A discussion of Radiative Transfer Models Thomas J

  • Slides: 34
Download presentation
A discussion of Radiative Transfer Models Thomas J. Kleespies NOAA/NESDIS

A discussion of Radiative Transfer Models Thomas J. Kleespies NOAA/NESDIS

Disclaimer • There will be some overlap between this presentation and this afternoon’s presentation

Disclaimer • There will be some overlap between this presentation and this afternoon’s presentation • There will be some overlap between these two presentations and other speakers • This is good! • This presentation will be a broad brush overview of radiative transfer models relevant to data assimilation (i. e. fast and with adjoint)

Outline • • • Scattering vs absorption models Line vs fast models Generally available

Outline • • • Scattering vs absorption models Line vs fast models Generally available line models Generally available fast models Generally available scattering models Generally available general purpose models

Scattering vs Absorption models • Full radiative transfer equation – Surface emission – Atmospheric

Scattering vs Absorption models • Full radiative transfer equation – Surface emission – Atmospheric emission – Surface reflection • Downward atmospheric • Downward solar • Downward CBR – Single scattering – Multiple scattering – Solar and Thermal

Single vs Multiple Scattering

Single vs Multiple Scattering

Atmospheric Emission Solar Single Scattering Atmospheric Multiple Scattering Surface Emission Surface Solar Reflection Surface

Atmospheric Emission Solar Single Scattering Atmospheric Multiple Scattering Surface Emission Surface Solar Reflection Surface diffuse reflection

General Types of Absorption Models • Line – by – Line models – Numerically

General Types of Absorption Models • Line – by – Line models – Numerically integrate individual absorption lines to produce optical depth/transmittance profile as a function of wavelength • Band Models – fit to LBL at narrow spectral interval • Fast Models – Statistical fit to line-by-line model transmittance for diverse atmospheres for specific instrument bandpass

General Types of Scattering Models • Discrete Ordinate Models – Radiative intensity not a

General Types of Scattering Models • Discrete Ordinate Models – Radiative intensity not a function of azimuth – N streams … bigger N, better result, more expensive • Adding/Doubling Models – Form of ray tracing with known transmission/ reflection of each layer. – Reflection and transmission of combined layers obtained by computing successive reflections between two layers. – If layers have same optical depth, referred to as doubling • Monte Carlo

Additional • Generally need a Mie code (spherical scatterers) or something like it to

Additional • Generally need a Mie code (spherical scatterers) or something like it to specify the scattering parameters • Need to pick a polydispersion of scatterers

 • Multiple scattering Models generally too expensive for data assimilation • When scattering

• Multiple scattering Models generally too expensive for data assimilation • When scattering is included in data assimilation RTM, it is generally single scattering, or a few streams approximation. • This is the topic of very active research

Generally Available Scattering Models • DISORT – Discrete ordinate, N-stream plane parallel code •

Generally Available Scattering Models • DISORT – Discrete ordinate, N-stream plane parallel code • Fu-Liou – 2 and 4 stream radiative transfer solver • SHDOM – 1, 2, 3 D RTM , including adjoint

Steps in absorption/emission modeling • Spectroscopic Data Base – Absorber, line position, line strength,

Steps in absorption/emission modeling • Spectroscopic Data Base – Absorber, line position, line strength, line half-width, etc. • Line model – Integrates individual lines over instrument spectral response. – Very expensive in infrared, almost trivial in microwave • Fast model – polynomial fit to LBL transmittances

Public molecular absorption databases • HITRAN – Rothman et al. – HITRAN is a

Public molecular absorption databases • HITRAN – Rothman et al. – HITRAN is a compilation of spectroscopic parameters that a variety of computer codes use to predict and simulate the transmission and emission of light in the atmosphere. The original version was created at the Air Force Cambridge Research Laboratories (1960's). The database is maintained and developed at the Harvard-Smithsonian Center for Astrophysics in Cambridge MA, USA. • GEISA - Jacquinet-Husson et al. – GEISA is a computer-accessible spectroscopic database, designed to facilitate accurate forward radiative transfer calculations using a line-by-line and layer-by-layer approach. Maintained and developed at Laboratoire de Météorologie Dynamique (LMD) in France.

Absorption Lines near 15 mm

Absorption Lines near 15 mm

Absorption Lines Near 4. 3 mm

Absorption Lines Near 4. 3 mm

Microwave line absorption

Microwave line absorption

Generally Available Line Models • LBLRTM – AER (Clough et al. ) – Based

Generally Available Line Models • LBLRTM – AER (Clough et al. ) – Based on FASCODE from AFGL – Supported by ARM – Widely used in US – Uses HITRAN • GENLEN 2 – Edwards – NCAR – Used in Europe – Europeans use GEISA or HITRAN

Generally available band models • MODTRAN – developed by AFGL – enhanced by private

Generally available band models • MODTRAN – developed by AFGL – enhanced by private industry – based upon FASCODE/LBLRTM/HITRAN – extremely flexible in atmosphere, absorbers, geometry – extremely cumbersome data input – no adjoint available, too slow for DA – Do. D support waning

Generally Available Fast Models • Community Radiative Transfer Model-CRTM (OPTRAN) – used at NCEP/EMC,

Generally Available Fast Models • Community Radiative Transfer Model-CRTM (OPTRAN) – used at NCEP/EMC, NASA, … • Radiative Transfer TOVS (RTTOV) – used at EUMETSAT, METOFFICE, others • AESFAST – used at Environment Canada • Optimal Spectral Sampling (OSS) – developed at AER, receiving some attention here

Why fast models? • Line-by-line models can take up to an hour to compute

Why fast models? • Line-by-line models can take up to an hour to compute a single channel/atmosphere • Fast models take maybe a millisecond • Line-by-line models don’t have adjoints, have to do finite differencing • Fast models have analytic adjoints. We will see how to write them this afternoon.

CRTM • Based upon Optical Path TRANsmittance (Mc. Millin et al) • Includes thermal

CRTM • Based upon Optical Path TRANsmittance (Mc. Millin et al) • Includes thermal emittance, clouds, simplified scattering, CBR, solar influence, surface emissivity models, support for many instruments • Yong Han will discuss in detail on 8 Aug

OPTRAN • Predicts absorption coefficient on the absorber level (product of which is optical

OPTRAN • Predicts absorption coefficient on the absorber level (product of which is optical depth, negative exponential of that is transmittance) – arbitrary input profile • Zenith angle implicitly included in optical path • Fitted to LBLRTM/HITRAN and Leibe(89, 93) • Adjoint well developed

RTTOV • Predicts optical depth on fixed pressure levels – must interpolate input profile

RTTOV • Predicts optical depth on fixed pressure levels – must interpolate input profile to the fixed levels • Must explicitly treat zenith angle • Includes thermal emittance, clouds, simplified scattering, CBR, solar influence, surface emissivity models, support for many instruments • Fitted to GENLN 2/MPM 89 -92/HITRAN (Kcarta for IASI) • Adjoint well developed

Which is better? • Depends on the situation/instrument • For a bit dated but

Which is better? • Depends on the situation/instrument • For a bit dated but still useful intercomparison of RTMs, see http: //collaboration. cmc. ec. gc. ca/science/ar ma/intercomparison/index. html

CRTM vs RTTOV • A friendly competition is very healthy, and results in improvements

CRTM vs RTTOV • A friendly competition is very healthy, and results in improvements in both codes • Institutions generally like to use locally developed codes because of control issues.

Innovation This is the de-biased difference between the radiance observations and the radiances computed

Innovation This is the de-biased difference between the radiance observations and the radiances computed from the NWP model background Following statistics are from global assimilation systems, clear only.

NCEP Temperature

NCEP Temperature

Metoffice Temperature

Metoffice Temperature

NCEP Water Vapor

NCEP Water Vapor

Metoffice Water Vapour

Metoffice Water Vapour

ITWG Intercomparison Bias against RTM

ITWG Intercomparison Bias against RTM

ITWG Intercomparison

ITWG Intercomparison

Summary • Overview of radiative transfer modeling • Emphasis on fast forward models •

Summary • Overview of radiative transfer modeling • Emphasis on fast forward models • This afternoon I will discuss how the Jacobians are computed • You will have the opportunity to perform simple coding exercises to generate tangent linear/adjoint-jacobian code