Tropical weather systems
within a global data
assimilation and forecasting
framework
Oreste Reale
NASA Goddard Laboratory for Atmospheres
and
GEST/UMBC
Motivation
 The deadliest natural events are related to tropical
weather systems (500,000 people died because of
1970 cyclone in Bangla Desh)
 Almost 40 years later (2008): Tropical Cyclone
Nargis killed at least 140,000 people
 Numerical weather forecasts in the tropics have
improved at a slower pace than mid-latitude
weather forecasts
 Acknowledging intrinsic predictability limitations,
there is vast room for improvement
Outline
 Part I: the global analysis in the tropics
 Part II: the representation of tropical cyclones in
global models
 Part III: improvements stemming from the
assimilation of AIRS-derived products
 Part IV: other improvements
 Conclusions and future
Outline of Part I
 AEJ representation in state-of-the-art reanalyses
 AEJ representation on weather-time-scales in
operational analyses during SOP-3 NAMMA
(2006)
 Vertical soundings during SOP-3 NAMMA (2006)
 Mid-tropospheric flow over the entire tropical
Pacific in August 2010 in NCEP operational,
ECMWF operational, and MERRA
AEJ representation in
state-of-the-art reanalyses
 Previously published work (Wu et al., 2009) shows
substantial differences between reanalyses in the
monthly mean representation of the African
Easterly Jet (AEJ)
 ERA-40, NCEP-R2, JRA-25 provide very different
descriptions of the AEJ structure, and of the
horizontal shear in the cyclonically-sheared portion
of the AEJ
M.-L. C. Wu, Reale, O., S. Schubert, M. Suarez,
R. Koster, P. Pegion, 2009:
African Easterly Jet: Structure and Maintenance.
J. Climate, 22, 4459-4480.
Large differences in
AEJ SHAPE, INTENSITY
VERTICAL STRUCTURE
and distribution of the
horizontal shear
in a 22-year average
performed on ERA-40,
NCEP-R2, and JRA-25.
From
Wu et al. (2009)
Fig 2
July zonal wind (m s−1, contours every 1 m s−1, 0 omitted, solid: positive,
dashed: negative) climatology (1980–2001) based on (top to bottom) ERA-40, NCEP R2,
and JRA-25 data: (left) meridional horizontal shear of the zonal wind at 600 hPa
and (right) meridional cross section at 0° longitude. (Wu et al., 2009, J.Climate)
AEJ and its instability properties in
state-of-the-art reanalyses
 Work recently submitted (Wu et al., 2011) shows
differences in the representation of the African Easterly Jet
(AEJ) seasonal instability properties between reanalyses
across a 22-year average
 Despite revealing some instability property of the AEJ that
appear data-independent, ERA-40, NCEP-R2, JRA-25 and
MERRA provide very different descriptions of the AEJ
horizontal structure, intensity, and of some properties that
control wave instability on a seasonal scale (JAS).
M.-L. C. Wu, Reale, O., S. Schubert,
M. Suarez, C. Thorncroft, 2011:
African Easterly Jet: barotropic instability, waves and
cyclogenesis.
Submitted to: J. Climate.
The analyses differ in terms of
strength and intensity of the low-level
monsoonal flow, slope of the
barotropically unstable part of the
AEJ, horizontal shear distribution.
All Figures show a
22-year JAS average
From
Wu et al. (2010)
Fig 2
Unexpected discrepancies between snapshots
of analyzed representation of the African
Monsoon-Eastern Tropical Atlantic regions
 The African Easterly Jet at about 600hPa, the lowlevel monsoonal flow (predominantly southwesterly
between 1000 and 800 hPa) and the Tropical Easterly
Jet (between 200 and 100 hPa) are the critical players
in Atlantic tropical development.
 Comparison between operational NCEP analyses and
GEOS-5-produced analyses reveal serious
discrepancies
 Validation agains the only vertical sounding in the area
at Cape Verde (15N, 23.5W) during the 2006 NAMMA
campaign, show that both analyses have large errors
Huge discrepancies between
GEOS-5 and NCEP operational analyses
Wind at 5-15N,
500-600 hPa, has
opposite direction!
Only in the tropics
the two analyses
differ substantially
Section at 23.5W
Largest differences between reanalyses are in the tropics,
at about 15N (on the order of 12m/s)
larger even than
discrepancies in the southern hemisphere jet stream
NCEP
GEOS-5
Huge differences in the entire tropical zonal flow
from 20S to 20N at all levels
Largest mid-tropospheric wind difference is in
the tropics, at 0-10N
GEOS-5 analyses
produce a weaker
easterly flow
than NCEP
GEOS-5
NCEP
Largest low-tropospheric wind difference is in
the tropics, between 10S and Equator
NCEP
GEOS-5
Opposite-sign
discrepancy
with respect to
previous slide:
GEOS-5 analyses
produce stronger
easterly flow than
NCEP)
Additional vertical soundings at Cape Verde during
SOP-3 provided the chance to validate operational
analyses in 2006
One of the rare cases in which NCEP
and GEOS-5 differ less than 5 m/s)
Both NCEP and GEOS-5 miss the
AEJ maximum at 600hPa. Error
larger than 10 m/s at AJE level!!!
obs
NCEP
vs
GEOS-5
obs
Catastrophic non-systematic differences
NCEP provides a good representation of
low-level and upper-level flows but misses
the AEJ. GEOS-5 has huge errors at all levels
except at 600hPa.
obs
NCEP and GEOS-5 both miss
the low-level flow, with NCEP having
larger errors.
NCEP
vs
GEOS-5
obs
Catastrophic non-systematic differences
NCEP produces a stronger AEJ.
GEOS-5 produces a stronger AEJ.
NCEP
vs
GEOS-5
Huge differences between operational ECMWF,
NCEP and MERRA over the entire tropical Pacific
during strong La Nina conditions (Aug 2010)
 Weather prediction over the tropical Pacific is
controlled by a good representation of the
predominantly easterly flow and periodic westerly
bursts along the Equator
 Large errors in the equatorial flow propagate
away from the Equator affecting TC genesis
prediction, and TC track forecast as far as 30N/S
Huge 600hPa zonal wind difference
affects the entire tropical Pacific in 2010
Speeds are very
comparable away
from the tropics.
Difference of
about 10m/s
over Eq.Pacific
Huge 600hPa zonal wind difference
affects the entire tropical Pacific in 2010
50% speed
Difference
Over Eq.
Pacific
Opposite sign
wind over,
and NE of,
Hawaii
Huge 600hPa zonal wind difference
affects the entire tropical Pacific in 2010
involving all 3 data sets
The largest 600hPa wind difference at 165W occurs
in the tropics, between 20S and 10N
ECMWF
MERRA
NCEP
Part II
The representation of tropical cyclones
in global models








The overall forecast quality is a blend of the impacts of initial conditions produced
by the Data Assimilation System -and- the forecast model capability
It is important to separate the intrinsic model capability from the impact of the
analysis
Less-than-optimal model performance with respect to TCs can be somewhat
improved with very good TC initialization
BUT less-than-optimal TC initialization can be somewhat compensated by a very
good model representation of the large scale forcing
What past and current models can produce in `free-running mode’ or in
`weather-forecasting mode’ concerning Tropical Cyclone vertical structure, scale,
intensity, track realism, genesis process, large-scale forcing
Model comparison in forecast mode: NMC MRF (1998), NASA GEOS-4 (2004),
NCEP GFS (2004); NASA GEOS-5 v2 (2009)
Long simulations (so as to free the model from the memory of the ICs, can be
performed to assess the intrinsic model capabilities with respect to TC structure
and realism) : ECMWF T511 Nature Run, NASA GEOS-5 (2009)
The problem of missing TCs in the operational ANALYSIS can deteriorate the
forecast of any good model
TCs in high-resolution global models
 It has been empirically noted in the operational wx forec.
community that at hor. res. of 1 degree one can start seeing
vertically aligned structures and an eye-like feature, at 0.5
degree the maximum winds begin to develop in the lower
levels (instead of the mid-troposphere, as observed in lower
resolution global models), at resolutions of few tens of
kilometers global models start displaying realistic radii of
maximum wind (e.g., Atlas et al., 2005; Shen et al., 2006;
Reale et al., 2007),
 But it takes cloud-resolving models at resolution of few
kilometers to detect eye-wall replacement cycles
 Accepting the limitation imposed by global models, it is
interesting to follow the representation of tropical cyclones
in global forecast models over the last 12 years.
Is high resolution always exploited?
 At any resolution, a wind speed vertical cross-section of a mature
tropical cyclone should present two approximately symmetric maxima
around a wind minimum.
 The compactness of this eye-like feature increases with resolution but
often high resolution models display structures that are much broader
and more diluted than what could be expected at that resolution.
 Unrealistically large eye-like features (on the order of hundreds of km,
encompassing several gridpoints) are common in GCMs even when
horizontal resolution is of a quarter of a degree.
 The optimal, theoretical representation that should be possible at a
given resolution, is NOT always reached.
 It is important to perform proper diagnostics that allow to assess the
quality of the representation of a TC at any given resolution
TCs in Global Operational forecasting models
track versus intensity forecast
 Forecast track failures in earlier global operational models were
generally assessed only from the point of view of large-scale forcings,
irrespective of how the TC structure was represented
 In the latest global operational models forecast, structure realism and
good forecast track appear to be connected (unlike the past, where
track and intensity were treated as completely separate problems)
 The quality of the representation of some large-scale forcings (i.e.
ITCZ position) appear to control part of the weather forecasting scales
involved with TC motion
 In the past (>10 years ago), TC representation in global operational
models was sporadic and very poor
 Bogusing was a necessity (now replaced by vortex relocation)
13 years ago: Bonnie (1998)
as seen by the MRF (ancestor of NCEP GFS)
850 hPa wind
Sea level pressure
NMC state-of-the-art representation of TCs in 1998: no more than 25 m/s,
excessively large scale (~1000km); center pressures above 1000 hPa
(despite containing Hurricane Hunters flight data). TCs away from
operational HH flights were often absent from analyses and forecasts.
Bonnie (1998) cont.
MRF (NMC-now NCEP) state-of-the-art representation of TCs in 1998:
no more than 25 m/s, unrealistically wide eye-like feature
(r~100km); very weak warm core
TC Structure: NASA GEOS-4 in 2004
Isidore (2002)
Modeled with GEOS-4 in 2004
Realistic deepening (center down to
960 hPa, unseen in any un-bogused
GCMs). The NCEP Analyses confirm
the position but are not as deep with
respect to observations.
wind speed,
temp, vort
Atlas, R., O. Reale, B.-W. Shen, S.-J. Lin, J.-D. Chern, W. Putman, T. Lee, K.-S. Yeh,
M. Bosilovich, and J. Radakovich, 2005: Hurricane forecasting with the high-resolution
NASA finite-volume general circulation model.
Geophysical Research Letters, 32, L03807, doi:10.1029/2004GL021513.
Example of a very realistic NASA GEOS4 simulation
in which track and intensity forecast go side-by-side
Ivan (2004)
Example of hurricane vertical structure as
modeled by the GEOS-4 (2004): Ivan
 The GEOS4 could produce a
very compact eye-like feature
throughout the troposphere, a
prominent warm core; wind
maxima located at about 850900 hPa and a radius of
maximum wind of about 50-100
km
 In this 66 hour forecast of
hurricane Ivan for
18z12Sep2004, initalized at
00z10September, the 900 hPa
wind is higher than 55 m/s
Wind speed, temp
Realistic Cyclogenesis in a global operational model
(NASA GEOS-4, 2004) WITHOUT BOGUSING
– Frances (2004): early
phase
 Example of rapid
deepening and good
forecast track despite poor
analyzed intensity
 One run reaches the
correct intensity (IC:
00z30Aug) and produces
the best forecast track as
well
 It takes four days for the
model to compensate the
deficient initialization
In free-running mode, the TCs are
spontaneously produced by the
model without memory of ICs.
 Seasonal runs or long runs exceed forecast capability but
statistical behavior of TC activity over
months/seasons/years, and realism of TC structure can be
inferred
 TC activity is controlled only by global forcings (SST)
 TC cyclogenesis and structure are produced by the model
alone without any contribution of Initial Conditions
 These are good tests to assess the intrinsic model
capability with respect to TC processes
T511 ECMWF Nature Run (2007)
 Free running model – no memory of initial
conditions – no additional data
 A long simulation is the only way to assess the
capability of a forecasting model – as opposed as
a DAS+forecasting model. No bogusing, vortex
relocation, targeted obs, can be added.
 13-month run, initialized May 2005
 Only SST (2005) and Sea-Ice as boundary
forcings
 Analysis published in Reale et al. (2007)
Reale, O., J. Terry, M. Masutani, E. Andersson, L. P. Riishojgaard, J. C. Jusem,
2007: Preliminary evaluation of the European Centre for Medium-Range Weather
Forecasts (ECMWF) Nature Run over the Tropical Atlantic and African Monsoon region.
Geophysical Research Letters, 34, L22810, doi:10.1029/2007GL31640.
EC T511 NR: realistic activity (9 strong TCs)
From Reale et al. (2007) GRL
EC T511: Realistic Variability of Atl. TC tracks
Looping and
Binary vortex
interaction
4 systems:
Looping,
Binary vortex
Interaction,
Extratropical
Transitions
and Extra-tropical
Re-intensification
Singuarities, binary vortex
Interactions, Intensity fluctuations
Due to large-scale forcing fluctuations
A long simulation must produce complex tracks
GEOS-5 with Stocastic Tokioka (2009)
Simulations by Myong-In Lee, PI S. Schubert (NASA):
Same experiment settings of ECMWF Nature Run
Behavior comparable to the EC T511
Control Run GEOS-5 0.25
(with rel. Arakawa-Schubert)
GEOS-5 (with Tokioka)
No cyclone reaches 1000hPa in the Control during September.
At least 7 cyclones below 1000 hPa in the GEOS-5 w.Tok. One hurricane
goes below 960hPa. Very realistic track variability, scale.
Even non-developing waves are well captured.
EC T511: Multiple simultaneous tropical cyclones can be
present in the Atlantic in very active seasons
Another important –realistic- capability of the ECMWF NR
500 hPa geop (m) and 900 hPa rel vort (s-1)
3 TCs simultaneously present in the
GEOS-5 w. Tokioka 11Sep
Simulations by Myong-In Lee, PI S. Schubert (NASA):
Slp (hPa) and 925 hPa wind (m/s)
Intensity
 In the operational forecasting environment, 10m observed
wind and center pressure are currently used
 PROBLEM: excessively high drag in the marine boundary
layer seems to occur in global models when winds exceed
30m/s: 10m wind often about 60% of the 850hPa wind
(unlike 90% in real world)
 Possibly due to unrealistically high roughness length over
oceans with wind speeds exceeding 30m/s
 As a consequence, it may better to use 850hPa or 900hPa
wind as intensity diagnostics in global models
 One simple way of assessing comprehensively the TC
intensity reached in a simulation is to produce the max
wind at 850hPa throughout the system’s lifespan
Example of Intensity inferred from 850hPa
wind max (Isabel, 2003)
Operational GFS and GEOS-4 have comparable intensity
Different degree of compactness
A possible metrics to assess how well the
horizontal scale is represented
Horizontal Compactness, ratio of radius of maximum
wind (rmw) over radius of wind greater than the
environmental wind of a given threshold, which we can
consider the radius of the tropical cyclone (TC) in the
model (rtc).
The wind magnitude of a modeled tropical cyclone
decreases from the center and is not distinguishable from
the large-scale wind at a certain distance. This distance
could be considered the tc-influenced domain in the model
and can be compared with the rmw. The smaller rmw with
respect to the rtc the more realistic the modeled cyclone
is. In low-resolution global models, the radius of maximum
wind occupies a large fraction of the domain affected by
the cyclone.
Example from older GEOS-5 v.2:
how compact is this 0.5 simulation of Helene?
Compactness evaluated in GEOS-5 simulation at .5
for Helene (2006)
850 hPa wind at 18.5N
Despite being a relatively
weak simulation,
the representation of the
system is quite compact
in the above sense
[RMW(l)+RMW(r)] / [RTC(l)+RTC(r)]=0.27
Compactness in the GEOS-5 w. Tokioka at 0.25
a much better rpresentation of a TC
At ~60W, a
RAINBAND
RMW
RTC(l)
RTC(r)
[RMW(l)+RMW(r)] / [RTC(l)+RTC(r)]=0.07
Very clear evidence of a rainband at 61W
RAINBAND
Warm Core Structure
One immediate, effective way of assessing if
a model produces a vertically aligned and
symmetric system, is to measure the
strenght of its warm core. One way is simply
to subtract a standardized zonal mean
intersecting the center of the storm.
Examples: GEOS-5 versus NCEP GFS
Examples of warm core (Helene, 2006)
48-h Fc
72-h Fc
GEOS-5
(0.25)
GEOS-5
(0.25)
GFS
GFS
48-h Fc
Ms. M. Fuentes, Ph.D. Thesis
72-h Fc
Vertical Structure inferred through zonal
and meridional vertical cross-sections of
wind speed and temperature of mature
TCs in the deep tropics
Desirable features:






Wind maximum at 900hPa or lower
Small radius of maximum wind
Perfectly vertically aligned low-speed column
Vorticity column with maximum in the lower levels
Low-level convergence confined below 800 hPa
Upper-level divergence confined above 200 hPa
Side by side comparison
ECT511 vs GEOS5 with Tokioka
GEOS runs by Myong-In Lee, PI S. Schubert (NASA):
EC T511 (2007)
Zonal
GEOS-5 with Tokioka (2009)
Meridional
Zonal
Meridional
GEOS-5 has slightly sharper warm core, better-defined eye, max wind at lower
elevation, slightly smaller radius of max wind. Intensity is about the same.
Hurricane in the Atlantic
(GEOS-5 long simulation w. Tok
by Myong-In Lee)
Winds up to 60m/s
Vorticity up to 3x10-3s-1
Warm core up to 10C!
Summary of TC features that can be seen in
operational global models
(GEOS-4, GFS, ECMWF T511)
observations, 0.25 GEOS-5 with Tokioka
 Horizontal scale ~wind speed comparable to the largescale environment (300-1000 km; 250-1000 km; 3001000km)
 Radius of max. wind (50-300 km; 20-100km; 50-100km)
 Low-level vorticity (10^-3 s-1, 3x10^-3 s-1 )
 850 hPa wind: above 60 m/s, above 100m/s, above 60m/s
 Warm core: 2-10 C; 6-14 C; 4-12C
 Horizontal compactness: 0.07-0.35; 0.05-0.15; 0.07-0.20
However, despite the current capability of global
models, state-of-the-art operational analyses can
completely miss existing Tropical Cyclones.
 Analyses are particularly deficient in the depiction
of developing, deepening and transitioning
tropical cyclones
 Analyses are deficient in representing
cyclogenesis and existing deepening cyclones in
the Eastern Atlantic
 Analyses are particularly deficient in representing
even fully-developed TCs over the Indian Ocean
TS Debby (2006) at 06z 24 Aug 2006
Obs center slp 999 hPa; Max wind 22 m/s
NCEP analyses
do not produce
a closed
circulation
GEOS-5 An.
200km
displacement
error for center
(obs. center X)
Conclusions (Part I and II)
 State of the art reanalyses (ERA-40, JRA-25, NCEP-R2 and
MERRA) show susbtantial differences in the seasonallyaveraged representation of the African Easterly Jet and more
generally of the circulation in the African Monsoon and tropical
Atlantic regions
 Operational analyses or reanalyses differ also at
instantaneous times in the tropical region. On the contrary,
away from the tropics, different analyses provide almost
identical representations of the wind field
 Despite changes in models and assimilation systems, and
increase in resolution, the representation of wind in the tropics
does not show much improvement from 2006 to 2010
 Major deficiencies appear on all 3 basins: Atlantic, Indian and
Pacific Oceans on scales spanning from storm-scale to
planetary, from weather to seasonal
Outline Part III
Improvements stemming
out of use of AIRS
 AIRS impact on midlatitude winter dynamics
 Global AIRS impacts in boreal winter, spring, summer and
fall conditions in five different years
 TC analysis. Improvement in tropical cyclone position and
structure, leading to improved forecast track, over all
basins
 AIRS impact on tropical cyclone Nargis
 AIRS impact on tropical cyclones in the Atlantic
 AIRS impact on cyclogenesis
 AIRS impact on extra tropical transitions
 AIRS Impact on Precipitation Analyses and Forecasts
Understanding and improving the impact of AIRS in the
GEOS-5 Data Assimilation and Forecasting System
 Previously published work (Reale et al., 2008) has shown
substantial improvement in analysis and forecasts over the
northern hemisphere extratropics in boreal winter
conditions, due to an improved representation of the
lower-mid tropospheric thermal structure in the high
latitudes, and consequently an improved polar vortex.
 The improvement comes from the assimilation of qualitycontrolled AIRS retrievals obtained under partially cloudy
conditions
Reale, O., J. Susskind, R. Rosenberg, E. Brin, E. Liu, L.P. Riishojgaard,
J. Terrry, J.C. Jusem, 2008: Improving forecast skill by assimilation of
quality-controlled AIRS temperature retrievals under partially cloudy conditions.
Geophys. Res. Lett., 35, L08809, doi: 10.1029/2007GL033002
 3 sets of 27 5-day forecasts, initialized at 00Z each day,
from 1/5/03 through 1/31/03:
– “CNTRL” set initialized from the control assimilation,
where all operational data (conventional and satellite)
are ingested except for AIRS
– “AIRS” set initialized from the assimilation in which
AIRS cloudy retrievals are ingested in addition to all
data used by the CNTRL
– “CUTF” set initialized from the assimilation in which
AIRS retrievals are ingested only above 200hPa
 All three sets verified against NCEP analysis
RESULTS:
500hPa geopotential height
anomaly correlation (AC) in the
Northern hemisphere extratropics of the average of all 27
forecasts verif. against NCEP
analysis as a function of
forecast time.
The day-5 500mb geopotential
height anomaly correlation
(AC5) for each of the 27
forecasts
AIRS forecasts demonstrated
superior skill over both the
CNTRL and CUTF in most of
the cases. Case 21 (init. 25
Jan) in which the CNTRL AC5
is high (.85) and the AIRS
forecast produces further
significant improvement
800hPa temperature anomaly
(AIRS minus CNTRL) analysis
(00Z 25 Jan). Notice the large
area of negative temperature
impact over northeastern Siberia,
Alaska and the Arctic region.
Analyzed emperature profiles, areaaveraged over the entire Arctic region
(70-90N), from 1000 to 100mb at the
initial forecast time (00Z 25 Jan) of:
CNTRL = black
AIRS = green
CUTF = red
and the AIRS minus CNTRL temperature
difference profile (orange).
The assimilation of AIRS cloudy
retrievals in the lower-mid troposphere
results in significantly colder
temperatures between 950 and 700mb,
with a peak at about 875mb.
From Reale et al. (2008)
500hPa geopotential height
anomaly (AIRS minus CNTRL)
at 00Z 25 Jan (analyses).
The hydrostatic adjustment
induced by lower temperatures
causes the 500mb geopotential
in the AIRS case to drop
substantially, modifying
drastically the structure of the
polar vortex.
From Reale et al. (2008)
Latitudinally averaged (40-80N) 500 hPa geopotential height
anomaly (AIRS minus CNTRL, shaded, and NCEP minus CNTRL,
solid black line) as a function of forecast time.
The initial negative anomaly
over Siberia and Alaska,
appears as a wave packet
undergoing dispersion,
amplifying and propagating
eastward . The AIRS minus
CNTRL anomaly observed at
day 5 over Canada and the
north Atlantic corresponds
well with the NCEP AN.
minus CNTRL in the same
region.
From Reale et al. (2008)
Global Impact of Clear-sky Radiances
versus
Quality Controlled cloudy Retrievals
 A small fraction of AIRS data is still retained in operational
weather systems, where the only AIRS data assimilated are
radiance observations of channels unaffected by clouds. This
imposes a severe limitation on the horizontal distribution of the
data.
 Susskind (2007, 2010) strategy, based upon previous work by
Chahine, allows improvement of soundings in partly-cloudy
conditions: a key element is the ability to generate case-bycase and level-by-level error estimates and use them for
quality control
 A very large number of experiments were produced, comparing
AIRS retrievals and radiances in all seasons, five different
years, with different quality controls, looking at both global
impacts and individual high-impact weather systems
AIRS Experiments settings
– GEOS-5 DAS: versions 2.0.2, 2.1.2, 2.1.4
– Control assimilation (CNTRL): assimilating all conventional
and satellite data, but no AIRS retrievals, from 8/10/06 to
9/15/2006 (NAMMA), 10/15/2005 to 11/15/2005 (Active TC
Atlantic season), 4/15/2008 to 5/15/2008 (Nargis), 7/1/2010
to 8/15/2010 (Pakistan floods)
– AIRS ``standard’’ QC RET: Same data as control plus AIRS
version 5 retrievals with “standard” quality control added as
rawinsonde temperature profiles.
– AIRS ``medium’’ QC RET: More restrictive QC for AIRS ret
– AIRS ``tight’’ QC RET: Most restrictive QC for AIRS ret
– AIRS RAD: AIRS clear-sky radiances from NESDIS
– Forecasts at 0.25 and/or 0.5 degrees
GEOS-5 2.0.2 Boreal Spring Conditions:
global impact of cloudy retrievals (tight QC) vs.
clear-sky radiances
Positive global impact of
AIRS retrievals (red).
Negative impact of AIRS
clear-sky radiances (green).
In addition, representation of
individual weather systems
in the tropics are strongly
impacted by AIRS.
Anomaly Correlations computed from 90S to 90N
GEOS-5 2.0.2 Boreal Summer Conditions:
global impact of cloudy retrievals (tight QC) vs.
clear-sky radiances
Strong global impact of
AIRS retrievals (red).
Smaller impact of AIRS
clear-sky radiances (green).
In addition, representation of
individual weather systems
in the tropics are strongly
impacted by AIRS.
Anomaly Correlations computed from 90S to 90N
GEOS-5 2.0.2 Boreal Fall Conditions:
global impact of cloudy retrievals (tight QC) vs.
clear-sky radiances
Strong Positive global impact of
AIRS retrievals (red).
Smaller positive impact of AIRS
clear-sky radiances (green).
In addition, representation of
individual weather systems
in the tropics are strongly
impacted by AIRS.
Anomaly Correlations computed from 90S to 90N
In addition to global skill, AIRS affects
the depiction of tropical weather systems
 AIRS cloudy retrievals change particularly the depiction of
developing and transitioning tropical cyclones
 AIRS impact on Tropical Cyclones in the GEOS-5 has
been studied over the Atlantic, Indian and Pacific Oceans
 AIRS improves the Tropical Cyclone ANALYSIS in
GEOS5-DAS in terms of intensity, confinement and
position
 The cause of the improvement arises from tight, strong
upper-tropospheric positive thermal anomalies detected
over organized convection
 No or minimal improvement derives from the assimilation
of clear-sky radiances
Published study on the impact of AIRS,
focused on a particulaly difficult tropical cyclone:
Nargis (2008)
 Work published in 2009 shows improvements in analysis over the
tropics in in the GEOS-5 DAS and forecasting model consequent to
assimilation of AIRS-derived information in CLOUDY areas. Case
chosen: catastrophic cyclone Nargis which hit Burma causing
devastating loss of life
 Tropical Cyclones in the Northern Indian Oceans are extremely difficult
to predict because of shorter lifespan and erratic tracks
 Operational global analyses often do not represent these cyclones’
position (or even the TCs’ very existence) accurately partly because of
strongly asymmetric data distribution geometry
 Forecasts are particularly penalized by analysis errors.
Reale, O., W. K. Lau, J. Susskind, R. Rosenberg, E. Brin, E. Liu, L.P. Riishojgaard, M.
Fuentes, R. Rosenberg, 2009: AIRS impact on the analysis and forecast track of tropical
cyclone Nargis in a global data assimilation and forecasting system.
Geophys. Res. Lett., 36, L06812, doi: 10.1029/2008GL037122
Complete miss of TC Nargis (2008) in both
operational NCEP and MERRA analyses at a
time when is declared having hurricane-level
winds by the JTPC and IMC
COMPLETELY
FLAT PRESSURE
FIELD
800x600km
Contours
every 1hPa
800x600km
Contours
every 1hPa
WINDS DO
NOT FORM
A CLOSED
CIRCULATION
X observed
cyclone’s center
WINDS DO NOT REACH 12m/s
WINDS DO NOT FORM A CLOSED CIRCULATION
Spectacular forecast track improvement for
tropical cyclone Nargis (2008) consequent to
qc-ed AIRS cloudy retrieval assimilation
Control
AIRS clear-sky radiances
AIRS cloudy retrievals
5 out of 7 forecasts initialized from the improved analyses
have a displacement error at landfall of about 50km
(Reale et al., 2009, Geophys. Res. Lett.)
Assimilation of clear-sky radiances produce minimal improvement
Improvement with AIRS cloudy retrievals
Analysis obtained
assimilating AIRS
cloudy retrievals
Well-defined
Cyclone
Green:
Observed
Track
108-hour
forecast (slp)
initialized from
improved
analyses
Green:
Observed
Track
CNTRL Analysis (above)
And forecast (below): No Cyclone
Accurate landfall is produced in the forecasts initialized
with AIRS: (Reale et al., 2009, Geophys. Res. Lett.)
Why AIRS radiances do not impact
the forecast for NARGIS?
USED
REJECTED
There are simply NO DATA accepted by the DAS in the area where NARGIS
developes, because the measurements are in cloudy areas.
QC-ed AIRS cloudy retrievals provide
substantial coverage over the area
The temperature information provided by cloudy AIRS retrievals where
the storm is developing leads to improved analyses and forecasts
How AIRS retrievals improve the analysis of a TC?
The localized, intense
Upper-Level heating
induced by AIRS data
in correspondence to
organized convection
deepens the low-level
cyclonic circulation of
TC Nargis
Shaded: 200 hPa AIRS minus CNTRL temp anomaly
Contour: AIRS minus CNTRL slp anomaly (Reale et al., 2009)
AIRS impact study in boreal summer
conditions
 AIRS improves the representation of the thermal structure
of the atmosphere in the tropics
 In particular, developing tropical lows are better defined
and confined with the ingestion of AIRS temperature
retrievals under partly cloudy conditions
 The improvement consists of a) more confined and tight
circulations b) more accurate center locations
 Experiments covering the NAMMA SOP-3 period (15Aug15Sep 2006) to investigate TC representation in the
Atlantic in response to AIRS data ingestion
AIRS TIGHT QC CLOUDY RET improves TC
position in the Analysis of Helene (2006)
Slp RAD analysis
(contour)
RAD slp impact
(shaded)
X: observed
Helene position
300 hPa Temp
Impact (ret minus
Control, shaded)
And slp impact
(contour)
Slp RET analysis (contour)
RETRIEVALS slp impact (shaded)
AIRS TIGHT RET produces a PERFECT
position for Helene and a deeper storm
Large improvement in the forecast of Hurricane
Helene’s genesis with `tight QC’ cloudy
retrievals
Comparison
Of 36-h
Forecasts
of AIRS TIGHT RET
(lower left)
with AIRS RAD
(upper right)
Forecasts from Analysis in which AIRS TIGHT RET are assimilated improve Helene’s
Formation as a hurricane (12z 16Sep). Improvement is minimal in RAD case
AIRS impact on extra-tropical
transitions: a difficult problem.




Rapid changes in dynamics from tropical to baroclinic
Very strong asymmetric vertical shear
Rapid acceleration of the systems
Small errors in TC location before transition lead to large
forecast track errors
 Small errors in the thermal structure of the atmosphere
before transition lead to large misrepresentations of storm
intensity
 AIRS quality-controlled cloudy retrievals positively impact
EXTRA TROPICAL TRANSITIONS
Extra-tropical transition of Hurricane Florence
(2006): improving the analysis before
transition 06z10Sep2006
RAD slp analysis (contour)
And difference from
CNTRL.. Negative impact
From assimilation of clear-sky
Radiances.
400 hPa AIRS RET-induced
Temp anomaly (shaded)
And impact on slp (contour)
AIRS RET slp analysis (contour)
and difference from CNTRL (shaded)
AIRS RET improves location and intensity also at
subsequent times; FORECASTS from the improved
analyses are much superior.
Predicting the ET intensification of Florence from
improved (before ET) analyses
The 60h forecast
Initialized from
Analyses in which
AIRS retrievals
are assimilated
produce a deeper
after ET-cyclone
with respect to
the RAD case,
in agreement with
observations.
Improvement in cloud structure caused by
AIRS cloudy retrievals
TS Helene Analysis at 06z 15Sep2006
30 hours before becoming a hurricane
800 hpa relative humidity, sea level pressure (hPa)
CNTRL
RETRIEVALS
Display an
Eye-like
feature
RADIANCE
NCEP
Operational
Analyses,
Very poor
The 36-hour forecast initialized from analyses in which AIRS
retrievals are assimilated is the only one that produce an eye,
a closed circulation, and a reasonable scale
Helene at 12z 16Sep2006, upgraded to hurricane
850 hpa relative humidity, sea level pressure (hPa)
CNTRL
RETRIEVALS
Clear eye-like
feature
RAD
NCEP
analyses
Too broad
wrt to obs
AIRS Impact on precipitation analysis and
forecast
 Weather-produced precipitation is generally poorly
predicted by global models in the tropics.
 In addition to the problems of convective parametrizations,
model resolution and physics improvements, a more
accurate thermal representation of the tropical atmosphere
can produce better precipitation
 While the next-generation improvement must come from
direct precipitation assimilation, some benefit also arise
from AIRS temperature and moisture profile assimilation
obtained under cloudy conditions.
 In the GEOS-5 experiments here described, the
`precipitation analysis’ described does not come from
precipitation assimilation but from a set of very short term
forecasts strongly constrained by observations (corrector
sequence)
OBS
CNTRL
RAD
Precipitation ``analysis’’ for
Helene (2006)
Not a true precipitation analysis since no
precip data are assimilated. Precip comes
from the `corrector sequence’ and is
essentially a set of very short term
forecasts strongly constrained by
observations.
The assimilation containing AIRS
retrievals, besides improving Helene’s
structure, also produces the best
accumulated precipitation
RET (st)
RET (t)
Zhou, Y., W. K. Lau, O. Reale, R.
Rosenberg, 2010: AIRS Impact on
precipitation analysis and forecast of
tropical cyclone in a global data
assimilation and forecasting system.
Geophys. Res. Lett., 37, L02806,
doi.1029/2009GL041494
Precipitation Forecast for
Helene
Precipitation forecast computed along
track and validated with SSM/I data.
Ingestion of AIRS retrievals cause the
GEOS-5 to have best skill. Improvement
with respect of CNTRL caused by AIRS
retrievals is about 30%,
radiances only 15% for 1-day forecasts.
Overall skill is very good in the 1-day
forecasts, reasonable at day 2, but drops
at day 3.
Zhou et al., (2010)
AIRS cloudy retrievals impact the forecast of
Nargis structure more than clear radiances
Radiances: very poor structure:
Two unconnected convective systems
without a deep circulation
Radiances
Very weak
System, low
Vorticity
Retrievals: Realistic 2-band
structure comparing well with
satellite
Retrievals
Much
higher
(100%)
Vorticity
OBS
CNTRL
AIRS RAD
AIRS RET
Precipitation
``Analysis’’
for Nargis
Not a true precipitation analysis
since no precip data are
assimilated. Precip comes from
the `corrector sequence’ and is
essentially a set of very short
term forecasts strongly
constrained by observations.
The assimilation containing AIRS
retrievals –which improves
Nargis structure- also produces
the best precipitation `analysis’.
Validation is made against
SSM/I, AMSU and TMI data
Zhou et al., (2010)
Precipitation Forecast for
Nargis
Forecasts computed along track and
validated with SSM/I data.
Ingestion of AIRS retrievals cause the
GEOS-5 to have better skill. Improvement
with respect of CNTRL caused by AIRS
cloudy retrievals (tight QC) is about 20%.
The impact of radiances is negligible.
Overall skill is very good in the 1-day
forecasts. Skill still reasonable at day 3.
Since the largest amount of
casualties caused by Nargis were
due to FLOODs,
this result has prominent
implications
Zhou et al., (2010)
also show consistent AIRS impact
on Wilma (2005)
Extreme precipitation over Indus River Valley
(July-Aug 2010 floods)
All operational
models missed
the precip max
over the Upper
Indus Valley
However,
AIRS cloudy retriev.
improve acc. prec.
along the central
part of the Indus
Valley with respect
to radiance assim.
Conclusions (Part III)
 Sets of data assimilation experiments without AIRS, with AIRS
cloudy retrievals (at two different quality controls) and with AIRS
clear-sky radiances were produced for boreal winter, spring, two
summers and fall conditions, for a total of about 600 days; 5- or
7-day forecasts are produced from all sets of analyses, for a
total of about 600 forecasts
 The overall skill of forecasts initialized from analyses in which
retrievals are assimilated is higher in every season
 Consistent improvements in the analysis of Tropical Cyclones
are noted as a consequence of AIRS retrievals ingestion
 The improvements affect FORECAST TRACK, TC structure
and EXTREME PRECIPITATION FORECASTS
 The importance of not rejecting AIRS-derived information
from cloudy areas becomes even more evident
Part IV: Other improvements.
One of many possible model improvements:
interactive aerosol in the GEOS-5
 An interactive aerosol capability based on Colarco et al
(2009, JGR) implemented in the NASA GEOS-5 by Arlindo
da Silva and collaborators
 Experiments to cover the NAMMA period (15Aug-15Sep
2006)
 Five sets of 30 5-day forecasts with a) no aerosol, b)
climatologically varyiing aerosol, c) interactive aerosol, at
two different global resolutions (0.5 and 0.25 deg)
 Results in: Reale, O. K. M. Lau, and A. da Silva (2011):
Impact of interactive aerosol on the African Easterly Jet in
the NASA GEOS-5 global forecasting system. Weather and
Forecasting, in press.
Interactive aerosol in the NASA GEOS-5
Good forecast of aerosol transport
Temperature impact (shaded) induced by
interactive aerosol (conc. solid black).
Vertical section at 10W
From Reale et al. (2011)
Wind forecast improvement up to 108-hours for the
same soundings in Cape Verde taken during
NAMMA SOP-3 (2006)
LEFT: Temper (obs,
verif analysis, contrl
fcst); induced temp.
anomaly (clim aer minus
cntrl, and inter. aer.
minus cntrl 108 hour
forecasts)
RIGHT:
Wind profiles:
Obs, verif analysis,
interactive aerosol,
climatological aer.
Reale et al. (2011)
Conclusions and Future
How can we improve
weather forecasts in the tropics?
 The analysis of the global atmosphere is still very deficient
in the tropics
 The largest number of victims by any natural catastrophe
are caused by extreme weather events in the tropics (firstly
freshwater floods caused by TCs).
 Current high-resolution global models start to resolve some
of the features of weather systems in the tropics
 Bad initialization hinders model performance
 Improvements in models (resolution, more sophisticated
treatments of processes, parametrizations) are important
 Improvements in analyses stemming from a more efficient
use of existing sensors (e.g. AIRS) are equally important
 Improvements deriving from future sensors (GPM, ISSWL,
etc) can be evaluated with OSSEs and could drive our
next-generation weather forecasting ability.
Acknowledgments
 Donald Anderson (NASA HQ) for past support to proposal ``Observing
System Experiments ’(OSE and OSSE) to evaluate and enhance the
impact of current and future satellite observations’ (PI: Oreste Reale,
2006-2009)
 Ramesh Kakar (NASA HQ) for current support to proposal
``Relationships among precipitation characteristics, atmospheric water
cycle, climate variability and change’’ (PI: W. K. Lau, 2009-2011)
 Ramesh Kakar (NASA HQ) for new support to proposal ``Using AIRS
data to understand processes affecting tropical cyclone structure in a
global data assimilation and forecasting framework’’ (PI: Oreste Reale,
2011-2013)
 Tsengdar Lee (NASA HQ) for generous allocations of NASA High End
Computer resources
 AIRS team at JPL and the Sounder Research Team at NASA GSFC
Reale, O., J. Terry, M. Masutani, E. Andersson, L. P. Riishojgaard, J. C. Jusem, 2007: Preliminary evaluation of
the European Centre for Medium-Range Weather Forecasts (ECMWF) Nature Run over the Tropical
Atlantic and African Monsoon region. Geophysical Research Letters, 34, L22810,
doi:10.1029/2007GL31640.
Reale, O., J. Susskind, R. Rosenberg, E. Brin, E. Liu, L. P. Riishojgaard, J. Terry, J. C. Jusem, 2008: Improving
forecast skill by assimilation of quality-controlled AIRS temperature retrievals under partially cloudy
conditions. Geophysical Research Letters, 35, L08809, doi:10.1029/2007GL033002.
Reale, O., W. K. Lau, J. Susskind, E. Brin, E. Liu, L. P. Riishojgaard, M. Fuentes, R. Rosenberg, 2009: AIRS
Impact on the Analysis and Forecast Track of Tropical Cyclone Nargis in a global data assimilation and
forecasting system. Geophysical Research Letters, 36, L06812, doi:10.1029/2008GL037122.
Wu, M.-L, O. Reale, S. Schubert, M. J. Suarez, R. Koster, P. Pegion, 2009: African Easterly Jet: Structure and
Maintenance. Journal of Climate, 22, 4459-4480.
Reale, O., W. K. Lau, K.-M. Kim, E. Brin, 2009: Atlantic tropical cyclogenetic processes during SOP-3 NAMMA in
the GEOS-5 global data assimilation and forecast system. Journal of the Atmospheric Sciences, 66, 35633578.
Zhou, Y., W. K. Lau, O. Reale, R. Rosenberg, 2010: AIRS Impact on precipitation analysis and forecast of
tropical cyclones in a global data assimilation and forecasting system. Geophysical Research Letters, 37,
L02806, doi.1029/2009GL041494.
Reale, O., and W. K. Lau, 2010: Reply to Comment on: `Atlantic tropical cyclogenetic processes during SOP-3
NAMMA in the GEOS-5 global data assimilation and forecast system.‘ Journal of the Atmospheric Sciences,
67, 2411-2415.
Reale, O., W. K. Lau, and A. da Silva, 2011: Impact of interactive aerosol on the African Easterly Jet in the
NASA GEOS-5 global forecasting system. In press on Weather and Forecasting.
Wu, M.-L, O. Reale, S. Schubert, M. J. Suarez, C. Thorncroft, 2011: African Easterly Jet: barotropic instability,
waves and cyclogenesis. Conditionally accepted on Journal of Climate.