USE OF PRECIPITATION INFORMATION FROM SPACEBORNE RADAR FOR VERIFICATION AND ASSIMILATION

1
USE OF PRECIPITATION INFORMATION FROM SPACEBORNE
RADAR FOR VERIFICATION AND ASSIMILATION IN THE
ECMWF MODEL A.Benedetti, P. Lopez, E. Moreau,
P. Bauer, F. Chevallier, M. Janiskova, A.
Tompkins
2
Outline

• Precipitation assimilation activities at ECMWF
• Brief overview of the Tropical Rainfall Measuring
  Mission (TRMM)
• Overview of the 1D-Var retrievals from the TRMM
  Microwave Imager (TMI)
• Validation of Rainrate/Brightness Temperature
  retrievals using the
• TRMM Precipitation Radar
• Outline of the 1D4DVar approach
• Use of radar reflectivities for assimilation
• Preliminary results
• Discussion and conclusions

3
Precipitation assimilation at ECMWF
Goal To assimilate observations related to
precipitation and clouds in ECMWFs
4D-Var system including parameterizations of
atmospheric moist processes.

• A bit of history
• Work on precipitation assimilation at ECMWF
  initiated by Mahfouf and Marécal
• 1D-Var on TMI and SSM/I rainfall rates (RR) (MM
  2000).
• Indirect 1D4D-Var assimilation of RR more
  robust than direct 4D-Var.
• 1D4D-Var assimilation of RR is able to improve
  humidity but also the dynamics in
• the forecasts (MM 2002).

4
TROPICAL RAINFALL MEASURING MISSION

• Operational since 1997 provides rain
  observations between 35S-35N
• Instruments on board (still working)
• - Microwave Imager (TMI) surface rainrate
  from Brightness
• 
  Temperatures (Tb)
• - Precipitation Radar (PR) rainrate
  profiles from Reflectivities (Z)
• - Visible and Infrared Scanner (VIRS)
• - Lightning Imaging Sensor (LIS)

PR IMAGE OF TROPICAL CYCLONE ZOE, December 2002,
165-180E/0-20S
http//trmm.gsfc.nasa.gov/
5
1D-Var retrievals from TRMM data
1D-Var on Brightness Temp.
1D-Var on TMI rain rates
TMI Brightness Temp (Tb)
Retrieval algorithm (2A12,PATER)
R E T R I V A L
Observed rainfall rates
Observations interpolated on models T511
Gaussian grid
moist physics radiative transfer
moist physics
1D-Var (TCWV, snow and rainfall rates)
V A L I D A T I O N
background T,qv
Radar Forward Model
background T,qv
PR reflectivity
6
Validation of 1D-Var retrievals of rainfall from
TMI radiances and TRMM Rainrates
Model FG T, q
Model FG T, q
Rainfall from TRMM Algorithms (2A12, PATER,
etc.)
Observed Radiances (TMI)
Moist physics radiative transfer
Moist physics
FG rain and snow rates
FG rainy radiance
1D-Var retrievals of rainfall and snowfall rate
1D-Var retrievals of rainfall and snowfall rate
1D-Var retrieval evaluation
TRMM-PR observations
Forward radar model equivalent reflectivity

-Based on Mie look-up tables for the computation
of reflectivity and extinction, assumes a
Marshall-Palmer distribution for rain and snow
particles -Includes treatment of bright band at
273K -Table entries are categorized according to
rain/snow contents, temperature, frequency and
hydrometeor category. -3D radar reflectivity at
14 GHz is computed via bilinear interpolation at
the given model temperature and rain/snow
content at each grid point and vertical
level -Model 3D rain/snow contents are computed
from precipitation fluxes assuming a fixed fall
velocity (see Excursus I).
7
Forward radar model

• Based on Mie look-up tables for the computation
  of reflectivity,
• assumes a Marshall-Palmer distribution for
  rain and snow
• particles and includes treatment of bright band
  at 273K
• 3D radar reflectivity at 14 GHz is computed via
  bilinear
• interpolation at the given model temperature
  and rain/snow
• content at each model grid point and vertical
  level
• Model rain/snow contents are computed from
  precipitation
• fluxes assuming a fixed fall velocity

8
1D-Var results
Background
PATER obs
1D-Var/RR
Case of tropical cyclone ZOE (26 December 2002
_at_1200 UTC) TMI data Surface rainfall rates (mm
hr-1)
9
1D-Var results
1D-Var/RR PATER
1D-Var/BT
Case of tropical cyclone ZOE (26 December 2002
_at_1200 UTC) Total Column Water Vapour increments
(top , kg m-2) and mean profiles of temperature
and specific humidity increments (bottom)
10
Evaluation of 1D-Var results using PR data
Background
PR obs
1D-Var/RR
1D-Var/BT
Case of tropical cyclone ZOE (26 December 2002
_at_1200 UTC) 14 GHz Radar Reflectivity at 2km
(dBZ)
11
Evaluation of 1D-Var results using PR data
Background
PR obs
1D-Var/RR
1D-Var/BT
Case of tropical cyclone ZOE (26 December 2002
_at_1200 UTC) 14 GHz Radar Reflectivity Cross
section (dBZ)
12
Evaluation of 1D-Var results using PR data
PR obs
Background
1D-Var/RR
1D-Var/BT
Case of tropical cyclone AMI (14 January 2003
_at_1800 UTC) 14 GHz Radar Reflectivity at 2km
(dBZ) and Mean Sea Level Pressure (hPa)
13
Evaluation of 1D-Var results using PR data
PR obs
Background
1D-Var/RR
1D-Var/BT
14 GHz Radar Reflectivity Cross Section (dBZ)
14
Statistical evaluation of 1D-Var results
Bias (solid) and rms (dashed) as a function of
reflectivity
Scatterplot of model Z vs obs

• Background has higher bias than retrievals
• Observations tend to show larger values (this
  could be also
• due to the fact that PR only sees rain )
• Little difference between 1D-Var/RR and 1D-Var/BT

Background
1D-Var/BT
1D-Var/RR

• PR Data from 21 tropical cyclones that were
  observed between January and April
• 2003) were used to evaluate the retrieval
  results.
• The 1D-Var/BT and 1D-Var RR were run for all
  cases and statistics were collected

15
Statistical evaluation of 1D-Var results
Heidke Skill Score
Probability distribution functions
HSS1 good skill HSS0 poor skill

• Retrievals are more skillful than background
• 1D-Var/BT slightly more skillful than 1D-Var/RR
  at
• large reflectivity values

PR obs
Background
1D-Var/BT
1D-Var/RR
16
Ongoing Research and Future Validation Work

• TRMM-Precipitation Radar data is a viable tool
  to make quantitative assessments regarding the
  quality of ECMWF precipitation retrievals.
• Global PR data analysis with an improved
  averaging to obtain more robust statistics is
  currently being investigated.
• PR data will be further used for evaluation of
  the TMI 1D4D-Var analysis
• and subsequent forecast
• Plans to use the PR data to study the spatial
  distribution of precipitation for
• verification of the forecast model are
  also ongoing research

17
1D4D-Var assimilation of TRMM data
1D-Var on TBs or reflectivities
1D-Var on TMI or PR rain rates
moist physics radiative transfer
or reflectivity model
moist physics
1D-Var (T,q increments)
background T,qv
background T,qv
18
1D-Var on TRMM/Precipitation Radar data
2A25 Rain
Background Rain
1D-Var Analysed Rain
2A25 Reflect.
Background Reflect.
1D-Var Analysed Reflect.
Tropical Cyclone Zoe (26 December 2002 _at_1200
UTC) Vertical cross-section of rain rates (top,
mm h-1) and reflectivities (bottom, dBZ)
observed (left), background (middle), and
analysed (right). Black isolines on right panels
1D-Var specific humidity increments.
19
Close-ups on 1D-Var using PR reflectivities with
different error assumptions on obs
1D-Var 25 error at all levels
1D-Var 50 error at all levels
20
1D-Var retrievals using PR observations at one
level only vs full profile
1D-Var obs at all levels
1D-Var obs at level 48 (2km)
21
Background and 1D-Var increments in Total Column
Water Vapour (pseudo-obs for 4D-Var) from PR
reflectivities
TCWV guess (kg/m2)
TCWV increments (kg/m2)
Increments indicate an overall moistening
confined along the satellite track
22
Forecast 26 Dec. 2003, 1200UTC
4D-Var differences in Total Column Water Vapour
and Mean Sea Level Pressure (MSLP) Between
experiment with PR data and control experiment
(no PR data)
Analysis 26 Dec. 2003, 0300UTC
Forecast 28 Dec. 2003, 1200UTC
No initial impact on the dynamics is evident in
the analysis. At 1200UTC, changes in Mean Sea
Level Pressure are developing and appear to
persist well into the forecast indicating a
shift in the location of the storm with
respect to the control run.
23
Comparison 1D4D-Var assimilation of TRMM-PR rain
rates/reflectivities Impact on analysed and
forecast TCWV and MSLP (Experiment
Control) (Tropical Cyclone Zoe, 26-28 December
2002)
Analysis at 300UTC, Dec 26
Forecast at 1200UTC, Dec 26.
Forecast at 1200UTC, Dec 28.
with PR rain rates
with PR reflectivities
24
1D4D-Var assimilation of TRMM-PR and TMI
observations Impact on tropical cyclone Zoe
track forecast (26-31 December 2002)

• Comparison of forecast tracks from
• control run (no TRMM data),
• observations,
• 1D4D on TMI TBs,
• 1D4D on TMI Rain Rates,
• 1D4D on TRMM/PR Rain Rates,
• - 1D4D on TRMM/PR Reflectivities
• Coloured labels indicate forecast times (in hours)

-As suggested by the MSLP changes, the track
forecasts are improved when TRMM observations
are assimilated in rainy areas especially when
using TMI Brightness Temperatures. -Despite the
smaller spatial coverage of TRMM/PR data (200-km
swath) compared to that of TMI data (780-km
swath), the impact of these type of observations
is non-negligible.
25
1D4D-Var assimilation of precipitation
preliminary conclusions

• Observations pros cons
• TMI RR computationally cheap
  only if rainy background over ocean
• algorithm-dependent (2A12,
  PATER,)
• TMI TB sensitivity to RR, cloud and
  WV computational cost of RTM
• flexibility of channels
  over ocean only
• TRMM/PR RR land and ocean, vertical info
  limited spatial coverage
• TRMM/PR Z land and ocean, vertical info
  limited spatial coverage
• All four methods manage to converge in various
  meteorological situations (large-
• scale/convective precipitation,
  tropics/mid-latitudes).

• 4D-Var is able to digest TCWV retrievals produced
  by 1D-Var on TMI and TRMM/PR observations in
  rainy areas.
• The significant impact on the humidity field seen
  at analysis time can be kept during the forecast,
  and the dynamics is affected accordingly.
• In the studied TC case, assimilating TMI and
  TRMM/PR observations improve the TC track and
  minimum MSLP forecasts.

26
1D4D-Var assimilation of precipitation
preliminary conclusions (2)

• TRMM/PR Rain Rates versus TRMM/PR Reflectivities
  ?
• Observational errors may be easier to
  prescribe for reflectivities than for 2A25
• derived rain rates.
• Inclusion of vertical correlations of
  observation errors has a marginal impact on
• the 1D-Var results.
• The extra computational cost for running the
  reflectivity model is reasonable.

• TMI versus TRMM/PR ?
• Including the information on the vertical
  distribution of rainfall contained in the TRMM/PR
  observations improves the 1D-Var retrieved rain
  rate profiles.
• Despite their smaller spatial coverage, the
  impact of TRMM/PR data is comparable to that of
  TMI data.
• TRMM/PR data can be used over land and ocean
  areas, whereas TMI data are currently restricted
  to ocean (surface emissivity over land).

27
1D4D-Var assimilation of precipitation prospects

• Cycle 1D4D-Var assimilation of TRMM and SSM/I
  observations in rainy areas over several months
• ? global scores, study of specific events,
  assessment of the different
• 1D-Var methods.
• Improve the determination of observation and
  model error statistics.
• Address the issue related to the use of satellite
  passive microwave data over land.
• Assess the potential of the assimilation of
  ground-based radar data, but problem of
  availability (non real-time, country-dependent)?
• Until when will TRMM observations be available?
• Looking forward to GPM (global coverage, better
  temporal resolution, information on atmospheric
  ice?).
• 1D4D-Var assimilation of SSM/I (and TMI data ?)
  expected to become operational in 2004.

28
Some statistics..
We defined a confusion matrix for grid points
where first guess and 1D-var BT and RR retrievals
hit/miss with respect to PR
Observed YES
Observed NO Predicted YES
A
C Predicted NO B
D Then we defined the
Heidke Skill Score (HSS)
2(AD-BC)
BB CC 2AD (BC)(AD)