Post on 30-Oct-2021
Develop an Advanced Radiative Transfer Modeling System
(ARMS) for Accelerated Uses of FengYun Satellite Data in
Numerical Weather Prediction Models
Fuzhong Weng
10th Asia-Oceania Meteorological Satellite Users’ Conference
Melbourne, Australia, December 2-7, 2019
Laboratory of Severe Weather, Beijing, China
2
ECMWF Global Medium-Range Forecast Skill
ECWMF started assimilation of satellite radiances in 1999 when the first microwave sounding data were assimilated.
The forecast score in SH and NH in terms of 500 hPa approached to the same level. In SH conventional data void
region, the score increase is largely attributed to uses of satellite observations
3
ECMWF Forecast of Hurricane Irma (2017)
Forecast with satellites
Forecast without satellites
700hPa humidity and wind initial conditions with satellites
700hPa humidity and wind initial conditions without satellites
Red shading humidity > 95%
Source: ECMWF
4
Number of Instruments Monitored and Assimilatedin European Global Forecast System
5
Radiative Transfer for TOVS (RTTOV) Model
Input profile on N
levels, view angles
and surface
parameters
Interpolate profile onto
54 fixed levels and check
Calculate predictors
on 53 layers from profile
Instrument
coefficients
Multiply predictors by
coefficients for each channel
=> layer optical depths for
each channel
Interpolate optical depths
to N user levels
Integrate RT equation
for each channel on user levels
Output radiances
and BTs/Reflectances
Optional surface
emissivity/reflectance
calculation
Optional cloud and
aerosol calculation
Optional solar scattering and
NLTE calculations
6
Community Radiative Transfer Model (CRTM)
• Atmospheric gaseous absorption – Band absorption coeff trained by LBL
spectroscopy data with sensor response functions
– Variable gases ( H2O, CO2, O3 etc) .
– Zeeman splitting effects near 60 GHz
• Cloud/precipitation scattering and emission– Fast LUT optical models at all phases
including non-spherical ice particles
– Gamma size distributions
• Aerosol scattering and emission– GOCART/WRF-CHEM (dust, sea salt,
organic/black carbon)
– Lognormal distributions with 35 bins
• Surface emissivity/reflectivity – Two-scale microwave ocean emissivity
– Large scale wave IR ocean emissivity
– Land mw emissivity including vegetation and snow
– Land IR emissivity data base
• Radiative transfer scheme – Tangent linear and adjoints
– Inputs and outputs at pressure level coordinate
– Advanced double and adding scheme
– Other transfer schemes such as SOI, Delta Eddington
“Technology transfer made possible by CRTM is a shining example for
collaboration among the JCSDA Partners and other organizations, and
has been instrumental in the JCSDA success in accelerating uses of
new satellite data in operations” – Dr. Louis Uccellini, Director of
National Weather Service
6
Fengyun Satellite Programs in Next Five Years
7
2012 FY-2F(Op)
2010 FY-3B (R&D)
2013 FY-3C(Op)2017 FY-3D(Op)
2014 FY-2G(Op) 2016 FY-4A (R&D)
2019 FY-3E(Op)
2021 FY-RM(Op)
2020 FY-4B (Op)
2020 FY-3F(Op)
2022 FY-4C(Op)
2018 FY-2H(Op)
2023 FY-3G(Op)
8
• In the geostationary orbit, FY-4 is carrying onboard the hyperspectral infrared
sounder, vis/IR imager, lightening mapper. The hyperspectral IR sounder is the
world first instrument in the geo-orbit. FY-4M will fly carrying onboard the geo-
microwave sounding imager in 2022.
• In the leo-orbits, FY-3 satellites are carrying onboard both hyperspectral infrared
sounder, microwave sounder and imager, radio occultation instruments, vis/infrared
imager, UV and space weather instruments.
• FY-3E will be the first early morning mission, in constellation with EUMETSAT
and US polar-orbiting satellite system.
• FY-3RM is a global precipitation mission and carries on board both active and
passive instruments to continue.
• Emerging small satellite missions from commercial sectors and non-CMA satellite
missions will require more mission agnostic approaches in developing the products
and applications.
• More importantly, a fast and accurate radiative transfer model must be developed to
support FY data assimilation
New Instruments on board FY Satellites Require
Intensive and Extensive Scientific Efforts
• Large uncertainties remain at large in simulating the radiances in scattering
atmosphere.
• Large biases (O-B) are observed for surface sensitive channels especially over
land at microwave frequencies, and the O-B angular-dependence behaves
differently from one model to another, and sometime is unphysical.
• Radiative transfer schemes solve for radiative intensity. The instruments of UV,
visible, and microwave imagers are sensitive to polarization and thus require a full
vector transfer scheme.
• A reference quality ocean emission and reflection model was a gap in our ability
to provide absolute calibration of the satellite-based observing system
9
Common Issues in Current Fast Radiative
Transfer Models
On April 29 to May 3, 2019, CMA,
EMCWF and JCSRA jointly hosted a joint
workshop on radiative transfer models for
satellite data assimilation at Tianjin, China.
More than 100 scientists from China, US,
UK, Germany and Japan attended the
workshop.
The participants reported the major
progresses in developing the fast radiative
transfer models for satellite data assimilations. In
past, the NWP community primarily uses RTTOV
(Europe) and CRTM (US). Now, China is
developing the Advanced Radiative Transfer
Modeling System (ARMS) for FengYun satellite
data applications. The SSC recognized the
significance of ARMS which will be the third
pillar in supporting NWP satellite data
assimilation after RTTOV and CRTM.
2019 International Workshop on Radiative Transfer Model for Satellite Data Assimilation, Tianjin, China
A science steering committee for radiative
transfer (SSC4RT) was formed and 10
distinguished scientists are selected as SSC
members. Several critical actions will be taken
after the workshop.
12
Advanced Radiative Transfer Modeling System (ARMS)
11
12
CRTM, RTTOV and ARMS
CRTM RTTOV ARMS
Radiative
transfer Solver
Advanced Adding
and Doubling
(ADA)
Delta-Eddington
Approximation,
DISORT, MFEAST
Polarization Two-
Stream, ADA, VDA,
VDISORT, SOI, 6S-V
Scattering
properties
Mie Table as a
function of
frequency,
temperature, radii,
and hydrometeor
type and density
Mie Table as a function
of frequency,
temperature, and
hydrometeor type and
density, Discrete
Dipole Approximation
Mie Table, T-Matrix
as a function of
frequency,
temperature, and
hydrometeor type and
density.
Cloud type Water, ice, rain,
snow, graupel, hail
Water, ice, rain, and
snow
Water, ice, rain, snow,
graupel, hail
Surface models NPOESS IR LUT,
Wu/Smith IR
Ocean EM, MW
LandEM, FASTEM
MW Ocean EM
UWisc IR Emissivity
Database, Cox/Munk
IR Ocean EM,
FASTEM,CNRW
MW,TELSEM MW
Uwisc IR Emissivity
Database, Wu/Smith
IR Ocean EM, MW
LandEM, FASTEM
MW Ocean EM, AIEM,
CNRW MW,TELSEM MW
13
ARMS Supported Instruments
• FY-3A MWTS
• FY-3A MWHS
• FY-3B MWTS
• FY-3B MWHS
• FY-3C MWTS-2
• FY-3C-MWHS-2
• FY-3D MWTS-2
• FY-3D MWHS-2
• FY-3 B/C/D MWRI
• FY-3 B/C VIRR
• FY-3C MERSI
• FY-3C IRAS
• FY-3D MERSI-2
• FY-3D HIRAS
• FY-4A GIIRS
• FY-4A AGRI
• FY-4M GMIS
• NOAA 15 to 19 AMSU-A
• NOAA 18-19 MHS
• NOAA 18-19 HIRS
• NOAA 15-19 AVHRR
• SNPP/NOAA-20 ATMS
• SNPP/NOAA-20 CrIS
• SNPP/NOAA-20 VIIRS
• METOP-A to C IASI
• METOP-A to C IASI
• METOP-A to C AMSU-A
• METOP-A to C AVHRR
• JAXA AMSR2
• NASA GMI
• EOS Aqua AIRS
• EOS Terra/Aqua MODIS
Radiative Transfer Solvers Used in ARMS
• Discrete Ordinate Radiative Transfer (DISORT)
• Vector DISORT (VDISORT)
• Matrix Operator (MO)
• Double and Adding (DA)
• Successive Order of Iteration (SOI)
• Polarization Two Stream Approximation
• Delta Eddington Scheme
• Second Simulation of a Satellite Signal in the Solar
Spectrum (6S)
• Line by Line Radiative Transfer Model (LBLRTM)
• Moderate Resolution Transmission (MODTRAN)
• Discrete Anisotropic Radiative Transfer (DART)
14
1exp[ ( )]l −= − +l l
I A C S
Sl=
m0{B(
l−1) +
B(l−1
) − B(l)
l−1
− l
[Al
−1 + ( −
l−1)]
+ 0[
0A
l+ E]
−1F
0
exp(−
0)}
IL
(L) = B(T
s) + RI
L(
L)
+ R0
F0
exp(−
L
0)
I
l(
l−1) = I
l−1(
l−1)
I
l(0) = I
0
Cloud Optical Property Library Used in ARMS
15
• Developed with the most accurate and
state-of-the-art light scattering
computation methods (T-Matrix [Bi et
al., 2014] and IGOM [Yang et al.,
1996]);
• Wide coverage of the spectrum from
0.2 to 100 um;
• Wide particle size range (maximum
dimension) from 2~104 um;
• Complete scattering phase matrix with
polarization
• Three degrees of ice surface roughness:
Completely Smooth, Moderately
Rough, Severely Rough;
• Extended to the microwave spectrum;
temperature dependence considered;
Ice particle single-scattering property database
Yang et al., 2013, JAS
Aerosol Scattering Database in ARMS
16
Infrared Line by Line (HITRAN)
Spectroscopy Data Base
17
MonoRTM Line by Line Spectroscopy Data Base
18
Atmospheric transmittance as a function of frequency in microwave region. The
black, blue, red and green curve represents the contribution of total, oxygen, water
vapor and ozone to the optical depth
Tra
nsm
itta
nce
Diverse Profiles from ECMWF Model Used
for Training Transmittance Models
19
ARMS Layer Optical Thickness Training
20
T: temperature (K); P : pressure (hPa)
Index w, o, d represent water vapor, ozone and dry gas, respectively
=
=
=
−−−−+=
−−−−+=
−−−+=
l
i
l
i
l
i
liAiAiiTiiTlT
liAiAiAiTiAiTlT
lAiAiAiTiTlT
AAA
A
1
322
1
2
1
))(2/())]1()()(1())(1()()([()(***
)),(2/())]1()()(1())1()()([()*(*
)),(2/())]1()())(1()([()(*
A(i) is the ith level integrated absorber amount for water vapor. The same equations applied to P*,p**,P***.
• A set of 6 predictors varying with
channel is selected from the predictors
pool;
• An exhausting search is performed for
each gas component and channel to
select the best set of predictors;
• The set of predictors with strong
correlations between the selected
predictors is not included.
=
+=6
1
,,0,, )()()())((j
ijiijiiiiich AxAcAcAkLn
)(, iich Ak is the absorption coefficient and Ln() is
the natural logarithm
21
HIRAS Simulation from ARMS vs CRTM
Surface Type: Water; The Local zenith angle is 50.208 degree. Surface temperature=288.0 K.
Wind_Speed: 10.0 m/s. Wind_Direction 10.0 degree. Salinity: 33 ppt
Wu and Smith.IR Water Emissivity
Performance of ARMS HIRAS Transmittance Module
22
Performance of ARMS GIIRS Transmittance Module
23
24
Surface Emissivity Models
Microwave land emissivity model (NESDIS model) (Weng et al, Yan, Grody, 2001), desert microwave emissivity library (Yan and Weng, 2011) TELSEM, and CNRM databases (Prigent, 200x)Vegetation (Chen and Weng, 2014)Surface roughness (Chen and Weng, 2015)
Ocean Sea Ice Snow Canopy (bare soil) Desert
Empirical snow and sea ice microwave emissivity algorithm (Yan and Weng, 2003; 2008)
FASTEM microwave emissivity model (Liu et al., 2010, English , 199x)
IR emissivity model (Wu and Smith, 1991; van Delst et al., 2001; Nalli et al., 2008)
NPOESS Infrared emissivity databaseIASI Land Infrared emissivity databaseUWIREMIS database
25
Group Photo at 2019 Bern Working Group Meeting on “A
reference quality model for ocean surface emissivity and
backscatter from the microwave to the infrared”
Major Gaps in Radiative Transfer Modeling:
A Reference Quality Ocean Emission and Reflection Model
• On November 20-22, 2019, a group of 15
scientists from US, Europe, China and
Japan had a meeting at Bern, Switzerland
to work on the European Commission
Horizon 2020 project (GAIA-CLIM).
• A reference quality ocean emission and
reflection model is needed to provide
absolute calibration of the satellite based
observing system. By reference quality it is
meant that the uncertainty is known, both
in terms of systematic and random error
and can be traced to SI standards.
• The gap and the need to work
collaboratively to address it has therefore
been well documented in many
international scientific fora.
26
Fast Ocean Emissivity Models
Foam-free Reflectivity in H-Polarization due to wind roughness
1 5 2 9
, ,
1.15 10 3.8 10 10
)
/
(
h foam free h calm s
tr w f
tr t
− − −
−
= +
= −
1 3 2 9
, ,
1.17 10 2.09 10 (7.32 10 ) 10
/
( )
v foam free v calm s
tr w exp f
tr t
− − − −
−
= −
= −
3 5 2 7 3 1.0 1.748 10 7.336 10 1.044 10g − − −= − − +
4 5 2 6 3 20 10 1.0 9.946 10 3.218 10 1.187 10 7 10g − − − −= − + − +
9
, 1.0 - (208.0 1.29 10 * ) / ( )
h foam sf t g
− = +
9
, 1.0 - (208.0 1.29 10 * ) / ( )
v foam sf t g
− = +
Foam Reflectivity in H-Polarization
Foam Reflectivity in V-Polarization
Foam-free Reflectivity in V-Polarization due to wind roughness
( ) , ,1p c p foam free c p foamf f− = − +
27
Refractivity or Emissivity for Calm Water Surface
For a specular surface, reflectivity can be calculated by Fresnel law:
( , )p
f Reflectivity
2
h,calm2
cos ( , ) sin
cos ( , ) sin
f
f
− − =
+ −
f Frequency Local zenith angle
2
,2
( , ) cos - ( , ) - sin
( , ) cos ( , ) - sinv calm
f f
f f
=
+
sea surface
28
Foam Coverage vs Wind Speed
Foam is a mixture of air and water
and has a higher emissivity than flat
water
Foam coverage:
231.3
0
61075.7
=
−
V
Vf c
5 2.551.95 10
cf u
−=
Stogryn, 1972
Monahan, 1986
29
Foam Emissivity vs. Incident Angle
30
Two-Scale Ocean Emissivity Model
The large-scale roughness is dependent on the gravity waves and whereas the small irregularities
is affected by capillary waves. There are coherent reflection and incoherent scattering associated
with the waves in both scales
Large scale
Small scale foam
coherent
incoherent
downwind upwind
crosswind
31
There are several well-established methods for
simulation of electromagnetic scattering from
randomly rough surfaces
❑ Kirchhoff Method (KM) based on the
assumption that the wavelength of the incident
wave is much shorter than the horizontal variations
of the surface so that the general solution can be
regarded as the integration of local plane-boundary
reflections.
Tangential Plane Approximation
Stationary Phase Approximation and Geometric
Optics (GO) (FASTEM)
Scalar Approximation and Physical Optics (PO)
❑Small Perturbation Method (SPM) based on the
assumption that the surface correlation length and
its standard deviation are smaller than the
wavelength (low frequencies).
❑Composite Two-scale Model based on the
separation of both the surface and the EM wave into
two distinct scales, e.g., Yueh et al., 1997
Two-Scale MW Emissivity Model
Relationship of BRDF, Bistatic Coeffs and Emissivity
10-3 10-2 10-1 100 101 10 2 103
105
100
10-5
10-10
10-15
Bjerkaas/Riedel (BR) Ocean Roughness Spectrum
32
(Elfouhaily et al., 1997).
33Yueh 1997
FASTEM6/5 and Two-Scale
Comparison of Model Simulations with JPL
WINDRAD Observations (theta=30o)
34
Multisensor Remote-sensing Testbed
Input
Satellite radiance or
Brightness temperature,
geolocation information
One Dimension
Variational (1DVAR)
for
sequential or
simultaneous retrievals
Output
Atmospheric
temperature, moisture,
hydrometeor, aerosol,
trace gases, profiles
Forward/Jacobian
Operators
CRTM
RTTOV
ARMS
Background
Atmospheric and
Surface Parameters
NWP model outputs or
climatology profile
• Algorithm valid in all-weather conditions, over all-surface types
• Model& instrumental errors are input to algorithm
• Background and observation error covariances are scene-dependent.
• Selection of background from climatology, NWP forecasts, and regressions
• Selection of channels to use and parameters to retrieve
35
Comparison of FY-3D and NOAA
Microwave Sounding Capability P
ress
ure
(h
Pa)
Weighting Function
ATMS (22 Channels)
Weighting Function
CMWS (30 channels)
Combined microwave sounding Suite (CMWS) from FY-3D MWTS and MWHS has
a better vertical resolution for atmospheric sounding comparing to ATMS
Comparison of ATMS and CMWS Warm Core
ATMS CMWS-20 CMWS-28
34
Comparison of Atmospheric Profiles Derived
from ATMS and CMWS(MWTS/MWHS)
Validation of atmospheric profiles derived from ATMS using GPS dropsondes for
hurricane Florence shows FY-3 microwave sounding system has better profiling
capabilities in hurricane conditions
HIRAS Channel Selection
38
• Principal component analysis is used for channel selection.
• 450 channels is selected,
39
FY-3D HIRAS Derived Atmospheric Profiles
• Data between June 2018 to May 2019 are used for retrieval, validation with ERA5 reanalysis
• 1DVAR is better than regression and machine learning
• The mean temperature RMS is about 1K between 200hPa and 700hPa
Slide courtesy of He Yanfeng, Anhui Meteorological Bureau
Summary and Conclusions
• Fast and accurate radiative transfer models are required for sensor
simulation, instrument calibration and validation, remote sensing and
NWP data assimilation, and thus are critical for satellite mission
successes
• Currently, RTTOV, CRTM and ARMS are now supporting operational
and research missions for critical satellite data assimilation, and they are
sharing some common modules for many applications
• ARMS is now being integrated with CMA NWP systems (e.g. 1dvar,
GRAPES-4dvar) with focuses on FY data assimilation
• ARMS team is working with the international community to accelerate
its science developments
40
Acknowledgement
• Chinese Academy of Meteorological Sciences (CAMS)
• National Satellite Meteorological Centre (NSMC)
• National Meteorological Centre (NMC)
• Chinese Academy of Sciences (CAS)
• Nanjing University of Information Science & Technology (NUIST)
• Nanjing University (NJU)
• Zhejiang University (ZJU)
• Sun Yat-sen University (SYSU)
• Fudan University (FDU)
• UK Metoffice
• ECMWF
• JCSDA
41