Progress in the joint HIRLAM/ALADIN mesoscale data assimilation activities

Nils Gustafsson and Claude Fischer

Mesoscale data assimilation –some initial considerations (1)

• During a planning meeting in Zürich (October 2006) a strong consensus among the participants from the 2 communities for development of A common HIRLAM/ALADIN mesoscale data assimilation system was established.

• 4 year planning horizon (2007-2010)• The question of dynamical/physical “balances” lies in

the heart of the data assimilation problem. We must be able to project the observed information on structures that will survive initial oscillations (adjustment processes).

• The present state of knowledge on adjustment processes and “balances” (including moisture) relevant for the mesoscale data assimilation is very limited.

Example: HIRLAM 4D-Var dependence on condensation/convection in the non-linear model

Some initial considerations (2)

• Variational techniques should continue to be the core of our strategy, in order to select challenges with the most likely substantial return of investments. However, we should be very open to improve the variational techniques by utilizing ensemble prediction input information.

• We need to pay more attention to surface and soil data assimilation, and for this purpose we need to utilize all available remote sensing data, in particular satellite data that will become available in the near future. Furthermore, as is the case with the surface parameterisation, it will (in the long run) be beneficial with an externalisation of the surface and soil assimilation

Example: Implicit flow-dependent structure functions

through 4D-Var

Retrieval of Microphysical Variables at T=65 min

Truth

Vr + Z

qc, qr, qi, qs, qh

RESULTS TOO GOOD! From Dale Barker 2007

Hybrid DA (Global) Single Observation Test• Temperature observation (O-B, =1K) at 50N, 150E, 500hPa.• Worst case scenario: Ensemble size N=1 (taken from KMA’s error breeding system).

T in

crem

ent

u in

crem

ent

Pure 3D-Var Pure Ensemble,No Localization

Pure Ensemble,With Localization

From Dale Barker 2007

Ensemble « climatological » statistics for very high resolution

AROME

ALADIN-FR

AROME

ALADIN-FR

List of actions to obtain code convergence1) Installation and testing of ALADIN 3D-VAR within the

HIRLAM (so far at met.no and SMHI)2) Possible adaptation of the extension zone treatment in ALADIN

(needed for efficiency of 4D-VAR) 3) Observation operator recoding and convergence (documentation

done and some comparison actions initiated)4) Comparison and validation of ALADIN and HIRLAM 3D-VAR

for synoptic scales. 5) Coding of ALADIN semi-Lagrangian TL and AD models (done)6) ALADIN 4D-Var “in a nutshell”7) Basic ALADIN 4D-Var

January 2007 – December 2008

Norwegian Domainand

single observation experiments

central close to boundary

Joint research – important work-packages (1)

WP1: Basic 3D-Var (2007-2008)

WP2: Further development of 3D-Var (2007-2010)• Wavelets • Water vapor control variable • Jb and Jk based on ensemble assimilations • Flow-dependency from ensembles • Improved balance constraints• Control of total water

WP3: 4D-Var (2007-2010)• ALADIN 4D-Var in “a nutshell”• Initial 4D-Var• TL and AD physics• “Moist physics” (research program)

Complex wavelets for LAM (A. Deckmyn)

• Wavelets are (partially) localised in both grid point space and Fourier space.

• Diagonalization of B in wavelet space can reproduce local variations in the structure functions and standard deviations.

• Current work is focussing on reproducing 3D structure functions for different variables (Alex Deckmyn, Tomas Landelius, Loik Berre)

Standard deviations of Temperature errorat the lowest vertical level: from data (left)and from wavelet-B (right).

Wavelet transform for Jb

• Variable substitution

• Fourier/Wavelet transform

)(

/ , /

)( )( )()(

**

1**

1

bo

b

obT

b

xuTJuu

ITBTuTxx

xJxxBxxxJ

*21* FDFBFDT

Correct treatment of the border – lifted wavelets

Joint research – important work-packages (2)

WP5: Assimilation of ground-based remote sensing data (2007-2010)

• Radar reflectivity and wind • GPS delays

WP6: Assimilation of high resolution satellite data (2007-2010)

• IASI• SSM/I• Clear and cloudy SEVIRI radiances• Binary cloudiness information

WP9: Surface data assimilation (2007-2010) – talk by J.-F. Mahfouf

Assimilation of radar reflectivities

Reference:

Haase, G., J. Bech, E. Wattrelot, U. Gjertsen and M. Jurasek, 2007. Towards the assimilation of radar

reflectivities: improving the observation operator by applying beam blockage information. Proc. 33rd

Conf. on Radar Meteorology. CD-ROM.

Antenna’s radiation pattern: 3d Gaussian power distribution (main lobe only)

Ray path: standard beam propagation (4/3 Earth’s radius)

Scattering by hydrometeors: Rayleigh scattering (attenuation neglected)

Mj hydrometeor contents (rain water, snow, graupel, pristine ice)

Observation operator for radar reflectivities

Courtesy O. Caumont

From model space to observation space

Interpolation on radar path:

part of the observation operator

horizontal bilinear interpolation from model grid to radar grid

vertical Gaussian-weighted interpolation from model levels to beam centre

Topographical beam blockage:

currently, not considered in the

observation operator

a beam propagation model (BPM)

simulates visibility maps for each

radar and each elevation angle

Radar challenges

Results

simulate visibility maps for three

French radars assuming standard

propagation conditions

integrate visibility maps into AROME

run an AROME experiment with the

improved observation operator for Z

and compare the results with a

control run

Results:

decrease of the absolute mean

departure (simulations are closer

to observations)

less rejected pixels after screening

Topography (radar Bollène) Visibility (BPM)

Observed reflectivities(14.05.2007, 09 UTC)

Difference of simulated reflectivities (VIS - CTL)

Summary

+

= HARMONIE !