Minutes of the sixth International Radiative Transfer Workshop,
June 2004
Monday
Tuesday
Wednesday
Thursday
Whole Week
BREDBECK 2004 Minutes
=======================
Monday, 21/06/04
================
Participants
============
AD: Adrian Doicu
AH: Arash Houshangpour
BR: Bengt Rydberg
CD: Cory Davis
CE: Claudia Emde
CJ: Carlos Jimenez
CM: Christian Melsheimer
CT: Claas Teichmann
CV: Carmen Verdes
EB: Emmanuel Brocard
FS: Franz Schreier
GH: Gang Hong
JM: Jana Mendrok
ME: Mattias Ekström
MK: Mashrab Kuvatov
MM: Mario Mech
NC: Nathalie Courcoux
NM: Nizy Mathew
OL: Oliver Lemke
PE: Patrick Eriksson
PM: Peter A. T. Mills
SB: Stefan Buehler
SE: Stephen English
SK: Susanne Korn
SR: Sreerekha Ravi
TK: Thomas Kuhn
UK: Una O'Keeffe
UL: Ulrich Löhnert
VJ: Viju Oommen John
Peter Mills Retrieval of water vapor isolines
==============================================
* Motivation
* contour advection
* how to retrieve isolines
- classification problem
* estimation of probability density function
* classification data - AMSU-B data
* training data - ECMWF
* RT model - ARTS
* Result - Contours are well retrieved - Confidence level worse where
gradient is high
* Error sources - Surface emissivity, collocation errors,
cloud contamination, measurement errors
* Simulations with collocated radiosonde profiles
Discussion: Gaussian PDF is an assumption
Scheme works best around water vapor line
Scheme similar to Carlos Monte Carlo Scheme ?
Arash Houshangpour: UTWV / UTH retrieval from AMSU radiances
============================================================
* Importance of UTWV and UTH
* Methodology - channel 18 and channel 19 are used
* Relationship between radiance and UTWV
* Retrieval of temperature parameters using AMSU-A 6 and 7
* Transformation of TB
* Determination of Model parameters on a Global Scale
- ECMWF dataset, ARTS simulations
* Fit parameters (Exponential Fit) for UTWV
* UTH retrieval
Discussion: Use of Channel 20
Adrian Doicu: Iteratively regularized Gauss-Newton method for
atmospheric inverse problems
=============================================================
* alternative approaches to OEM
* Criteria for selection of appropriate regularization method
* OEM: fully stochastic approach
* alternative: semi-stochastic approach
* Criterium 1: Tikhonov Regularization (not efficient)
Iteratively regularized Gauss-Newton Method
* Criterium 2: Iteratively regularized Gauss-Newton Method with simple bounds
* Criterium 3: Multi-parameter regularization method
* Criterium 4: Error analysis
==============================================================================
BREDBECK 2004 Minutes
=======================
Tuesday, 22/06/04
===================
9:30--10:00 Presentation for Cory Davis
=======================================
Modeling the effect of Cirrus on Microwave Limb sounder radiance
* introduction on EOS/MLS
* influence of the clouds on MLS measurements
* example of the simulation
* performance of the Monte Carlo method
- error and CPU time depend on the ice path
- randomly oriented particles are simpler to deal than the oriented particles
* summary
Discussions:
AD: For the scattering properties: Do you use the Mischenko program?
CD: Mischenko constructed program. This is used only for the size
distributions
PE: The shape does not matter so much
CM: Do you use Monte Carlo to have single scattering
properties?
10:15 -- Franz Schreier
=======================
* Spectral Grid Optimization Scheme for line-by-line cross section
* Definition for the cross section
* presentation of the algorithm: sample line center in a fine grid
* and coarser in the wings
* use a function decomposition: smooth + rapid contribution
* Clough and Kneizys approach
* Uchiyama approach
* Performance of the 2 and 3 grid algorithm.
* examples of different schemes (for CO2)
* new scheme is much faster (scale of 10)
* Sparks scheme: performance (JQSRT 97)
* Conclusions: speed up of about a factor of 300
* Independent of the line shape
* Fair intercomparison with Sparks algorithm
SB: why the value of the cut off does not have influence on speed.
FS: the coarse grid is much larger then the fine one.
SB: How do you subtract the line center?
11:15-- Claudia Emde
====================
Cloud contaminated mm-wave spectra using ARTS-1-1
* scattering model: presentation of the algorithm
* cloud-box definition
* discretisation of radiation field
* problems arisen:
* sequential updated of the grid is required
* zenith angle grid optimization: more points around 90 deg (specially for
limb sounder)
* presentation of the test cases setup
* 3D simulations-LOS: presentation of the results for different position of
the sensor.
* difference of the 3D vs 1D. The results given by 1D are totally different.
* 3D model is strictly necessary.
Discussions:
A.D: Picard iteration to solve the radiative transfer equations
Which are the boundary condition?
CE: compute the incoming radiation from all direction
PE: use one frequency and then use it as starting point
AD: Initial point important to reduce the number of iterations
CE: the first guess is not used for many points
AD, SB, PE,...: discussion on the boundary conditions and the way of
implementation.
11:50-- Claas Teichmann
========================
Single scattering properties
* Definitions and set-up
* Example of the polarization difference field
* example of the effect of the gas absorption and scattering properties:
changes only due to the gas absorption (for spherical particles)
* results similar for the cylindrical particles
* consistence check
* exemplification for the LOS inside the cloud
* more deformation at higher or lower aspect ratios
* strong differences in the signal for p20 and p30.
* for 500 GHz it does not work well (investigate)
* Summary and outlook
CM: What do you mean by sing scat. prop. change? Why does it depend on the
gas absorption?
CT: discuss in the working group
CD: problem at 500GHz. Does it change with the aspect ratio?
CT: no, it is bad for all the aspect ratio.
CD: Ideas about what's going wrong?
CT: numerical problems..
CE: the idea is to calculate with Monte Carlo method and see if it is also
a problem
PE: normalization problem?
SB: advertising on the polarization working group
CM: Deformation ratio: how did you define it?
SB: publication from Georg Heygster on the single scattering properties
12:30: P.E., C.E., organization for the working group
=====================================================
==============================================================================
BREDBECK 2004 Minutes
=======================
Wednesday, 23/06/04
Group Work summary
==================
SSP Interface (Rekha)
=====================
* PyArts will do averaging from DDA and puts the data into the database
* PyArts module to export data from the database to arts
Limb sounder group(Carmen)
==========================
* input for limb sounding instruments - odin, smiles, mls
* discussed instrumental characteristics, vertical and horizontal
resolution, orbits
* frequency optimization
SE - information about working groups at EUMETSAT of IASI for nadir
instruments
Sensor polarization (Mattias)
=============================
* surface reflection and polarization
- Christian to finish section on polarization in AUG
- check the angle dependence of rotating mirror for AMSU-B
SE - (90 - scan angle)
SE - from vertical to horizontal as you go from nadir to limb
* sensor demonstration
* reconsider mixer part - dual mixers
* put example control files in ARTS package
SB - follow the convention in ARTS for example files (.arts.in)
PE - a documentation for the example files in AUG.
Retrieval algorithms (Adrian)
=============================
* discussion on different retrieval approaches
- Patrick showed Bayesian approach
- peter brought forward stochastic approach
- levenber-marquardt and other methods were discussed
- Regularization methods also were discussed
Matlab group (Oliver)
=====================
* ATOVS part - reading routines in Matlab are much slower
- conclusion - compiler has to be checked
ME, OL, CJ - discussion on Matlab compilation
CE - implementing GMT in atmlab
OL - It is already decided to use GMT for AMSULAB
Polarization (Claas)
====================
* discussed claas's results
- 2 main sources
- negative Q, the main source is coming from below
- positive Q, polarization coming from sides
- An orientation distribution will give the same effect as using a
horizontal particle with a different aspect ratio
CM - justification - randomly oriented particle is similar to
spherical particle
Size distribution (Bengt)
=========================
* particle size distribution in cirrus clouds
- cory has used heymsfields distributions on
- further look is to look at effect of this different distributions
on radiative properties
- using ARTS to study this
* can give an estimate on the retrieval error
UTH retrieval (Carlos)
======================
* AMSU data to derive a long range UTH climatology
* 3 different approaches in Chalmers and Bremen
- simple linear regression with 1 amsu-b channel, 5-9%
- complicated NN regression with amsu-a and, 7.5%
- multi physical regression with amsu-a and amsu-b, 3-8%
* 3 approaches use different definitions for UTH
- weighted UTH, mean UTH, mean UTH resp.
* different data - ECMWF, radiosonde, ECMWF resp.
* issues
- which training dataset is better?
- why present regressions seems to over-estimate low UTH values
and under-estimate high UTH values
* future
- 3 articles on all methods
- testing all regression to evaluate relative performance using ECMWF data
- AMSU synthetic database including more realistic - clouds, ground
PE - Peter's work of retrieval of isolines also could be added
SE - why not using Bayesian algorithm, regression will carry on the
errors in the training data set to the retrieval
CJ - speed is an issue
SB - climate studies, basically we want to look at radiances
-how sensitive is the radiance to upper troposphere humidity
SE - This would bring the same biases in the ECMWF to the retrieval
A long discussion on this followed.
GMT plots (Oliver)
=================
- Nice map plots
- GMT is free
==============================================================================
BREDBECK 2004 Minutes
=======================
Thursday, 24/06/04
Group Work summary
==================
Surface emissivity group (Steve)
================================
* current status in arts
- general solution for full stokes hemispherical integration
including separate handling of emission and reflection
- no physics for any actual surfaces
* what is available
- FASTEM2 in RTTOV7 - calculates v,h given wind speed and skin temp
for ocean.
- land model is purely empirical, atlases(C. Pringent, t. Hewison)
- FASTEM3 in RTTOV8 - added a correction to emission term for
azimuthal variations for all azimuthal variations (Poe.G via
Weng.F).
- f.weng - NOAA snow + sea ice emissivity model
- IOMASA snow + sea ice emissivity model
- c.matzler - snow emissivity - complicated
* recommendations
- high priority
- incorporate FASTEM3 from RTTOV8 ocean only.
- this is easy- land code less valuable
- adapt arts input interface to accept wind vector and skin temp.
- retain direct input of pre-calculated emissivity.
- add no other surface variables yet
- medium priority
- consider replacing FASTEM3 with full GO model exploiting
general solution already in ARTS
- info from AAPP or other sources to pre-calculate emissivity
- low priority
- when schemes for land, ice, snow emissivity mature and simplify
- incorporate in arts
- also take care of different definitions of stokes vector
- look at windsat data
Research ideas (Cory)
=====================
* writing papers looking at science issues
* looking at cloud micro physics
* real ice water content field
* find a cloud field scenario
* comparison of RT models - MC and DOIT
* discussed Claas' problem at 500 GHz
- cory will do it for MC
- problem with scattering integral for 1D DOIT method for higher stokes component.
SB - something to look into in at Bremen
SB - any ideas for the 3D scenario?
CD - output from RT study, TRMM (last slide of cory's presentation)
SB - good thing to look at RAL scenario explained in UTLS study
RTTOV-ARTS comparison (Una)
===========================
* clear sky results - quite good agreement
* rttovscatt doesn't simulate lowest TB's
* ARTS vs obs good(re = 100 um)
* micro physics - step back from RT - agree on suitable size distr param
- set distr. by max crystal size rather than mean crystal size
* surface emissivity
* 3 way model intercomparison
- to involve al gasievsky's model
General RTM (Franz)
===================
* discussed parameterization of RT models
PyARTS group (Cory)
===================
* short presentation of what PyARTS can do
* grid optimization - patrick's numerical method was discussed
- one has to derive the optimized grid from a set of different scenarios
- scattering integral calculation- improvement using gauss-legendre method
SB - isn't that difficult?
CD - if the function can be represented by a polynomial, you can get accu
rate results
SB - if the phase matrix is peaked, one has to be careful
CD - trapezoidal integration is not sufficient
AMSU working group (Nathalie)
=============================
* ozone line problem
- influence?
- 10 years ago, metoffice did some calculation and found negligible
differences
- about 0.3 K difference was found during some recent calculation at Bremen
- some calculations will be done again including stratospheric and
excluding and then comparing.
AI - send the fascod profiles to Steve to perform similar calculation
* scan angle dependence on AMSU-B
- amsu-a biases are not symmetric and can be found from NCEP
- amsu-b biases are symmetric
- steve can also give info on values regarding biases
*NOAA 15 and 16 BT diff
- a 1K difference was found in the calculation at Bremen
AI - to compare NOAA 15 and NOAA 17
* side bands
- rectangular pass band for ARTS and RTTOVB
- diff are negligible with Gaussian passbands
* ARTS - RTTOV comparison
- differences are larger than the mean difference at some places
AI - checks whether the profiles are within RTTOV limits
AI - check temp profiles - strong inversion
AI - extract those profiles and perform with a fine grid using ARTS
* AMSU -B channel 18
AUG group (Patrick)
===================
* overview of documentation structure
* control file examples - AI to remove old files and update them
with working new ones
* details by command line help
* code documented by doxygen and code comments
* no link between arts-xml-data and arts1-1
- arts-xml-data only for internal usage
* went through the AUG
- no obsolete text and FIXME
* doxygen doc- highly appreciated
- command line description of WSMs will be put into doxygen
Future plans
============
SE - surface data from some low-frequency instruments
SB - lower frequencies were not the priority
- in the framework of the new instruments, sub-mm frequency are preferred
SE - but one can test the model for low frequencies
PE - single scattering properties - IR?
- is the database by Bonn going to be general? to be useful for IR.
CD - it is intended to be more general for any RT calculation
SB - use mailing lists with suggestions
- Adrian for ex. can communicate about his single scattering code.
CM - working group summaries, presentations on the web page
Everyone agreed
SB - if somebody need papers that were discussed, can be obtained.
CM - Dietrich's paper on intercomparison on JPL and HITRAN.
==============================================================================
Monday
Tuesday
Wednesday
Thursday
Whole Week
