On models of stochastic condensation in clouds
Gustavo Abade
University of Warsaw · Institute of Geophysics · Faculty of Physics
Atmospheric Physics Seminar March 5, 2021
condensation cloud
condensation level
coagulation Microphysical processes
Cloud condensation nuclei (CCN)
condensation cloud
condensation level
coagulation Microphysical processes
Cloud condensation nuclei (CCN)
Narrow size distribution - stable cloud
r
size distribution
Broad size distribution - unstable cloud
r
size distribution
Narrow size distribution - stable cloud
r
size distribution
Broad size distribution - unstable cloud
r
size distribution
Which is the most likely distribution?
Maximum entropy principle
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Which is the most likely distribution?
Maximum entropy principle
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Which is the most likely distribution?
Maximum entropy principle
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Closed cloud parcel
Given:
I N - number of droplets
I M - total mass of liquid water
Which is the most likely distribution f (r)?
Closed cloud parcel Given:
I N - number of droplets
I M - total mass of liquid water
Which is the most likely distribution f (r)?
Closed cloud parcel Given:
I N - number of droplets
I M - total mass of liquid water
Which is the most likely distribution f (r)?
Closed cloud parcel The most likely f maximizes the spectral entropy:
H =− Z
f (x)[ln f (x)] dx
x - droplet mass
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Closed cloud parcel The most likely f maximizes the spectral entropy:
H =− Z
f (x)[ln f (x)] dx
x - droplet mass
+ constraints
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Closed cloud parcel The most likely f maximizes the spectral entropy:
H =− Z
f (x)[ln f (x)] dx
x - droplet mass
f (x)dx = f (r)dr
Liu, Atmos. Res., 12 (1995); Yano, J.-I., JAS, 76 (2019)
Weibull distribution
Closed cloud parcel f (r)∼ r2exp(−βr3)
r f
Liu, Atmos. Res., 12 (1995)
Which is the least likely distribution?
Maximum energy principle
Which is the least likely distribution?
Maximum energy principle
Closed cloud parcel The least likely f maximizes the populational energy:
E = Elatent+ Esurface+ . . .
I constraints
Liu, Atmos. Res., 12 (1995)
Closed cloud parcel The least likely f maximizes the populational energy:
E = Elatent+ Esurface+ . . .
I constraints
Liu, Atmos. Res., 12 (1995)
Monodisperse cloud
Closed cloud parcel f (r)∼ δ(r − hri)
f
hri r
Liu, Atmos. Res., 12 (1995)
Most likely
r f
Least likely
f
hri r
Many issues: - “no dynamics”
- equilibrium×non-equilibrium - closed system× open system
Most likely
r f
Least likely
f
hri r
Many issues:
- “no dynamics”
- equilibrium×non-equilibrium - closed system× open system
Most likely
r f
Least likely
f
hri r
Many issues:
- “no dynamics”
- equilibrium×non-equilibrium - closed system× open system
Most likely
r f
Least likely
f
hri r
Many issues:
- “no dynamics”
- equilibrium×non-equilibrium
- closed system× open system
Most likely
r f
Least likely
f
hri r
Many issues:
- “no dynamics”
- equilibrium×non-equilibrium - closed system× open system
Let us describe the condensation process
Diffusional growth
Droplet growth dr dt = 1
r DhSi
hSi - mean-field supersaturation
LES grid box
r
water droplet moist air
∆∈ inertial range
Equation for f (r, t)
∂f
∂t =− ∂
∂r(˙rf ) +· · ·
Advection in radius space with velocity
˙r = DhSi r
Narrow size distribution!
t = 0
t = 103 s
f
r
Equation for f (r, t)
∂f
∂t =− ∂
∂r(˙rf ) +· · ·
Advection in radius space with velocity
˙r = DhSi r
Narrow size distribution!
t = 0
t = 103 s
f
r
Equation for f (r, t)
∂f
∂t =− ∂
∂r(˙rf ) +· · ·
Advection in radius space with velocity
˙r = DhSi r
Narrow size distribution!
t = 0
t = 103 s
f
r
Eulerian stochastic model
Numerical simulation
Lasher-Trapp et al., QJRMS, 131 (2005)
Kinetic equation for f (r; x, t)
∂f
∂t +∇ · (uf)=−∂
∂r( ˙rf ) +· · · ˙r = DS r
x ∆
Kinetic equation for f (r; x, t)
∂f
∂t +∇ · (uf)=−∂
∂r( ˙rf ) +· · · ˙r = DS r
x ∆
Kinetic equation for f (r; x, t)
∂f
∂t +∇ · (uf)=−∂
∂r( ˙rf ) +· · · ˙r = DS r
x ∆
dependent
variable = mean + fluctuation
Kinetic equation for f (r; x, t)
∂f
∂t +∇ · (uf)=−∂
∂r( ˙rf ) +· · · ˙r = DS r
x ∆
f =hfi + f0
u =hui + u0
˙r =h ˙ri + ˙r0
S =hSi + S0
∂hfi
∂t +∇ · [huihfi] = − ∂
∂r[h ˙rihfi]
+ ∇ ·hu0f0i+ ∂
∂rh ˙r0f0i+· · ·
Mean growth rate (narrowing):
h ˙ri = DhSi r
Turbulent effect (broadening):
hu0f0i = ? h ˙r0f0i = ?
∂hfi
∂t +∇ · [huihfi] = − ∂
∂r[h ˙rihfi]
+ ∇ ·hu0f0i+ ∂
∂rh ˙r0f0i+· · · Mean growth rate (narrowing):
h ˙ri = DhSi r
Turbulent effect (broadening):
hu0f0i = ? h ˙r0f0i = ?
∂hfi
∂t +∇ · [huihfi] = − ∂
∂r[h ˙rihfi]
+ ∇ ·hu0f0i+ ∂
∂rh ˙r0f0i+· · · Mean growth rate (narrowing):
h ˙ri = DhSi r
Turbulent effect (broadening):
hu0f0i = ? h ˙r0f0i = ?
Slow microphysics hu0f0i = −K∇hfi
K− turbulent diffusivity
Cannot explain the observed spectrum broadening
Buikov, M. V. (1960’s)
Slow microphysics hu0f0i = −K∇hfi
K− turbulent diffusivity
Cannot explain the observed spectrum broadening
Buikov, M. V. (1960’s)
Slow microphysics hu0f0i = −K∇hfi
K− turbulent diffusivity
Cannot explain the observed spectrum broadening
Buikov, M. V. (1960’s)
Fast microphysics
hu0f0i = −K ∇hfi + ?
h ˙r0f0i = ? + ?
supersaturation
fluctuations ↔ vertical velocity fluctuations
Fast microphysics
hu0f0i = −K ∇hfi + ?
h ˙r0f0i = ? + ?
supersaturation
fluctuations ↔ vertical velocity fluctuations
S(t) w(t)
τ∼ 1/nhri
dS dt =−1
τ [S− Seq(t)] Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
S(t) w(t)
τ∼ 1/nhri
dS dt =−S
τ + aw(t)
τ − phase relaxation time
dS dt =−1
τ [S− Seq(t)] Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
S(t) w(t)
τ∼ 1/nhri
dS dt =−1
τ [S− Seq(t)]
Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
S(t) w(t)
τ∼ 1/nhri
dS dt =−1
τ [S− Seq(t)]
Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
S(t) w(t)
τ∼ 1/nhri
dS dt =−1
τ [S− Seq(t)]
Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
S(t) w(t)
τ∼ 1/nhri
dS dt =−1
τ [S− Seq(t)]
Seq(t) = aw(t)τ
w(t) S(t)
Slow microphysics (τ big) hS0w0i = 0
Fast microphysics (τ small) S(t)∼ w(t)
Condensation reversibility
S(t)∼ w(t)
B A
h
˙ m∼dr3
dt = C1w(t) m(z = h)− m(z = 0) = C2h
A. Khain et al., Atmos. Res., 55 (2000)
Condensation reversibility
S(t)∼ w(t)
B A
h
˙ m∼dr3
dt = C1w(t) m(z = h)− m(z = 0) = C2h
A. Khain et al., Atmos. Res., 55 (2000)
r
Growth rate
˙r = DhSi r ∼ 1
r
Mass rate
˙ m∼dr3
dt ∼ r2 ˙r∼ r
Supersaturation absorption is faster for larger droplets
r
Effective microscale supersaturation
Seff≈hSi hri−1r
˙reff= DSeff
r ≈DhSi hri
Fluctuation in growth rate
˙r0 ≈DS0 hri
Suppress the growth of smaller droplets
r
Effective microscale supersaturation
Seff≈hSi hri−1r
˙reff= DSeff
r ≈DhSi hri Fluctuation in growth rate
˙r0 ≈DS0 hri
Suppress the growth of smaller droplets
Direct Numerical Simulations
S0= S− hSi
r0 = r− hri
Paoli and Sharif, JAS, 66 (2009)
hu0f0i = − K∇hfi− K1∂rhfi
h ˙r0f0i = − K2∇hfi − K3∂rhfi
Ki− effective diffusion coefficients
hu0f0i = − K∇hfi− K1∂rhfi
h ˙r0f0i = − K2∇hfi − K3∂rhfi
Ki− effective diffusion coefficients
hu0f0i = − K∇hfi− K1∂rhfi
h ˙r0f0i = − K2∇hfi − K3∂rhfi
Ki− effective diffusion coefficients
Gamma distribution
f (r)∼ rpexp(−βr)
I power law p∼ 5 − 10 I exponential tail I only one mode!
Khvorostyanov and Curry, JAS, 66 (1999)
Gamma distribution
f (r)∼ rpexp(−βr)
I power law p∼ 5 − 10
I exponential tail I only one mode!
Khvorostyanov and Curry, JAS, 66 (1999)
Gamma distribution
f (r)∼ rpexp(−βr)
I power law p∼ 5 − 10 I exponential tail
I only one mode!
Khvorostyanov and Curry, JAS, 66 (1999)
Gamma distribution
f (r)∼ rpexp(−βr)
I power law p∼ 5 − 10 I exponential tail I only one mode!
Khvorostyanov and Curry, JAS, 66 (1999)
Lagrangian stochastic model
Lagrangian model
hW i Wi′
Si
T and pdecreasing
dSi0 dt =−Si0
τc − Si0
τm+ aWi0(t)
τc∼ 1
Nhri (condensation)
τm∼ eddy turnover time (mixing)
I W0(t): prescribed stochastic process.
Celani et al., EPL, 70 (2005); Grabowski and Abade, JAS, 74 (2017)
Lagrangian model
hW i Wi′
Si
T and pdecreasing
dSi0 dt =−Si0
τc − Si0
τm+ aWi0(t)
τc∼ 1
Nhri (condensation)
τm∼ eddy turnover time (mixing)
I W0(t): prescribed stochastic process.
Celani et al., EPL, 70 (2005); Grabowski and Abade, JAS, 74 (2017)
Kinematic framework Synthetic turbulent-like flow
Turbulent-like flow
u = (u, w) = ∂ψ
∂z,−∂ψ
∂x
ψ(r, t) =X
random harmonics
0 200 400 600 800 1000 1200
0 500 1000 1500 2000 2500
z [m]
x [m]
hw2i =σw2(z) hw(x0, z)w(x0+ x, z)i =Cˆw(x)σ2w(z)
Pinsky et al., JAS, 65 (2008), ..., Magaritz-Ronen et al., ACP, 16 (2016)
Vertical profiles
Size distribution at different heights
I broader distribution in the presence of turbulence
I bimodal distributions
—no turbulence —turbulence (model 1) — turbulence (model 2)
Size distribution at different heights
I broader distribution in the presence of turbulence
I bimodal distributions
—no turbulence —turbulence (model 1) — turbulence (model 2)
Size distribution at different heights
I broader distribution in the presence of turbulence
I bimodal distributions
—no turbulence —turbulence (model 1) — turbulence (model 2)
Size distribution at different heights
—no turbulence —turbulence (model 1) — turbulence (model 2)
Summary
I Problem of narrow × broadsize distributions in clouds
I Subgrid fluctuations in Eulerianmodels are tricky
I Lagrangian approach is more natural
THANK YOU!
Summary
I Problem of narrow × broadsize distributions in clouds
I Subgrid fluctuations in Eulerianmodels are tricky
I Lagrangian approach is more natural
THANK YOU!
Summary
I Problem of narrow × broadsize distributions in clouds
I Subgrid fluctuations in Eulerianmodels are tricky
I Lagrangian approach is more natural
THANK YOU!
Summary
I Problem of narrow × broadsize distributions in clouds
I Subgrid fluctuations in Eulerianmodels are tricky
I Lagrangian approach is more natural
THANK YOU!