An overview of cloud and precipitation microphysics and its parameterization in
models
Hugh Morrison
NCAR*Mesoscale and Microscale Meteorology Division, NESL
WRF Workshop, June 21, 2010
*National Center for Atmospheric Research is
sponsored by the National Science Foundation
Outline
• Background/introduction
• Basics of microphysics parameterization- Liquid schemes
- Ice microphysics and mixed-phase schemes
- Multi-moment schemes
• Microphysics and data assimilation
• General thoughts on use of microphysics
schemes
Cloud particles (microphysics) Individual clouds
0.5 mm
Mesoscale cloud systemsWeather systems
1000 km
“microphysics”
processes controlling formation of cloud
droplets and ice crystals, their growth and
fallout as precipitation
0.5 mm
“microphysics parameterization”
“parameterization”
grid-scale microphysics
Predicted temperature,
moisture, wind, etc.
latent heating,
drying/moistening
“macrophysics” schemes are often used in
larger-scale models (cloud fraction, PDF cloud
schemes) to drive microphysics
Microphysics plays a key role in cloud, climate and weather models
Stephens (2005)-Latent heating/cooling(condensation, evaporation, deposition,
sublimation, freezing, melting)
-Condensate loading (mass of the condensate carried by the flow)
-Precipitation(fallout of larger particles)
-Coupling with surface processes (moist downdrafts leading to surface-wind
gustiness, cloud shading)
-Radiative transfer(mostly mass for absorption/emission of LW, particle size also important
for SW)
-Cloud-aerosol-precipitation interactions(aerosol affect clouds: indirect aerosol effects, but clouds process aerosols
as well)
Microphysics schemes can be broadly
categorized into two types:
N(D)
Diameter (D)
N(D)
Diameter (D)
Detailed (bin) bulk
Representation of particle size distribution
Size distribution
assumed to follow
functional form
Size distribution
discretized into
bins
Bulk schemes predict one or more bulk quantities
(e.g., mixing ratio) and assume some functional form
for the particle size distribution, e.g., gamma
distribution:
n(D) = N0 Dm e-lD
If N0 and m are specified, then l can be obtained from
the predicted mixing ratio q:
Equations for
isometric particle
shapes
Liquid microphysics – Kessler (1969)
• Separate liquid into cloud water and rain
N(D)
N(D)
N(D)
Droplet mass distribution
evolution during rain
formation using a detailed bin
model.
Grabowski and Wang 2009
Time
N(D)
Diameter (D)
N(D)
N(D)
Accretion of cloud water by existing rain
Autoconversion of cloud water to form rain
Diffusional growth of cloud water
• Key liquid microphysical conversion processes
Liquid microphysics – Kessler (1969)
• Separate liquid into cloud water and rain
• Marhsall-Palmer distribution for rain
qr
Rain
(Prognostic)
qc
CloudWater
(Prognostic)
qv
Water Vapor
(Prognostic)
Sedimentation
Autoconversion/
Accretion
Evaporation/
Condensation
Evaporation
Marshall-Palmer (1948) rain drop distribution
- N0 = 8 x 106 m-4
- m = 0
N(D)
Log Particle Diameter (D)
N0
Slope l
Extension to ice phase
…subsequent studies extended the Kessler
approach to include ice (e.g., Koenig and Murray
1976; Lin et al. 1983; Rutledge and Hobbs 1984; Lord et al.
1984; Dudhia 1989)
Ice microphysical processes
• Diffusional growth/sublimation
• Aggregation (autoconversion, accretion)
• Collection of rain and cloud water (riming)
• Melting
• Freezing
• Ice particle initiation (nucleation)
• Sedimentation
Ice microphysics has important impacts on
dynamics and surface precipitation due to:
• slower fallspeed of snow compared to rain
• extra latent heating (cooling) due to freezing
(melting)
(e.g., Leary and Houze 1979; Lord et al. 1984; Fovell and Ogura 1988;
Zhang and Gao 1989; McCumber et al. 1991; Liu et al. 1997; McFarquhar
et al. 2006)
Example: Impact of ice microphysics on 2D
tropical squall line
Liu et al. 1997
Liquid only (Kessler)
Ice + Liquid (Koenig and
Murray)
However, there is strong
case dependence of effects!
However, ice microphysics is significantly more
complicated because of the wide variety of ice
particle characteristics…
Pristine ice crystals,
grown by diffusion of
water vapor
Snowflakes, grown by
aggregation
Pruppacher and Klett
Different types of ice (small ice, snow, graupel, hail,
etc.) are typically parameterized by partitioning ice
into different species whose characteristics (N0,
particle density, fallspeed) are determined a priori.
• “Effective density” of ice particles is typically
expressed by a mass-size relationship of the form:
m = aDb
• In many schemes, b = 3 (corresponding with
spheres) and a ~ 0.1 g cm-3 (snow), ~ 0.4 g cm-3
(graupel), or ~ 0.9 g cm-3 (hail).
• More recently, schemes have been developed that
assume b ~ 2 (Thompson et al. 2008; Milbrandt et
al. 2010; Morrison and Grabowski 2008), which is
closer to observationally- and theoretically-derived
values.
How ice is separated into different species
(cloud ice, snow, graupel, hail, etc.) can have a
large impact on simulations.
from Biggerstaff and Houze (1993)
3D mid-latitude squall line simulations
Morrison and Bryan (2010, in prep)
It seems likely that “optimal” parameterization settings in terms of number and type of ice species are case dependent. Even within a given species, there is large variability - in general the boundaries between different species are not obvious.
-
From A. Heymsfield
• Recent work has attempted to move away from the
paradigm of separating ice into different species w/
fixed characteristics, and instead allow particle type
to vary as a function of the rime and vapor deposition
ice mixing ratios or process rates, which are predicted
or diagnosed separately (Stoelenga et al. 2007;
Morrison and Grabowski 2008; Lin and Colle 2010).
-
• Vapor depositional
growth • Riming of crystal
interstices
• Vapor depositional growth
• Further growth by
riming and vapor
deposition
• Complete filling-in
of interstices with
rime
Stage 2: Partially-rimed crystalStage 1: Unrimed crystal Stage 3: Graupel
D
Morrison and Grabowski
2008
Multi-moment versus single-moment schemes
• Single-moment – predict mixing ratio only for each
species
• Multi-moment – predict additional quantities for
each species (number concentration, reflectivity)
Prediction of additional moments allows greater
flexibility in representing size distributions and
hence microphysical process rates.
N(D) = N0 Dm e-lD
• Prediction of 2nd moment (number concentration N)
allows N0 to vary with q and N, giving scheme more
flexibility (e.g., Koenig and Murray 1976; Ferrier 1994; Meyers et al.
1997; Seifert and Beheng 2001; Milbrand and Yau 2005; Morrison et al.
2005)
• Prediction of 3rd moment (reflectivity Z) allows N0
and m to vary with q, N, and Z (e.g., Milbrandt and Yau 2005;
Gilmore and Straka 2009)
Key impacts of single vs. double-moment:
• Sedimentation (treatment of size sorting)
• Evaporation of rain - 2-moment schemes have a more
flexible treatment of rain drop mean size
Example: Impact of single vs. double-moment
on idealized 2D squall lineMorrison et al. 2009
2-moment
1-moment
Precip
Reflectivity
t = 6 hr
Example: Impact of single vs. double-moment
on idealized 2D squall line
Small N0, low evaporation rate in 2-
moment simulationWeaker cold pool in 2-moment
simulation
Spatial structure of N0 in 2-moment scheme is consistent with observations
(e.g., Waldvogel 1974; Tokday and Short 1996).
Morrison et al. 2009
• Other parameters also impact rain drop size
distribution and hence evaporation rate (rain
drop breakup, rain size distribution width or
shape, etc.).
Example: parameterization of rain drop
breakup in simulations of tornadic
supercell thunderstorms, Dx = 1 km
Morrison and Milbrandt (2010)
Microphysics and data assimilation
• Assimilation of radar reflectivity, satellite radiances,
etc.- Requires reasonable level of complexity of microphysics for
forward operator to “correctly” partition increments
• Issues with development of tangent linear and
adjoint models for microphysics
- Nonlinearity of microphysics (e.g., rain evaporation)
- Complex interaction (e.g., 3-species interaction)
- Conditionals/branches (e.g., autoconversion thresholds)
Microphysical parameter estimation using 4DVAR
(e.g., Zhu and Navon 1999) or ensemble Kalman filter
(e.g., Tong and Xue 2008)
From J. Sun et al.)
Key uncertainties – microphysical parameters
• Depends on type of microphysics scheme (number of
species and moments)
- One-moment schemes – N0 (but predicted in 2-moment
schemes…)
• Ice microphysics – density, fallspeed, etc.
• Conversion parameters – snow to graupel, liquid
and ice autoconversion, etc.
Case dependence of parameters??
Large uncertainties remain in our basic
understanding of the physical processes
of ice particles!
- Nucleation
- Particle shape (habit)
- Diffusional growth
- Aggregation, breakup
- Riming
How does one choose which type of microphysics scheme to use?
Many factors to consider:
Computational cost – number of species and predicted
moments is key (can be few % increase in run time with
each additional prognostic variable in WRF)
Appropriateness for application (e.g., real time forecasting
vs. research)
Appropriateness for case (liquid vs. mixed-phase, 3-
species ice vs. 2-species ice)
There has been a trend toward the use of more
complex microphysics schemes (i.e., more species
and more predicted moments) given:
- desire for better physical realism and representation of
microphysical processes
- need for more flexibility to cover a large number of
different applications
The development of more complex schemes has been
possible because of:
- increasing computer power
- improvements in understanding underlying physical
processes (theory and observations)
While more complex schemes improve physical
realism and are able to simulate microphysical
processes more realistically, they may not necessarily
lead to consistently better results (especially w/o
tuning).
- more realistic microphysics may expose other model deficiencies
(e.g., resolution, initialization/forcing, etc.)
- other physical parameterizations (e.g., PBL, radiation) have
often been developed and tuned to work with simpler
microphysics schemes
Differences are often as large or larger amongst
simulations using more complex schemes than
amongst those using simpler schemes, and while
complex schemes are useful as benchmarks for
testing simpler schemes, this should be done with
care.
- more degrees of freedom in complex schemes means divergence
of results is more likely
- key uncertainties related to the underlying physics often can’t be
addressed by increasing complexity of the scheme, and therefore
still require arbitrary or tuned parameter settings
Summary
• Microphysics parameterization is key in weather and climate
models because of interaction with dynamics, radiation,
aerosols/chemistry, etc.
• Microphysics parameterizations vary widely in complexity,
with a general trend toward use of more detailed multi-
moment and multi-species schemes.
• Numerous uncertainties remain in our understanding of the
basic physics, especially for the ice phase. Progress depends on
improvements in theory and especially observations (e.g., lab
studies).