Bias correction of window channels on microwave and ... · Email:[email protected] 0.0 0.2 0.4 0.6...
1
LCC(Low Cloud Cover,p>800hpa) Bias correction of window channels on microwave and Infrared sounders Wei HAN 1 and Tony McNally 2 1.Chinese Academy of Meteorological Sciences, CMA 2. ECMWF ABSTRACT The This study is focused on the bias correction of window channels on microwave and infrared sounders (e.g. AMSUA and HIRS). These channels are primarily used to quality control other sounding channels by virtue of their enhanced sensitivity to cloud contamination. In general these channels may be susceptible to systematic errors like any other channel and as such must be bias corrected before they are used. It will be shown that their sensitivity to clouds poses particular problems for the evaluation and correction of systematic errors and some potential solutions are proposed. 1. Interaction between bias correction and quality control -The are two primary sources of interaction a) outliers (e.g. cloudy data) -b) asymmetric quality control (AQC) (e.g. warm/cold) -These sources can influence bias estimation in a static scheme -These sources can lead to a feedback loop and a drift in an adaptive bias estimation(e.g. VARBC) 2. Mean F-G departures (bias corrected) -Clear radiances over sea -Clean observation bias? Or dragged by cloudy populations? 3. Theoretical study using the offline toy model -Mode, Mean and Bias - Window channel of AMSUA and HIRS: A4 and H8 - Robust mode estimation in VarBC: two ways Tight QC: will give the mode by box-car QC if there is a dominant mode Robust pseudo-mode estimation( Weighting function) 4. IFS experiment of tight QC 5. Conclusions and discussions 1. Mode,Mean and Bias Reference Auligne, T. and A.P.McNally (2007). Interaction between bias correction and quality control. Quart. J. Roy. Meteorol. Soc. 133, 643–653. Han Wei and Anthony McNally, 2008: Bias correction of window channels on microwave and infrared sounders, NWP-SAF Visiting Scientist Report NWPSAF-EC-VS-016. 2. Mean F-G Departures (2007D2008JF,bias corrected) 3. Theoretical study using the offline toy model 2008DJF(2007120100-2008022800) O-B, Window Chs(A4 and H8) NOAA15 A4 NOAA18 A4 METOP A4 METOP H8 (Auligne and McNally 2007) Mean,Pseudo-Mode and Mode estimated bias correction will be“dragged” towards the outliers Pseudo-mode is not sensitive to QC 2008011200, 12h cycle(-9,+3) IFS CY32R3 Bias estimated in VarBC tends to be the mode: AMUSA Ch4 NOAA15: -0.3 f0u3: tight QC ezht: control HISTOGRAM: VarBC.cycle.2008011200 METOP: +0.2 NOAA18: -0.2 IFS CY32R3 Operational FG Depar. MODE BC FG Depar. NOAA15 A4 NOAA18 A4 METOP A4 Mode is the clear radiance bias if background and RT model are bias free The Toy model Agree well with f0xx 2 2 1 1 ( ) ( )( ) 2 n i i J m d m w n = − ∑ 4. IFS experiment Email:[email protected] 0.0 0.2 0.4 0.6 0.8 1.0 -135 -90 -45 0 45 90 135 -135 -90 -45 0 45 90 135 -60 -30 0 30 60 -60 -30 0 30 60 NOAA-15 NOAA-18 METOP-2 TCC(Total Cloud Cover) Impact of tight QC for BC of WinCH on the usage of AMSUA Ch5,6,7,8 Use more data in CLEAR region Use less data in CLOUDY region It seems that Bias Corrected FG-Depar. of CH4 in clear cases tends to be 0 -0.10 -0.05 0.00 0.05 0.10 Normalised difference in RMS error RMS forecast errors in Z(f0u3-ezht), 12-Jan-2008 to 22-Mar-2008, from 61 to 71 samples. T+12 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+24 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+48 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+72 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+96 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+120 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+144 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+168 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+192 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa T+216 -90 -60 -30 0 30 60 90 Latitude 1000 800 600 400 200 Pressure, hPa f0u3 better ezht better TQC better TQC worse TQC better TQC worse RMS AC 500 hPa geopotential ANCF and REF (61 cases, 20080112-20080312, own analysis) Acknowledgements to NWP SAF VS project
Transcript of Bias correction of window channels on microwave and ... · Email:[email protected] 0.0 0.2 0.4 0.6...
LCC(Low Cloud Cover,p>800hpa)
Bias correction of window channels
on microwave and Infrared soundersWei HAN1 and Tony McNally2
1.Chinese Academy of Meteorological Sciences, CMA2. ECMWF
ABSTRACTThe This study is focused on the bias correction of window channels on microwave and infrared sounders (e.g. AMSUA and HIRS). These channels are primarily used to quality control other sounding channels by virtue of their enhanced sensitivity to cloud contamination. In general these channels may be susceptible to systematic errors like any other channel and as such must be bias corrected before they are used. It will be shown that their sensitivity to clouds poses particular problems for the evaluation and correction of systematic errors and some potential solutions are proposed.
1. Interaction between bias correction and quality control-The are two primary sources of interaction
a) outliers (e.g. cloudy data)
-b) asymmetric quality control (AQC) (e.g. warm/cold)
-These sources can influence bias estimation in a static scheme
-These sources can lead to a feedback loop and a drift in an adaptive
bias estimation(e.g. VARBC)
2. Mean F-G departures (bias corrected) -Clear radiances over sea-Clean observation bias? Or dragged by cloudy populations?
3. Theoretical study using the offline toy model-Mode, Mean and Bias
- Window channel of AMSUA and HIRS: A4 and H8
- Robust mode estimation in VarBC: two ways
Tight QC: will give the mode by box-car QC if there is a dominant mode
Robust pseudo-mode estimation( Weighting function)
4. IFS experiment of tight QC
5. Conclusions and discussions
1. Mode,Mean and Bias
ReferenceAuligne, T. and A.P.McNally (2007). Interaction between bias correction and quality control. Quart. J. Roy.
Meteorol. Soc. 133, 643–653.
Han Wei and Anthony McNally, 2008: Bias correction of window channels on microwave and infrared sounders, NWP-SAF Visiting Scientist Report NWPSAF-EC-VS-016.
2. Mean F-G Departures (2007D2008JF,bias corrected)
3. Theoretical study using the offline toy model
2008DJF(2007120100-2008022800)O-B, Window Chs(A4 and H8)
NOAA15 A4 NOAA18 A4
METOP A4 METOP H8
(Auligne and McNally 2007)
Mean,Pseudo-Mode and Mode
estimated bias correction will be“dragged”towards the
outliers
Pseudo-mode is not sensitive to QC
2008011200, 12h cycle(-9,+3)
IFS CY32R3
Bias estimated in VarBC tends to be the mode: AMUSA Ch4
NOAA15: -0.3
f0u3: tight QC
ezht: controlHISTOGRAM:
VarBC.cycle.2008011200
METOP: +0.2
NOAA18: -0.2
IFS CY32R3
Operational FG Depar.
MODE BC
FG Depar.NOAA15 A4
NOAA18 A4
METOP A4
Mode is the clear radiance bias if background and RT model are bias free
The Toy model Agree well with f0xx
2 2
1
1( ) ( ) ( )2
n
i iJ m d m wn
= −∑
4. IFS experiment
Email:[email protected]
0.0
0.2
0.4
0.6
0.8
1.0
-135 -90 -45 0 45 90 135
-135 -90 -45 0 45 90 135
-60
-30
030
60
-60-30
030
60
NOAA-15
NOAA-18
METOP-2
TCC(Total Cloud Cover)
Impact of tight QC for BC of WinCHon the usage of AMSUA Ch5,6,7,8
Use more data in CLEAR region
Use less data in CLOUDY region
It seems that Bias Corrected FG-Depar.
of CH4 in clear cases tends to be 0
-0.10
-0.05
0.00
0.05
0.10
Nor
mal
ised
diff
eren
ce in
RM
S e
rror
RMS forecast errors in Z(f0u3-ezht), 12-Jan-2008 to 22-Mar-2008, from 61 to 71 samples.
T+12
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+24
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+48
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+72
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+96
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+120
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+144
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+168
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+192
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
T+216
-90 -60 -30 0 30 60 90Latitude
1000
800
600
400
200
Pre
ssur
e, h
Pa
f0u3 better
ezhtbetter
TQCbetterTQC
worse
TQCbetterTQC
worseRMS
AC
500 hPa geopotential ANCF and REF(61 cases, 20080112-20080312, own analysis)
Acknowledgements to
NWP SAF VS project
