Motivation

1
The difficult art of evaluation clouds and
convection representation in GCMs

• Motivation
• Configuration
• Results

Roel Neggers Pier Siebesma siebesma_at_knm thanks
to many others at KNMI
2
Evaluation Strategy
Large Eddy Simulation (LES) Models Cloud
Resolving Models (CRM)
Single Column Model Versions of Climate Models
3d-Climate Models NWPs
Development
Testing
Evaluation
3
Potential issues i) How representative are
these idealized situations? ii)
Parameterizations might get calibrated to rare
situations iii) Do the available cases
represent those situations where GCMs
have most trouble / uncertainty ?
Q How can we improve/ensure the statistical
significance and relevance of SCM
simulations?
4
Cant we do more between case studies and global
evaluation?
Large Eddy Simulation (LES) Models Cloud
Resolving Models (CRM)
Single Column Model Versions of Climate Models
3d-Climate Models NWPs
Development
Testing
Evaluation
5
Continuous SCM evaluation The Cabauw SCM Testbed
Purpose Daily SCM simulation at Cabauw for
long, continuous periods of time Evaluation of
long-term statistics against observational
datastreams
6
The idea
Short-range (3 day) SCM simulations are
generated daily for Cabauw Method a
combination of prescribed large-scale forcing and
nudging towards a background
state (observed/forecast/reanalysis) Build
up a long (multi-year) archive of simulations
Allows diagnosing monthly/yearly statistics
i) improved statistical significance
(representativeness) ii) many
different weather regimes are automatically
captured Comprehensive evaluation of the
complete parameterized system against
Cabauw observations Covering
thermodynamics, momentum, radiation, clouds,
soil, etc. Allows constraining all
parameterizations simultaneously ?
should reveal compensating errors in GCMs
7
The Cabauw site
remote sensing in situ (in tower) in situ (ground)
wind profiler SJAC 2m meteo
CT75 ceilometer LAS-X rain gauges
ir-radiometer optical particle counter disdrometer
3 GHz radar FSSP-95 TDR
35 GHz radar nephelometer BSRN station
10 GHz scanning radar sonic anemeter
backscatter lidar gas analyzer
GPS-receiver aetholometer
HATPRO MWR sun photometer
UV radiometer humidograph
scintillometer wind sensors
pyranometer nubiscope temperature sensors
Operated by KNMI Operational since 1972 Tower
height 213m Main scientific goals
Atmospheric research (PBL) Climate
monitoring Air pollution monitoring Model
evaluation Cabauw is a
site http//www.cesar-observatory.nl/
8
What are the strong points of Cabauw for model
evaluation?

• The number of operational instruments
• Continuity of measurement
• Long time-coverage
• High sampling frequency
• A well-organized data archive that is easily
  accessible (CESAR)
• Web browser to confront models (SCM, LES) with
  observations

9
Testbed infrastructure the interactive browser
New!
10
Individual cases
Example a diurnal cycle of shallow cumulus
convection
? CT75 lowest cloud base
11
Evaluation strategy
1) Statistically identify a problem in a GCM
Long-term GCM statistics guide the
evaluation effort 2) Assess if the problem is
reproduced by the corresponding SCM
Exactly matching the GCM statistics
(monthly/yearly means) 3) If so, identify
which individual days contribute most to the
error Selected individual cases are
guaranteed to matter 4) Study those days in
great detail, using a variety of statistical
tools 5) When the cause is identified and
understood, formulate a solution 6)
Re-simulate and re-evaluate the modified SCM
7) Rerun the GCM including the improved physics
3D
1D
3D
12
Example Addressing a summertime diurnal warm
bias over land in a GCM
An issue encountered during the implementation of
a new shallow cumulus scheme into the ECMWF IFS
EDMF-DualM Eddy Diffusivity Mass Flux
scheme
Teixeira and Siebesma, AMS BLT proceedings, 2000

Siebesma et al., JAS 2007
Dual mass flux framework
Neggers et al., JAS 2009,
June issue
13
Convective Mass flux decreasing with height
mass flux cloud core fraction core velocity
LES clouds in silico
Siebesma et al JAS 2003
x
x

14
Recently validated for Clouds in vivo (Zhang,
Klein and Kollias 2009)
clouds in vivo
ARM mm-cloud radar
Updraft mass flux updraft fraction updraft
velocity
15
Siebesma Holtslag 96
LES clouds in silico
old
new
Implication for new EDMF scheme flexible
decreasing mass flux
clouds in vivo
16
Different tendency to form cumulus anvils is
caused by differences in the vertical structure
of model mass flux
Non-mixing Fixed structure
Mixing Flexible structure
M
M
Tiedtke (1989) in IFS
EDMF-DualM
17
Characterizing the differences in global cloud
occurrence
Evaluation of IFS cloud top heights against
Cloudsat / Calypso
Not so nice!
Nice!
Nice!
Figure courtesy Maike Ahlgrimm, ECMWF
18
1 The GCM problem
ECMWF IFS difference in summertime diurnal
cloud cover between
CY32R3 EDMF-DualM and CY32R3
stnd
new
free climate run, June-July 2008
Thanks to Martin Köhler, ECMWF
19
Along with a daily mean 2m temperature bias over
land
free climate run, June-July 2008
20
Suggesting
SW
1. less PBL clouds
4. low level warming
2. larger SW down
3. larger H
Q Can this hypothesis be tested at a local
atmospheric profiling station (i.e Cabauw)?
21
Step 1 Can this bias be reproduced by Single
Column Model at a local site (Cabauw)?

• Run Single Column Model versions in a 3-day
  forecast mode over the Cabauw atmospheric
  profiling site for the same period with the stnd
  and the new scheme forced by the same large scale
  forcing (ERA-Interim)

• Make a composite over the diurnal cycle

22
Step 1 Can this bias be reproduced by Single
Column Model at a local site (Cabauw)?

• Run Single Column Model versions in a 3-day
  forecast mode over the Cabauw atmospheric
  profiling site for the same period with the stnd
  and the new scheme forced by the same large scale
  forcing (ERA-Interim)

• Make a composite over the diurnal cycle

stnd
new
x
obs
New scheme has similar bias of 0.5 K at local
noon.
23
Step 1 Can this bias be reproduced by Single
Column Model at a local site (Cabauw)?

• Run Single Column Model versions in a 3-day
  forecast mode over the Cabauw atmospheric
  profiling site for the same period with the stnd
  and the new scheme forced by the same large scale
  forcing (ERA-Interim)

• Make a composite over the diurnal cycle

stnd
new
x
obs
New scheme has similar bias of 0.5 K at local
noon.
Conditional sampled clear sky days
No bias!
So it must be the clouds!
24
Step 2 take a 2 year period and compare monthly
mean cloud fraction and downwelling SW radiation
with observations.
25
Step 2 take a 2 year period and compare monthly
mean cloud fraction and downwelling SW radiation
with observations. (still consecutive 3
day-forecasts in a scm)
SW-down
Cloud fraction
model
Stnd
new
observations
New scheme too few clouds (bias -15) and too
much downwelling SW radiation (bias 51W/m2).
26
Step 3 Zooming in on days where this bias is
most prominent
stnd
new
27
Step 3 Zooming in on days where this bias is
most prominent
Constant mass flux Fixed structure
stnd
Tiedtke (1989) stnd
M
new
Mixing Flexible structure
EDMF-DualM (new)
M
28
stnd
new

• New scheme has more realistic mixing
• New scheme has a better cloud fraction profile
• But. Systematic too low cloud cover??

29
stnd
new

• New scheme has more realistic mixing
• New scheme has a better cloud fraction profile
• But. Systematic too low cloud cover??

30
Cloud Overlap functions at present maximum
overlap for BL-clouds (in all GCMs!)
Is this a realistic assumption?
height
cfmax
cftot
Cloud fraction
Implies total cloud fraction cftot cfmax
31
Cloud Overlap functions at present maximum
overlap for BL-clouds (in all GCMs!)
Is this a realistic assumption?
height
cfmax
cftot
LES revisited
Cloud fraction
Implies total cloud fraction cftot cfmax
cftot/cfmax 23 depending on shear, depth of
cloud layer
32
Cloud Overlap functions at present maximum
overlap for BL-clouds (in all GCMs!)
Is this a realistic assumption?
height
cfmax
cftot
LES revisited
Cloud fraction
Implies total cloud fraction cftot cfmax
cftot/cfmax 23 depending on shear, depth of
cloud layer
This number is enough to correct the bias in
cloud cover and short wave radiation!
cftot
33
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES).
34
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES). Power of shortrange NWP forecasts to
identify errors in cloud related processes.
35
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES). Power of shortrange NWP forecasts to
identify errors in cloud related processes. All
ingredients (radiation, convection and cloud
parameterization) usually matters.
36
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES). Power of shortrange NWP forecasts to
identify errors in cloud related processes. All
ingredients (radiation, convection and cloud
parameterization) usually matters. Be aware of
compensating errors (convection scheme vs cloud
overlap assumptions in this case). Many GCMs are
currently optimized on their radiative
properties.
37
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES). Power of shortrange NWP forecasts to
identify errors in cloud related processes. All
ingredients (radiation, convection and cloud
parameterization) usually matters. Be aware of
compensating errors (convection scheme vs cloud
overlap assumptions in this case). Many GCMs are
currently optimized on their radiative
properties. All the shown tools (SCM, obs,
plotting) will be made available for all climate
models of the EUCLIPSE partners for the Cloudnet
and some ARM sites (thanks to Roel Neggers, EU,
ARM).
38
Concluding Thoughts
Strength of combining different models (GCM, SCM,
LES). Power of shortrange NWP forecasts to
identify errors in cloud related processes. All
ingredients (radiation, convection and cloud
parameterization) usually matters. Be aware of
compensating errors (convection scheme vs cloud
overlap assumptions in this case). Many GCMs are
currently optimized on their radiative
properties. All the shown tools (SCM, obs,
plotting) will be made available for all climate
models of the EUCLIPSE partners for the Cloudnet
and some ARM sites (thanks to Roel Neggers, EU,
ARM). At present ECMWF, ECHAM and AROME are in
the testbed (DWD-model will follow soon)
39
Thank You