representing cloud and precipitation in nwp models … · 2016-01-23 · representing cloud and...
TRANSCRIPT
.
REPRESENTING CLOUD AND PRECIPITATION
IN
NWP MODELS IN CANADA
(Peter) M.K. Yau1 and Jason Milbrandt2
1McGill University, Montreal, Canada2Environment Canada [RPN], Dorval, Canada
Environment Canada's forecast model
GEM (Global Environmental Multiscale)
Global Uniform Global Variable Limited Area (LAM)
Grid configurations:
• medium-range (10-d)
• x = 35 km → 25 km
• t = 15 min
• short-range (48-h)
• x = 15 km → 10 km
• t = 7.5 min
• experimental
• short-range (24-h)
• x = 2.5 km → 1 km
• t=1 min (t=30s)
Simple Cloud SchemeDetailed Microphysics
Scheme
The simple cloud scheme (Sundqvist)
• Cloud-cover fraction is diagnosed (function of RH)
• Condensation occurs when RH exceeds a threshold (80% near
surface)
• Total condensate (cloud water/ice) is prognostic (advected)
• Precipitation falls instantly to the ground – there is no
advection of precipitation
Global Uniform Global Variable
Multi-moment scheme
Milbrandt and Yau (JAS 2005 a,b)
Milbrandt and Yau (JAS, 2006 a,b)
Gultepe and Milbrandt
(Pure App. Geoph.,2007)
Milbrandt et al. (MWR, 2008)
Milbrandt et al. (MWR, 2010)
Dawson et al. (MWR, 2010)
Six hydrometeor categories:
2 liquid: cloud, rain
4 frozen: ice, snow, graupel, hail
Scheme implemented in
GEM-LAM, Global variable (Canada)
ARPS (U Oklahoma, US)
WRF 3.2 (US)
RAIN
GRAUPEL HAIL
SEDIMENTATIONSEDIMENTATION
VAPOR
ICECLOUD
VDvr VDvs
NUvi,
VDvi
CLci, MLic, FZci
CLcs
CNig CNis,
CLis
CLri
CLih
CLsh
CLir-g
CLsr-h
CLir-g
CLsr-g
CLch
CNsg
CNgh
MLgr
CLcg
VDvg
CLir
VDvh
self-
collection
self-
collection
CLrh,
MLhr,SHhr
NUvc,
VDvc
CNcr,
CLcr
CLsr CLrs
MLsr, CLsrSNOW
The detailed microphysics scheme
Limited Area Model
•Overview of the scheme
•Testing and improvement in IMPROVE-2
(GEM-LAM)
•Forecast in winter Olympics 2010
(GEM-LAM)
•Testing over Arctic (GEM-Global Variable)
ANAYLTICAL FUNCTION
BULK METHOD
Representing the size spectrum
N(D)
D [ m]
100
[m-3 m-1]
20 40 60 800
101
100
10-1
10-2
1 m3
Gamma Distribution Function:
DeDNDN 0)(
* Q = r q (mass content)
INCREASING
VALUES(of , N0 and )
log
N(D
)
log N
(D)
log N
(D)
D [mm] D [mm]D [mm]
Varying (slope) (N0 and constant)
Varying (shape) (Q* and N0 constant)
Varying N0 (intercept) ( and constant)
BULK METHOD
pth moment:(
xp
x
xxx
p
x
pNdDDNDpM
10
0
1)()(
Size Distribution Function:
D
xxxx eDNDN
0)(
Total number concentration, NTx
)0()(0
xxTx MdDDNN
Radar reflectivity factor, Zx
)6()(0
6
xxx MdDDNDZ
Mass mixing ratio, qx
densityairDcDmwhere
Mc
dDDNDc
q
xx
xx
xx
x
r
rr
,)(
),3()(
3
0
3
Predict evolution of
specific moment(s)
e.g. qx, NTx, ...
Implies prediction of evolution
of parameters
i.e. N0x, x, ...
BULK METHOD
pth moment:(
xp
x
xxx
p
x
pNdDDNDpM
10
0
1)()(
Size Distribution Function:
D
xxxx eDNDN
0)(
Predict evolution of
specific moment(s)
e.g. qx, NTx, ...
Implies prediction of evolution
of parameters
i.e. N0x, x, ...
For every predicted moment, there
is one prognostic parameter.
The remaining parameters are
prescribed or diagnosed.
Two-moment scheme:
qx and NTx are predicted;
x and N0x are prognosed;
(x is specified)
Three-moment scheme:
qx, NTx and Zx are predicted;
x, N0x and x is prognosed
One-moment scheme:
qx is predicted;
x is prognosed
(N0x and x are specified)
e.g.
CLOSURE OF SYSTEM
( densityairandcDDmwhere
Gq
ZNc T
r
r
,
,)1)(2)(3(
)4)(5)(6()(
)(
3
2
2
Solve for shape parameter α from
3
1
)1(
)4(
r
q
cNT
Solve for slope parameter λ from
)1(
1
0
TN
N
Solve for intercept parameter N0 from
→ NT and q vary monotonically in a 1-moment scheme
Diagnostic closure for α in 2-
moment scheme
)(
,3
1
m
T
m
Df
cN
qD
r
r
r = M (p,αest) / M(p,αcorr)
Verification and
improvement of Multi-
moment scheme in GEM-
LAM (1 km) in IMPROVE-2
CASE STUDY
November-December 2001: IMPROVE-2 Observational Campaign
Improvement of Microphysical Parameterization through
Observational Verification Experiment
Cresswell
Sounding
150 km
13-14 Dec 2001 case:
• chosen for study at W.M.O. International Cloud Modeling Workshop,
Hamburg (July 2004)
• special issue of J. Atmos. Sci. (October 2005) dedicated to IMPROVE-2
GOES – IR: 2239 UTC 13 Dec 2001
CASE STUDY
Precipitation in IOP region:
• prefrontal showers;
• moderate to heavy stratiform rain
(associated with mid-level baroclinic
zone);
• surface frontal rain-band;
• transition to sporadic showers
Characteristics:
• large-scale baroclinic system
• strong low-level cross-barrier flow
OBSERVED PRECIPITATION1600 UTC 13 Dec – 0800 UTC (18 h)
BIAS SCORES4-km MM5 Simulation
UNDER-Predicted
OVER-Predicted
CASE STUDY: MM5 Simulations
Source: Garvert et al. (2005a) [J. Atmos. Sci.]
13-14 Dec 2001 case:
• MM5 runs at 4-km and 1.3 km exhibited errors in surface precipitation
attributed to problems associated with the microphysics (SM Reisner-2)
S-Pol
1.5 PPI
R 150 km
Portland
0.5 PPI
R 200 km
4km-GEM
700 hPa
1-km GEM
E. Reflectivity
No pronounced over prediction along lee side of Cascade
Source: Stoelinga et al. (2003) [Bull. Amer. Meteor. Soc.]
MICROPHYSICS: Observations
Aircraft flight tracks (2200 – 0200 UTC)
Convair-580
NOAA P-3
MICROPHYSICS: Observations
Source: Wood et al. (2005) [J. Atmos. Sci.]
NO
AA
P-3
Co
nva
ir-5
80
Mean size inc. with dec. height
Z>4.5 km - single ice xtal
3<Z<4 km – dendrite
2<Z<3 km – column &
aggregate
13-14 Dec 2001 CaseA B
MICROPHYSICS: Observations
Source: Garvert et al. (2005b) [J. Atmos. Sci.]
Combined Observations for 2200–0200 UTC
Cloud liquid water along P-3 flight legs
Under prediction of vertical
extent of cloud water
Ice/snow content along Corvair flight legs
Over prediction of concentration of snow
mass
→ too large deposition and/or riming
IMPROVEMENTS OF SNOW CATEGORY
•Diffusional growth
•Growth by riming
“The electrostatic analogy of the capacitance theory of
ice crystal growth is highly flawed and does not produce
the observed growth rates of ice crystals.
It severely overpredicts the growth rates in almost all
cases [by a factor of 3 to 8+ for plates and 2 to 4 for columns]
involving even simple hexagonal shapes.”
Bailey and Hallet (2006)
Electrostatic Analogy for
Diffusional Growth of Ice Crystals
i
i
AB
SC
dt
dm )1(4
i
i
AB
SC
dt
dm )1( 4
Add CORRECTION FACTOR to DIFFUSIONAL GROWTH EQUATION
i
icorr
AB
SfC
dt
dm )1( 4
where fcorr must be < 1, with value justified by results
fcorr = 0.50
With decreasing fcorr,
SNOW content (qs) is reduced
and
CLOUD LWC (qc) is increasedfcorr = 0.25
fcorr = 1.0Sensitivity Tests for
IMPROVE-2:qs
qc
g kg-1
g kg-1
g kg-1
C = 0.5D
C = 0.25D
C = 0.125D
Other evidence:
Field et al. (2008)
Westbrook et al. (2008)
• For the collection efficiency, Ecs = 1 is often assumed (for collection of cloud by snow)
• If Ecs < 1, the snow riming rate will be overestimated
( xyxxyyyxxyyyyxyyxxyx dDdDDNDNDDEDmDDDVDVCL )()(),()()()(
4
1 2
0 0
r
RIMING of SNOW
Stochastic collection equation: (for category x collecting category y)
COLLECTION
EFFICIENCY
Approximation:
• Works for Dc ~ 15-30 m,
and Ds ~ 150-1500 m
• Reduces riming rate 10-80%
(vs. Ecs = 1)
5.0
1000
1000,min(
30
)30,min(),(
m
mD
m
mDDDE sc
sccs
500
m
225
m
150
m
700
m
950
m
Ds
0.5Dc
Ec
s(D
c,D
s)
17
00
m
RIMING of SNOW
*Wang and Ji, 1992
Test of 2-moment microphysics in
Vancouver Olympics 2010 in 1 km
GEM-LAM
1.0 km
Whistler
• 3 nested LAM integrations twice daily
from 0000 and 1200 UTC GEM-Regional
forecasts:
LAM-15 km → 2.5 km → 1 km
Vancouver
15 km
2.5 km
Nesting strategy for
LAM-V10 system
Verification for
LAM-V10
Olympic Autostation Network (OAN):
• approx. 40 standard and special surface observing
sites (hourly or synop available on GTS)
• large number (relatively) of surface stations
• concentrated in small region
Verification Examples
Observations courtesy of George Isaac
SNOW
PELLETS
FLUFFY
SNOWFLAKES
Observed:*
*Forecaster:
Michael Gélinas
Experimental field:Solid-to-Liquid ratio
Testing of 2-moment microphysics in
Global GEM variable 15 km over the
Arctic
30 day simulation – July 2008 over Arctic
Polar-GEM:
•x = 15 km
Sundqvist
Two-MomentOne-Moment
GPCP merged obs
PRECIPITATION
GPCPmerged
obs
Sundqvist
Two-Moment
One-Moment
Cloud Scheme:
SENSITIVITY TO TIME STEP
t = 450s
t = 225s
t = 120s
t = 60s
Sundqvist
Two-Moment
One-Moment
Cloud Scheme: GPCPmerged
obs
60-h Simulation
Sundqvist
Two-Moment
One-Moment
Cloud Scheme:
GPCPmerged
obs
SENSITIVITY TO TIME STEP
60-h Simulation (t = 60 s)
SUMMARY
1) Multi-moment mixed phase bulk cloud microphysical
schemes have been developed and implemented in
GEM-LAM and GEM-Global Variable
2) Comparison with in-situ field measurements allows
improvements in the scheme
3) Implementation in GEM-Global Uniform is planned but
still needs work to address
a) time splitting for microphysics
b) subgrid scale cloud fraction
c) simplification to allow for a mixture of
higher and lower moment hydrometeor
categories
( z
Vq
t
q xqx
SEDI
x
rr
SEDIMENTATION: Bulk scheme
xqV = mass-weighted fall velocity
SM
( z
VN
t
N xNx
SEDI
x
xNV = number-weighted fall velocity
DM
( z
VZ
t
Z xZx
SEDI
x
xZV = reflectivity-weighted fall velocity
TM
For a given size distribution, xNxqxZ VVV
ANA
TM
DM0SM
Effects on sedimentation terms
(Q = r q)
z
[km]
Q [g m-3]
TM better than DM0 better than SM
DIFFERENCE RELATED TO SIZE SORTING
Disadvantages of 1-moment scheme
a) Inconsistency in modeling physical processes
From closure relation, NT and q vary monotonically → NT increases
or decreases with q, but
in breakup, NT increases but q = constant, and
in diffusional growth, q increases but NT = constant.
c) Inconsistency in modeling size sorting in sedimentation
→ mean size increases with decreasing height, but not necessarily
true in 1-moment as mean diameter is
3
1
T
mcN
qD
r
Disadvantages of 2-moment fixed α scheme in
sedimentation
Rate of change of Dmx (size sorting)
proportional to fallspeed ratio
From TRIPLE- MOMENT sedimentation profiles:
Diagnosed α → sedimentation results in larger mean size (larger Dm) but
narrower spectrum (larger α )
0
2)(
4dDDNDqE
dt
dqxb
cxc
CL
x r
...21
proc
x
proc
x
S
x
dt
dq
dt
dq
dt
dq
MICROPHYSICAL
PROCESSES
How well do the various bulk scheme
predict sources/sinks?
0
)()(
dDDNdt
Ddm
dt
dq
CLCL
x CONTINUOUS COLLECTION
OF CLOUD WATERe.g.
xb
cxccxc
CL
DqEqEDVD
dt
Ddm
22
4)(
4
)(r
r
pth moment:
0
)()( dDDNDpM x
p
x
)2( xx
CL
x bMdt
dq
xb
xx DaDV )(
How well do the various bulk scheme predict
sedimentation and sources/sinks?
TM and DIAG DM schemes
better than
SM AND FIXED DM schemes