Diagnostics for the Tropics:

1
Diagnostics for the Tropics Some (Cautious)
Uses of Satellite Data

• Duane Waliser/JPL
• Jui-Lin (Frank) Li/JPL
• Baijun Tian/JPL-UCLA
• Xianan Jiang/JPLUCLA
• Terry Kubar/JPL
• Anne Chen/JPL
• Chris Woods/JPL

 ECMWF Annual Seminar Diagnosis of Forecasting
and Data Assimilation Systems 7 - 10 September
2009
2
Outline
MJO Diagnostics applications to contemporary
GCMs (traditional measures traditional
answers good/bad) AIRS Applied to the MJO TRMM
Applied to the MJO CloudSat Applied to the
MJO CloudSat Cloud LiquidIce Considerations
Multi-Sensor/Parameter Considerations MJO
Forecast Metric Update
Newer Process and Vertical Structure Information
3
Problems in the Tropics?
Results from the Aqua-Planet Experiment (APE)
Courtesy, Dave Williamson, NCAR
4
Madden-Julian Oscillation (a.k.a. Intraseasonal,
40-50, 30-60 Day Oscillation)

• Intraseasonal Time Scale 40-60 days
• Planetary-Scale Zonal Wavenumbers 1-3
• Baroclinic Wind Structure
• Eastward Propagation (5-10 m/s)
• Strong Seasonal Dependence
• Significant Interannual Variability
• Potential Role of Ocean/SST Feedback
• Convection Has Multi-Scale Structure
• Significant Remote and Extra-Tropical Impacts
• Predictability 2-4 Weeks -gt Seamless
  Prediction
• Models have chronic problems with MJO.
• Consistent, comprehensive and in-depth diagnostic
  analysis of models is needed.

1987/88
Madden Julian, 1972
5
A Typical MJO in N.H. Winter

• Composite rainfall maps derived from merged
  satellite and in-situ measurements are separated
  by 10 days.
• Rainfall anomalies propagate in a eastward
  fashion and mainly affect the Tropical eastern
  hemisphere.
• These anomalies are accompanied by anomalies in
  wind, solar radiation, sea surface temperature,
  etc.

6
A Typical MJO in N.H. Summer

• Composite rainfall maps derived from merged
  satellite and in-situ measurements are separated
  by 10 days.
• Rainfall anomalies propagate in a northeast
  fashion and mainly affect the Tropical eastern
  hemisphere.
• These anomalies are accompanied by anomalies in
  wind, solar radiation, sea surface temperature,
  etc.

7
MJO Simulation Diagnostics

• Developed by US CLIVAR MJO Working Group
• For recipes, plots and codes, see
  www.usclivar.org/mjo.php
• For more motivation and details, see CLIVAR MJO
  Working Group, 2009 MJO Simulation Diagnostics,
  Journal of Climate, J. Clim., 22, DOI
  10.1175/2008JCLI2731.1.
• NCAR NCL has incorporated these into their latest
  release.
• These diagnostics are mostly for indicating and
  quantifying the degree a model represents the
  MJO, but wont give you much indication of why
  your models representation is lacking.
• There is little vertical structure information
  included as confidence in thermodynamic and cloud
  profiles still needed from new satellite data
  sets and recent analyses.

8
MJO Diagnostics Recipe for Calculating Diagnosti
cs ____________ Calculation Codes Available
9
MJO Diagnostics
10
MJO Diagnostics Equatorial Space-Time
Spectra U, Rain, OLR ____________ NCEP1, NCEP2,
ERA40
NCEP1
ERA40
11
MJO Diagnostics Time Series Spectra U, Rain,
OLR ____________ Domains of Interest
OLR-IO
U850 - EP
12
MJO Diagnostics Life-Cycle Composites U, Rain,
OLR, SLP, SF
Rainfall
U850
Satellite Rain/Cloud AVHRR, GPCP, TRMM Analysis
Data NCEP1,NCEP2
13
Mean SST
MJO Diagnostics Important Mean State Quantities
Mean Zonal Wind Shear
Mean 850 hPa Zonal Wind
14
MJO Diagnostics Adopted into NCAR NCL Graphics
15
Applying MJO Diagnostics

• MJO WG application of MJO Simulation Diagnostics
  to a set of contemporary coupled and uncoupled
  GCM simulations
• CAM3.5, CAM-3Z, SP-CAM, SNU, GEOS5, GFDL,
  ECHAM4/OPYC, CFS
• For details, see Kim, D., et al. 2009
  Application of MJO Simulation Diagnostics to
  Climate Models, J. Climate, In Press.
• Most models look relatively poor still although
  better than a decade ago (e.g., Slingo et al.
  1997). The two best representations were from
  SP-CAM (uncoupled MMF) and ECHAM4/OPYC (coupled
  GCM).

16
MJO Simulation Diagnostics Mean Precip U850
Figure 1 November-April mean precipitation
(shaded) and 850hPa zonal wind (contoured) of a)
CMAP/NCEP1, b) CAM3.5, c) CAM3z, d) CFS, e)
CM2.1, f) ECHAM4/OPYC, g) GEOS5 h) SNU and i)
SPCAM. Contours of mean 850hPa zonal wind are
plotted every 3 ms-1, zero line is represented by
thick solid line. Unit is mmday-1 for
precipitation and ms-1 for 850hPa zonal wind.
17
MJO Simulation Diagnostics Corr RMSE
Figure 2 Scatter plot of pattern correlation
and normalized RMSE for Nov-Apr mean a)
precipitation b) 850hPa zonal wind, c) zonal wind
shear (200hPa-850hPa), d) outgoing longwave
radiation, e) 200hPa zonal wind and f) sea
surface temperature. The region for pattern
correlation and normalized RMSE is 40E-220E,
25S-15N. RMSE is normalized by standard deviation
of observed value.
18
MJO Simulation Diagnostics Variance Precip U850
Figure 3 As in Figure 1, except for variance of
20-100 day band pass filtered precipitation and
850hPa zonal wind. Contours of 850hPa zonal wind
variance are plotted every 3 m2 s-2, 9 m2 s-2
line is represented by thick solid line. The unit
is mm2 day-2 for precipitation and m2 s-2 for
zonal wind.
19
MJO Simulation Diagnostics W-F Precip U850
Figure 4 November-April wavenumber-frequency
spectra of 10oN-10oS averaged precipitation
(shaded) and 850hPa zonal wind (contoured). a)
CMAP/NCEP1, b) CAM3.5, c) CAM3z, d) CFS, e)
CM2.1, f) ECHAM4/OPYC, g) GEOS5 h) SNU and i)
SPCAM. Individual November-April spectra were
calculated for each year, and then averaged over
all years of data. Only the climatological
seasonal cycle and time mean for each
November-April segment were removed before
calculation of the spectra. Units for the
precipitation (zonal wind) spectrum are mm2 day-2
(m2 s-2) per frequency interval per wavenumber
interval. The bandwidth is (180 d)-1.
20
MJO CCEWsMODELINGinIPCC Models
Difficult to get all Parts of the Variability
Right
Lin et al., 2005
21
MJO Simulation Diagnostics Precip LH Flux
Figure 10 Phase-longitude diagram of OLR
(contour, interval-5, green-positive/purple-negati
ve) and evaporation (shaded). Phases are from MJO
life-cycle composite and values are 5S-5N
averaged. The unit of OLR and evaporation is W
m-2.
22
Multi-Scale Structure
Nakazawa 1988
How Important is This Finer Structure To The
Phase Speed, Eastward Propagation, etc Need
Diagnostics for this?
MJO
23
MJO Newer Satellite Data

• A-Train and TRMM offers some new opportunities
  for examining the vertical structure of the MJO.
• AIRS Tian et al. (2006) with NCEP/NCAR and
  recent update with ERA-Interim Tian et al.
  (2009 in prep).
• TRMM Latent Heat Jiang,..Tomkins, et al. (2009).
• CloudSat Boreal Summer - Jiang et al. (2009)
  in prep) Boreal Winter Tian et al. (2009 in
  prep).
• Successes and Cautions for Satellite Data and
  Re-analyses

24
Extended EOF of Rain
MJO Temperature and Moisture Vertical Structure
Composite Methodology and Events Tian et al. 2006
Were AIRS q T channels used in ERA-Interim?
Longer Record -gt 18 Events Now
8 Events in First Study
25
Pressure-Longitude Diagrams of Temperature
Anomaly Along Equator for the MJOTRMM Rainfall
Anomaly Shown as Line Plot (right axis) Panels
Separated by 10 Days
Tian et al. (2006)
AIRS
NCEP/NCAR
-20 Days
-10 Days
0 Days
10 Days
20 Days
26
Vertical Profiles of Temperature Anomaly In the
Indian W.Pacific Ocean for the MJO
Tian et al. (2006)
Indian Ocean
Western Pacific Ocean
NCEP
AIRS
AIRS
NCEP
The plot on the left shows the profiles over the
Indian Ocean for Lag 2 pentads (disturbed)
minus Lag -2 pentads (suppressed). The plot on
the right shows the profiles over the western
Pacific Ocean for Lag 4 pentads (disturbed) -
Lag 0 pentads (suppressed).
27
RMS DIFFERENCE BETWEEN AIRS NCEP FOR MJO
Averaged over 200-1000 hPa
Tian et al. (2006)
28
Pressure-Longitude Diagrams of Water Vapor
Anomaly Along Equator for the MJOTRMM Rainfall
Anomaly Shown as Line Plot (right axis) Panels
Separated by 10 Days
AIRS
NCEP/NCAR
Tian et al. (2006)
-20 Days
-10 Days
0 Days
10 Days
20 Days
29
Pressure-Longitude Diagrams of Water Vapor
Anomaly Along Equator for the MJOTRMM Rainfall
Anomaly Shown as Line Plot (right axis) Panels
Separated by 10 Days
AIRS
ERA-INTERIM
-20 Days
-10 Days
0 Days
10 Days
30
Pressure-Longitude Diagrams of Water Vapor
Anomaly Along Equator for the MJOTRMM Rainfall
Anomaly Shown as Line Plot (right axis) Panels
Separated by 10 Days
AIRS
ERA-INTERIM
-20 Days
-10 Days
0 Days
10 Days
31
MJO Activity During 9899 Winter
Examining MJO Latent Heat Profiles Two TRMM
(Tao et al. CSH Olson et al. TRAIN) and Two
ECMWF (ERA40 IFS CY31r1) Products Jiang et
al. 2009
Rainfall (TRMM 3B42, 10oS-10oN)
Rainfall (seasonal mean removed)
32
EC vs TRMM Q1
CY31r1
(70-90oE 10oS-10oN)
Rainfall
TRMM 3B42
ERA40
TRAIN
CSH
33
EC vs TRMM Q1 Anomalies
(70-90oE 10oS-10oN)
TRMM 3B42
ERA40
TRAIN
CSH
34
Q1 Decomposition EC vs. TRMM (70-90E 10S-10N)
TRMM/Train
EC-IFS
Convective
Stratiform
Radiative
35
Decomposition of Total Rainfall
Total
Convective
ERA-40
Stratiform
Caution is warranted regarding model-data
differences in representation of stratiform and
convective
TRMM/TRAIN
Stratiform Rainfall Percentage
TRMM/TRAIN
FOLLOW-ON WORK UNDERWAY WITH 3 LONG-TERM TRMM
RECORDS AND NEW ANALYSES
ERA-40
36
CloudSat Application MJO/ISV-driven Monsoon
Onset Breaks

• Cloudsat (Jun Sep 2006, 2007, 2008)
• Horizontal resolution 1x1 degs
• Variables
• Cloud liquid water content (LWC)
• Ice water content (IWC)
• Cloud types
• High Cirrus
• Middle Altocumulus (Ac), Altostratus (As)
• Low Stratocumulus (Sc), Stratus (St),
  Nimbostratus (Ns)
• Vertical Cumulus (Cu)
• ERA-Interim (Jun Sep 2006, 2007, 2008)
• Horizontal resolution 1.5x1.5 degs 4x daily
• Variables LWC, IWC
• TRMM rainfall (3B42 1997-2008)
• horizontal resolution 0.25x0.25 deg., 20-70-day
  band-pass filtered

37
Hovmöller diagram of TRMM precipitation
(20-70-day filtered 75-95oE)
2006
2007
2008
(mm/day)
Jiang et al. (2009)
?
?
?
?
?
?
?
?
?
?
Time series of EEOF1 of 1-D 20-70d filtered GPCP
rainfall (5oS25oN, averaged over 75-95oE sector)
for MJJAS, 1996-2007. The EEOF12 basically
captures northward propagation of the BSISO.
38
Composite BSISV Evolution (10 events)
-10day
TRMM rainfall
-5
Northward propagation
0
Time-latitude evolution (75-85oE)
5
10
(mm/day)
15
20
(mm/day)
Jiang et al. (2009)
39
Composite LWC IWC relative to convection center
(gm/m3 80-95oE average)
IWC
LWC
(gm/m3)
hPa
CloudSat
Phase and magnitude difference for IWC have
likely explanation - Although for LWC???
ERA-Interim
S
N
S
N
TRMM rainfall
TRMM rainfall
(mm/day)
convection center
convection center
deg
deg
N
S
N
S
Convection Center
Convection Center
Jiang et al. (2009)
40
Cloud Ice and Liquid

• A-Train MLS and CloudSat offer new opportunities
  for examining the vertical structure of cloud
  water.
• Cloud Ice Li et al. 2005 Li et al. 2007
  Waliser et al. 2009 Woods et al. 2009 -
  submitted
• Cloud Water Li et al. 2008 Li et al. 2009
  submitted
• Careful regarding model satellite
  representations
• Cloud Ice Obs ok Cloud Liquid Obs - caution

41
Model Uncertainties Cloud Ice
Mean IWP from 16 IPCC Contributions of 20th
Century Climate Color scale log10 Raises
Uncertainty about Cloud Feedback Representation
42
Observational Uncertainties Cloud Ice
Annual Mean IWP MODIS - Courtesy S. Platnick
Annual Mean IWP CERES/MODIS - Courtesy P. Minnis
There has been very little agreement between
available cloud ice products and none provided
vertical structure information, i.e. ice water
content Vertical information requires sounding
techniques
Annual Mean IWP NOAA/Microwave - Courtesy H. Meng
Annual Mean IWP ISCCP - Courtesy W. Rossow
43
Cloud Ice Water Path Values from CloudSat
Annual Mean MODIS - Courtesy S. Platnick
Annual Mean IWP CERES/MODIS - Courtesy P. Minnis
Annual Mean IWP NOAA/Microwave - Courtesy H. Meng
Annual Mean IWP ISCCP - Courtesy W. Rossow
44
A-TRAIN IWC Values MLS and CloudSat
Vertically-Resolved IWC
Annual Mean IWC at 215 hPa
MLS 1.5
MLS 8/2004-7/2006
CloudSat R04
CloudSat 8/2006-7/2007
45
ECMWF IMPROVEMENTS
ECMWFR30
May-July Mean Cloud IWC 215 hPa
ECMWFR31
ECMWFR30
Change
MLS
R31 changes involved allowing ice-phase
supersaturation to moisten upper troposphere and
reducing ice crystal sedimentation and snow
auto-conversion rates to increase cloud IWC
(Tompkins et al. 2007). These changes were
motivated in part by the newly available MLS IWC
data that indicated increases in iwc may be
warranted in the model.
46
What do GCM representations imply by cloud ice
and other frozen hydrometeors
More Complex Model
Typical GCM
e.g. fvMM5, DARE/RAVE
e.g., ECMWF, GEOS5, NCAR/CAM
47
GCM Cloud Ice Water Content (IWC) Annual Mean
Values
CAM3
GEOS5
ECMWF
fvMMF
DARE
CloudSat is sensitive to larger/precipitating
hydrometeors and thus is not - as is - an
appropriate validation field for GCM IWC
48
DARE
fvMMF
All make order one contributions to total ice
water path?
graupel
Can CloudSat Be Used as a Preliminary Estimate
of the Total IWC Field to compare to GCMs that
represent this?
ice
Cloud Ice 1/3
Can we judiciously sample/filter Cloudsat to use
for GCM Cloud IWC?
snow
all
49
NP - Non-Precipitating at Surface NC -
Non-Convective
CloudSat has cloud classification and surface
precipitation flags Can we put these to use to
get an estimate of cloud iwc?
Total
50
Back to IPCC Models - CloudSat Cloud IWP
Mean IWP from 16 IPCC Contributions of 20th
Century Climate Color scale log10 Raises
Uncertainty about Cloud Feedback Representation
51
Composite LWC IWC relative to convection center
(gm/m3 80-95oE average)
IWC
LWC
(gm/m3)
hPa
CloudSat
ERA-Interim
S
N
S
N
TRMM rainfall
TRMM rainfall
(mm/day)
convection center
convection center
deg
deg
N
S
N
S
Convection Center
Convection Center
Jiang et al. (2009)
52
Getting Better Species Constriaints Using
CloudSat-Determined PSD information
Example CloudSat IWC PSD
fvMMF PSDs
Lognormal Dg 69.0 µm, slog 0.38 NT 39800 m-3
exponential snow
exponential graupel
monodispersed cloud
0.1 mm threshold between large and small ice
Woods et al. 2009
Divide ice into large and small and integrate
Sum species, then divide ice into large and small
and integrate
53
Getting Better Species Constriaints Using
CloudSat-Determined PSD information
Estimates of
precipitating ice mass cloud
ice mass using 100 um
threshold (e.g., Ryan 2000)
54
Getting Better Species Constriaints Using
CloudSat-Determined PSD information
Estimates of
precipitating ice mass cloud
ice mass Comparison
to fvMMF -gt too little (much) big (small)
particle mass
fvMMF IWClt100µm
fvMMF IWCgt100µm
55
Getting Better Species Constraiints Using
CloudSat-Determined PSD information
Estimates of
precipitating ice mass cloud
ice mass Comparison
to ECMWF -gt good agreement on cloud
mass
ECMWF C31r
56
Satellite Estimates of Liquid Water Path
CERES/MODIS LWP
ISCCP LWP
SSM/I LWP
CloudSat non-precip LWP
CloudSat precip LWP
CloudSat total LWP
Li et al. 2008
57
GMAO/MERRA
ECMWF R30
GEOS5
CAM3
fvMMF
GCM values of cloud Liquid Water Path
Mean122.8
LWC NEEDS MORE WORK IN MODELS AND OBS
Li et al. 2008
58
GCSS Pacific Cross Section Intercomparison
JJA 2006
IWP
LWP
Li, et. el.. 2009b
59
GCSS Pacific Cross Section Intercomparison
LWC IWC Differences -gt Model or Data?
CloudSat Total
LWC
a1
IWC
a2
JJA 2006
a3
Li, et. el.. 2009
LWCIWC

• Ice Magnitude difference due to issue discussed
  above
• Poor Agreement near surface in Subtropical East
  Pacific
• Poor Agreement in coastal stratocumulus region

60
GCSS Pacific Cross Section Intercomparison
Cloud Cover/Frequency
ECMWF
(a)
(b)
CloudSat
JJA 2006
fvMMF
GEOS5
(c)
(d)
Note fvMMF contains cloud cover from graupel,
snow and cloud.
Li, et. el.. 2009
What Can Multi-Sensor Analysis Show Us?
61
Multi-Sensor Analysis YOTC CloudSat-Centric
A-Train Dataset
Covering YOTC Period 5/2008-4/2010 1st Version
Expected November 2009
62
Single-layer uniform low cloud frequency is a
maximum in the heart of the subtropics near
25N-30N at 70 in JJA (also single-layer low
clouds are maximum during JJA, consistent with
many previous studies) Joint LidarRadar agrees
with MODIS remarkably well, except Joint
LidarRadar sees slightly more low clouds in the
tropics (perhaps these are trade cumuli) and also
sees fewer clouds than MODIS at 35N CloudSat
alone misses MANY shallow subtropical low clouds
(especially when tops are under 1 km, not shown)
Multi-sensor examination of low-cloud systems
MODIS
Inversion Frequency
LidarRadar
JJA 08
Radar
63
Multi-Sensor Application Convection (cloudsat
2km), rainfall (AMSR 30km) and tropospheric
Humidity (ECMWF 25KM)
IWC, Reff, Nc from CloudSat increase with
increasing AMSR Rainfall Rate for Convective
Clouds However, as Rain Rate increases, upper
(lower) trop becomes drier (moister). (similar
to Kahn et al. 2009 w/ AIRS)
IWC
Reff,
Nc
64
explanations
CloudSat 2km
Particle Growth Drying
ECMWF (or AIRS) 25km




More Rain Subsidence Drying
Evaporation Moistening
Evaporation Moistening
65
In any case, our next step is to coarsen
resolution of CloudSat to ECMWF/AIRS grid and
redo analysis with significantly more data than
the 1 month (albeit 60,000 profiles) shown here.
66
Operational MJO Forecast Metric

• Need for a common MJO forecast metric
• quantitative forecast skill assessment.
• targeted model improvements.
• even friendly competition to motivate further
  improvements.
• developing a multi-model ensemble forecast of the
  MJO.

ENSO Nino 3.4 Index Weather 500 mb
heights MJO - ?
67
Operational MJO Forecast Metric

• US CLIVAR MJO Working Group and WGNE Project.
• Derived from Wheeler and Hendon (2004) RMM
  indices but designed for operational
  application.
• A number of operational centers participating in
  real-time. http//www.cpc.ncep.noaa.gov/products/p
  recip/CWlink/MJO/clivar_wh.shtml
• Described in Gottshalck et al. 2009 BAMS
  submitted

68
Contributors, Contents and Status Courtesy of Jon
Gottschalck and CPC/NCEP/NOAA
W forecast sent only once per week
See web page for key to Product IDs
http//www.cpc.ncep.noaa.gov/products/precip/CWlin
k/MJO/clivar_wh.shtml
69
MJO Forecast Metric Display Format
Observational RMM1 / RMM2 values for the past 40
days 15-day model forecasts --Green
line Ensemble mean week 1 (thick), week 2
(thin) --Ensemble members light gray shading
90 of forecasts
dark gray shading 50 of forecasts
70
MJO Forecast Metric
Project Website at CPC/NCEP
Example Comparison
71
SUMMARY

• There is a need for developing and applying
  standardized metrics and diagnostics, especially
  useful are those that are relevant for both
  research and operation communities.
• Diagnostics for describing how bad/good a model
  is are more forthcoming than those that can
  elucidate why the model is bad/good. Objectives
  of WWRP/WCRP MJO Task Force include developing
  more process-oriented diagnostics.
• New satellite estimates/products are starting to
  provide new constraints on quantities and aspects
  of vertical structure.
• MJO Forecast Metric is functioning in an
  operational setting across a number of
  forecasting centerswork is now needed on
  validation analyses.

72
EXTRA
73
Application of above hydrometeor
considerations Issues of GCM specification of
particle type and sizes have important
implications for radiation calculations
Often in GCMs water mass in blue is ignored in
radiation calc. - convection thought to be of too
small area and precip too large of particles to
matter much.
74
Using 2B-FLXHR Algorithm (LEcuyer et al. 2008)
we compare Exp control where control is
complete ice from cloudsat to Exp which has an
estimate of the precipitating and convective ice
removed.

• IMPACTS on the net TOA SW and LW are considerable
  20 W/m2
• IMPLICATIONS
• models not accounting for this are getting TOA
  balance incorrectly, i.e. compensating errors in
  quantities such as cloud cover, cloud particle
  effective radius and/or cloud mass,
• implied errors in the vertical distribution of
  heating which can negatively impact the
  circulation as well as the model response to
  external forcings such as changes in GHGs

Waliser, Li, LEcuyer et al. 2009, GRL, In Prep