surface data assimilation at ecmwf
DESCRIPTION
Surface data assimilation at ECMWF. [email protected]. ECMWF turned 30 last week. European Centre for Medium range Weather Forecast. weather forecasting : 10 days deterministic forecast (resolution 40 km, soon 25 km) 10 days Ensemble forecast Monthly forecast Seasonal forecast - PowerPoint PPT PresentationTRANSCRIPT
European Centre for Medium range Weather Forecast
• weather forecasting :– 10 days deterministic forecast (resolution 40 km, soon 25 km)– 10 days Ensemble forecast– Monthly forecast– Seasonal forecast– Reanalyse (ERA40)
• Annual Training course : data assimilation (see website)• Additional mission to ECMWF (2005) :
"To develop, and operate on a regular basis, global models and data assimilation systems for the dynamics, thermodynamics and composition of the Earth's fluid envelop and the interacting part of the earth-system".
• GEMS project (Global and regional Earth-system Monitoring using Satellite and in-situ data)– Greenhouse gases, reactive gases, air quality, aerosols.
• Atmospheric CO2 concentration assimilation => need for CO2 surface fluxes
Operational Forecast System
Data assimilation in 2 steps
• 1) Atmospheric variables– 4D VAR assimilation (since 1999)
• 12 h windows
• 23 satellites sources
• adjoint and tangent linear models
• 2) Surface Variables – Analysis of snow
– Analysis of sea-ice concentration and SST
– Land-surface analysis (soil moisture)
OPERATIONAL SYSTEM : 4D-VAR
• the goal of 4D-Var is to define the atmospheric state x(t0) such that the “distance” between
the model trajectory and observations is minimum over a given time period [t0, tn]
finding the model state (at the initial time t0) that minimizes the cost-function :
bb00
100
1
00 2
1y)(y)(
2
1xxBxxxRxx
TT
o
iiii
n
i
oiii HH
xi is the model state at time step ti such as:
00ii t,tM xx
M is the nonlinear forecast model integrated between t0 and ti
H is the observation operator (model space observation space)
From Philippe Lopez
INCREMENTAL FORMULATION OF 4D-VAR
ii1
i
n
0iii0
10 )()(
2
1
2
1dxHRdxHxBxx '
i
T'i
T0
boiiii Hy xd
• 4D-Var can be then approximated to the first order by minimizing:
where is the innovation vector
00ii t,t x Mx '
• In incremental 4D-Var, the cost function is minimized in terms of increments:
with the model state defined at any time ti as: bbb00, , xxxxx ttM iiiii
ii1
i
n
0i
i0i01 )(t,t
2
10
dxHRHMxB 'i
T'T'x
• Gradient of the cost function:
0x
id
ix
0x
computed with the nonlinear model at high resolution using full physics M
computed with the tangent-linear model at low resolution using simplified physics M’
computed with a low resolution adjoint model using simplified physics M’T
Adjoint operators
Tangent-linear operators
From Philippe Lopez
TESSEL scheme in a nutshell
High and lowvegetationtreated separately
Variable root depthNo rootextraction ordeep percolationin frozen soils
snow underhigh vegetation
+ 2 tiles (ocean & sea-ice)
• Tiled ECMWF Scheme for Surface Exchanges over Land
Limitations : single soil type No seasonal cycle of LAI
P. Viterbo
SURFACE ASSIMILATION (1)
• Lower troposphere is sensitive to land surface/soil specification (i.e evaporation and transpiration respond to soil moisture)
• To initialise prognostic variables of land surface parameterisations in NWP
• Forecast drifts are possible due to:– Atmospheric forcing (radiation, rainfall) deficiencies, that may
trigger positive feedback loops• i.e : Positive feedback : lower soil moisture /decrease evaporation/
higher temperature, drier air, reduced precipitation
– Misrepresentation of land surface processes
From Janneke Ettema
Optimal Interpolation at ECMWF
• No routine measurement of soil moisture. -> indirect estimation
• The soil moisture is updated by a linear combination of the forecast errors of the parameters T2m and RH2m.
• Benefits:– It prevents drifts of land surface variables– No use of climatology
• Drawbacks:– Increments smaller than (but of the order of) seasonal variability– Run at synoptic time only– No handling of biases– Focus on a correct evaporative fraction, not necessarily on a correct
land surface state– A rigid framework; difficult to add different observation types or to
change the land surface model
From J, Ettema
ELDAS: Soil moisture analysis systems
Optimal Interpolation:
• Used in the operational ECMWF-forecast since 1999 (Douville et al., 2000)
• Fixed statistically derived forecast errors
• Criteria for the applicability of the method
- atmospheric and soil exceptions
- By design, corrections when T and RH error are negatively correlated
Extended Kalman Filter:(single column model)
• Used in the operational DWD- forecast since 2000 (Hess, 2001) *
• Updated forecast errors
• Criteria for the applicability of the method- Reduced set of exceptions
* Changes:• Assimilation of 2m- T and RH, μw-Tb,
TIR Tb• Model forecast operator accounts for
water transfer between soil layers
From Janneke Ettema
Extended Kalman Filter
Timet+24ht0t+9h t+12h t+15h
Minimization3 perturbed forecasts for each state variable
Forecast(first guess)
Analysed forecast for new soil moisture at t+24h
Comparison with observationsT2m,RH2m,Tb
SimulatedT2m,RH2m,Tb
Opt. Soil moisture
Linearity of observation operator allows a simple minimisation
Observatory Natural Carbon Fluxes
Jean-Christophe Calvet
Météo-France
Overview
Modelling of the carbon cycle in the geoland project
geoland
05.09.2005
The Observatory of Natural Carbon Fluxes of geoland
Partners
• Research partners: KNMI, LSCE, ALTERRA
• Service providers: ECMWF, Météo-France
• Associated user: LSCE
Objectives
• Kyoto protocol
• Transpose the tools used for weather forecast to the monitoring of vegetation and of natural carbon fluxes:
Near real-time monitoring at the global scale (ECMWF) based on modelling,in situ data,assimilation of satellite data.
• Scientific validation of the system
Observatory Natural Carbon Fluxes
Jean-Christophe Calvet
Météo-France
Models
Modelling of the carbon cycle in the geoland project
geoland
05.09.2005
ISBA-A-gs / C-TESSEL
Met. forcing LAI
LE, H, Rn, W, Ts…
Active Biomass
CO2 Flux[CO2]atm
ISBA / TESSEL
Met. forcing LAI
LE, H, Rn, W, Ts…
ISBA-A-gs / C-TESSEL are CO2-responsive land surface models, new versions of operational schemes used in atmospheric models
Prescribed
INTERACTIF
Motivation for assimilation• Again Forecast drifts are possible due to:
– Atmospheric forcing (radiation, rainfall) deficiencies, that may trigger positive feedback loops
– Misrepresentation of vegetation process (phenology, photosynthesis).
• Control variable : LAI• Use of remote sensing observation to constrained the LAI values.
– 10 days window, (En?)KF, land-surface only• (Land surface model are cheap to run )
– Obs: LAI, • Dataset : mean LAI + (N, STD) PER TILE• resolution 0.5/0.5 • from spot4/VEGETATION • Processed by POSTEL, Toulouse
• Operational dataset after 2007 ?: MODIS ? VIIRS ?• fAPAR ?• Cloudy area, Missing data ?
Future of land surface data assimilation system
• 1st tier: Soil wetness/water fluxes– 24-hour window assimilation system:
• Post-ELDAS KF analysis, coupled surface-atmosphere
• Obs: Ta, RHa, heating rates, MW data (?)
• Forcing: Precipitation, radiation fluxes
• 2nd tier: Carbon/water fluxes and green biomass– 10 days window, (En?)KF,
• land-surface only
• Obs: NDVI, LAI, (fPAR ?), tiled
• Forcing: Precipitation, radiation fluxes, temperature
Conclusions
• Soil moisture assimilation tested with EKF. – EKF and IO gives similar result (Seuffert et al.) but EKf is more flexible
(new observations types)– Studies (Seuffert et al.) have shown the synergy of new observation types
(TIR Tb, microW Tb)– Production system need to be developed – Model hydrology need to be improved
• Surface scheme TESSEL is being upgraded to C-TESSEL– Description of the carbon cycle– On going 1D test– Global runs soon – Assimilation scheme planned for next year
• 2D-Var Assimilation currently on-going at Météo-France on a similar model (ISBA-A-gs) (Jarlan and Calvet)
Extended Kalman Filter for soil moisture
[ , ] [ , ] [ , ] ,
, ( , , )
,
Background soil moisture Observations
Background and observations error covariance
( ) ( ) ( ) [ ( )M ] [inimize ( )]
( ) ( ) (
b m m b
T Tb b
a b i i i i i i b
y T RH T
J y H y H
J i i y
2 2
1 1
24 24 24
x
B R
x x x B x x x R x
0 x x K H x [ , ]
,
[ , ]
Observation operator and linearized observation operator
Quasi-linear assumption, approximated by one-sided finite difference, using a perturbed run
24-h window
)i i
H
i i
24
H
24
[ , ] [ , ] [ , ] [ , ]
[ , ]
Gain matrix (model space)
Background error covariance evolu
[ ( ) ( )]
tion
Analysis error covar
( ) ( ) ( )
( ) [ ( )iance
T Ti i i i i i i i
Ti i i i
Ti i
i i
i i i
i i
1 1 1 124 24 24 24
24 24
124
K B H R H H R
B 1 M A M Q
A B H [ , ]
Full (non-linear) model
Model error covariance
( )]
( )Next day backgroun )d (
i i
b i i a
M
i
i M i
1 124
24
Q
R H
x 24 x
forecast (2 x) forecast
• From the SSM/I instrument ECMWF currently assimilates rain-free radiances and Total Column Water Vapour Retrievals. Rain affected radiances are monitored passively.
• The AMSU-A is a 15-channel microwave temperature/humidity sounder that measures atmospheric temperature profiles and provides information on atmospheric water in all of its forms (with the exception of small ice particles). The first AMSU was launched in May 1998 on board the National Oceanic and Atmospheric Administration's (NOAA's) NOAA 15 satellite.
• HIRS is a twenty channel atmospheric sounding instrument for measuring temperature profiles, moisture content, cloud height and surface albedo.