Impact of AIRS and AMSUA
over the Antarctic region
Tom Auligné, Dale Barker, Zhiquan Liu, Hui-Chuan Lin, Hui Shao
National Center for Atmospheric Research
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Introduction: AMPS
The Antarctic Mesoscale Prediction System (AMPS)
• Real-time, experimental NWP modeling capability for Antarctica
• Purposes: Support of weather forecasting and scientific activities
AMPS Forecasts
Frequency:
2 / day
Initializations: 00 & 12 UTC
Duration:
36–120 hours
.
.
60-km
20-km, 6.7-km, 2.3-km
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Introduction: WRF-Var
• Variational DA system (3DVar)
• Lateral Boundary Conditions from FNL
• Specific Background Error Covariances
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
AMSUA Impact: Forecast Score vs. RS
• Horiz. resolution = 60km
• 57 Levels, Model top = 10hPa
• Full cycling
• NOAA 15/16/18 AMSU-A channels 4 to 9
• Radiance over ocean only
• Static Bias Correction (Harris and Kelly, 2001):
4 predictors
• Thinning 120km
• QC = thresholds on innovations
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
AIRS innovations: Channel Selection
Model
Top
Ozone
Model
Top
Solar
contamination
RTTOV
RTTOV
CRTM
CRTM
AIRS
AIRS
TT Jacobians
Jacobians
T
Surface
O3
Q
T
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
AIRS innovations: QC & Thinning
• Pixel-level QC
–
–
• Channel-level QC
Reject limb observations
Reject pixels over land and sea-ice
–
–
Gross check (innovations <15 K)
First-guess check (innovations < 3σo).
Error factor tuned from objective method
(Desrozier and Ivanov, 2001)
• NESDIS Cloud detection
–
–
• Imager AIRS/VIS-NIR
LW window channel > 271K
Thresholds on model SST minus SST
from 4 AIRS LW channels
Day only (cloud coverage within AIRS pixel <5%)
• Thinning
Thinning
Thinning (120km)
(120km)
Warmest
Warmest FoV
FoV
345
345 active
active data
data
696
696 active
active data
data
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
AIRS innovations: Cloud detection strategy
From « hole hunting »
(identifying clear pixels)…
… to identifying clear channels
(insensitive to the cloud).
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Cloud Detection: MMR scheme
RTM
•k
Rν
= Radiance calculated in clear sky
Minimum Residual:
RνCld = (1 − N )Rνo + NRν•k
= Radiance calculated for overcast black cloud
at level k
Rν• k / Rνo
Nk3
Vertical Level
Rνo
Nk2
Channel Number (LW band)
Multivariate Minimum
Residual (MMR):
Nk1
No
n
Rν = N Rν + ∑ N k Rν•k
Cld
Pixel
o
o
k =1
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Cloud Detection: MMR scheme
n
RνCld = N oRνo + ∑ N k Rν•k
k =1
Cloud fractions Nk are ajusted
variationally to fit observations:
J (N ) =
1 ⎛ Rν − Rν
⎜⎜
∑
2 ν ⎝
Rνo
Cld
Obs
⎞
⎟⎟
⎠
2
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
0 ≤ N k ≤ 1, ∀k ∈ [0, n]
with
n
N + ∑ Nk =1
o
k =1
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Cloud Detection: comparison with MODIS
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
”Cloud
Clearing”
RνObs − Rνo
RνObs − RνCld
« Retrieved » innovations
« Clear » innovations
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Bias Correction: Variational BC
Modeling of errors in satellite radiances:
ε =0
y = H (xt ) + B(β ) + ε
Parameters
Predictors:
N
B(β ) = ∑ βi pi
i =1
• Offset
• scan, scan2, scan3
Bias parameters can be estimated within the variational assimilation, jointly with the
atmospheric model state (Derber and Wu 1998) (Dee 2005) (Auligné et al. 2007)
Inclusion of the bias parameters in the control vector : xT º [x, β]T
Jb: background term for x
Jo: corrected observation term
−1
J(x, β ) = (x b − x)T B−1
x (x b − x) + [y − H (x) − B( β )] R [y − H (x) − B( β )]
T
+ (βb − β )T B−1
β (β b − β )
Jβ: background term for β
«Optimal
«Optimal »» bias
bias correction
correction
considering
all
available
considering all available information
information
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Bias Correction: VarBC Timeseries
No Inertia Constraint
Inertia Constraint
Innovations for AIRS window channel #787
After BC
Before BC
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Inverse Modeling: Adjoint Parameter Estim.
atm
Rν
⎡ dτ ( z , θ ) ⎤
dz
= ∫ Bν (T ( z )) ⎢ ν
⎥
z0
⎣ dz ⎦
∞
∞
τν ( z1 , θ ) = exp{−γ ν sec θ ∫ kν ( z )c( z ) ρ ( z )dz}
z1
B = Planck function
τ = transmittance
T = temperature
θ = incident angle
z = altitude coordinate
kv = absorption coefficient
c = mixing ratio
Ρ = atmospheric density
γ modulates atmospheric absorption to compensate for:
• poor knowledge of gas concentrations (CO2, …)
• errors in definition of ISRF
• errors in mean absorption coefficient
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Inverse Modeling: Parameter Estimation
Sensitivity to model error
Timeseries of γ estimations
γ-1
Analysis cycle
Auligné (2007)
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Feedback: Bias Correction and Cloud Detection
Cloud
Cloud Detection
Detection Independent
Independent from
from VarBC
VarBC
Number of active observations
Before bias correction
After BC
Bias
Cloud
correction detection
Auligné and McNally (2007)
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
AIRS Impact: Forecast Score vs. Radiosondes
Whole Domain
High Latitudes (> 60S)
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Model Error: Major Limitation
Mean Analysis
Increments
Mean Forecast Error
(24h-12h)
Surface
50 hPa
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008
Conclusions and future work
•
Encouraging results
–
–
•
Short-term improvements
–
–
–
–
•
Conservative implementation
Slightly positive impact on forecast quality
Raise model top
Use more AIRS channels
Test simple hybrid DA schemes (random+systematic errors)
Compare impact to retrievals
Longer-term developments
–
–
Improved estimation of surface emissivity over land
Assimilate cloud-contaminated radiances
International TOVS Study Conference XVI. Angra dos Reis, Brazil. May 7-13 2008