Robin Hogan ECMWF and University of Reading Contributions from Sophia Schäfer Christine Chiu University of Reading Carolin Klinger Bernhard Mayer LMU Munich Susanne Crewell ID: 555869
Download Presentation The PPT/PDF document "Representing 3D radiative effects in wea..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Representing 3D radiative effects in weather and climate models
Robin Hogan
ECMWF and University of Reading
Contributions from:
Sophia
Schäfer
, Christine
Chiu (University of Reading)
Carolin
Klinger, Bernhard
Mayer (LMU Munich)
Susanne
Crewell
(University of Cologne)
Maike
Ahlgrimm
(ECMWF)Slide2
Motivation and overview
No current radiation scheme that represents all 3D effects in shortwave and
longwave
is fast enough to use in weather and climate models
Therefore, we have no reliable estimates of their impact on fluxes globally, or the impact on temperature and other model variables
Solar energy industry requires forecasts of direct and diffuse solar radiation separately, which can be
significantly biased
when 3D effects are neglected
This talk introduces a radiation scheme that can fill this gap
Matrix-exponential method for solving a new form of two-stream equations
How important are
longwave
3D effects?
Observational evidence for 3D effects in direct/diffuse ratio
What is the effective cloud edge length for 3D radiation?
OutlookSlide3
Shortwave 3D radiative effects
Useful to consider mechanisms for 3D effects (
Varnai
and Davies
1999)The two main mechanisms give errors of opposite sign (Hogan & Shonk 2013):
Side Illumination
Direct-beam effect
Can be captured to some extent by modifying cloud overlap (Tompkins &
DiGiuseppe 2007)
Side escape
Diffuse effect
Cannot be captured by modifying cloud overlapSlide4
Longwave 3D radiative effects
Very little literature; most radiation people seem to assume it’s negligible
Heidinger
& Cox (1995) estimated 30% increase in surface
cloud forcing at 11
m
Thought experiment: consider a cubic isothermal optically thick cloud in vacuumEach face emits same, and half of radiation from horizontal faces goes downTherefore surface downwelling radiation must be three times what would be calculated neglecting 3D effects
What about more realistic clouds with absorption by atmospheric gases?Slide5
SPeedy
Algorithm for
Radiative
TrAnsfer
through
CloUd Sides
SPARTACUS!Slide6
SPARTACUS approach
Follows from Hogan &
Shonk
(2013), but:
Represents
longwave
3D effects, including longwave scatteringRepresents cloud inhomogeneity using “Tripleclouds” approachA
more elegant and accurate solver using matrix exponentialsClouds represented by three regions at each height Sufficient to represent cloud structure (Shonk & Hogan 2008)Extra terms added to two-stream equations:
a
a
c
b
c
b
a
Source terms
Shortwave: direct solar beam
Longwave
: Planck function
New terms
Exchange between regions
Hogan &
Shonk
provided formulas for
f
xy
in terms of
cloud edge length
a
u
a
v
a
u
b
v
bSlide7
Matrix solution in a single layer (shortwave)
Define diffuse upwelling, diffuse
downwelling
and direct
downwelling
as vectors u, v and s:
Write two-stream equations as: where 9x9 matrix is composed of known terms analogous to g1-g4 in the standard two-stream equations: (coupled linear homogeneous ODEs)Solution for layer of thickness z
1:
Matrix exponential
Waterman (1981),
Flatau
& Stephens (1998)
Can compute using
Padé
approximant plus scaling & squaring method (Higham 2005)Slide8
Reflection and
t
ransmission matrices
We want relationships between fluxes of the form:
Transmission matrix for 2 regions given by
and likewise for
R and S±
If matrix exponential is decomposed as: then reflection and transmission matrices given by:For scalars, we get same answer as Meador & Weaver (1980) formulas For speed, only use matrix exponential for partially cloudy layers
u
(0)
u
(
z
1
)
v
(0)
s
(0)Slide9
Extension to multiple layers: the adding method
The adding method (e.g.
Lacis
and Hansen 1974) can be used to combine the reflectance and transmittance matrices of pairs of layers
In
N-stream radiative transfer (e.g. N
=16), the elements of the flux vector would represent different streams, but the method works just as well for different regionsWe work up from the surface and compute the albedo of the whole atmosphere below each half-levelAlbedo
After this we can head back down again to compute the fluxes
For one region, this is exactly the same as solving a tridiagonal system with forward elimination followed by backsubstitution
A
aa
A
bb
A
ba
A
abSlide10
How do we deal with cloud overlap?
Following Edwards-
Slingo
code method: overlap matrices
Downward overlap
(similarly for upward overlap
U
)Matrix elements calculated from a decorrelation length following Shonk et al. (2010)Albedo just above a half level (
A) is related to albedo just below a half level (B) by A=UBVTwo-stream equations now look like this:
V
aa
V
ba
V
bb
V
ab
Half-level
Extension to
longwaveSlide11
Broadband shortwave SPARTACUS vs MYSTIC (I3RC case 4)
SPARTACUS coded up in Fortran 90 with RRTM-G for gas absorption
Use “
ellipsified
” cloud edge length (see later)
Compare to full 3D Monte Carlo calculation from MYSTIC in cumulus
Mean of 4 solar azimuths, error bar indicates standard deviation due to sun orientation
Good match!3D effect up to 20 W m-2, similar to inhomogeneity effectLarge difference in direct surface flux at large solar zenith angleSlide12
3D effects in observations of direct/total downwelling flux
Troccoli
&
Morcrette
(2014
) reported biases in ECMWF direct solar radiation from, important
for solar energy industryBin observations and model by solar zenith angle and cloud fraction, considering only cases of boundary-layer clouds:
Next step: apply new 3D radiation scheme to the ECMWF cloud fields to verify that differences are due to 3D effectsECMWF modelARM SGP (13 yrs
)Slide13
What about cloud edge length?
SPARTACUS takes
cloud edge length per unit area of
gridbox
as inputWill need to be parameterized in the GCM as an effective cloud sizeE.g. use shallow/deep cumulus schemes to diagnose when clouds with strongest 3D effect are presentFor cubes, longwave
SPARTACUS matches SHDOM/MYSTIC well, but not for realistic cloudsHypotheses:Small-scale structure of a cloud does not matter for radiation; the effective edge length is that of an ellipse with the same area and aspect ratio Clouds tend to cluster, but SPARTACUS assumes random distribution
Horizontal cross section through a cloud
“
Ellipsified
” cloudSlide14
Four experiments manipulating the I3RC cumulus field…
Isolated cloud Original
Original “
Ellipsified
” cloudsSlide15
Longwave downwelling flux: SPARTACUS versus SHDOM
Excellent match with ICA, but SPARTACUS overestimates 3D effect
SPARTACUS overestimation is removed for isolated,
ellipsified
clouds
Parameterization will need to account for clustering and effective edge length
Independent column approx 3D radiationSlide16
Longwave presents additional challenges
To compute exchange between cloud and clear-sky, shortwave SPARTACUS assumes cloud-edge flux equal to in-cloud mean flux
In optically thick clouds, scattering
reduces
emitted flux below black-body value (emissivity effect)
In optically thin clouds, lateral flux
builds up
towards cloud edge
F
Using thought experiment for a cube, we have parameterized these effects
Parameterization strictly only applicable for clouds with an aspect ratio of around 1Slide17
SPARTACUS uses ellipsified edge length but no proximity/lateral effects yet
3D effects increase surface CRF by 29% in MYSTIC and 36% in SPARTACUS
Also differences in 1D calculations that need to be investigated
Surface 3D effect of 4 W m-2 smaller than shortwave maximum
Partly just because cumulus
clouds have
smaller CRF in longwave than shortwaveBut constant over diurnal cycle so might integrate to a larger effect?
Broadband longwave SPARTACUS vs MYSTIC (I3RC case)Down
UpSlide18
Summary
SPARTACUS is a promising method for representing 3D effects efficiently in a GCM radiation scheme
Radiatively
effective cloud edge length is approximately equal to perimeter of a fitted ellipse, although cloud clustering is important as well
Longwave
3D effects systematically increase CRF and shouldn’t be neglected Incorporate
longwave parameterizationsImplement online in the ECMWF modelHow can we parameterize cloud edge length from model fields?What is the impact of 3D radiation on global fluxes and temperatures?N
ext stepsSlide19Slide20
Shortwave results
Coded up in Fortran 90 with RRTM-G for gas absorption
Good match with cumulus case of
Pincus
et al. (
2005): cloud cover 0.22, edge length calculated from aspect ratio of 0.73D effect similar size to inhomogeneity effectLarge difference in direct surface flux at large solar zenith angleSlide21
Impact on flux and heating rate profiles
Longwave
Shortwave,
q
0
=70°
Heating rate
Downwelling Upwelling