Minutes of the IOMASA Final Presentation,
6-7 October, 2005
held at IUP, Bremen, Germany

Start of meeting: 6 October, 9:15 CEST
End of meeting: 7 October, 12:00 CEST

Participants:

IUP: 
Jens Borgmann                     
Sven-Erik Ehlers                  
Georg Heygster                    GH
Christian Melsheimer              CM


DTU-DCRS: 			    
Leif Toudal                       LT
Roberto Saldo                     RS

DMI: 				    
Søren Andersen                    SA
Rasmus Tonboe

met.no: 			    
Frank Thomas Tveter               FT
Harald Schyberg                   HS
Vibeke Thyness                    VT
				    
SMHI:
Per Dahlgren                      PD
Nils Gustavsson                   NG

UAG:
Stephen English                   SE
Carl Fortelius                    CF

1. Introductory items:

• Welcome address.
• Overview of project schedule: 3 weeks before end of project

2. Results of IOMASA

2.1 Part 1 (IUP): Remote sensing of atmospheric parameters

Total water vapour (TWV) from AMSU-B:

• Standard algorithm, adapted to AMSU-B and Arctic: Use AMSU-B water vapour channels (18, 19, 20) and 150 GHz window channel (channel 17), retrieve TWV up to about 7 kg/m², independent of surface emissivity.
• Extended algorithm: Use lowest AMSU-B channels (channel 16, 89 GHz) to retrieve TWV above 7 kg/m²
• Retrieval then depends critically on ratio of reflectivities r (reflectivity = 1 - emissivity) at 150 and 89 GHz.
• Emissivity over ice determined from SEPOR/POLEX data (N. Selbach's work at IUP, 2 years ago): from linear fit we get emissivity at 89GHz as a function of emissivity at 150 GHz
• If we impose the condition epsilon₈₉ = 1 when epsilon₁₅₀ = 1the reflectivity ratio r is constant.
• Comparison and regression with ECMWF ERA-40 reanalysis data show
  • high correlation (0.9) and small mean difference (AMSU-TWV minus ECMWF-TWV) for winter months, when extended algorithm is not used because of low humidity in the entire arctic
  • lower, but still reasonable correlation (0.61) for month with small ice cover and high humidity (August) and extensive use of extended algorithm; quite high mean difference
• Almost no concurrent data with Polarstern cruise radiosonde measurements.

Cloud liquid water

• R-factor is logarithm of ratio of polarisation differences at two SSM/I of AMSR(-E) channels
• Directly related to cloud liquid water, but contaminated by water vapour signal and ground contribution, especially over ice
• water vapour contribution can be corrected for, using, e.g., reanalysis data.
• Background contribution can be estimated by averaging over several days (at least as long as typical weather system time scale, i.e., more than 1 week)

Emissivity for AMSU channels:

• Use measured AMSU-A brightness temperatures at on of the window channels (1,2,3, 15), ECMWF reanalysis data for atmospheric temperature and humidity for simulating corresponding brightness temperatures (MWMOD) for surface emissivity zero and one: Ground emissivity retrieved from that.
• Algorithm working: Comparison of emissivities retrieved over open water with open water emissivities from FASTEM model mostly consistent. Exception: 50 GHz channel
• Comparison of this emissivity algorithm with emissivity algorithm at DTU: They are equivalent, but DTU uses HIRLAM data instead of ECMWF data, and RTTOV instead of MWMOD.
• Correlation of retrieved emissivities at the various AMSU A and B channels.
• various maps of emissivity, e.g., of the entire Arctic, can be produced: daily/monthly means etc.

C     SE: As to emissivity; how many degrees of freedom does
C         emissivity at those AMSU channels have, and how many degrees
C         of freedom of data do we have. 

2.2 Part 2 (met.no, SMHI): Improving numerical weather prediction models

Overview: Improved use of satellite data in NWP

• Status
  • HIRLAM 3DVar adapted to use AMSU-A data assimilation
  • Impact study
  • New quality control threshold instead of cloud-screening
  • Problem with weighting - partial rerun.
  • Various reports and publications, one journal paper planned.
  • Improved sea ice emissivity values used in NWP model: using different values for open water (epsilon_w), FYI (epsilon_f) and MYI (epsilon_m), from DTU
    epsilon = c_wepsilon_w + c_fepsilon_f + c_mepsilon_m
    where Sum c_i = 1
• Further work in DAMOCLES:
  • more tuning and adjusting of emissivity
  • seasonal and angular variation of emissivity from IUP, DTU
  • Emissivity in control variables
  • Feedback of observation departure
  • Correlation between emissivities

New Quality control method

• QC (quality control) based on Bayesian risk
• reject observations with positive risk
• risk is a function of the innovation, i.e., f(y - Hx_b) where y is the observation, x_b the first guess, Hx_b the first guess equivalent
• risk plots: risk vs. y - Hx_b
  ⇒ skewed: rejection closer to zero innovation on one side.

C   GH: Why innovation curve not maximum at
C       y - Hxb = 0? 
C   FT: There is usually a bias between RTTOV result and measured Tb.

• Innovation probability: Fitted Gaussian (p_fit)
• optimal fitted one is the one where ∫_Innov.(p_fitrisk) is minimum.

C   GH: Impact of new quality control checked?
C   FT: Being done now.
C   SE: Equation of risk function?
C   FT: In Deliverable report (here).
C   GH: Innovation thresholds different over open water, ice etc.?
C   FT: Not yet.
C   NG: Compare old cloud contamination check with this new method.

Impact study

• Setup:
  • Feb-Apr 2005
  • Experimental run:
    • normal assimilation
    • + AMSU-A ch.1-10 over ice
    • + Ice emissivity from DTU
  • Reference run: No AMSU-A data assimilation
  • Verify against EWGLAM
  • HIRLAM 6.2.1, 20 km resolution
  • Quality control based on Bayesian risk as described above.
• Results:
  • Largest model difference (Exp. - Ref.) of 500 hPa surface over Arctic sea ice
  • Almost no difference over central Europe (dense radiosonde network)
  • +48hour forecast: difference advected to Europe
  • Vertical profiles: good effect
  • MSLP (mean sea level pressure): std. dev. and bias increase slightly with forecast time
  • 10m-temperature: almost no effect
  • Accumulated RMSE (root mean square error): at times, exp. run much better than ref. run
• Summary:
  • Neutral to positive verification score
    ⇒ recommend using AMSU-A upper channels over ice operationally
  • New QC leads to use of more observations

C   GH: 2 changes: New QC, and better ice emissivities from DTU?
C   VT: Yes.
C   SE: Exp.: AMSU-A assimilation only over ice, not over water;
C       Ref.: no AMSU-A assimilation at all?
C   VT: Correct.

• What happened during the 2 periods of significantly positive impact:
  • related to cases where reference run has problems over Arctic areas
  • Weather situation: flow from Arctic (ice) towards Atlantic/Europe, e.g.,
    • Location of low between Norway and Scotland predicted significantly better in exp. run (far from sea ice!)
    • Low over Barents Sea much better captured by exp. run

C   NG: Would be interesting to see which AMSU-A observation in
C       which region contributes to the improvement
C   GH: 3DVar?
C   HS: Yes, the 4DVar HIRLAM is not yet operational.
C   SE: Is this a direct effect of AMSU-A assimilation, over indirect
C       effect by "supporting" some sparse observations
C       (Bjørnøya) that would otherwise have been rejected?
C   HS: Both.
C   SE: You could systematically check forecast against analysis.

R-factor (cloud signature) in forecast

• Included R-factor (cloud signature) in forecaster visualisation system (called DIANA)
• compared with HIRLAM TCLW (total column liquid water) and AMSU-A TCLW (method by Grody)
• ⇒ could be useful to locate fronts over the ocean!

C   SE: How do forecasters locate fronts?
C   HS: for short range forecast: AVHRR
C   CF: Note: surface fronts not equal to cloud patterns
C   HS: ⇒ R-factor: deep clouds, i.e., height-integrated; AVHRR: cloud tops

• Over sea ice, not much to see from R-factor because of spatially variable surface term
  ⇒ subtracting background might help.

Assimilation of retrieved TWV (from IUP)

• Setup:
  • 3DVar
  • Area ≥ Swedish HIRLAM area
  • period: Jan 2005
  • Reference run: no TWV assimilation
  • Experimental run: Assimilation of TWV retrieved according to algorithm from IUP
• Exp. - Ref. at 850 hPa: increase in
  • specific humidity
  • cloudiness
  • temperature
• Verification against observation (only 5 stations in Arctic):
  • 2m-temperature (T2m): bias significantly reduced
  • RMSE reduced
  • MSLP: less effect
• TWV retrievals increase WV ⇒ warming ⇒ better T, MSLP

Assimilation of raw AMSU-B radiances

• Assimilation of raw AMSU-B radiances:
  • over open water: easier, commonly done by other centres
  • over ice: unexplored
• over open water:
  • get rid of cloud/rain contamination (scatter index T₁₆-T₁₇)
  • T_skin important in dry cases, but not properly handled yet ...
• over ice:
  • no scatter index possible
    ⇒ use OSI SAF AVHRR cloud mask: compare scatter index and cloud mask (look at innovation statistics)
    • over water, cloud mask fine
    • over ice, cloud mask also makes innovation statistics less skewed

C   GH: Would you rather assimilate AMSU TWV or AMSU-B radiances?
C   PD: Not decided yet.
C   SE: Is the cloud mask over ice bad? Did you check that yourself?
C   PD: No, the info is from experience at other institutes.
C   GH: Can we say something general about the two opposite strategies
C       of (1) assimilating direct data (radiances) and (2)
C       assimilating retrieved quantities?
C   SE: Past experience: the smaller the model bias, the less useful
C       is the assimilation of retrieved quantities.
C   NG: But assimilating retrieved quantities helps identifying
C       biases!
C   GH: Is this worth a publication?
C   NG/SE: Yes.
C   NG: Some money in DAMOCLES is planned for that.

• NWP improvement by
  1. New surface flux scheme
  2. including OSI SAF ice concentration
• NWP experiments:
  1. HIRLAM reference model
  2. (a.) and new surface flux scheme
  3. (a.) and new turbulence scheme
  4. (a.) and OSI SAF ice concentration
• Verification of new flux and turbulence scheme: Positive impact
• Limitations:
  • not yet implemented over sea ice
  • poor performance of moisture surface flux
  • need to adapt to new snow/ice parametrisation
• Verification of OSI SAF sea ice data assimilation: positive impact, mainly in bias (low level T in Scandinavia), hints at model bias problem

C   LK: In the turbulence scheme: what is the roughness length over sea
C       ice?
C   NG: Roughness length depends on surface type, wind speed (water);
C       over ice, it is needed, but not done yet.
C   NG/GH/LT: There is a WP on sea ice albedo in DAMOCLES.

Part 3 (DTU): Empirical model for emissivity and backscatter of sea ice

Emissivity (AMSU-A, AMSU-B)

• Theory: Emissivity calculated from measured AMSU brightness temperature and from simulated up- and downwelling radiances given the atmospheric state.
• T_up, T_down from RTTOV 8.5
• Atmosphere from HIRLAM colocation files
• AMSU-A data including scan angle θ
• OSI-SAF ice concentration and type
• ⇒ Emissivity for FYI and MYI for all AMSU-A (met.no collocation files: only for AMSU-A) frequencies and scan angles
• cloud screening wit SAF cloud mask and HIRLAM clouds
• To be continued in DAMOCLES
• Should be used in NWP to have an improved first guess of emissivity.

C    CM: How about emissivity at AMSU-A at non-window channels (50-60 GHz)?
C    LT: Approximately equal to emissivity at 50 GHz

Statistical Retrieval of Atmospheric and Oceanic Parameters

• Theory: Optical estimation method (OEM).
  • Input data: AMSR(-E)
  • Retrieved quantities: SST, T_ice, wind speed, TWV, CLW, IC, MY ice fraction
• Some results:
  • Atmospheric parameters (TWV, CLW) over ice pretty bad because of unknown emissivity
    but:
  • Atmospheric fields over open ocean very good
  • Atmospheric influence on low IC area much reduced

C   NG: Ice type: reliable?
C   LT: it is still susceptible to atmospheric changes/variability
C   SA: Use MY IC for NWP?
C   HS: In fact, SSM/I MY IC used as a predictor for surface
C       emissivity
C   SA: Note that SSM/I MY ice is very unstable, it is only something
C       like 90 GHz emission.
C
C   LT: How many degrees of freedom are there in emissivities of AMSU
C       channels?
C   HS: Regional differences: ice physically different; would be good
C       to know more.
C   LT: Distinction into FYI and MYI explain 97% of the variability of
C       emissivity ...
C   CM: ... in other words, FYI and MYI are something like the first two
C       principal components (thus about >2 degrees of freedom??)
C  GH: Regional differences are related to different meteorol. history 

Part 5(DTU): Demonstration of real-time processing and user interface

• Distribution and processing has been running for more than one year now.
• Some examples [...]
• Compare TWV from OEM with TWV from AMSU
• Compare CLW from OEM with R-factor from SSM/I

C   LK: What does CLW over ice show?
C   LT: part of it is just ice emissivity variations, we include
C       Rasmus Tonboe's emissivity model (meteorol. history)
C   LK: Over snow-covered land?
C   LT: A snow emissivity model would be needed for that...

Part 4 (DMI): Construction of algorithm for sea ice concentration retrieval

Emissivity modelling

• MEMLSI: vertical structure, density contrasts, layering
• comparison satellite (AMSR) brightness temperature vs. in-situ measured; problems:
  • satellite footprint large
  • in-situ data quite variable within satellite footprint
• investigated sensitivity of IC algorithms to 2 most important parameters:
  1. upper layer density
  2. grain size of the layer above ice
  -- all IC algorithms behave differently, NT2 is worst.
• simulated correlation between window and sounding frequencies, e.g., 89 and 150 GHz, can be pretty low (hoar formation etc.)
• right input: good emissivity result, but validation data are a problem

Ice concentration algorithm validation

• Statistics for 31 Oct - 31 Mar over high IC (>90%) areas: NT2 has high variability
• validation methods with ...
  1. ship observations
  2. aeroplane observations
  3. satellite observation
  ... have complementary advantages and disadvantages as to spatial and temporal coverage and resolution.
• RGPS deformation data: No correlation between convergence/divergence and IC change! (reason: sensitivity to thin ice)
• Ship data: Ships will steer through low IC areas
• Comparison with Polarstern data:
  1. Cruise ARK XIX-1 ("Winter cruise 2003")
  2. Cruise ARK XVII-2 (Aug-Sep 2001)
• SAR validation: Correlation SAR - SSM/I
  • 0 (zero) at high IC !
  • ≈0.8 at low IC
• Algorithm recommendations: TUD, near90GHz, Bristol

Recommendations

• Emissivity modelling:
  • continue modelling the state of snow/ice
  • parameterisation of sea ice state / RT
  • validation with concurrent radiometer data needed!
• Ice concentration retrieval
  • Use high IC (from MW sensors) with caution
  • use 85 GHz channel at high.intermediate IC
  • if no 85 GHz channel: use Bristol algorithm

Additional Feedback from User Advisory Group

S. English, MetOffice

NWP problem in polar areas have been demonstrated: there is enormous potential for improvement

On biases: biases often reduced by assimilation, but: bias must be pinned down, model induced bias must be avoided/corrected first, where possible (there will be an ECMWF workshop on bias correction in November)

C. Fortelius, FMI

Don't stop when you have demonstrated positive impact, but make the improvements enter into HIRLAM reference system.

Review of Projects: all tasks completed?

• Part 1: o.K., only CLW has been behind expectations.
• Part 2: o.K., except for partial rerun of calculations after problems with weighting (see above)
• Parts 3/5, 4: o.K.

Publications planned

1. TWV: IUP
2. Emissivity retrieval and impact on assimilation (see Vibeke Thyness' poster at ITSC (Int. TOVS Study Conference) in China): IUP, DTU, met.no
3. Paper on Impact Study of emissivity on HIRLAM: met.no
4. New surface flux scheme in HIRLAM (V. Perov's work): SMHI
5. Impact of AMSU TWV assimilation into HIRLAM; conference paper: SMHI
6. Validation of microwave ice concentration algorithm: DMI
7. Rasmus Tonboe's simulated TB using coupled thermodynamic and dynamic model: DMI
8. Book-keeping model for sea ice: DMI
9. IOMASA in general (for BAMS): all

Project Management

Reports, Deliverables

• Upcoming Final Report due (in December). That means
  • Management report over last 6 months (person months and money planned/used, a few lines about progress, per work package, see previous management reports)
  • Periodic reports over last year (method used/developed, achievements, relevance, per work package, see periodic report last year)
  • Updated list on publications published, submitted, planned
  • The Final Report, covering all three years, in a publishable form, up to 50 pages (this means about 10 pages per project part). Suggested structure, for each of the 5 parts:
    1. Background (description of the problems to be solved)
    2. Scientific/technological and socio-economic objectives
    3. Applied methodology, scientific achievements and main deliverables
    4. Conclusions including socio-economic relevance, strategic aspects and policy implications
    5. Dissemination and exploitation of the results
    6. Main literature produced
  • Please forward input (format as usual) to coordinator at the beginning of November!
• Final cost statements are also due, covering year 3 of the project.

GLOSSARY/ACRONYMS: