Transcript Presentation - Copernicus.org

New optical package and algorithms
for the accurate cloud cover
estimation for short wave
parameterization.
Mikhail Krinitskiy, Alexey Sinitsyn, Sergey Gulev
P.P.Shirshov Institute of Oceanology, RAS, SAIL,
Moscow, Russia
The problem
• Short-wave parameterization by cloud cover
(Alexey Sinitsyn, since 2004)
– Cloud cover estimation using handy optical fish-eye
camera sky images with the set of most common
cloud cover estimation algorithms
We present now:
The package for automatic cloud cover estimation using
fish-eye optical module operated by PC.
+ new algorithm for automatic cloud cover estimation
2 / 12
Optical package. Previous revision.
- Nikon D80 10MPx
- Sigma 8mm F3,5 Circular-Fisheye
- View angle 180 ° х 120 °
- Projection of fisheye lens:
equidistant projection
- Manual shooting
PROBLEMS to solve:
Image resolution:
3872x2592 px
- Cuts 20% of image
- Requires operator
assistance
3 / 12
Optical package. New revision.
Lab view
CONTAINS:
- Mini-computer with
positioning sensors (GPS,
accelerometer and others)
- Hi-res. fish-eye camera
Mounted assembly
- Connected to the PC,
monitoring the position
and managing shooting
process
- No operator assistance
required
4 / 12
Optical package. New revision.
SKY IMAGES: over 18000 shots in 9 days
- 1920 x 1920px
- 24-bit RGB
- 180x180 view angle
Observations: North Atlantic
17.09.2014 - 25.09.2014
5 / 12
Optical package. New revision.
SOFTWARE
Outdoor
part
minicomputer firmware
• collecting sensors data
• communicating
the PC
with
Windows-based PC software
• communicating with the outdoor
set (computer & camera)
• monitoring
positioning
and
managing shooting process
• processing sky images
6 / 12
Processing images: the algorithm
Most common processing schemes
Si = (R-B)/(R+B)
(Yamashita, Yoshimura, Nakashizuka,
2004)
Si = R/B
(LONG and DELUISI, 1998)
then threshold using fixed value
Sim
PROBLEMS to solve:
- Sun disk counted as cloudy area
- Cloud cover systematic underestimation (up to 2-3
Octas)
- Strong thresholding value dependency of the result7 / 12
Processing images: the algorithm
Our improved scheme: Grayness rate Index
R,G,B – color components
Y – brightness of pixel
• Depends (negatively)
only of color
saturation.
• Highly sensitive to
every gray colors
(white, gray or black)
0.5
0.0
Cloud cover systematic
underestimation – solved
(caught thin clouds)
8 / 12
GrIx
• Independent of Y
1.0
Processing images: the algorithm
Our improved scheme: Sunburn effect suppression
1.0
0.9
source field
GrIx
0.8
0.2
Sun disk as cloudy area solved
Thresholding value
dependency of the result solved
sunburn suppressed
0.1
0.0
0
100
200
x, px
300
400
500
9 / 12
Processing images: the algorithm
Our improved scheme: SOME RESULTS
Cloud cover estimation (Grayness Index - based with sunburn effect suppression)
Deviation of human-observed values.
60
2007yr. campaign
80
2010yr. campaign
50
60
40
30
40
20
20
10
0
-8
-6
-4
-2
0
+2
+4
+6
+8
0
-8
-6
-4
-2
0
+2
+4
+6
10 / 12
+8
Processing images: the algorithm
Our improved scheme: SOME RESULTS
Source sky image
Visual observed cloud
cover: 3 (of 10) – 30%
Common schemes
result
Cloud cover = 13%
AND EVEN MORE!
Our algorithm
result
Cloud cover = 31%
11 / 12
One more thing
Lower clouds margin height: basics
L = 17,162 m
H
L
12 / 12