEP
ww.
CAMERA SIGNAL
..
..
.
.
.
....
studimas permanent
Marwah
Marturma
256
512

PICTURE ELEMENT (PIXEL) NUMBER
Fig. 4-4. Measured signal (from radiance at 650nm) as measured by the EO system on 3 SEP and 4 SEP 1984
at 1400 LST along an east-west arc passing near zenith. 3 SEP had broken cumulus, 4 SEP was clear.

Fig. 4-5. TEST OF RADIANCE THRESHOLDING FOR CLOUD DETECTION. 3 SEP 84 1401
ORIGINAL IMAGE 09/0385 1401
DETECTION BY RADIANCE THRESHOL
K50 = SKY > 50 CLD
(a) Original image.
(b) Image with all signal levels less
than 50 color-coded blue.
Fig. 4-6. GRADIENT INFORMATION IN CLOUD DETECTION TEST IMAGE. 3 SEP 84 1401
PSEUDO GRADIENT DERIVED FI
CLOUD FIELD IMAGE
PSEUDO SRADIENT BERAGE FRA
THRESHOLDED SKY
(<903 BLUE - SKY
WITH SUPER IMPOSED GRADIEN
AS
DELL
21
(a) Pseudo gradient derived from image.
(b) Pseudo gradient color-coded red,
superimposed on threshold color-
coded image.
-24-

Fig. 4-7. TEST OF CLOUD DISCRIMINATION USING RATIO OF TEST IMAGE TO CLEAR IMAGE DATA.
RATIO OF CLD IMACE TO CLEAR SK
RATIO 1.8
RAT10<1.4
BINARY CLOUD DISCRIMINATION
BASED ON RATIO OF CLOUD TO CLEA
IMAGES, THRESHOLD USE
no
(a) Ratio image. Ratios <1 coded blue,
ratios <1.4 coded pink, ratios >1.4
coded grey.
(b) Cloud discrimination result using a
ratio threshold of 1.4.
RATIO OF 2 IMAGES, BOTH AT 1055 ON 09/11/8
ONE WITH ND FILTER 9.7 LOGS, OTHER 1.39 LOG
DIFFERENCE IMAGE IN XORIG IMAGE 09/03/84 AT 14015
EACH PIXEL CONTAINS SIGNAL AT X MINUS SIONAL AT PIXEL X
FILTER RATIO IS 4.90 ABSOLUTE
OBSERVED RATIO COLOR SCALE
DIFFERENCE SCALE
Fig. 4-8. Ratio of images taken with
0.7 log filter and 1.39 log filter, both
acquired 11 SEP 84 1055.
Fig. 4-9. Difference image, shows
signal (x) - signal (x+1) 3 SEP 84 1401.
-25-

Fig. 4-10. CLOUD DISCRIMINATION FOR 1 ROW WITH HYBRID ALGORITHM, UTILIZING DIFFERENCE
INFORMATION AS WELL AS SIGNAL LEVEL
CLOUD IMAGE FROM @903/84 AT 1401
WITH ROW 282 PROCESSED USING DIFF AND THRESH ALGOR.
LATE AFTERNOON CLOUD IMAGE FROM 09/03/84 AT 1657
WITH ROW 282 PROCESSED USING DIFF AND THRESH ALGOR
BLUE = NO CLOUD
RED = CLOUD
BLUE = NO CLOUD
RED. CLOUD
VISIBILITY LAB
visiBILITE LAB
(a) Cross-sun case. 3 SEP 84 1401
(b) Upsun case. 3 SEP 84 1657.
-26-

START
TESTCLD
Tapes of
EO images
512x512
Read image, extract 1 row,
associate with time/date
file CLOUD. X
file of ident info; ie time, etc
file CLDARC
1 row from each image,
with ident info
END
TESTCLD
Read CLDARC, apply cloud
algorithm, cid=1, no cld = 0
START
CLDSTATS
CLDBIN
apply computed e for each x
o carpente tereta
binary file, 0 & 1,
cld results as a function of o
-
Accumulate results for each
row, compute freq.
CFLOS vs 0
file CLDSTAT
Detect longest cld free arc,
each image, accumulate
results of cum freq of
cld free arc length
CFLOS VS 0;
cumfreq, cld free
arc length vs arc length
END
CLDSTATS
Fig. 5-1. Conceptualized chart of EO Camera data processing program.
-27-

0021-84 b
Fig. 5-2. Sample imagery utilized for geometric calibration.
-28-
FREQUENCY CLD FREE LINE OF SIGHT
ter
FREQUENCY (%)
AUTOMATIC
- como MANUAL
-- MODEL
tot
OL
-80
-60
-40
-20
0
20
40
60
80

OBSERVATION ZENITH ANGLE
Fig. 5-3. Cloud free line of sight, frequency of occurrence vs zenith angle.

E
CUM FREQ CLD FREE ARC LENGTH
---
CUMULATIVE FREQ (%)
-30
AUTOMATIC
MANUAL
-
0
20
40
120
140
160
60 80 100
ARC LENGTH (DEGREES)
.
Fig. 5-4. Cloud free arc length, cumulative frequency of occurrence vs arc length.

29 AUG - 20 SEP
0700 - 1600 LST
1951 - 1973
All Hours
RELATIVE
FREQUENCY
%
CLEAR
ABOVE 9/10
1/10-5/10 6/10-9/10
SKY COVER
Fig. 5-5. Relative frequency of cloud cover at C Station, White Sands Missile Range.
-31-
APPENDIX A
TABLE A: WSMR DEMONSTRATION EQUIPMENT LIST
L.
BACK-UP SYSTEM, E/O CAM I
A. Detector Assembly
1. GE TN2200, 128X128 Solid State Camera, w/Controller.
2. C-130 Filter Assy. #8, (450nm, 605nm) - see AV84-063t (23 AUG 84).
3. Bicar Fisheye Lens Assembly, manual IRIS control.
4. Manual Solar Disc Occultor.
5. Prototype Irradiometer attachment (Cosmicor 19° FOV Back-up).
B. Monitor and Control Assemblies
1. C-130 Filter Control Panel.
2. Panasonic, model TR-920M Solid State T.V.
3. Dylon, model 1015B, Magnetic Tape Controller.
4. Cipher, model F880640-90-10250, Magnetic Tape Drive.
5. Hewlett-Packard 9825 Computer.
6. Computerware 5809-based Computer/Controller.
C. Inter-connecting Cabling
I.
PROTOTYPE SYSTEM, E/O CAM II
A. Detector Assembly
1. GE TN2505, 242x370 Solid State Camera, w/power supply.
2. C-130 Filter Assy. #12, (650nm) - see AV84-0631 (23 AUG 84).
3. Bicar Fisheye Lens Assembly, prototype remote control Iris.
4. Prototype remote control Solar Disc Occultor.
B. Monitor and Control Assemblies
1. CRM Filter Control Panel.
2. Panasonic, model TR-920M Solid State T.V.
3. Sony, model V05600, 3/4" VCR.
4. Inter-connecting Cabling.
III. PHOTOGRAPHIC CAMERA SYSTEMS (See AV84-052118 July 84)
A. Primary Camera, Automax, Model G1
1. Traid model 735, 180° Lens.
2. 100 ft. magazine.
3. Black and White Film, Kodak Plus-X.
B. Secondary Camera, Automax, Model G1
1. Traid model 735, 180° Lens.
2. 100 ft. magazine.
3. Color Film, Kodak VPS-3.
C. Back-up Camera, 35mm SLR Hand-held
1. Bicar Fisheye Adapter.
2. 36 exposure individual toll film
3. Black and White - Color.
D. Support Equipment
1. Automax Remote Control Panel.:
2. Time/Date Bracket for Automax.
3. Automax Control Cables.
IV. PERIPHERAL SUPPLIES
A. Wratten N.D. Filters
B. Filter Cutters
C. Clear Cover Plates
D. Minor Repair Tool Kit
E. Optical Cleaning Kit
F. Magnetic Tape, 2400 ft. reels, 30
G. Tape Mailer Case
H. Data Logs
L Procedural Check List
.32-
APPENDIX B
This section lists those EO Camera II images which have been extracted for data analysis. Whereas all of the fisheye
images have been digitized and may be computer accessed, some of the fisheye images were special purpose images,
such as extra data taken under certain cloud conditions. For the statistical analysis, one image for each hour has been
extracted and included in the computations. The following table lists these images. In addition, on certain days when
the sky was clear for several hours continuously, data were acquired every second hour. In these cases, the statistics
program has treated the times with no data as having a clear sky. Those cases with clear sky but no data are listed in
the second table.
In the first table, the date, time, F-stop, and neutral density filter were extracted from manual data logs. The tape file
refers to the location on tape of the digitized image. The last column indicates the clear data image number which is
closest in time, where 1 represents the image near 0800, 2 is 0900, and so on through clear image number 12 near
1900 hours.
eo Camera II Images Extracted for Analysis
__
CLOUD
IMAGE#
DATE
TIME
F-STOP
N.D.
FILTER
TAPE FILE
CLEAR
COMP#
16
18
co van A WNA
16
16
1
N
16
16
11
A
11
11
19
11
20
21
22
082984
082984
082984
082984
082984
082984
082984
082984
082984
082984
083084
083084
083084
083084
083084
083084
083184
083184
083184
083184
083184
083184
083184
083184
083184
083184
083184
083184
083184
090184
090184
090184
090184
090184
090184
090184
090184
090184
090184
090184
090284
090284
090284
090284
0902
1010
1100
1201
1300
1401
1500
1607
1702
1802
0804
0901
1001
1105
1203
1302
0808
0904
1000
1112
1202
1259
1411
1507
1607
1701
1753
1902
1927
0816
0901
0952
1054
1156
1301
1358
1512
1602
1656
1754
0753
0900
1002
1115
Awwa uwwawAAAAAAaaAAAAAAAnauwwa tuua AAAuuuaaa
AWNOOoo van AWNDROoooaan AWNman AWN-FOOoo van AWN
23
11
11
24
25
26.
27
11
28
OvauAWN
vo
...
CLOUD
IMAGE#
DATE
TIME
F-STOP
N.D.
FILTER
TAPE FILE
11
11
11
11
11
22
11
26
090284
090284
090284
090284
090284
090283
090284
090384
090384
090384
090384
090384
090384
090384
090384
090384
090384
090384
090484
090484
090484
090484
090484
090484
090484
090484
090484
090484
090484
090584
090584
090584
090584
090584
090584
090584
090684
090684
090684
090684
090684
090684
090684
090684
090684
090784
090784
090784
090784
090784
090784
090784
090784
090784
090784
090884
090884
090884
090884
1159
1302
1356
1458
1549
1655
1757
0757
0855
1001
1057
1151
1255
1401
1453
1556
1657
1756
0750
0853
0950
1100
1153
1248
1352
1503
1558
1654
1748
0814
0853
0954
1053
1204
1259
1356
0756
0858
1011
1150
1257
1359
1458
1555
1657
0815
0900
0958
1054
1156
1257
1359
1455
1551
1655
0856
1000
1056
1259
AAAA AWAAAAAAWA Awwwwwwwa Wa wwAAAAAWA A Wawwa A A wwwwwwAAAA Awwwa Aw
Dovau AWNO
75
80
85
87
88
89
20
21
93
11
11
11
25
26
96
27
97
11
11
8
11
CLEAR
COMP#
a AWNOOoo van A WN A Boo wauwawan AWNEO van AWNoooo van AWN-Oooo van
101
102
104
au AWN
11
.34-
......
CLOUD
IMAGE#
DATE
TIME
F-STOP
N.D.
FILTER
TAPE FILE
CLEAR
COMP#
105
106
107
108
109
111
115
116
117
118
119
120
121
122
123
124
125
126
oo u au AWNAN
127
128
129
130
ovau AWNAN
131
132
133
134
135
136
137
138
139
140
141
090884
090884
090884
090884
090984
090984
091084
091084
091084
091084
091084
091084
091084
091084
091084
091084
091084
091084
091184
091184
091184
091184
091184
091184
091184
091184
091184
091184
091184
091284
091284
091284
091284
091284
091284
091284
091284
091284
091384
091384
091384
091384
091384
091384
091384
091384
091384
091384
091484
091484
091484
091484
091484
091484
091484
091584
091584
091584
091584
1356
1429
1551
1701
0857
1054
0820
0849
0951
1052
1155
1257
1404
1453
1607
1702
1758
1850
0754
0854
1000
1055
1154
1300
1355
1457
1551
1654
1752
0801
0855
0951
1058
1155
1252
1353
1450
1555
0754
0851
0952
1054
1153
1306
1354
1548
1700
1759
0806
0852
0953
1043
1150
1248
1346
0758
0850
0948
1050
WA A A A AWAA ANUNN wwwwwAAN N N N A WA AuW NN NNA NA wwwuA AWAAAAAANA A A WNN
ܝܚ ܝܕ ܢܨ ܚ
142
ܬ ܟ
ܗ ܢ
143
144
145
146
147
148
149
150
ܣ ܩ ܝܚܕ ܢ ܚ
U
151
152
153
ܬ ܟ ܘܙ ܘ
154
155
11
11
156
157
158
11
159
11
11
11
160
161
162
163
A WNnau AWN-
11
5.6
164
16
165
166
167
CLOUD
IMAGE#
DATE
TIME
F-STOP
N.D.
FILTER
TAPE FILE
CLEAR
COMP#
..................
....
....
..
16
ا ب
16
ب ب ب
ب
1148
1300
1346
1454
1547
1647
1747
0754
0851
0955
1154
1351
1450
1552
11
11
ب ب
ب
168
169
170
171
172
173
174
175
176
177
179
181
182
183
184
185
186
187
189
16
ب ب
ب
16
ب ب
ب
1649
16
ب ب
ب ب
lo
ب
191
ب
192
ب
193
ب
194
195
ب ب
ب
ه
091584
091584
091584
091584
091584
091584
091584
091684
091684
091684
091684
091684
091684
091684
091684
091684
091784
091784
091784
091784
091784
091784
091784
091784
091784
091884
091884
091884
091884
091884
091884
091884
091884
091884
091984
091984
091984
091984
091984
091984
091984
091984
091984
092084
092084
092084
092084
092084
092084
092184
092184
092184
092184
092184
092184
092184
092184
092184
bbbbbb
196
197
199
201
202
203
204
205
206
207
208
209
211
213
214
215
216
217
218
219
220
16
1747
0755
0852
1049
1251
1353
1449
1554
1652
1754
0750
0953
1156
1254
1350
1456
1550
1653
1746
0753
0849
1053
1248
1351
1449
1552
1658
1757
1252
1355
1452
1551
1651
1754
0752
0854
0954
1049
1144
1248
1350
1450
1552
ه با یا بیا بیا بیا بیا بیا با ه
5.6
5.6
5.6
ه
ه ب
ه
ه
ه یا بیا بیا با
221
bbbn4
هم به در بره ه
OOoo van
.
o van AWNA Ovao Ooo vaan se oooo vanrw Ooo va ANO ooo vuWN
222
223
224
225
226
227
228
229
230
231
232
233
lo
ه ه ه ه
el
16
ا »
Clear Sky Incidents without recorded data
CLOUD
IMAGE#
DATE
TIME
103
110
112
113
114
178
180
090884
090984
090984
090984
090984
091684
091684
091784
091784
091884
091884
091984
091984
1200
1000
1200
1300
1400
1100
1300
1000
1200
0900
1100
1000
1200
188
190
198
200
210
212
-37.