Platzhalter für (Titel-) Bild - Options and extensions for the stochastic shallow convection scheme in ICON - DWD
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Platzhalter für (Titel-) Bild
MODIS Aqua 20130505
Options and extensions for the stochastic shallow convection scheme in ICON
Maike Ahlgrimm, Daniel Klocke, Ekaterina Machulskaya, Mirjana Sakradzija, Axel Seifert…
Why stochastic shallow convection?
Traditional closure assumptions for convection no longer hold at high resolution
Grid box area too
small to contain a Convection is not in
complete equilibrium with the
ensemble of large-scale state
convective clouds (closure)
M: mass flux of the ensemble mi: mass flux of an individual cloud
The resolved atmospheric state no longer predicts a unique (deterministic)
convective state – there are many possible realisations!
Graphic: M. Sakradzija ICCARUS 2021 2
The idea: Predict the cloud ensemble
The mass flux on large scales (where traditional
assumptions are a good approximation) is determined with
the classical parameterisation (Tiedtke-Bechtold/IFS)
At individual grid points, a stochastic cloud
ensemble is generated whose mass flux
(averaged across larger scales) converges to
that of the classical parameterisation
Bonus: The ensemble automatically adapts
to the grid resolution. The smaller the grid
spacing, the greater mass flux departures from
the cloud ensemble mean
But: mass flux limiters interfere
Graphic: M. Sakradzija ICCARUS 2021 3
How is the mass flux at a single grid point calculated?
2) Construct the mass flux
1) Mass flux representative distribution. Distribution
of large scale is calculated parameters include and
using the Tiedtke-Bechtold the Bowen ratio
scheme
first call to 3) Draw number of newly
convection generated clouds from
Poisson distribution
5) Call Tiedtke-Bechtold scheme a
second time (this time using the 4) Add up mass flux (m) of individual
second call to stochastically perturbed mass flux M) clouds to get the grid box mean
convection to generate convective tendencies mass flux (M)
ICCARUS 2021 4
Status
Can demonstrate positive features of the scheme:
Improved ITCZ/precip location in tropical Atlantic (Sakradzija et al. 2020)
Improved SW bias over maritime cumulus areas (EUREC4A domain)
Mass flux limiters not needed for stable operation
Mass flux distributions adapt to resolution as designed
Stochastic in space, continuous in time (memory effect)
Scheme captures temporal evolution of convection, and delays convection by approx. 1h
Tropical Atlantic SW bias, EUREC4A marine cumulus
zonally avg precip vs. TRMM
ICCARUS 2021 5
What’s new?
typical cloud fraction
Approximation using Stochastic Differential Equations
profile from T-B scheme
(SDE) implemented
use either, or use in piggy-backing mode
Spinup/decay options
let the cloud ensemble evolve gradually, or spin-up instantly to
be in equilibrium with forcing?
Representation of the updraft core
can we really assume that the updraft fraction is small relative to radiation “sees” consider
the grid size, and is irrelevant for radiation? only anvil updraft core
Lateral entrainment/detrainment profile
derive individual cloud’s maximum height and use for
construction of detrainment profile representative of the cloud
ensemble within the grid box
ICCARUS 2021 6
Mass flux closure
concept reality
default (with MF
Occurrence
limiter) M M
T-B without MF “moist static energy equilibrium closure”
limiter
1
Cloud base mass flux (kg m-2 s-1)
Problem 2:
Problem 1:
The current closure is ill-defined at a significant number
MF limiter not only catches pathological cases, but
of grid points (around 10%)
significantly determines magnitude of MF (used as
Options: ignore points without closure (no conv), use
tuning parameter)
alternative closure at these points (CAPE, Boeing), use
-> convective activity becomes too great without
alternative closure at all times (Boeing)
limiter
ICCARUS 2021 7
Summary
Hindcast scores promising, but not (yet) matching performance of parallel routine
alternative to tuning via mass flux limiter is required
cp/cv bug fix has required a new “fix” (grayzone tuning) to achieve former performance, which
uses combination of shallow and deep convective schemes to parameterize shallow clouds – not
desirable
Impact of convection scheme on radiation is “filtered” through subgrid diagnostic cloud scheme, and a
lot can be “lost in translation” – crucial to consider the interaction of the convection scheme with the
rest of ICON
Scheme has the potential to generate realistic spread of convective tendencies in ensemble settings –
first tests being run
ICCARUS 2021 8
Alternative Version: Stochastic Differential Equations
ICON D2 simulation of a single day 20130505 with shallow convection
Diurnal cycle of domain total cloud numbers and mass flux are comparable:
SDE version uses 4 prognostic variables to track state of cloud ensemble, can be saved for restart/cycling
Explicit version keeps track of up to 5000 individual clouds per grid cell – easy to extend
ICCARUS 2021 9
Spinup/decay of the cloud ensemble
Cloud base mass flux lifecycle of each cloud is modelled m
BOMEX, 10km, SCM (32 column torus)
lifetime
Without spinup With spinup
kg m-2 s-1
BOMEX, 10km time step
Side note:
Typical for mass flux schemes to be intermittent – too much mass flux exhausts subcloud layer, convection
does not trigger next time step until BL has regained some energy/moisture
Stochastic scheme is less intermittent in time: more gradual increase of MF, continuous in time MF
evolution
ICCARUS 2021 10Vertical transport by convection
Idealised example: BOMEX SCM Cloud fraction
Convective clouds are produced almost exclusively via detrainment at Default ICON
plume top
A realistic cloud profile is only achieved through compensation in time
Pressure hPa
(on-off behaviour of default cloud scheme, with varying plume depths)
This compensation is absent in the stochastic scheme – consistency in
time (memory effect), highlighting problematic plume-top detrainment
stoch conv
Time
can calculate updraft core fraction from
explicit cloud ensemble
11
ICCARUS 2021Lateral entrainment/detainment profiles
• Use explicit scheme to develop more realistic detrainment profile
Cloud fraction
m
Pressure hPa
lifetime
• draw mi
• assign lifetime
• assign max depth depth
• assume cylindrical
updrafts Time
lifetime
ctop
? contribution to cloud fraction due
mi= large variance in turbulent scheme
cbase
ICCARUS 2021 12You can also read
