Direct assimilation of all-sky
microwave radiances at ECMWF
Peter Bauer, Alan Geer, Philippe Lopez,
Deborah Salmond
European Centre for Medium-Range Weather Forecasts
Reading, Berkshire, UK
Slide 1
17th ITSC 04/2010
ⒸECMWF
1
Microwave imager ‘all-sky’ approach
MHS Obs-Model mean
and standard deviation
Clear-clear
Cloud-cloud
Obs Clear-FG cloud
Obs Cloud-FG clear
37 GHz (v) Obs - FG
All-sky
Slide 2
CTRL
In global Numerical Weather Prediction (NWP) systems, satellite observations
provide 90-95% of the actively assimilated data. However, over 75% of satellite
observations are discarded due to cloud contamination and unknown land, snow
and sea-ice surface emissivity. Cloud-affected data is difficult to deal with
because of the limited accuracy of moist physics parameterizations in numerical
models and the need to model radiative transfer through clouds and precipitation.
Errico et al. (2007, JAS, 65, 3348-3350) summarise the most important issues
related to the assimilation of cloud and precipitation observations. One prominent
problem is the non-linearity (and non-continuity) of models, i.e. the sensitivity of
the parameterizations to input parameter perturbations is strongly state
dependent. This can cause data assimilation systems to produce erroneous
analysis increments in model state variables affecting balance and stability.
Slide 3
17th ITSC 04/2010
ⒸECMWF
3
Moreau et al. (2003, QJRMS, 130, 827-852) and Bauer et al. (2006, QJRMS,
132, 2277-2306) have shown that for global models it is preferable to assimilate
microwave radiances rather than rain rates because:
• the observation operator always produces non-zero gradients (in the absence
of signal saturation) whether a scene contains clouds or not;
• an observation operator which combines moist physics and radiative transfer
model behaves more continuously over its dynamic range than does moist
physics alone;
• the statistics of observed-minus-modelled radiance departures have mostly
Gaussian characteristics, and this also facilitates bias correction;
• observation operator errors can be derived from spatial departure covariance
statistics.
Bauer et al. (2010, QJRMS, submitted) and Geer et al. (2010, QJRMS,
submitted) describe improvements in the use of microwave satellite observations
in the ECMWF forecasting system. Since March 2009, the operational system
has assimilated all-sky microwave radiances directly into 4D-Var.
Slide 4
17th ITSC 04/2010
ⒸECMWF
4
All-sky radiance assimilation:
• direct radiance assimilation over ocean only;
• no spatial interpolation – observations super-obbed to grid points, operators
(moist physics and radiative transfer model) applied at grid-point locations;
• “all-sky” – all observations (clear, cloudy, rainy) assimilated through the
same operator:
• this gives balanced sampling,
• and allows to create/dissipate/modify clouds similarly.
Control Variable /
State vector
Forecast
model
State at
time i
Radiative
transfer
Radiance
observations
x0
Mi
xi
Hi
yi
J
x 0
M Ti
J
x i
H Ti
J
y i
Wind and mass
Humidity
Clouds and rain
Dynamics
Moist physics
Wind and mass
Humidity
Clouds and rain
Clear,
cloud
Clear
sky and
Slide 5
rain including
scattering
Clear
All-skies:
sky clear,
cloudy and rainy
Histograms of SSM/I first-guess
radiance departures for samples
where both model and observations
contain clouds (dotted) or only clearsky (solid), where
the observations contain clouds but
the model is cloud-free (dash-dotted)
and vice versa (dashed) for
channels 19v (a), 19h (b), 22v (c),
37v (d), 37h (e), 85v (f),
85h (g). Data drawn from a 12-hour
assimilation on 15 September 2007
00 UTC.
Slide 6
17th ITSC 04/2010
ⒸECMWF
6
Linearity – single observation test:
• A: Convection in the inter-tropical convergence
zone (ITCZ): model has too much rain
compared to the observation.
• B: Mid-latitude cold front: model has too little
moisture and cloud.
• C: Convection in the ITCZ: a near neighbour to
case A, but chosen to illustrate a situation with
poor minimisation quality.
SSM/I channel 19v first guess departures
through the 4D-Var minimisation for singleobservation test cases A, B and C using the allsky approach. Squares indicate departures
calculated using the nonlinear T511 forecast
model in the `outer loop'; crosses indicate the
departure at the end of each minimisations or
`inner loop', calculated using the incremental
Slide 7
method.
Linearity - full system
Non-linear vs. linear
departures
in first inner-loop
Rainfall radar
(research only)
All-sky microwave
imagers
Slide 8
Clear-sky satellite
and conventional
Linearity - full system
Non-linear vs. linear
departures
in last inner-loop
All-sky microwave
imagers
Slide 9
NWP model biases – Example: cold sectors
Mean SSM/I channel
37v first-guess
departures
(1st July 2009)
ECMWF model cloud
liquid water fraction,
(liquid + ice)/total
Slide 10
NWP model biases – Example: location/width of fronts
First guess
Observation
Departure
Slide 11
Mean SSM/I channel 37v radiance departures
Mean TCWV increments
All-sky (total): 4004937
AllSky - Control
90
a
135
0 15 30 45
-45 -30 -15
45
90
135
6
All-Sky (cloudy): 2603074
b
-135
-90
-45
0
45
90
135
-45 -30 -15
0 -15 -30 -45
0 15 30 45
45 30 15
-45
0
45
90
4
-135
-90
-45
0
45
90
b
135
135
-45 -30 -15
0 -15 -30 -45
0 15 30 45
45 30 15
-135
-90
-45
0
45
90
-2
-4
135
-135
-6
-135
-90
-45
0
45
90
0 15 30 45
-45 -30 -15
0 -15 -30 -45
-90
-45
17th ITSC 04/2010
0
45
90
0
45
90
135
6
4
2
135
-90
-45
0
45
90
135
-90
-45
0
45
90
135
135
45 30 15
-135
-45
8
Mean SSM/I channel 37v first-guess departures highlight
Slide 12
systematic model cloud deficiencies. The mean TCWV
increments show that the moisture climate is similarly
affected by 1D+4D-Var (Control) and the all-sky system.
ⒸECMWF
12
-2
-4
-6
-135
Control (clear): 763185
d
-90
10
0
0
-135
135
AllSky - AllSkyOff
2
All-Sky (clear): 1401863
c
90
0 -15 -30 -45 -60 -75
-90
45
75 60 45 30 15
-135
0
-75 -60 -45 -30 -15
0
FG departure [K]
-45
-45
0 -15 -30 -45 -60 -75
0 -15 -30 -45
-90
-90
75 60 45 30 15
45 30 15
-135
-135
-8
-10
TCWV difference [%]
45
15 30 45 60 75
0
0
-45
15 30 45 60 75
-90
0
-135
-75 -60 -45 -30 -15
a
Impact verification - satellites
a
b
12
183±1
HIRS channel number
Channel frequency [GHz]
11
183±3
7
6
5
183±7
4
1.0
1.2
1.4
1.6
Std. dev. [K]
1.8
2.0 -1.0
-0.5
0.0
Mean [K]
0.5
1.0
0.4
0.5
0.6 0.7 0.8 0.9
Std. dev. (normalised)
1.0
1.1 -0.6
-0.4
-0.2
0.0
0.2
Mean [K]
0.4
Verification of short-range forecasts (solid) and analyses (dashed) against MHS (left) and
HIRS (right) radiance observations over period 1 July-31 October 2009:
• no microwave imagers
• 1st version of all-sky system
Slide 13
• revised version of all-sky system
0.6
Impact verification - sondes
c
d
100
150
300
200
400
250
300
500
Pressure [hPa]
Pressure [hPa]
400
700
500
700
850
850
1000
1000
0.6
0.7
0.8
0.9
1.0
Std. dev. (normalised)
1.1 -0.0010
-0.0005
0.0000
0.0005
-1
Mean [kg kg ]
0.0010
0.7
0.8
0.9
Std. dev. (normalised)
1.0
-0.0001
0.0000
0.0001
0.0002
-1
Mean [kg kg ]
Verification of short-range forecasts (solid) and analyses (dashed) against drop-sonde
(left) and radio-sonde (right) observations over period 1 July-31 October 2009 (standard
deviations have been normalized with „no-imager values‟):
• no microwave imagers
Slide 14
• 1st version of all-sky system
• revised version of all-sky system
17th ITSC 04/2010
ⒸECMWF
14
0.0003
Summary:
The all-sky microwave imager radiance assimilation system operated at ECMWF since
March 2009 applies a unified observation operator to clear-sky, cloud and precipitation
affected radiances. It mostly constrains moisture analyses and short-range forecasts over
the oceans and improves the fit to other moisture-sensitive instruments such as HIRS, MHS,
drop and radio-sondes. Despite initial concerns, operator linearity, Gaussian error statistics,
and minimization performance were similar to a system with clear-sky observations. The
main outstanding issues with clouds and precipitation are systematic NWP model biases
and a better specification of model background error statistics in these conditions.
Operational implementations:
• June 2005 – Operational 1D+4D-Var of rainy/cloudy SSM/I
• June 2008 – Addition of AMSR-E and TMI to 1D+4D-Var
• March 2009 – All-sky direct 4D-Var of SSM/I and AMSR-E
• September 2009 - Direct assimilation of selected cloudy scenes from IR radiances (AIRS
and IASI) – with cloud as a sink variable
Forthcoming changes:
• Autumn 2010 – All-sky upgrade: Complete revision of observation handling, errors, bias
correction, and quality control, using a “symmetric total error” approach
Slide 15
In research:
• Extension to MHS and AMSU-A, infrared radiances; radar rain-rate assimilation.