Cloud Detection

1
Cloud Detection

• 1) Optimised CI Microwindowscnc
• 2) Singular Vector Decomposition
• 3) Comparison of Methods fffffffffff

2
1) CI Microwindow Optimisation
3
Currently MW1 788.2, 796.25 cm-1 MW2
832.3, 834.4 cm-1 CI LMW1 / LMW2 If CI lt
threshold ? cloud If CI gt threshold ? clear
operational threshold 1.8
CRISTA experiment
Aim Find a better pair of MWs, and/or a better
threshold value, using objective criteria based
on simulated spectra with known cloud amounts
4
(No Transcript)
5
(No Transcript)
6
Best MWs are those which best correlate CI with
CEF Current MWs show linear relationship for
a,b minimumizing
7
Iterative approach (Desmond) Search through MWs
with integer wavenumber boundaries and then, for
each 'coarse' MW, iterate moving each boundary
one grid point at a time.
MW1
MW2
RMSE Current MWs 788.2, 796.25 832.3,
834.4 0.181 Optimised MWs 774.075, 775.0
819.175, 819.95 0.157
8
Monte-Carlo approach Randomly-selecting MWs from
the domain (specified by mid-point and width) and
iterating from these to adjust the boundaries
10000 different MW pairs randomly selected from
the entire 750970 cm-1. Select region of lowest
RMSE and do another 10000 iterations. Repeat.
9
Another criterion Best MWs will have large
relative distance between clear and cloudy
distributions of CI RelDist (mean CIclear
mean CIcloudy) / (stddevclear stddevcloud)
10
Current MWs have RelDist 2.03
11
Summary and Future Work
MW1
MW2
RMSE RelDist Current MWs 788.2, 796.25
832.3, 834.4 0.181 2.03 Desmond MWs
774.075, 775.0 819.175, 819.95 0.157
na M.C. RMSE MWs 777.0, 779.0
819.0, 820.0 0.156
na M.C. RelDist MWs 800.0, 802.0
831.0, 832.0 na 2.77

• In future
• Iterate within M.C MWs to find exact location of
  min/maximum
• See how the two agree
• Test to see how rigorous each set of MWs is at
  cloud detection and EF estimation

12
2) Singular Vector Decomposition
13

• Singular Vector Decomposition SVD
• is statistical technique used for finding
  patterns in high dimensional data
• mn matrix A can be decomposed into
• AV DU
• V mm left-singular vectors
• U mn right-singular vectors
• D mm singular values
• transforms a number of potentially correlated
  variables into a smaller number of uncorrelated
  variables (SINGULAR VECTORS)

orthonormal matrices
diagonal matrix
14
In this case A is a set of m spectra each of
length n Each row of U is a singular vector with
n spectral points Singular value Dii weights
the Uj singular vector.
Idea is to find singular vectors that describe
clear and cloudy atmospheres and use them in
cloud detection
15
Calculate N clear singular vectors SVclear
16
Calculate M cloudy singular vectors SVcloudy
17
Use SVclear and SVcloud to do a Least Squares Fit
of arbitrary signal L(?) ?Ni ci SVclear i ?Mj
dj SVclear j
15km
12km
9km
6km
18
Chi-Squared Ratio Test
, and then threshold
19
Integrated Radiance Ratio Test
20
Summary and Future Work

1. Have successfully calculated SVs to represent
   atmospheric constituent variability (SVclear) and
   SVs to capture variability in cloud spectra
   (SVcloud)
2. Have implemented two detection methods and have
   defined thresholds using simulated and real MIPAS
   data
3. Have tested proficiency using simulated data

• Complete full comparison of different cloud
  detection methods used to date.

21
3) Comparison of Detection Methods
22
Comparison of Detection Methods
1. Current Operational CI 2. Optimised CI
microwindows 3. SVD chi-squared ratio 4. SVD
integrated radiance ratio 5. Simple radiance
threshold
Idea Compare retrievals (using MORSE) of
'well-mixed' gases assuming that using spectra
with residual cloud will result in retrievals
which deviate significantly from climatology
23
Analysis done on cases where Different
cloud-detection methods disagree over whether it
is clear/cloudy and only use the clear cases
24
Summary and Future Work

1. Std. Deviations in VMRs from climatological means
   for retrieved well-mixed trace gases from MORSE
   should give measure of strength of each detection
   method
2. No clear winner yet

• Continue testing and comparing CIRA
  climatology??