Minutes of the IOMASA Progress Meeting 3,
3-4 March, 2005
held at DTU, Lyngby, Denmark

Start of meeting: 3 March, 13:00 CET
End of meeting: 4 March, 15:30 CET

Participants:


IUP: 
Georg Heygster                    GH
Christian Melsheimer              CM


DTU-DCRS: 			    
Leif Toudal                       LT
Roberto Saldo                     RS

DMI: 				    
Søren Andersen                    SA

met.no: 			    
Harald Schyberg                   HS
Vibeke Thyness                    VT
				    
SMHI:
Nils Gustafsson                   NG
Per Dahlgren                      PD

User Advisory Group (UAG):
Stephen English                   SE    
Carl Fortelius                    CF
Helge Tangen                      HT
Morten Lind                       ML

Thursday, 3 March: IOMASA consortium

1. Introductory items:

Welcome address.
Overview of project schedule: Phase 2 (algorithm development) ends in March, the next two phases deal with data processing and near real time experiment.

Emissivity

Harald Schyberg, met.no

including typical values from surface emissivity into AMSU-A data assimilation in HIRLAM
emissivities from Leif Toudal, DTU (see below)
not much effect in channel 5 (53.6 GHz), emissivity=1 yields almost same results, since in channel 5, all radiation is from the atmosphere
To Do:
- seasonal dependence of epsilon(sea ice): from IUP?
- Include water vapour (WV) info?

Leif Toudal, DTU

Approach to estimate emissivity from AMSU-A data:
1. Determine atmospheric transmission: assuming isothermal atmosphere, temperature from nearby weather station; using MWMOD
2. Get emissivity, using brightness temperature (Tb) measured by AMSU.
Calculated time series of some test regions with known ice type,
Averaged the time series over periods of little variability (high variability means likely weather influence) and cold temperatures (little water vapour influence)
Results: Cross correlation of daily means of emissivity in different frequencies
Notes about the brightness temperature of the ground:
- T_B(ground) = epsilon(ground) T_skin
- but the skin temperature T_skin depends on the frequency!
- larger penetration depth for lower frequency => T_skin is integrated from emissions of a thicker layer, so it is from "further down" and thus higher.

Discussion

No incidence angle dependence (yet) in DTU method - IUP could provide approximate incidence angle dependence for first-year ice (FYI) and multi-year ice (MYI) to DTU and met.no
How about using colocated HIRLAM data? - Leif (DTU) can try that, while IUP sticks to ECMWF data, which might be slightly better, but coarser (HIRLAM: 20 km grid, ECMWF: 1.5° grid)

Other

Nils Gustafsson, SMHI
- HIRLAM runs with Ocean and Sea Ice Satellite Application Facility (OSI SAF) ice concentrations (IC) instead of ECMWF's, and the improved surface flux scheme: Significant effect on SST!
Søren Andersen, DMI
- R-factor(Deliverable 4.2): How to handle the surface term over sea ice?
SMHI (NG), DMI (SA): Assimilation experiment with sea ice: Use February 2005. Ice concentrations (IC) from DMI; should be ready by end of April.

Project Management

Next Meetings

PM4: Oslo (met.no), week from June 13-17; preferably June 16-17 (Thursday/Friday), noon to noon.
Final Presentation (FP): Bremen (IUP), October 6-7 (Thursday/Friday)

Reports, Deliverables

Upcoming management/progress report for the period Nov 2004 - April 2005: Send input to coordinator in May.
Material for Final Report should be forwarded to coordinator by end of September, so that the Report can be compiled before the project ends (end of October)

Timing of Phase 3 (Production experiment on 2-year historic data set) and Phase 4 (Validation and real time experiment)

Originally:
- Phase 3: March-June 2005 (month 29-32)
- Phase 4: July-October 2005 (month 33-36(
However: We are going to start Phase 4 at least a month earlier, in order to have more time for it, since the data production is mostly in an advanced stage already.
The "real time" should not be taken too literally for the NWP and ice tasks (Phase 4 is in summer): rather months January and/or February 2005 should be taken since summer

Friday, 4 March: IOMASA consortium and UAG

2. Progress of Phase 2: Status and Results of Phase 2 of each Partner:

2.1 Part 1 (IUP): WP 2.1: Atmospheric remote sensing algorithms

(C. Melsheimer):

Total water vapour (TWV) from AMSU-B:

J. Miao' algorithm (using ratios of brightness temperature differences of 3 frequencies).

  Freq. [GHz]:  89 150  183.31+/-7  183.31+/-3  183.31+/-1
  channel no.:  1   2       3          4            5
  Channel 1, 2: Window channels, 
  Channel 3,4,5: around water vapour line.

TWV = ( C₀ + C₁ * log [(T_i-T_j-F_ij)/(T_j-T_k- F_jk)] ) * cos(theta)
4 "calibration parameters" determined from regression of radiosonde data and simulated Tbs: C₀, C₁, F_ij, F_jk
Status: relaxing saturation condition to T3-T4 > F_3,4 works very well. Can now produce
- Swath data (one TWV value for each AMSU-B pixel)
- Daily/monthly average maps, in various formats (gridded NetCDF, PostScript, PNG etc)
Swath data for years 2001/2002 almost complete. Access at http://www.seaice.dk/iomasa/vip/water-vapour, user name is partner, password: ask Roberto Saldo or Christian Melsheimer
Comparison with NCEP reanalysis data shows agreement to within about 1 kg/m², but much higher discrepancies (AMSU TWV about 3 kg/m² higher than NCEP's) at some particular places, e.g. Baffin Bay/Labrador
Working on the extension of the algorithm by using the 89 GHz channel along with some information on the emissivity of the ground. First results.

%   LT: Using constant rj/ri =1.22 for ice is not good, since rj/ri
%       increases for emissivity(150) above 0.8, and that is of course common
%       for ice!

Surface emissivity at AMSU-A frequencies

Purpose:
- Improve assimilation of AMSU-A radiances into NWP models (IOMASA)
- Improve temperature profile retrieval near surface
Method: Use
- AMSU-A data,
- colocated radiosonde profiles
- simulated brightness temperatures with atmospheric profiles (temperature, humidity) as input
- ice concentrations from SSM/I
Then:

epsilon = [T_b(AMSU-A channel) - T_b(epsilon=1) ]/[T_b(epsilon=1) - T_b(epsilon=0)]
- T_b(epsilon=1) and T_b(epsilon=0) simulated using MWMOD (MicroWaveMODel)
- atmospheric profiles (input for MWMOD) from Polarstern cruises or ECMWF model profiles
- Polarstern data only for summer; in order to have complete year-round time series, ECMWF model profiles have to be taken.
Results:
- Emissivity of sea ice (Arctic, Antarctic) and land ice (Antarctica) dependence on frequency and incidence angle.
- Time series of emissivity (monthly means): shows cycle of open water (emissivity low), new ice (emissivity very high), ice (emissivity high).
- Trying to check viability of the algorithm by comparing retrieval of open water emissivities with modelled ones (inside MWMOD): more complex than expected, some agreement.

% SE: You assume specular reflection (thus 1 - epsilon for the ground
%     reflectivity), which is not really true. There is a paper by
%     C. Maetzler (TGARS-L, submitted 2004) on non-specular reflection
%     - you should check the implications for your algorithm.

2.2 Part 2 (met.no/SMHI): WP 2.2 Improve Arctic high-resolution NWP

H. Schyberg (met.no)

Temperature data assimilation (AMSU-A)

HIRLAM 3DVar adapted to using AMSU A brightness temperatures over sea ice
Testing and impact studies ongoing
DL report drafts DL2.2.2, 2.2.3
AMSU-A preprocessing chain

Using improved ice emissivities

OSI SAF FYI and MYI concentrations c_i
Instead of ice emissivity=1 use typical emissivities of FYI and MYI from DTU (see below)
Using fractions of ice and open water (OW): eps = Sum_{OW,FYI,MYI} eps_ic_i
Possible improvements
- further tuning and adjusting emissivity using background departure statistics
- include scan angle dependence
- regional/seasonal dependence (IUP data?)
- emissivity as control variable
- correlation of emissivity between channels
- feedback of departures?
- Other ideas?

(V. Thyness):

Assimilation Experiment

First experiment

Establish reference with "upper" AMSU-A channels (6,7,... non-surface channels)
allow also passive monitoring of lower channels
reference run: no AMSU-A data assimilation
experimental run: with AMSU-A assimilation over sea ice
- verify against EWGLAM (European Working Group on Limited Area Modelling) stations
- first use epsilon = 1
- then use DTU's emissivities
- Results:
  - strong improvement of correlation between T_b(observed) and T_b(modelled) at 23 GHz, less improvement at 53.6 GHz
  - Little change in mean sea level pressure as function of forecast time

Further

Fixed number of iterations (150)
Include lower channels (4 and 5), same as for open sea

% SE: Channel 5: std dev of 1K (no bias correction, no cloud check) -
%     1K is quite high. What std dev do you get if you include bias
%     correction and cloud check
% VT: Not yet.
% SE: You should get that number
% GH: Plans to include scan angle dependence of emissivity
% HS/VT: Yes. Till now, some of that is corrected by bias correction,
%     but it is better to do that in the forward model 
% SE: Compare to analysis.

(P. Dahlgren):

Humidity Assimilation (AMSU-B):

compare direct assimilations of AMSU-B radiances with assimilation of TWV data from IUP
originally AMSU-B was new to HIRLAM at SMHI

AMSU-B over sea

Cirrus and rain contamination
AAPP has algorithm to spot such
Bias correction with linear regression, 7 predictors (same as AMSU-B)
Dry radiances a problem, ground contributes a lot => skin temperature needed
Get this from NWP RT model: It is done for AMSU-A, should also work for AMSU-B

%  SE: Using skin temperature as predictor over sea ice not
recommendable. Check

AMSU-B over sea ice

cirrus and precipitation: no detection algorithm over ice from AAPP!
=> OSI SAF cloud mask (needs colocation code)
AVHRR VIS/NIR cloud mask
after cloud masking: bias correction coefficients
then: assimilate radiances with skin temperature and emissivities (channel 3,4,5) as free parameters
real time AVHRR cloud mask for operational use would be good, but is it feasible?

Assimilation of TWV

TWV from DTU (IUP algorithm)
over ice in winter: promising
over sea: skewed and bias

N. Gustafsson (for S. Gollvik and V. Perov):

Impact of OSI-SAF sea ice information and improved surface flux calculations on NWP

HIRLAM modifications and impact tests
- A: Put OSI-SAF sea ice fraction into HIRLAM surface analysis; HIRLAM calculates surface fluxes in tiles (water, ice, land types)
- B: Improved surface flux calculations in stably stratified boundary layers (Silentikievic, Perov)
- Run parallel data assimilation and forecast experiments (HIRLAM 22 km, 40 levels, for January 2005) with:
  1. Reference
  2. Reference + A
  3. Reference + B
  4. Reference + A + B
Modified HIRLAM surface analysis with OSI-SAF sea ice
1. Run HIRLAM SST analysis with ECMWF (NCEP) SST as pseudo-observations; use a previous analysis as first guess: Diagnose sea ice fraction
2. Run sea ice fraction analysis with OSI-SAF input in the form of super-observations at 22 km horizontal scale; use a previous analysis as first guess.
3. Correct SST analysis with sea ice info
- [Plot: HIRLAM sea ice from SST versus HIRLAM sea ice with OSI SAF input] - discrepancies mainly near ice edge, and along Labrador coast, Gulf of St. Lawrence
```
%    HS: Blue dots on land = lakes?
%    NG: Yes, but not relevant and meaningful.
    
```
- [Plot: Reference HIRLAM SST versus HIRLAM SST with OSI SAF sea ice input] - Small differences near ice edge, and strange large discrepancy far away from ice on open water, south of Newfoundland
- [Plot: HIRLAM 2 meter temperature versus HIRLAM 2 meter temperature with OSI SAF sea ice input] - improvement, localised areas with high impact (10 K)
Summary:
- Software for introducing OSI SAF ice fraction into HIRLAM developed
- Local impact of OSI-SAF sea ice on low level temperature can be significant (0-15° C). Synoptic scale impact and impact on forecast verification scores: to be examined.
- some improvement/tuning needed (interface etc.)

%  LT: OSI SAF data: 12h or 24h?
%  HS: 24 hrs.

2.3 Part 3 (DTU): WP 3.2: Construction of sea ice emissivity forward model:

(L.Toudal):

1. Sea Ice Emissivity at AMSU-A and B frequencies

Starting point: Radiative transfer (RT) equation for non-scattering, isothermal atmosphere
Estimate atmospheric absorption (using RT model MWMOD) from estimate of surface temperature (from nearby weather station):
Then: Retrieve surface emissivities from AMSU window channels:
epsilon = [T_b(AMSU-channel) - aT_a - (1-a)aT_a - (1-a)²T_CB] / [(1-a)T_s T - - (1-a)aT_a - (1-a)²T_CB]
where a is 1-exp(-opacity); T_a is the atmospheric temperature T_s is the surface temperature T_CB is brightness temperature of the cosmic background
Assumption: Oxygen absorption dominant, ignoring water vapour (dry arctic winter conditions), can be improved by taking TWV into account; nadir-looking AMSU
Used this algorithm for a few sites with well-known ice type (FYI, MYI): Time series of emissivity for several weeks in winter, at times strong variabilities; because of
- wrong surface temperature (stations some distance away)
- ice variability
- weather influence
For each of the 2 ice types, averaged times series over a period when it shows little variability => typical emissivities of FYI and MYI at AMSU frequencies
Correlations between emissivities of different AMSU channels:
- low correlations between emissivities at 23 and at 150/183 GHz: shows that assumption of low WV influence is o.K.
- surprisingly low correlation between 89 GHz channels of AMSU-A and B
Standard deviation of averages emissivities: 0.01 to 0.04
To do:
- Improved TWV correction (HIRLAM)
- Improved T_a(HIRLAM)
- Include angular dependence (IUP)

%  SE:  Emiss. at 23 and 37 GHz are very close (almost identical):
%       puzzling since the first is sensitive mainly to WV, the second mainly
%       to liquid water.
%  LT: Downwelling radiation (simulated with MWMOD, used for
%      estimating atmospheric absorption) is very similar for both (8 K
%      and 10 K)

Statistical retrieval of surface and atmosphere parameters from AMSR-E data

Linearisation of RT equation and surface contribution (Frank Wentz' model + ice part)
=> T_antenna = F (SST, WV, CLW, T_air, ...)
OEM (optimal estimation method) retrieval.
Example maps:SST: errors mainly where surface emissivity is too poorly known
Data (maps) accessible in IOMASA interactive ice browser(IIIB)
In addition in IIIB: Data from other retrieval algorithms (e.g., TWV algorithm from IUP) and OSI SAF ice concentrations, met.no ice drift data
Comparison OEM-retrieved cloud liquid water with R-factor: Mostly matching, some small errors
Comparison OEM-retrieved TWV from AMSR-E and TWV from AMSU-B: nicely complementary (AMSR-E TWV works well over water for high TWV, AMSU-B TWV works well over ice and for low TWV); however: probably some systematic offset.
OSI SAF ice concentrations (IC): spurious IC near Newfoundland in January for some days.

%  LT:  Might cause the problem with SST at SMHI (see above).
%  SA:  You should throw away ice concentrations below 5%.
%  GH:  ASI weather filters?
%  SA: Would also remove some real low IC

Website usage statistics: steady increase over last year (but not all because of IOMASA data products)

2.4 Part 4 (DMI): WP 4.2: Construction of algorithm for sea ice concentration retrieval:

(S. Andersen):

Goals:
- Improve description of leads and polynyas
- Improvements for low IC range and at ice edge
Status:
- Results likely to enter operations soon:
  - Ice/snow emission modelling
  - Concentration algorithm evaluation
  - AMSR processing
  - Routine SAR classification for validation
- Longer operational scope (still experimental):
  - Bookkeeping
  - Thermodynamic model
  - IC system
Example: Warm air intrusion: Melt and refreeze
- common in winter near ice edge
- causes problems in IC retrieval, especially NASA Team 1 (NT1) algorithm (used by OSI SAF): IC underestimated by 20%, persistently until up to three months after the melt event.
- Bootstrap and 85 GHz algorithms less sensitive to melt events than NT
- 90 GHz algorithm: better ???, but almost as good as NT (OSI SAF should update algorithm; NT2 very stable
Emission modelling
- MEMLS (Microwave Emission Model of Layered Snowpacks; Mätzler)
- added sea ice module => MEMLSI
- In-situ ice profiles from Polarstern March 2003, but:
  - grain size poorly documented
  - Lacks snow/ice density data for many stations
  - Representativeness/varying spatial scales?
- Initial sensitivity study: Draft report in progress
Validation
- Validation with classified SAR data
- 30 scenes completed, 20 more to be completed through April
- Match-up with SSM/I or AMSR data relatively simple
- result will help in final algorithm selection
- Accuracy: Comparing result from two independent human operators: Ice/water confusion 2 to 3%

Basic conclusions: Findings directly applicable operationally after validation:

OSI SAF algorithm needs maintenance:
- Bootstrap and/or NT2
- Possibly Near-90-GHz with SSMIS
- To be confirmed in validation against SAR data

Algorithm pros and cons so far

	90 GHz (pol)	Bootstrap	NT	NT2
Pro	resolution	weather insensitive	temperature insensitive	surface insensitive (seemingly)
Contra	weather, surface/snow	temperature	surface/snow	some weather, surface-atmosphere confusion

Advanced Developments (rather experimental developments with uncertain operational scope, could be considered e.g. in next phase of OSI-SAF):
- Bookkeeping model:
  - Today's IC, c from satellite
  - Yesterday's IC updated with drift information from satellite: uc
  - Compare c and uc to keep track of new ice formation, ridging, melting
  - Can keep track of
    - ice age distribution
    - surface melt history
    - deformation/ridging
- Thermodynamic model:
  - Input: Ice/snow profile and meteorological data
  - Thermodynamic and ice mass model
  - Backscatter/Emission model
- Ice concentration system:
  - Combine virtues of 89 GHz and low resolution algorithms
  - Integrate atmospheric data from other satellites via assimilation into NWP model
  - Flags to indicate, e.g., R-factor, history from book-keeping or emissivity anomaly from thermodynamic model
  - No general merging method exists yet for different ice data and atmospheric data
Overall Results:
- Likely impacts on OSI-SAF operations:
  - Revised OSI-SAF algorithm
  - Higher resolution
  - Use of 85/89 GHz
- Improved understanding of surface radiative processes:
  - Better sea ice retrieval
  - Foundation for improving a range of parameters retrieved over sea ice

3. Review of Phase 2

Phase 2 will be completed as planned at the end of March
Phase 3 (data production) has partly been started already.
Phase 4 (near-real time experiment) should be started earlier than planned, best: June/July.

Feedback from UAG

Helge Tangen

What you are thinking to provide is useful to many
Data for users ..?..:
- Formats! (easy access)
- Quality
- Validation
can give spin-offs of ice service: more specific products and improved products

Stephen English

How is AMSU forward-modelled over ice?
Difference first guess (can be based on observation) - background (known, independent of observation)
- when background too far => too nonlinear
- background must be independent of observations
- first guess closer to solution than background
- when there is no good background (e.g., ice emissivity = 1), first guess can be used
- =>first guess emissivity from IOMASA (DTU or IUP) algorithms, then retrieve emissivity as well
- this is the best role for emissivity
- Also: improve background for emissivity (constraints for emissivity) using IOMASA emissivity result. Current background information (ice emissivity = 1) is poor.
- for background, you need error correlation between emissivity at different frequencies.
Fuzhong Weng: Also similar emissivity scheme being used at ECMWF: Compare!

% GH:  How to get that model?
% SE:  Can provide that.
% SA:  Will this yield good emissivity? But T_skin is
%      different for different frequencies.
% SE:  ..then you would retrieve some effective emissivity, but NWP
%      allow only one T_skin.
%      This effective emissivity can be very different from real,
%      forward-modelled one. But it would certainly help the analysis
%      (NWP).

alternatively (second option): simplified model of emissivity, based on simple geophysical parameters (ice, snow info) and include that into NWP

% LT:  2nd option would be better (more physics) but not within IOMASA.

Morten Lind

Big issue found in IOMASA: (un)reliability of IC during melt events!
Uncertainty in IC (IOMASA aimed at halving that) - have you achieved that?
How much did IOMASA improve IC retrieval?
Make sure results also enter operation, e.g., OSI SAF
compare ECMWF TWV and IOMASA TWV! (OSI SAF uses ECMWF TWV for atmospheric corrections, how about IOMASA TWV?)

% HT: Since Sweden, Norway and Denmark are in OSI SAF consortium, it
%     should be possible to get IOMASA results into OSI SAF
%     operational service
% GH: But IOMASA mainly aims at providing methods that can
%     improve operational services; maybe ICEMON or Polarview can do
%     that (making the methods enter operational services)
% SA: Same applies for HIRLAM.
% CM: All this should be put into TIP (Technology Implementation Plan)
% SE: Can IOMASA-developed  results/methods be redistributed, e.g. for
%     NWP SAF, AAPP?
% CM: Yes; deliverable reports contain good descriptions, some code
%     etc.
% GH: Then maybe we should open our member web site to the public
% CM: But only after removing our e-mail addresses from it.

Carl Fortelius

To prevent HIRLAM from diverging (national versions), try to do your changes with the reference system
```
% NG: This is taken care of already - of course, experimentation is
%     done on local version.
```
Putting together modellers and remote sensing people proved to be very useful. Will IOMASA formulate some recommendations on combination of remote sensing and NWP?

%  GH: Yes.
%  NG: Maybe there should be a workshop for NWP modellers?
%  GH: Whom to invite: ECMWF, MétéoFrance etc.
%  NG: Combine that with IOMASA Final Presentation? Or rather some
%      DAMOCLES meeting (but that will only take place next year, if
%      at all)

Other Issues

Reminder: Publications! What? Where? (EuroGOOS, e.g., has peer-reviewed proceedings). Start soon.

GLOSSARY/ACRONYMS:

AAPP	ATOVS and AVHRR Processing Package
AVHRR	Advanced very high resolution radiometer (Vis./NIR sensor on NOAA satellites)
ATOVS	Advanced TIROS Operational Vertical Sounder (passive microwave sensor on NOAA satellites)
ASI	ARTIST sea ice algorithm
CLW	cloud liquid water
CWV	column water vapour (= TWV = PWV)
DMI	Danish Meteorological Institute
DTU-DCRS	Technical Univ. of Denmark, Danish Center for Remote Sensing
ECMWF	European Centre for Medium-range Weather Forecast
epsilon, eps	Surface emissivity
EWGLAM	European Working Group on Limited Area Modelling
FP	Final Presentation; Framework Programme
FY	first-year (ice)
IC	ice concentration
ICEMON	Sea ice monitoring in the polar regions
IR	infra-red
IUP	Institut für Umweltphysik (Environm. Physics), Univ. Bremen
MEMLS	Microwave Emission Model of Layered Snowpacks
met.no	Norwegian Meteorological Institute
MWMOD	MicroWave MODel (by Fuhrhop et. al)
MY	multi-year (ice)
NCEP	National Centers for Environmental Prediction
NT	NASA TEAM (algorithm for sea ice concentration retrieval)
NT2	NASA TEAM 2 (algorithm for sea ice concentration retrieval)
NWP	numerical weather prediction
OEM	optimal estimation method
OSI SAF	SAF on Ocean and Sea Ice
PWV	precipitable water vapour (= TWV = CWV)
RT	radiative transfer
SAF	Satellite Application Facility
SMHI	Swedish Meteorological and Hydrological Institute
SST	sea surface temperature
T	temperature
Tb	brightness temperature
TCW	total cloud water
TGARS-L	IEEE Transactions on Geoscience and Remote Sensing, Letters
TWV	total water vapour (= CWV = PWV)
WV	water vapour

Minutes prepared by Christian Melsheimer

Minutes of the IOMASA Progress Meeting 3, 3-4 March, 2005 held at DTU, Lyngby, Denmark