Ecmwf presentation

1
GPM for Operational Numerical Weather
Prediction Peter Bauer European Centre for
Medium-Range Weather Forecasts (ECMWF) with
contributions from Philippe Lopez, Alan Geer,
Sabatino Di Michele ECMWF Véronique Ducrocq
CNRM Rémy Montroty Météo-France Arthur Hou
NASA Rossella Ferretti U. LAquila Russ
Treadon NCEP Kostas Lagourvados NO
Athens Sue Ballard UK Met Office
2
Data assimilation problem
Operator
Observations
Background
Errors
Errors
Errors
Assimilation System
Analysis
Bayes theorem
Errors
(Courtesy F. Chevallier)
3
Microwave imager data Humidity analysis in
cloud-free areas

• EUMETNET Composite Observing System (EUCOS)
• Observing System Experiments (OSE)
• ECMWF operational 4D-Var system
• T511/L60 12-hour
• Control operational analysis
• Baseline all conventional observations used in
  NWP (radiosondeaircraftprofiler network
• surface landbuoyship data)
• Reference BaselineAtmospheric Motion
  Vectors (AMVs) from GEOMODIS

Forecast RMSE 850 hPa rel. humidity
Forecast RMSE 500 hPa rel. humidity
Forecast RMSE 200 hPa vector wind
(Kelly and Thépaut 2006)
4
Model issues Physics
Comparison of forecasts from 5 CRMs (at 1.25 km
resolution) for a TOGA-COARE squall line
Hydrometeor profiles
Evolution of surface precipitation
6h total amounts of hydrometeors
(Redelsperger et al. 2000)
5
Model issues PhysicsResolution
WGNE assessment of short-term quantitative
precipitation forecasts Relative error
contribution Tropics, Summer SE,
Winter for rain systems over Australia
(Ebert et al. 2003)
6
Model issues PhysicsResolution
UM
Méso-NH
ARPEGE
Integrated cloudprecipitation water contents
kg/m2 UM UK Met Office unified
model ARPEGE Météo-France global model Méso-NH
CNRM non-hydrostatic regional model 12-hour
forecast initialized on 17/02/1997 at 00
UTC Resolution UM, Méso-NH 11km, ARPEGE 30 km
7
Sensitivity Time evolution of Jacobians 2D-Var
Maximum sensitivity of simulated 3-hour
accumulated precipitation to initial T and q
over 2D-Var 12-hour assimilation window (from
rain-gauge assimilation exp.)

• In 2D-Var, the sensitivity of the simulated
  precipitation to initial T and q decreases in
  time and does not necessarily depend on the
  amount of rainfall in the model.
• The weight given to rain-gauge observations
  during the 2D-Var minimization decreases in time
  through the assimilation window.

(Courtesy P. Lopez)
8
Sensitivity Time evolution of Jacobians 3D-Var
AD-integration
1200UTC 14 September 2003
12-hour surface rainfall (RR12h) and MSLP
48-hour T95L60 forecast started at 1200 UTC 12
September 2003.
Time (h)
TC Isabel
(Courtesy P. Lopez)
9
Data assimilation implementations
Physical initialization Donner (1988)
experimental, global, CCM0B Krishnamurti et al.
(1991) experimental, global, FSU
T106L14 Nudging (latent heating) Heckley et
al. (1990) experimental, global (tropical),
ECMWF MacPherson et al. (1996) operational,
regional, UKMO Wang and Warner (1988)
experimental, regional, MM5 Ducrocq et al.
(2000) experimental, regional, Méso-NH Lin et
al. (2002) experimental, regional,
NCEP Variational techniques (3D/4D-Var) Increme
ntal Mahfouf and Marécal (2002) experimental,
global, ECMWF T511L60 Treadon (2003)
operational, global, GDAS, T254L64 Koizumi et
al. (2005) operational, regional, JMA,
MSM Bauer et al. (2006) operational, global,
ECMWF T799L91 Non-incremental Zupanski et al.
(2002) experimental, regional, NCEP ETA Zou
and Kuo (1996) experimental, regional,
MM5 Tsuyuki (1997) experimental, global, FSU
T42L10 Sun and Crook (2001) experimental,
local, MM5/WRF Vukicevic et al. (2004)
experimental, regional, RAMDAS Other Hou et
al. (2002, 2004), experimental, global, GEOS
only affordable system for global operational
applications ECMWF, NCEP, JMA, MSC, UKMO,
Météo-France
10
Global operational systems GDAS
NCEP Global Data Assimilation System (GDAS),
3D-Var Products assimilated 1º superobs of
SSM/I (FNMOC) and TMI (2A12) precipitation
rates 02/21/2001 operational use of
SSM/I 10/16/2001 operational use of TMI
Impacts Greatest impact on reducing areas of
spurious precipitation Reduction in overall
impact as increase amount of assimilated radiance
data Future direction Moving towards radiance
assimilation with cloudy fields of view
GFS T254L64 Rain - NoRain
(Courtesy R. Treadon)
11
TCWV increments 1-cycle, Rain - NoRain 2005080100
1D4D-Var Assimilation system (operational since
June 2005)
Operational system 1D-variational retrieval of
TCWV in cloudsrain with SSM/I radiances 4D-vari
ational assimilation of TCWV Operational
system 4D-variational assimilation of SSM/I
radiances in cloudsrain
Direct 4D-Var Radiance assimilation system
(experimental)
kg m-2
12
Global operational systems ECMWF
Mean Relative Humidity Forecast RMSE Difference
08-10/2004 Rain No Rain
90 Statistical significance Improvement
Deterioration
13
Global experimental systems GEOS
Improved MJO signals in GEOS analysis over
tropical oceans (10N-10S) Jan-Dec 2001
Improved hurricane forecast
Hurricane Bonnie Experiment
5-day track forecast from 12UTC 8/20/98
Day-3 precipitation threat score
Rain Threshold (mm/day)
Blue Control forecast Red Forecast with
TMISSM/I rainfall data in initial
condition Green NOAA best track
Precipitation assimilation using the VCA
procedure in the GEOS DAS replicates the observed
MJO rainfall patterns
(Hou et al. 2004, 2006)
14
Regional experimental systems UK Met Office
UK Met Office model suite Global, 40 km,
4D-Var Europe, 12 km, 4D-Var UK, 4 km, 3D-Var
(FGATRH, latent heat nudging)
MOPS Radar No MOPS
Accumulated precipitation (mm) 19-20 UTC 27th
April 2004 from 18 UTC analysis
(Courtesy S. Ballard)
15
Regional experimental systems CNRM
Extreme flash-flood event over Southern France in
September 2002 12-h accumulated rainfall from 12
UTC, 8 Sept to 0 UTC, 9 Sept 2002

• Aladin
• 9.5 km resolution
• parameterized convection
• 1 prognostic water variable (water vapour).

• Meso-NH/Arome
• 2.5 km resolution
• explicit convection
• advanced microphysic scheme 6 prognostic water
  variables (water vapour, cloudwater, rainwater,
  snow, graupel, primary ice)
• mesoscale data assimilation.

(Courtesy V. Ducrocq)
16
Regional experimental systems CNRM
Gauge network

Nîmes
Nîmes radar
Initial conditions Large-scale ARPEGE analysis
for 12UTC, 8th Sept. 2002

12-h accumulated rainfall 08/09/2002 12 UTC
09/09/2002 00 UTC
(Courtesy V. Ducrocq)
17
Météo-France
(Courtesy Rémy Montroty)
18
National Observatory of Athens
The 4-6/12/2002 Antalya-Turkey case Humidity
adjustment of initial conditions based on
satellite estimated precipitation Kostas
Lagourvados
MM5-CNTRL 66 mm OBS 122 mm
Control run 6-hour accumulated precipitation (at
10 mm interval) valid at (a) 1200 UTC 5 December
2002, (b) 1800 UTC 5 December 2002 and (c) 0000
UTC 6 December 2002 (a) as predicted by MM5 Grid
2.
MM5-ASSIM 91 mm OBS 122 mm
Assimilation run 6-hour accumulated
precipitation (at 10 mm interval) valid at (a)
1200 UTC 5 December 2002, (b) 1800 UTC 5 December
2002 and (c) 0000 UTC 6 December 2002 (a) as
predicted by MM5 Grid 2.
(Courtesy Kostas Lagourvados)
19
University LAquila et al.

• Assimilation of conventional/non-conventional
  data in MM5
• 3 two way nested domains at 27km (D1), 9km (D2),
  and 3km (D3)
• GPS ZTD (3DVAR/FDDA), SSMI and SYNOPSOUNDINGS
• CASE FLOOD in SICILY 16 September 2003

(Courtesy Rossella Ferretti)
20
University LAquila et al.
21
Data assimilation trade-off for instrument
definition
Derive post-EPS mission requirements related to
the observation of clouds and precipitation with
a polar microwave radiometer

• Project
• Identify those microwave frequencies between
  5-200 GHz that are optimally suited for cloud and
  precipitation remote sensing (and that are
  protected by International Telecommunication
  Union, ITU, regulations).
• Consider ocean/land surfaces and all weather
  conditions.
• Develop framework for the estimation of
  potential hydrometeor retrieval accuracies given
  the identified channel selection.
• Assess retrieval accuracy based on user
  requirements defined for post-EPS.

P. Bauer and S. Di Michele, ECMWF
22
Information Content
Signal to noise can be characterized as x/? or ?x
/? with x atmospheric/surface
variable ?x standard deviation of xs
variability ? noise Information content of a
measurement factor of x-knowledge improvement
when making observation(s) often as log
(factor) Linear Gaussian case Hs S P(x)
S P(xy) Entropy reduction A Analysis
error covariance matrix P (joint) pdf of
x(y) Iterative method for channel
optimization A-1 B-1 hhT Improvement of
A over B with H B Background error covariance
matrix H Jacobian matrix with columns h,
normalized with R R ObservationModelling
error covariance matrix 1. calculate Hs for all
channels and select highest 2. update B with
A 3. calculate Hs for remaining channels and
select highest
P. Bauer and S. Di Michele, ECMWF
23
Ingredients
x - (T,q)-profiles from short-range ECMWF model
forecasts, - hydrometeor profiles from
application of observation operator, - surface
emissivities from modelling (ocean) or
climatologies (land). H - linearized
large-scale condensation convection scheme, -
multiple scattering radiative transfer model
(conical scanner ensuring constant viewing
geometry and polarization usage). B - for (T,q)
from operational model formulation, - for
hydrometeors from HBHT. R - HBHT calibrated
with results from observational
method. Objective Identify channels that
provide highest information content in most of
the cases Normalize Entropy reduction with
number of cases for which reduction exceeds a
threshold
P. Bauer and S. Di Michele, ECMWF
24
Definition of passive microwave imager specs for
post-EPS

• Eumetsat Polar System (EPS) follow-on, 2020
• Study on dedicated specifications for clouds and
  precipitation channel selection
• Channel identification and hydrometeor retrieval
  accuracy estimation (also with AMSR-E as
  baseline)

Rain Snow
ocean land
ocean land
AMSR-E
(Channel priorities for land/ocean surfaces,
global profile datasets, optimal estimation
theory, x-axis mean entropy reduction)
P. Bauer and S. Di Michele, ECMWF
25
Issues

• Precipitation specific scoring
• Analysis system performance
• Independent, system-wide, satellites
  (radiance)
• 5

• Sub grid-scale processes
• Consistency b/w NL, TL, AD
• Sensitivity w/r/t control variables
• 30

Physical parameterizations Observation
operators Analysis system Regional
model initialization/forcing Data
assimilation experimentation Land surface
analysis system/modelling

• Choice of observable RR vs. TB/Z
• Observationmodelling errors
• Accuracy vs. efficiency
• Regularity/linearity
• 5

Model component/experiment evaluation

• Operational vs. experimental
• Spatial/temporal scales
• Balance b/w moisture dynamics
• Model errors (background), spatial
  structure
• 50

• Operational vs. experimental
• Objective, statistically significant
• Cases where moisture analysis critical
• Consistency b/w impact of clear-sky
• AND cloud/precip observations
• 10

26
Expected developments relevant to GPM-era
ECMWF now - 25 km, 91 model levels
(T799L91) - 2 analysis
suites (6-hour, 12-hour window)
- 2 10-day forecasts initialized at 00 and 12
UTC - 50-member EPS (65 km,
T319) - Radiances/products
from 20 different satellite sensors
assimilated - Assimilation
of rain-affected radiances operational since
28/06/2005. ECMWF 2010-15 - deterministic 15
km, EPS 30 km - towards
longer assimilation window analyses
- towards unified ensemble prediction
systems (medium-range, monthly, seasonal)
- towards coupled data
assimilation (land-ocean-atmosphere)
- towards environmental monitoring
- towards focus on severe weather
forecasting. General - NWP-systems will become
much better in (physically) resolving even
meso-scale synoptic systems.
- NWP-systems will become much better in
assimilating cloud and rain
affected observations (see recent JCSDA
workshop) Regional NWP - Cloud/precipitation
observations will become standard observations
(NWC) for constraining moist physics
on short time-scales. - Deterministic/ensemble
model cascades global NWP regional NWP -
hydrology
GPM data must be made available in near real-time
(even more stringent for meso-scale modelling
with short cut-off times)