Assimilation of rain and cloudaffected microwave radiances at ECMWF

1
Assimilation of rain and cloud-affected microwave
radiances at ECMWF

• Alan Geer, Peter Bauer, Philippe Lopez
• Thanks to Deborah Salmond, Niels Bormann, Bill
  Bell, Chris ODell, Graeme Kelly

2
Rain and cloud in NWP

• Improved initial conditions lead to improved
  forecasts
• Variational assimilation (e.g. 4D-Var) is used to
  generate these initial conditions by combining a
  first guess forecast with observations
• Conventional Weather stations, radiosondes,
  aircraft
• Satellite Infrared, microwave, scatterometer,
  atmospheric motion vectors
• Need more information on temperature, pressure,
  winds and humidity everywhere, but particularly
  in cloudy and rainy regions
• Need information on the cloud and rain
  themselves. But arent clouds and rain transient
  phenomenon?
• Directly useful for short range forecasting
• The presence or absence of cloud or rain can be
  used to help infer the temperature, pressure,
  wind and moisture structure of the atmosphere to
  benefit longer term forecasts
• If comparison to the observations reveals
  shortcomings in the cloud and rain models, they
  will have to be improved

3
Assimilation of microwave imagers at ECMWF

• 1998 - SSM/I TCWV assimilation from 1D-Var in
  clear skies over oceans
• 2003 - Direct 4D-Var of clear-sky SSM/I
• 2005 - 1D4D-Var of rainy SSM/I
• Bauer et al. , QJRMetS, 2006a,b,c
• 2007-8 - AMSR-E, TMI added in rainy and clear sky
• 2009? - direct assimilation of all-sky radiances
  in 4D-Var

4
ECMWFs current rain and cloud assimilation
approach 1D4D-Var

• Clear sky SSM/I radiances are directly
  assimilated in 4D-Var
• Cloudy and rainy SSM/I radiances have been
  assimilated operationally at ECMWF since 28th
  June 2005, over sea only, using a 1D4D-Var
  method
• 1D-Var retrieves T and q profiles and surface
  windspeed
• 1D-Var observation operator includes
• simplified large-scale and convective cloud
  schemes
• Microwave radiative transfer
• TCWV retrievals are assimilated in 4D-Var

5

Tephigram temperature and humidity
Cloud ice / water
Cloud fraction
Rain/snow
6
Quality of 1D4D-Var rain retrievals
near-instantaneous colocations
First guess
Retrieval
SSM/I retrieval compared to mean of PR footprints
within 7.5 minutes and 25km
Geer, Bauer, Lopez, QJRMetS, latest issue, 2008
7
Quality of 1D4D-Var rain retrievals correlation
coefficients
Geer, Bauer, Lopez, QJRMetS, latest issue, 2008
8
Forecast scores 1D4D-Var rainy assimilation
RMSE against operational analyses
Vector wind
Relative humidity
South
Tropics
North
Limited observing system Limited observing system
plus 1D4D-Var Full observing system without
1D4D-Var Full observing system
Kelly et al., Mon. Weath. Rev., July, 2008
9
Emissive reflector biases

• All conical-scanning microwave imagers (TMI,
  SSMI, SSMIS, AMSR-E ) incorporate a spinning
  reflector
• If the reflector is emissive
• Unfortunately a common situation
• SSMIS Bill Bell, 2008, IEEE
• TMI Frank Wentz, 2001, IEEE

Reflector emissivity
10
SSM/I
AMSR-E
First guess departure K
TMI
11
TMI reflector temperature estimated from first
guess departure biases
Estimated reflector temperature K
12
Radiative transfer biases in cloud and rain
Modelled cloud liquid water (at SSM/I observation
locations 12hrs of data)
37v Obs FG departure K (after moist physics
improvements CMAX cloud overlap)
37v Obs FG departure K (after moist physics
improvements CMEAN cloud overlap)
13
RTTOV-SCATT Two independent column
approximationTb (1 - Cmax ) Tb(clear)
Cmax Tb(cloudy)
RTTOV fast radiative transfer
Forecast model 1 grid point
Cloudy column
Clear column
TOA
Cmax
Cloud
Cmax
Surface
14
RTTOV-SCATT revised versionTb (1 - Cmean )
Tb(clear) Cmean Tb(cloudy)
RTTOV fast radiative transfer
Forecast model 1 grid point
Cloudy column
Clear column
TOA
Cmean
Cloud
Cmax
Surface
15
Modelled cloud liquid water (at SSM/I observation
locations 12hrs of data)
37v Obs FG departure K (after moist physics
improvements CMAX cloud overlap)
37v Obs FG departure K (after moist physics
improvements CMEAN cloud overlap)
16
All-sky, direct 4D-Var assimilation

• In contrast to 1D4D-Var, the full information
  content of the observations is assimilated
• Surface temperature and winds
• Cloud and precipitation
• Total column water vapour
• A unified assimilation
• All sky conditions (rainy, cloudy, clear) are
  treated in the same assimilation stream

17
RMS forecast errors relative humiditynormalised
difference (all-sky 4D-Var - 33r1 control)
degradation
10th Aug to 4th Sept 2007 18 to 26 samples
verified against own analyses.
improvement
18
Departure statistics SSM/I obs- FG mean
19
Summary 1 - issues

• Emissive reflectors
• TMI suffers from emissive reflector bias
• AMSR-E also?
• Be careful when creating multi-instrument
  products
• Recommendations for instrument builders
• Need to build non-emissive reflectors
• Need for accurate measurements of reflector skin
  temperature
• Radiative transfer in rain and cloud
• Move from maximum to weighted average cloud
  fraction
• Better agreement with 10 independent column
  approach and with observations

20
Summary 2 - assimilation

• 1D4D-Var assimilation of rain- and cloud-
  affected SSM/I radiances
• Operational since June 2005 but only the TCWV
  information content is currently used
• Positive impact on forecast scores for tropical
  moisture and winds
• Impact is comparable to clear sky microwave
  imager assimilation.
• Good quality rain retrievals (compared to PR)
• Direct 4D-Var assimilation of all-sky radiances
  (clear, cloudy, rainy)
• Full information content of the observations is
  assimilated
• Improved forecasts compared to previous system
• To be made operational early 2009