Analysis of ATMS and AMSU Striping Noise from Their Earth Scene Observations

advertisement
Analysis of
ATMS and AMSU Striping Noise
from Their Earth Scene Observations
Zhengkun Qin1, Xiaolei Zou1, Fuzhong Weng2
1. Center for data assimilation in research and applications Nanjing
University of Information Science & Technology, China
2. National Environmental Satellite, Data & Information Service, National
Oceanic and Atmsopheric Administration, Washington, DC, USA
March 26, 2014
Outline
•
•
•
•
•
ATMS Striping Phenomena
Striping analysis based on PCA and EEMD
Striping noise of water vapor channels
Striping analysis of Maneuver data
Summary
Suomi NPP ATMS Striping
It was first identified by NWP user community
through the global distribution of the bias (O-B)
between ATMS observations and simulations. Unlike
AMSU-A, ATMS O-B displays a distinct O-B
irregularity in its along-track direction. NWP users refer
this irregularity as striping brightness temperatures or
radiances.
Global Distribution of ATMS Ch10 O-B
February 24, 2012
Obvious Stripping noise can be found in the O-B figure of Ch10
K
K
Principle Component Analysis (PCA)
Data Matrix:
A96N
 TB1,1  TB1, N 


 

 
 TB

 96,1  TB96, N 
TBk, j : Brightness temperature at the kth FOV
and the jth scanline
k=1,2,…,96; j=1,2,…,N
j
k
Principle Component Analysis (PCA)
The covariance matrix: S  AA T


Sei  i ei
 
 T T
u1 u2  u96 E A

ei   e1,i


e2,i  e96,i 
: the principle component (PC) for the ith mode.

ui   ui ,1 ui ,2  ui , N 
: the PC coefficient for the ith mode.
The ATMS data can be then expressed as:

A   ei u i
96
i 1
PCA modes of TB for ATMS Ch10
TB
Scan angle dependence of the
brightness temperature
Variations same for all FOVs are
tend to appear in the 1st mode
K
 
e2 u 2

e1 u1
K
 
e3 u 3
K
K
PCA modes of TB for NOAA-18 AMSU-A Ch9
TB
Same features as ATMS Ch10, but
the first mode is more smooth.
K
 
e2 u 2
 
e1 u1
K
 
e3 u 3
K
K
Brightness Temperature at nadir
ATMS Ch10
TB(1) (K)
TB
nadir

 e48,1 u i
Observation
Simulation
NOAA-18 AMSU-A Ch9
Obvious oscillations can
be found in ATMS, but not in
TB(1) (K)
Number of Scanline
AMSU-A
Number of Scanline
EEMD Method
Raw data
R0 t   Tb (t)
Intrinsic Mode Function (IMFs)
Cn 1 
Mean of the envelopes of Rn-1
Rn t   Rn 1 (t)  Cn 1
n
Tb t    C j t   Rn t 
j 1
Huang and Wu (2008) and Wu and Huang (2009)
Brightness temperature
for the first PCA mode at nadir
TB-1st IMF
e48,1 u1,j (K)
e48,1 u1,j (K)
TB at Nair
TB-first two IMFs
TB-first three IMF
e48,1 u1,j (K)
Scanline
e48,1 u1,j (K)
Scanline
Scanline
Scanline
C1 (K)
Amplitude (K)
Power Spectral of the first three IMFs
C3 (K)
C2 (K)
Frequency (s-1)
Scanline
Power spectral density distributions
Power Spectrum Density
SNPP ATMS channel 10
Before
After
Frequency (s-1)
Power spectral is modified after
striping noise removed for ATMS,
but there are just a little differences
for AMSU-A before and after
EEMD smoothing.
Power Spectrum Density
NOAA-18 AMSU-A channel 9
Frequency (s-1)
Global distribution of Stripping noise
Before-After
Of O-B for
ATMS Ch10
K
After
Before
K
K
Obs and O-B of the TB
O-B ATMS Ch22
Obs ATMS Ch22
LWP
K
K
Obs MHS Ch3
Kg/m2
IWP
O-B MHS Ch3
K
K
Kg/m2
PCA modes of TB for ATMS Ch22
TB
K
 
e2 u 2

e1 u1
K
 
e3 u 3
K
K
Eigenvector and TB at nadir
NOAA-18 MHS Ch3
e1
e1
ATMS Ch22
FOV
TB(1) (K)
TB(1) (K)
FOV
Scanline
Scanline
Striping noise can be found in both ATMS Ch22 and NOAA-18 MHS Ch3
Global distribution of stripping noise
SNPP ATMS Ch 22
NOAA-18 MHS Ch3
NOAA-16 AMSU-B Ch3
K
Stripping noises are stronger in
ATMS and MHS, but weaker in
AMSU-B.
Striping noise of maneuver TB
ATMS Ch22
Scanline
Scanline
ATMS Ch10
K
Scanline
Scanline
K
Stripping noise
can be removed in
maneuver data also.
FOV
K
K
K
FOV
FOV
K
FOV
Striping Index (SI) of maneuver TB
Valong
Valong
1 N  1
 
N j1  M
Vcross
1

M
Vcross
s
Vcross
2
 
 Tb (k, j) 
k1

M
2
1 N 
 
1 N
  N   Tb (k, j)  N  Tb (k, j) 
k1 
j1
j1

M
SI 
Valong
Vcross
along-track variance Valong
cross-track variance Vcross
Channel
Stripping Index (SI)
is reduced to close to one.
It shows that our method
can removed stripping
noise reasonably.
Striping Index
Variance (K2)
s
Valong

1
  Tb (k, j)  M
k1
M
Channel
Summary
• For tempearture channels, ATMS’s striping noise
is much stronger than AMSU-A
• For water vapor channels, both ATMS and MHS
(AMSU-B) have striping noise
• Method based on PCA and EEMD can filter the
striping noise obviously and flexibly
• Root-cause and more accurate filters need further
investigation
Thank you !
Questions ?
Download