JCSDA Community Radiative Transfer Model (CRTM)

1
JCSDACommunity Radiative Transfer Model (CRTM)

• Paul van Delst (CIMSS)
• Yong Han (NESDIS)
• Quanhua Liu (QSS)

5th MURI Workshop 7-9 June 2005 Madison WI
2
JCSDA Mission

• Accelerate and improve the quantitative use of
  research and operational satellite data in
  weather and climate prediction models

JCSDA Partners
NOAA/NWS/NCEP Environmental Modeling Center NASA/GSFC Global Modeling Assimilation Office
NOAA/NESDIS Office of Research and Applications NOAA/OAR Office of Weather and Air Quality
US Navy Oceanographer of the Navy Naval Research Laboratory (NRL) US Air Force Air Force Director of Weather Air Force Weather Agency
3
JCSDA Goals

• Reduce from two years to one year the average
  time for operational implementation of new
  satellite technology
• Increase uses of current satellite data in NWP
  models
• Advance the common NWP models and data
  assimilation infrastructure
• Assess the impacts of data from advanced
  satellite sensors on weather and climate
  predictions

4
CRTM Contributors

• JCSDA/CIMSS. Framework, CloudScatter, IR
  SfcOptics for water, RTSolution for validation
  (VDISORT)
• AER Inc Jean-Luc Moncets group. OSS
  AtmAbsorption.
• NOAA/ETL Al Gasiewskis group. CloudScatter (?W)
  and RTSolution.
• NASA/GSFC Clark Weaver. AerosolScatter (IR).
• NOAA/NESDIS Fuzhong Wengs group. Microwave
  SfcOptics for land, snow, ice.
• AOS/SSEC/CIMSS/UWisc-Madison Ralf Bennartzs
  group. SOI RTSolution.
• UCLA Kuo-Nan Lious group. ?-4 stream
  RTSolution.

5
CRTM Schematic
6
What is the CRTM Framework?

• At the simplest level, its a collection of
  structure definitions, interface definitions, and
  stub routines.
• There are User and Developer interfaces, as well
  as Shared Data interfaces and I/O.

Why do this?

• The radiative transfer problem is split into
  various components (e.g. gaseous absorption,
  scattering etc) to facilitate independent
  development.
• Want to minimise or eliminate potential software
  conflicts and redundancies.
• Components developed by different groups can
  simply be dropped into the framework.
• Faster implementation of new science/algorithms.

http//cimss.ssec.wisc.edu/paulv/CRTM
7
AtmAbsorption (1)

• Two methodologies
• OPTRAN. Polychromatic (two versions). Adjoint
  Jacobians.
• OSS. Monochromatic. Analytic Jacobians.

OPTRAN OSS
Total channel resolution transmittance Predict band transmittance for each absorbing gas from absorption coefficient, ?, predicted from regression fits Select the regression coefficients, cijk, for each gas that minimises transmittance errors. Channel radiances are obtained from a weighted sum of monochromatic radiances for a set of predefined nodes, The monochromatic Rn are obtained from the OSS monochromatic optical depth profiles for the selected node frequencies. Nodes are selected and weights calculated for a channel to satisfy a specified accuracy (e.g. 0.05K). Higher accuracy ? more nodes ? longer computation times.
8
AtmAbsorption (2)Computation and Memory
Efficiency
Time needed to process 48 profiles with 7
observation angles
Instrument OPTRAN-V7 Adjoint CompactOPTRAN Adjoint OSS Adjoint
AIRS 22m36s 35m12s 3m10s
HIRS 13s 17s 9s
Memory resource required (Megabytes)
Instrument OPTRAN-V7 Double precision CompactOPTRAN Double precision OSS Single precision
AIRS 66 5 97
HIRS 0.5 0.04 4
9
CloudScatter (1)

• NESDIS/ORA lookup table (Liu et al, 2005).
• Mass extinction coefficient
• Single scatter albedo
• Asymmetry factor
• Legendre phase coefficients and delta-truncation
  factors for 2-, 4-, 6-, 8-, and 16-streams
• Analytic phase functions (HG and Rayleigh)
• Infrared (Intensity only)
• Spherical particles for liquid water and ice
  cloud (Simmer, 1994)
• Non-spherical ice cloud (Liou and Yang, 1995
  Macke et al,1997 Baum et al, 2001)
• Microwave (including polarisation)
• Spherical particles for rain drops and ice cloud
  (Simmer, 1994)
• The number of streams is determined using Mie
  size parameter, 2?reff/?

10
CloudScatter (2)

• ETL library is microwave only, fixed stream (8)
• Mie spherical scattering model
• Five hydrometeor phases, exponential size
  distributions
• Phase functions
• Currently, Henyey-Greenstein phase function
  matrix.
• Being extended to incorporate full Mie scattering
  phase functions via an exact Mie library.
• Currently, no polarisation capability.
• Scattering matrix Jacobians are analytical.

11
AerosolScatter

• Currently only handles aerosol absorption.
  Aerosol scattering is planned.
• Seven aerosol types
• Dust
• Sea salt
• Dry and wet organic carbon
• Dry and wet black carbon
• Sulfates
• Multiple size distribution modes.
• Uses same scattering structure definition as
  CloudScatter routines

12
SfcOptics

• Microwave
• Land (Weng et al, 2001) Snow and sea ice (Yan
  and Weng, 2003)
• Ocean
• Wind vector dependent (Liu and Weng, 2003)
• Wind speed dependent (English, 1998 FASTEM-3)
• Infrared
• Land
• Measurement database for 24 surface types in
  visible and infrared (NPOESS Net heat Flux ATBD,
  2001)
• Regression methods
• Retrieval methods
• Ocean
• IRSSE (van Delst, 2003). Based on Wu-Smith (1997)
  model.
• Nick Nalli (NESDIS/ORA) also working on sea
  surface emissivity and reflectivity model.
• New surface optical models are also being
  developed by other groups (Land data assimilation
  folks)

13
RTSolution (1)

• SOI, Successive Order of Interaction
  (UWisc-Madison)
• Truncated doubling technique for layer
  transmission, reflection and source functions.
  Successive order of scattering (SOS) used to
  integrate emission and scattering events from
  surface and atmosphere. IR and ?W. (Heidinger et
  al, 2005)
• Vector delta 4-stream (UCLA)
• Delta truncation is applied to reduce phase
  functions to four expansion terms. The optical
  depth and single scattering albedo, as well as
  the expansion coefficients, are rescaled to take
  account of strong forward scattering. The layer
  transmission, reflection, and source functions
  are solved analytically. Adding method is used
  for vertical integration. IR and ?W. (Liou et al,
  2005)

14
RTSolution (2)

• DOTLRT, Discrete ordinate tangent linear raditive
  transfer (NOAA/ETL)
• Layer transmission and reflection computed using
  a matrix operator method. Symmetric phase matrix
  is used to simplify matrix manipulation. Layer
  source function is obtained from the transmission
  and reflection. Adding method is used for
  vertical integration. IR and ?W. (Voronovich et
  al, 2004)
• No polarisation capability.
• VDISORT (NOAA/NESDIS/ORA)
• Used for validation.
• Valid for visible, infrared and microwave
  sensors fully polarimetric (all Stokes vectors).
• Matrix operator method is used for layer
  transmission, reflection and source functions.
  Vertical integration is performed with linear
  algebra system where continuity at boundaries is
  ensured. (Weng and Liu, 2003)

15
Current Status

• Currently we are working on the integration of
  the various CRTM components.
• Goal is to produce a working version of an
  OSS-based CRTM by end of June (!).
• An OPTRAN-based model is also being worked on for
  migration purposes in the GFS.
• The development process was deliberately informal
  so there is a bit of integration work still to
  do.
• Some codes have to be modified from a
  channel-based form (e.g. IRSSE model) to a
  frequency-based form for use with OSS.
• Other outstanding issues are that some codes
  produce analytic Jacobians rather than coding
  adjoints via the forward ? tangent linear ?
  adjoint ? K-matrix methodology.

16
Jacobians. Analytic or adjoint?

• Here Im talking about the use of Jacobians in
  NWP. Other applications may have other
  requirements.
• Analytic Jacobians are generally not suitable for
  variational analysis. This is difficult to prove,
  but experience in NWP has shown this to be the
  case.
• The Jacobian has to take into account any
  numerical approximations used in the forward
  model, e.g. quadrature, regression fitting,
  interpolations, etc.
• The Jacobian has to be entirely numerically
  consistent with the forward model.
• Minimisation algorithms in NWP are sensitive to
  very small inconsistencies or errors.