Minutes of the IOMASA Progress Meeting 2,
30-31 October, 2003
held at SMHI, Norrköping, Sweden

Start of meeting: 30 October, 13:00
End of meeting: 31 October, 12:00

Participants:

IUP: 
Georg Heygster                    GH
Christian Melsheimer 		  CM

DTU-DCRS: 			    
Leif Toudal Pedersen 		  LTP
Roberto Saldo 			  RS

DMI: 				    
Søren Andersen 			  SA
Thomas Bøvith			  TB 

met.no: 			    
Harald Schyberg 		  HS
Frank Thomas Tveter 		  FT
				    
SMHI: 				    
Nils Gustafsson			  NG
Tomas Landelius 		  TL

1. Introductory items:

• Welcome address.
• Introductory round.
• Overview of project schedule: We are amidst Phase 2, i.e., algorithm development.

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

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.

• Using radiosonde data to get the constants C0, C1 for the algorithm
  TWV * sec(theta) = C0 + C1 * log [(Ti-Tj)/(Tj-Tk)],
• Use different channel combinations (i,j,k = 3,4,5, or 2,3,4) depending on saturation of channel 5
• C0, C1 depends weakly on the temperature and water vapour profile of the atmosphere, therefore:
• Use different sets of C0,C1 for different seasons and regions
• Assumption of surface emissivity independent of frequency is o.k. for channels 3,4,5, but not for 2,3,4 => higher error.
• Educated guess of emissivity, using, e.g., ice type information, could help improve accuracy.
• Related work: N. Selbachs PhD thesis: emissivity at 157GHz and 183GHz is quite well correlated, emissivity at 89GHz and 183GHz less well correlated.

//    
//    LTP: If partitioning into regions is done, the region
//       borders have to be treated carefully (discontinuities) 
//    NG: Why not use temperature profile from satellite
//       (AMSU-A) to decide which bias to use?
//    LTP: As to emissivities - some info on them is available...
//    NG: We use emissivity as a control variable, therefore 
//       knowledge about the correlations between emissivities at different
//       frequencies can be very useful.
//    NG: When do you have TWV fields - we want to see the difference
//       between assimilating AMSU-B radiances and assimilating TWV (from
//       AMSU-B); we'll be ready for assimilating AMSU-B radiances in about
//       half a year.
//    CM: Depending on how well the algorithm is calibrated, 
//       first fields can be ready by end of this year.
//

Cloud liquid water (CLW) over ice from SSM/I

• CLW influence radiative fluxes, sea ice balance,sea ice retrieval, its parameterization critical to climate models.
• Problem: high and varying surface signal, low thermal contrast
• Solution: Use R-factor, calculated from horizontal and vertical polarisation brightness temperatures (T_h, T_v) at two frequencies f1, f2, from SSM/I: R(f1,f2)=log[(T_v(f1)-T_h(f1))/(T_v(f2)-T_h(f2)], i.e. logarithm of ratio of polarisation differences
• R(1,2) = R(surface) + B*(CLW + A*TWV) + R(dry air); B is related to the frequency difference of the liquid water absorption coefficient (k(f1)-k(f2)), A is the ratio of the frequency difference for liquid water and water vapour absorption coefficients
• Choose suitable frequency pair to make B large and A small in order to retrieve CLW, e.g. (18GHz,23GHz) and (23GHz,36GHz)
• Surface contribution R(surface) investigated using clear-sky (CLW=0) cases
• AMSR(-E) data:
  • can be used instead of SSM/I; higher resolution
  • additional channels at 6GHz and 10GHz can yield information on surface contribution

//
//    NG: Can you retrieve total water, CLW + PWV?
//    GH/HS: Difficult to retrieve them simultaneously
//       since they have very different radiative behaviour.
//    Cloud ice is an error source because of scattering. 
//    HS: Can CLW be retrieved using other sensors?
//    GH: Yes, e.g. Polder; needs sunlight.
//    HS: Can we also get AMSR data in near real time
//    LTP: Yes, possible (within about 4-6 hours)
//    GH: We use AMSR for daily sea ice maps (Gunnar Spreen, Lars Kaleschke),
//       sometimes, AMSR data are there faster than SSM/I
//

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

NWP assimilation activities:

SMHI: AMSU-B, humidity

met.no: AMSU-A, temperature

• NWP fields for use by project partners
  • code/script to extract 2 years (2003-2004) NWP data implemented
  • deliverable report 2.2.1 sent
• extracts from HIRLAM 20 of 6hr forecast:
  • total CLW
  • CWV
  • 2 m temperature
  • 10 m wind speed
  • precipitation last 6 hrs (NEW)
• 4 analyses daily (0h, 6h, 12h, 18h)
• Data format etc. documented in Deliverable Report 2.2.1

//
//    NG: Precipitation 0-6hrs dangerous, in first hour (0h-1h), values
//       might be erroneous
//    LTP: Use  2h-6h instead of 0h-6h?
//    NG: Yes, for example, but difficult to achieve. One possibility
//       is to use, e.g., 6h-12h.
//

Setup of operational data stream for assimilation

• done: EARS (Eumetsat ATOVS Retransmission Service)
  • observations received within 30 min
  • reception facility installed, AMSU-A 1c
  • interface developed
• AMSU-A collocation chain:

    
       EARS + HIRLAM20 + OSI SAF
      ---------------------------
      = Tb's, RTTOV simulated Tb's, profiles, ice data
    

WP 2.2

• Experimental OSI SAF chain for production of additional ice products for IOMASA under development:

                   Input   ->         Output 
    Operational:   SSM/I         Total ice concentration (and more)
                               (HDF, GRIB + quicklooks)
    Experimental:  QuikSCAT,     MY ice fraction (new for IOMASA),
                   AVHRR         later: Better emissivity information
  

• Subroutine to import SAF ice data at AMSU footprint
• will be basis for producing statistics for AMSU observations vs. forward-modelled (RTTOV) observations from HIRLAM fields
• statistics will aid in developing emissivity formulation for AMSU-A channels
• will aid in quality control formulation (problem: deep [= thick] water clouds)

Humidity Assimilation (AMSU-B):

• done:
  • outer minimization loop
• ongoing:
  • new humidity variable (rel. humidity)
  • skin temperature and emissivity as control variable
• planned:
  • visit at UK MetOffice (Per Dahlgren)
  • emissivity, first guess
  • AMSU-B cloud mask
  • bias correction scheme (between radiative transfer model and observation)
  • AMSU-B into HIRVDA (= HIRLAM variational data assimilation)

Surface heat flux modeling:

• done:
  • HIRLAM snow scheme
  • Flux formulation for stable conditions
• ongoing:
  • Flux implementations
• planned:
  • validation of new flux and snow
  • HIRLAM ice scheme
  • validation of new ice

//       
//    GH: There are activities on snow depth with AMSR, and snow/ice
//       interface temperature
//    TL/GH Priorities: which of the planned items first?
//    LTP: In winter, snow/ice interface temperature from radiometer
//       (AMSR), since snow quite transparent then
//       <- snow depth from HIRLAM? 
//       -> snow surface temperature
//    TL: Emissivity at AMSU (A,B) channels?
//    LTP: OSI-SAF: ice concentration & ice type => we => emiss. at AMSU
//       freq.
//    HS: ...or directly from SSM/I (no detour via ice concentration/type)
//    GH: Emissivity. important, needed by many => we need to coordinate that
//          (DTU group is working on that) 
//    LTP: Most crucial: AMSU-A channels; AMSU-B channels less influenced
//    SA: In the end, all improvements should be put into one model
//    HS/NG: That is the plan, and then we can assess the overall impact
//       of IOMASA
//    NG: This can take very long ...
//    CM/GH: ...follow-on projects? Discuss that later (e.g., next meeting) 
//    GH: The CARE project should be made more aware of importance of
//       surface for NWP
//

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

• IOMASA Interactive Ice Browser on DTU web site: http://www.seaice.dk/iomasa
• AMSR-E data:
  • ice products:
    • 89 GHz algorithm
    • bootstrap algorithm
    • combined algorithm (similar to ARTIST algorithm)
  • a number of regular users
• Advanced (experimental) statistical retrieval: SST, WS (wind speed), WV(z), CLW(z), T_air(z), T_ice, C (ice conc.), epsilon(f) (emissivity)
  • Statistical retrieval:
    • measured radiance T_a = F(p), p=(SST, WS, WV...)
    • linearise radiative transfer equation: T_a = M p
      • Wentz's water emissivity model
      • Empirical ice emissivity model for AMSR channels

      //
      //    LTP/GH: Problems with calibration of AMSR data (used to calibrate
      //       ice emissivity model.
      //    GH: There are two different calibrations (a Japanese one and
      //       one from F. Wentz), they are expected to converge; more info
      //       on that in March 2004.
      //
      

    • retrieval of p using optimal estimation method (OEM), as, e.g. described in LTP's PhD thesis.
• example TWV, CLW, wind, SST retrieval from North Atlantic (Sep 26, 2003, 4:30 UTC):
  • looks reasonable; note: even SST under clouds, because of the use of microwaves
  • small problem: low SST values at left edge

  //
  //    GH: possible explanation: SST mainly from 6 GHz channel (least
  //       sensitive to atmosphere /  most sensitive to SST), has widest
  //       opening angle => sees part of e.g., spacecraft
  //    HS: Ice surface temperature?
  //    LTP: Does not work yet, but this can be sorted out.
  //    SA: Area north-east of Greenland looks strange also in TWV
  //    LTP: Yes, that's an area of 30%-60% ice concentration
  //       => different channels (having different footprints) see different
  //       things => trouble
  //
  

• Comparison SST 27 Oct 2003: DTU AMSR statistical retrieval vs. OSI-SAF SST (based on IR):
  • correlation (scatter plot) looks good
  • problem at coast because algorithm does not understand land => just a question of a good land mask
• Comparison SST 27 Oct 2003: DTU AMSR statistical retrieval vs. Wentz's AMSR algorithm:
  • correlation good except one area with very poor result

  //
  //    LTP: DTU algorithm is related to Wentz's
  //    SA: There is an alternative algorithm from Japan, completely
  //       independent of Wentz's
  //
  

• Wind estimate:
  • high CLW causes errors
• Comparison Wind 27 Oct 2003: DTU AMSR statistical retrieval vs. Wentz's
  • maybe calibration problem/discrepancies
• Comparison TWV [27 Oct 2003?]: DTU AMSR statistical retrieval vs. Wentz's:
  • good correlation
  • standard deviation of WV over ice higher (interesting to compare this with IUP WV activities)

  //
  //    GH: You could compare with GPS-WV and Polder WV
  //    NG: We are also assimilating ground-based GPS-WV estimates
  //
  

• CLW:
  • Comparison with AVHRR cloud image looks reasonable
• Comparison Ice concentration: DTU AMSR statistical retrieval vs. Bootstrap algorithm
  • ice edge reasonable
• maybe include emissivity (10 channels) in statistical retrieval (needs a priori emissivity info)
• MY ice: enhanced at (SSM/I) ice edge (footprint size problem: different channels see different areas)

---

• AMSR retrieval of all that was done to check the forward model (emissivity, radiative transfer)
• the retrieved data would need proper validation

//
//    LTP: we should make comparison with IUP's CLW, TWV
//    GH: Advantage of Wentz's ocean emissivity model over MWMOD?
//    LTP: Wentz's model orders of magnitude more computationally
//       efficient because it is tuned to SSM/I and AMSR frequencies.
//    GH: AMSR: There is a correction for land contamination.
//

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

• Objectives: scatterometer, radiometer, NWP, validation data => (1) atmosphere, (2) surface effects, (3) improved C_T (total ice conc.) algorithm, (4) validation
• (1) Atmosphere data:
  • Scatterometer, Radiometer, NWP =>(evaluate)=> WV, wind speed, LWC, temperature
  • awaiting input from partners 1 and 2
  • choose best source
• (2) Surface effects:
  • which ones are important?
    • time series of scatterometer & SSM/I data
    • time series of concentration algorithms (near 90 GHz, NASA TEAM, NASA TEAM 2, bootstrap)
  • develop detection scheme
  • correct/flag data
  • example: Melt Events:
    • melt and refreeze affects ice concentration (IC) retrievals
    • detectable in time series of NASA TEAM (NT) alg. retrievals over FY and MY ice
    • does not affect all algorithms equally (bootstrap almost immune)
  • Melt events have typical signatures in time series of T, GR(19-37) , PR(19), Tb(19), Tb(37)-Tb(89), sigma0, APR (active pol. ratio from scatterometer), bootst. IC, NT IC, near90 IC [for plots of these signatures see PDF file of presentation]; different signatures for FY and MY ice
  • e.g., NT IC decreases during melt event and stays low afterwards
  • the melt signatures of the various quantities help discover synergies

  //
  //    GH: Stefan Voss (PhD thesis about sea ice, IUP, 2002) did not
  //       observe reduced FY NT IC after melt events.
  //    SA: Surprising because we have definitely observed that.
  //    GH/SA: Not clear why the findings are so different...
  //           
  

• (3) Algorithm development:
  • Scatterometer => surface; Radiometer => ice conc.; NWP => history => improved ice concentration retrieval
  • focus on surface effects (atmospheric correction may use existing techniques)
  • Initial phase: data sets online
  • Near future:
    • identify
      • problematic conditions
      • algorithm differences
      • active/passive "fingerprints"
    • feedback to part 2 (surface properties)
    • correlation between orbits (met.no)

      //
      //    HS: Ocean: No correlation between 2 orbits
      //          Ice: Some correlation
      //          This might help algorithms
      //
      

    • Tuning of Seawinds statistics (met.no), see PM1

      //    
      //    HS: Not ideal yet
      //    
      

    • Coinciding interests with EuroClim project
• (4) SAR classification (T. Bøvith):
  • Code from Lars Kaleschke (IUP),
    • streamlined/optimized
    • adapted to operational use
    • parallelised
    • interface to ice analyst's workstation
  • Idea: provide tool to ice analyst to delineate ice in great detail
  • Ice Analyst Interface
    • view/select features
    • select/manipulate training areas
    • review classification results
    • mask untrusted areas
  • Draft report available
    • 5 representative scenes:
      • worst recognition accuracy: 80%
      • typically > 90%
      • some features are consistently helpful

    //
    //    LTP/RS: Problems with ENVISAT AO data: ordered data not recorded,
    //       wrong scene on CD (DVD? tape?), ...
    //    
    

    • SAR features (9 different texture measures):
      • homogeneity measures: angular 2nd moment (ASM), entropy (ENT)
      • smoothness measures: contrast (CON), dissimilarity (DIS), homogeneity (HOM), inverse difference moment (INV)
      • others: correlation (COR), mean (MU)
    • using too many features makes the result worse
    • consistently useful: COM, ASM, ENT, Lee
    • occasionally useful: INV, COR, HOM, DIS, MU (~Lee)

  //
  //    GH: How can too many features reduce the accuracy? Is it
  //       "overlearning" (too specific training)?
  //    SA/TB: No, rather: some features are not very well correlated with
  //       what we want to see => they introduce noise
  //    RS: How do you know the "consistently useful features" are just the
  //       ones that work best?
  //    SA: Our results are based on 5 scenes of different physical
  //       setting. So we can not say that choosing only the
  //       "consistently useful features" will give the best results,
  //       the "occasionally useful" features should be checked as
  //       well.
  //
  
    

• Examples for classification:
  (a) Turbulent sea limit (E Greenland, Scoresby sound area):
  • turbulent (=wind-roughened) water and ice are indistinguishable
  • overestimation of turbulent water area => training areas in very turbulent area to be omitted; user must know the limitations of the data

  //
  //    RS: Can you tell if a simple approach (thresholding) works as well?
  //    SA: Why not, but quite likely it won't work well. Several of the
  //      features are certainly best described by a combination of
  //      texture  and amplitude. How non-linear this is, we do not know
  //      for sure.  
  //
  

  (b) Turbulent sea limit (Baffin bay, Disko Island area, 16-26 Dec 2001):
  • Melt and refreeze
  • series of 3 RADARSAT scenes

  //
  //    CM: Water between ice dark?
  //    SA: Typically yes, but theta-dependent.
  //       SSM/I IC show the typical melt problems, but SAR always shows
  //       contrast. This has to do with the radar wavelength. It can be
  //       shown that C-band is largely immune to the larger snow grains,
  //       whereas Ku band will be affected. Present SAR's operate in C-band.
  //
  

• Plans:
  • Test dependence of result on ice expert
  • training by ice analysts
  • compare method to 2001 Odden and 2003 Polarstern
  • initiate routine processing
  • Test Gamma and PMR features
  • Coincident RADARSAT/ENVISAT data

//    
//    SA: Do we try to make microphysical description of ice on which
//       algorithm is based? or rather go on with empirical algorithm
//    LTP: A physical algorithm is nice, but we should have a fallback
//        option
//    GH: Now we have 2 totally different ice algorithms: 
//       DTU's statistical (OEM) retrieval, and DMI's.
//       Should one of them be the "standard" one in IOMASA?
//    SA: Ice service should not rely on just one algorithm (e.g., switch
//       to bootstrap during melt events)
//    LTP: We should continue with both approaches; OEM does not always
//       work fine.
//    HS: And it's 2 different sensors: AMSR and SSM/I
//    LTP: Yes, but OEM algorithm with SSM/I works almost as fine as with
//       AMSR; however, SST retrieval works much better with AMSR.
//       Also consider: there are 3 SSM/I's but only 1 AMSR (used to be 2
//       until recently...)
//

2.5 Part 5 (DTU-DCRS): WP 5.1 Real time processing and user interface: Define interfaces and formats:

3. Review of Phase 2:

4. Inter-task communication/coordination

DTU-IUP: compare CLW fields (from OEM and AMSU) for some example situations; IUP sends specification of example (cloud systems across ice edge) to DTU.

5. Project management

5.1. Report duties

• Periodic report after each project year (year one ends now), due within 2 months(compiled and sent to EC by coordinator)
• Input to coordinator by November 20 (LaTeX, ASCII, RTF, (DOC))
• Kind of input needed:
  • for Section 1 (Management Report):
    • planned/used manpower (person-months) and funds; by WP
    • scientific/technical progress
    • milestones/deliverables obtained
    • deviations from work plan
    • meetings, conferences, cooperations with other projects
  • for Section 2 (Executive Summary about main results achieved, to be made public by EC (1 to 2 pages)):
    • publications by project partners from the last 12 months
    • publications in queue
    • 2 hard copies of each publication
  • for Section 3 (Detailed scientific/technical progress, per WP, max. 4 pages per WP):
    • Objectives
    • Methodology and scientific achievements related to WP including contribution from partners
    • Socio-economic relevance and policy implication
    • Discussion and conclusion
    • Plan and objectives for the next 12 months

5.2. Next meetings

All project members agreed to shift Midterm Review (MTR) from end of month 22 to end of month 18, i.e. May 2004:

• MTR: Monday 10 May - Tuesday 11 May, 2004 at met.no, Oslo, Norway
• PM3: Feb/Mar 2005 at DTU, Copenhagen, Denmark

6. Action items for PM2

6.1.1 Deliverables 3.1.2: HIRLAM data year 1

6.1.2 Deliverable 5.1: formats/interfaces for real time production

6.2. Input for first Periodic Report

7. Additional topics

IOMASA brochure

• not complete yet, should be done this year.
• 3 nice images needed:
  1. Ice concentration
  2. Atmospheric field (TWV or CLW or so)
  3. Ice Browser screenshot
• comments etc. till 20 November

8. Any other business

IOMASA web pages

• IOMASA web pages, now hosted by IUP, have recently often been unreachable (construction works, moving of servers to new building etc.)
• we should consider moving them to DTU web server that is maintained by Roberto Saldo

Presentations at this meeting

All participants of this meeting are to send electronic copies of their presentations to the coordinator, preferably in PDF format.

DMI report on melting events

put link to online version of report (on DMI website) to IOMASA member web site

GLOSSARY/ACRONYMS: