3D Radiative Transfer in Cloud Resolving Models
Fabian Jakub, Carolin Klinger
LMU - Meteorological Institute Munich
November 16, 2015
1 / 26
Does 3D Radiative Transfer impact cloud evolution?
Earlier studies suggest radiation may affect
I cloud evolution and lifetime
I microphysical processes (condensation, nucleation)
I precipitation onset/amount
I convective organization
Can we model radiative transfer in the atmosphere?
2 / 26
Does 3D Radiative Transfer impact cloud evolution?
Earlier studies suggest radiation may affect
I cloud evolution and lifetime
I microphysical processes (condensation, nucleation)
I precipitation onset/amount
I convective organization
Can we model radiative transfer in the atmosphere?
2 / 26
HD(CP)2 project (www.hdcp2.eu)
I run hindcasts over Central Europe
I 100m horizontal resolution
I grids consisting of 10.000 x 15.000 x 300 voxels
I first develop a model capable of running it (ICON)
I . . . with the goal to develop improved parameterizations for
weather and climate predictions
3 / 26
History of Radiative Transfer
Radiative Transfer theory well established
I radiative transfer equation (1960 Chandrasekhar)
dL
kext · ds= −L+
ω0
4π
∫4π
p(Ω′,Ω) L(Ω
′)dΩ
′+(1−ω0)BPlanck(T )
I surprisingly well working 1D approximations
I sophisticated 3D models since the 90’s (e.g. MonteCarlo)
I . . . but orders of magnitude too slow to run in atmospheric
models
4 / 26
History of Radiative Transfer
Radiative Transfer theory well established
I radiative transfer equation (1960 Chandrasekhar)
dL
kext · ds= −L+
ω0
4π
∫4π
p(Ω′,Ω) L(Ω
′)dΩ
′+(1−ω0)BPlanck(T )
I surprisingly well working 1D approximations
I sophisticated 3D models since the 90’s (e.g. MonteCarlo)
I . . . but orders of magnitude too slow to run in atmospheric
models
4 / 26
History of Radiative Transfer
Radiative Transfer theory well established
I radiative transfer equation (1960 Chandrasekhar)
dL
kext · ds= −L+
ω0
4π
∫4π
p(Ω′,Ω) L(Ω
′)dΩ
′+(1−ω0)BPlanck(T )
I surprisingly well working 1D approximations
I sophisticated 3D models since the 90’s (e.g. MonteCarlo)
I . . . but orders of magnitude too slow to run in atmospheric
models
4 / 26
Approximations for Radiative Transfer
Radiative transfer describes photon interactions with atmosphere.MonteCarlo modeling of scattering and absorption:
simplify to solve:
I Plane Parallel approx.
I Independent Column approx.
I Twostream solvers
I diagonal band-matrix (5)
5 / 26
Approximations for Radiative Transfer
Radiative transfer describes photon interactions with atmosphere.MonteCarlo modeling of scattering and absorption:
simplify to solve:
I Plane Parallel approx.
I Independent Column approx.
I Twostream solvers
I diagonal band-matrix (5)
5 / 26
Approximations for Radiative Transfer
Radiative transfer describes photon interactions with atmosphere.MonteCarlo modeling of scattering and absorption:
simplify to solve:
I Plane Parallel approx.
I Independent Column approx.
I Twostream solvers
I diagonal band-matrix (5)
5 / 26
A walk through the ages
3D parameterizations – a tradition at MIM:
I Tilted-ICA (Gabriel and Evans 1996,
Varnai 1999)
I Non-local ICA (Marshak 1998)
I displaced shadow regions (Schumann 2002)
I TICA in EULAG (Wapler 2008)
I NICA with automatic kernel
size (Wissmeier 2012)
6 / 26
A walk through the ages
3D parameterizations – a tradition at MIM:
I Tilted-ICA (Gabriel and Evans 1996,
Varnai 1999)
I Non-local ICA (Marshak 1998)
I displaced shadow regions (Schumann 2002)
I TICA in EULAG (Wapler 2008)
I NICA with automatic kernel
size (Wissmeier 2012)
6 / 26
A walk through the ages
3D parameterizations – a tradition at MIM:
I Tilted-ICA (Gabriel and Evans 1996,
Varnai 1999)
I Non-local ICA (Marshak 1998)
I displaced shadow regions (Schumann 2002)
I TICA in EULAG (Wapler 2008)
I NICA with automatic kernel
size (Wissmeier 2012)
6 / 26
Why care for 3D radiation now? – a matter of resolution
Complex cloud radiation interaction
Visualization done with libRadtran.org/MYSTIC (Montecarlo code for the phYSically correct Tracing of photons In Cloudy atmospheres)
Mayer,
B., 2009. Radiative transfer in the cloudy atmosphere (EPJ Web of Conferences)
7 / 26
Why care for 3D radiation now? – a matter of resolution
Global models
Visualization done with libRadtran.org/MYSTIC (Montecarlo code for the phYSically correct Tracing of photons In Cloudy atmospheres)
Mayer,
B., 2009. Radiative transfer in the cloudy atmosphere (EPJ Web of Conferences)
7 / 26
Why care for 3D radiation now? – a matter of resolution
Weather models today
Visualization done with libRadtran.org/MYSTIC (Montecarlo code for the phYSically correct Tracing of photons In Cloudy atmospheres) Mayer,
B., 2009. Radiative transfer in the cloudy atmosphere (EPJ Web of Conferences)
7 / 26
Why care for 3D radiation now? – a matter of resolution
Next-gen models
Visualization done with libRadtran.org/MYSTIC (Montecarlo code for the phYSically correct Tracing of photons In Cloudy atmospheres) Mayer,
B., 2009. Radiative transfer in the cloudy atmosphere (EPJ Web of Conferences)
7 / 26
Why care for 3D radiation now? – a matter of resolution
Next-gen models
Visualization done with libRadtran.org/MYSTIC (Montecarlo code for the phYSically correct Tracing of photons In Cloudy atmospheres) Mayer,
B., 2009. Radiative transfer in the cloudy atmosphere (EPJ Web of Conferences)
7 / 26
The Tenstream solver
A new concept for a solver – what do we want?
I3RC cloud scene, benchmark heating rate
calculation with MYSTIC (MonteCarlo code)
I accurately approximate 3D
effects
I has to be several orders of
magnitude faster than state
of the art 3D solvers
I parallelizable on modern
machines
8 / 26
The TenStream solver
Finite Volume formalism:Discretization of energy transport – spatially and by angle
ST↓
SL→
SR→
SB↓
(a) direct streams (θ=40)
EB↑
ET↓
ERւEL
ց
ERտEL
ր
(b) diffuse streams
Fabian Jakub and Bernhard Mayer, 2015. A three-dimensional parallel radiative transfer model for atmospheric
heating rates for use in cloud resolving models – The TenStream solver (JQSRT)
9 / 26
The TenStream solver
Setup equation system for one voxel:
ET↑
EB↓
EL
ER
EL
ER
SB↓
SR→
=
γ1 γ2 γ3 γ3 γ4 γ4 β01 β11
γ2 γ1 γ4 γ4 γ3 γ3 β02 β12
γ5 γ6 γ7 γ8 γ9 γ10 β03 β13
γ5 γ6 γ8 γ7 γ10 γ9 β04 β14
γ6 γ5 γ9 γ10 γ7 γ8 β05 β15
γ6 γ5 γ10 γ9 γ8 γ7 β06 β16
0 0 0 0 0 0 α00 α10
0 0 0 0 0 0 α01 α11
EB↑
ET↓
ER
EL
ER
EL
ST↓
SL→
Coupling voxels in 3
dimensions. . .
. . . gives huge but sparse matrix.
=⇒ solve with PETSc!
10 / 26
The TenStream solver
Setup equation system for one voxel:
ET↑
EB↓
EL
ER
EL
ER
SB↓
SR→
=
γ1 γ2 γ3 γ3 γ4 γ4 β01 β11
γ2 γ1 γ4 γ4 γ3 γ3 β02 β12
γ5 γ6 γ7 γ8 γ9 γ10 β03 β13
γ5 γ6 γ8 γ7 γ10 γ9 β04 β14
γ6 γ5 γ9 γ10 γ7 γ8 β05 β15
γ6 γ5 γ10 γ9 γ8 γ7 β06 β16
0 0 0 0 0 0 α00 α10
0 0 0 0 0 0 α01 α11
EB↑
ET↓
ER
EL
ER
EL
ST↓
SL→
Coupling voxels in 3
dimensions. . .
. . . gives huge but sparse matrix.
=⇒ solve with PETSc!
10 / 26
Energy transport coefficients
We need to determine the energy transport
from one stream to another:
EB↑
ET↓
ERւEL
ց
ERտEL
ր
→ solve radiative transfer equation with
MonteCarlo method
. . . and put them into LookUpTable
11 / 26
Energy transport coefficients
We need to determine the energy transport
from one stream to another:
EB↑
ET↓
ERւEL
ց
ERտEL
ր
→ solve radiative transfer equation with
MonteCarlo method
. . . and put them into LookUpTable
11 / 26
Does it work?
3D MYSTIC 1D independent-column Twostream
Computations done with libRadtran (Library for Radiative Transfer, libradtran.org)
12 / 26
Does it work?
3D MYSTIC TenStream
Computations done with libRadtran (Library for Radiative Transfer, libradtran.org)
12 / 26
Thermal Radiative Transfer
Thermal spectral range (Wallace and Hobbs 2006, p.114).
13 / 26
3D Solar vs. Thermal Radiative Transfer
3D Solar 3D Thermal
Heating Rate [K/d]Surface Flux [W/m2]
0 30 60 90 120700 595 463 283 0 - - - -
14 / 26
Monte Carlo Variance Reduction Techniques
Emission–Absorption
dT
dt= − 1
ρ cp(qem−qabs)
0.0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
Sta
ndard
Devia
tion
[K/d
]
1 2 5 10 2 5 102
Optical Thickness
(1) Emission - Absorption
15 / 26
Monte Carlo Variance Reduction Techniques
Emission–Absorption
dT
dt= − 1
ρ cp(qem−qabs)
0.0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
Sta
ndard
Devia
tion
[K/d
]
1 2 5 10 2 5 102
Optical Thickness
(1) Emission - Absorption
(1b) Emission - Absorption (Variance Reduction)
15 / 26
Monte Carlo Variance Reduction Techniques
Net Flux Divergence
dT
dt= − 1
ρ cp∇ ~Enet
0.0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
Sta
ndard
Devia
tion
[K/d
]
1 2 5 10 2 5 102
Optical Thickness
(1) Emission - Absorption
(1b) Emission - Absorption (Variance Reduction)
(2) Net Flux Divergence
15 / 26
Monte Carlo Variance Reduction Techniques
CombiningEmission–Absorptionand Net FluxDivergence toHYBRID
0.0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
Sta
ndard
Devia
tion
[K/d
]
1 2 5 10 2 5 102
Optical Thickness
(1) Emission - Absorption
(1b) Emission - Absorption (Variance Reduction)
(2) Net Flux Divergence
(3) HYBRID
15 / 26
3D vs. 1D Thermal Radiative Transfer
16 / 26
Neighboring Column Approximation NCA
MYSTIC NCA ICA
17 / 26
Neighboring Column Approximation NCA
MYSTIC NCA ICA
17 / 26
Neighboring Column Approximation NCA
MYSTIC NCA
18 / 26
Neighboring Column Approximation NCA
MYSTIC NCA
18 / 26
Neighboring Column Approximation NCA
MYSTIC NCA
18 / 26
Neighboring Column Approximation NCA
MYSTIC NCA
18 / 26
NCA vs. 1D - NCA vs. 3D MYSTIC
19 / 26
NCA vs. 1D - NCA vs. 3D MYSTIC
19 / 26
Couple Paramterizations to Atmospheric Model
TenStream solver and NCA coupled to the UCLA-LES
I use LES to model atmospheric flow with resolutions from
10 m to 1 km
I consider dynamics, turbulence, microphysics and radiation
I TenStream solver increases total model runtime by factor 3-5
I NCA only factor 1.5-2 more expensive
20 / 26
Cloud Evolution With 3D Thermal Radiation
Shallow Cumulus Cloud Field
21 / 26
Cloud Evolution With 3D Thermal Radiation
Shallow Cumulus Cloud Field
21 / 26
Cloud Evolution With 3D Thermal Radiation
Single CloudNo Radiation 1D Thermal ICA 3D Thermal NCA
22 / 26
Cloud Evolution With 3D Thermal Radiaton
23 / 26
Flying through a Cloud Field
24 / 26
Current state and a glimpse at whats to come..
Conclusions
I Development of 2 parallel solvers
I Implemented in UCLA-LES
I Unprecedented large and extensive simulations
Outlook
I Further investigation of simulations (cloud organization)
I RCE simulations
I Get ready for large scale computations in ICON – HD(CP)2-Project
I Systematic analysis of 3D effects and impact on large scale flow –
W 2W -Project
25 / 26
Thank you!
Carolin Klinger and Bernhard Mayer, 2014. Three-dimensional Monte Carlo calculation of atmospheric thermal heating rates (JQSRT)
Fabian Jakub and Bernhard Mayer, 2015. A three-dimensional parallel radiative transfer model for atmospheric heating rates for use incloud resolving models – The TenStream solver (JQSRT)
Carolin Klinger and Bernhard Mayer, 2015. The Neighboring Column Approximation (NCA) - A fast approach for the calculation of 3Dthermal heating rates in cloud resolving models (JQSRT)
Fabian Jakub and Bernhard Mayer, 2015. 3-D radiative transfer in large-eddy simulations - experiences coupling the TenStream solver tothe UCLA-LES (GMDD)
26 / 26