1 the future of nwp stephen j. lord ncep environmental modeling center emc senior staff fred toepfer...
Post on 19-Jan-2016
219 Views
Preview:
TRANSCRIPT
1
The Future of NWP
Stephen J. LordNCEP Environmental Modeling Center
EMC Senior StaffFred Toepfer John Derber
Hua-Lu Pan Ken MitchellGeoff DiMego Naomi Surgi
D. B. Rao
2
Overview• Why have we been so successful?
• What can we do for an encore?
• Shortfalls (or what do we need to do better)?
Caveat
• Mostly a personal perspective, colored by experience at EMC, NCEP…NOAA
3
Why have we been so successful?
• Improved technology (computing, data assimilation & modeling techniques, obs)
• Societally-relevant products with a demand for– Improved product performance– Increased product areas
• Focused goals with quantitative scores– Systems evaluated every day
• Vs obs by weather & climate experts
• By diverse users with a lot at stake
Save Lives & PropertyWeather-Sensitive Commerce ($2+ T)
4
Precipitation Forecast Day 1 1"
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Th
rea
t S
co
re
NGM
ETA
GFS
HPC
CorrelationsOf HPC with:
Eta: 0.99GFS: 0.74NGM: 0.85
1 day of QPFskill gainedevery 25 years
Impact of NCEP Models on HPC Precipitation Forecasts
5
Why have we been so successful? (cont)
• Competition– With ourselves
– Across international, operational and research, weather & climate forecast centers
• Diverse approaches– Highly accurate NWP analysis & forecast systems with
different approaches
– No single solution (normalized for available resources)
– Very few “breakthroughs” (although 4D-VAR, “physics”, “vertical coordinate” are some individual reasons for success)
6
What can we do for an encore?
1. Continue to exploit the systems we have built*
2. Increase the rate of development for possible operational implementation*
*To be discussed further
7
What can we do for an encore?1. Continue to exploit the systems we
have built. Increase the– Range of Skillful forecasts*– Number of Useful products*– Increase the
• Available information used (data assimilation)• Useful information produced
– Probabilistic information (e.g. ensembles)
• Information used from all products– Product accessibility* – User education & training
8Tra
nsp
ort
atio
nT
ran
spo
rta
tion
Tra
nsp
ort
atio
nT
ran
spo
rta
tion
Fo
reca
st
Lea
d T
ime
Fo
reca
st
Lea
d T
ime
Warnings & Alert Warnings & Alert CoordinationCoordination
WatchesWatches
ForecastsForecasts
Threats Assessments
GuidanceGuidance
OutlookOutlook
Pro
tect
ion
of
Pro
tect
ion
of
Life
& P
rop
ert
yL
ife &
Pro
pe
rty
Pro
tect
ion
of
Pro
tect
ion
of
Life
& P
rop
ert
yL
ife &
Pro
pe
rty
Sp
ace
S
pa
ce
Op
era
tion
Op
era
tion
Sp
ace
S
pa
ce
Op
era
tion
Op
era
tion
Re
cre
atio
nR
ecr
ea
tion
Re
cre
atio
nR
ecr
ea
tion
Eco
syst
em
Eco
syst
em
Eco
syst
em
Eco
syst
em
Sta
te/L
oca
l S
tate
/Lo
cal
Pla
nn
ing
Pla
nn
ing
Sta
te/L
oca
l S
tate
/Lo
cal
Pla
nn
ing
Pla
nn
ing
En
viro
nm
en
tE
nvi
ron
me
nt
En
viro
nm
en
tE
nvi
ron
me
nt
Flo
od
Miti
ga
tion
F
loo
d M
itig
atio
n
& N
avi
ga
tion
& N
avi
ga
tion
Flo
od
Miti
ga
tion
F
loo
d M
itig
atio
n
& N
avi
ga
tion
& N
avi
ga
tion
Ag
ricu
lture
Ag
ricu
lture
Ag
ricu
lture
Ag
ricu
lture
Re
serv
oir
Re
serv
oir
Co
ntr
ol
Co
ntr
ol
Re
serv
oir
Re
serv
oir
Co
ntr
ol
Co
ntr
ol
En
erg
yE
ne
rgy
En
erg
yE
ne
rgy
Co
mm
erc
eC
om
me
rce
Co
mm
erc
eC
om
me
rce
Benefits
Hyd
rop
ow
er
Hyd
rop
ow
er
Hyd
rop
ow
er
Hyd
rop
ow
er
Fire
We
ath
er
Fire
We
ath
er
Fire
We
ath
er
Fire
We
ath
er
He
alth
He
alth
He
alth
He
alth
Forecast Forecast UncertaintyUncertaintyForecast Forecast UncertaintyUncertainty
MinutesMinutes
HoursHours
DaysDays
1 Week1 Week
2 Week2 Week
MonthsMonths
SeasonsSeasons
YearsYears
Initial Conditions
Boundary Conditions
Seamless Suite of Forecasts
Range of skillful forecasts
NOW
Increase the Range of Skillful Forecasts
9Tra
nsp
ort
atio
nT
ran
spo
rta
tion
Tra
nsp
ort
atio
nT
ran
spo
rta
tion
Fo
reca
st
Lea
d T
ime
Fo
reca
st
Lea
d T
ime
Warnings & Alert Warnings & Alert CoordinationCoordination
WatchesWatches
ForecastsForecasts
Threats Assessments
GuidanceGuidance
OutlookOutlook
Pro
tect
ion
of
Pro
tect
ion
of
Life
& P
rop
ert
yL
ife &
Pro
pe
rty
Pro
tect
ion
of
Pro
tect
ion
of
Life
& P
rop
ert
yL
ife &
Pro
pe
rty
Sp
ace
S
pa
ce
Op
era
tion
Op
era
tion
Sp
ace
S
pa
ce
Op
era
tion
Op
era
tion
Re
cre
atio
nR
ecr
ea
tion
Re
cre
atio
nR
ecr
ea
tion
Eco
syst
em
Eco
syst
em
Eco
syst
em
Eco
syst
em
Sta
te/L
oca
l S
tate
/Lo
cal
Pla
nn
ing
Pla
nn
ing
Sta
te/L
oca
l S
tate
/Lo
cal
Pla
nn
ing
Pla
nn
ing
En
viro
nm
en
tE
nvi
ron
me
nt
En
viro
nm
en
tE
nvi
ron
me
nt
Flo
od
Miti
ga
tion
F
loo
d M
itig
atio
n
& N
avi
ga
tion
& N
avi
ga
tion
Flo
od
Miti
ga
tion
F
loo
d M
itig
atio
n
& N
avi
ga
tion
& N
avi
ga
tion
Ag
ricu
lture
Ag
ricu
lture
Ag
ricu
lture
Ag
ricu
lture
Re
serv
oir
Re
serv
oir
Co
ntr
ol
Co
ntr
ol
Re
serv
oir
Re
serv
oir
Co
ntr
ol
Co
ntr
ol
En
erg
yE
ne
rgy
En
erg
yE
ne
rgy
Co
mm
erc
eC
om
me
rce
Co
mm
erc
eC
om
me
rce
Benefits
Hyd
rop
ow
er
Hyd
rop
ow
er
Hyd
rop
ow
er
Hyd
rop
ow
er
Fire
We
ath
er
Fire
We
ath
er
Fire
We
ath
er
Fire
We
ath
er
He
alth
He
alth
He
alth
He
alth
Forecast Forecast UncertaintyUncertaintyForecast Forecast UncertaintyUncertainty
MinutesMinutes
HoursHours
DaysDays
1 Week1 Week
2 Week2 Week
MonthsMonths
SeasonsSeasons
YearsYears
Initial Conditions
Boundary Conditions
Range of skillful forecasts
FUTURE
Seamless Suite of Forecasts
Increase the Range of Skillful Forecasts
10
Increase the Range of Skillful ForecastsS/I Climate
The new NCEP Coupled atmosphere-ocean Forecast System (CFS)
Componentsa) T62/64-layer version of the current NCEP atmospheric GFS (Global Forecast System) model and
b) 40-level GFDL Modular Ocean Model (MOM, version 3)
c) Global Ocean Data Assimilation (GODAS)
Notes:• CFS has direct coupling with no flux correction• GODAS
– Implemented September 2003, runs daily– Salinity analysis, improved use of altimeter data– Real time global ocean data base in WMO standard format– Ready for GODAE
11
12
AC=.86
AC=.80
AC=.43
Peitao Peng CPC
Tropical Precipitation Performance
13
Increase the Number of Useful Products
• Real-time Ocean• Air Quality• Fire Weather• Homeland Security• Seasonal• Monthly
• Systems sensitive to environmental parameters
– Ecosystems– Disease vectors– Agriculture– Routine Reanalysis and assessment
NCEP Production SuiteWeather, Ocean & Climate Forecast Systems
Version 3.0 April 9, 2004
0
20
40
60
80
100
0:00 0:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 4:30 5:00 5:30 6:00
6 Hour Cycle
Per
cent
Use
d
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
US GODAE: Global Ocean Prediction with HYCOM
• Goal: to develop and demonstrate real-time, operational, high resolution ocean prediction systems for the Global Oceans and Basins
• NCEP Partners with
• University of Miami/RSMAS
• NRL Stennis, NRL Monterey, FNMOC
• NOAA PMEL, AOML
• Los Alamos National Laboratory
• Others (international, commercial)
• Hybrid isopycnal-sigma-pressure ocean model (called Hybrid Coordinate Ocean Model – HYCOM)
• Funded FY 2003-2007 by NOPP
Chesapeake Bay
ScheduleNorth Atlantic
World Oceans
North-East Pacific
Hawaii
2005 2006 2007
Global atmosphere-ocean Coupling and Hurricane-Ocean Coupling
Initiate interactions with NOS on bay and estuary model boundary conditions; Initiate wave-current interactions.
16
Future Enabling Architectures
• Adding Model Components (to increase useful products)– ESMF
• Complexity vs computational efficiency
• Product accessibility*– NOMADS
17
ESMF Architecture
Components Layer:Gridded ComponentsCoupler Components
ESMF Superstructure
ESMF Infrastructure
User Code
BLAS, MPI, NetCDF, …
Low Level Utilities
Fields and Grids Layer
Model Layer
External Libraries
1. ESMF provides an environment for assembling geophysical components into applications.
2. ESMF provides a toolkit that components use to
i. increase interoperability
ii. improve performance portability
iii. abstract common services
18
RT-NOMADSDistribution of
Real-Time and Retrospective NCEP Model Data Sets
Jordan C. Alpert jordan.alpert@noaa.gov 4/22/03
NOMADSNOAA Operational Model Archive
and Distribution System
On demand access to (x, y, z, t, product) spacedownloaded in user-defined format
19
What can we do for an encore? (cont)
1. Continue to exploit the systems we have built
2. Increase the rate of development for possible operational implementation
– Improvements must occur simultaneously for many more applications (waves, hurricanes, precip, aviation, week-2)* • Each improvement gives rise to increasing expectations
– The problems are getting tougher• As perfection is approached• Forecast system output increasingly resembles the atmosphere• Forecast and delivery deadlines shrinking
– It is more difficult to predict the expected improvement from each proposed change
20
2001 GFS Implementation• Improved model climate in tropics
– Prognostic liquid water– Radiation interactive with condensed water & cloudiness– Full simulation of water transport
• PBL Convective clouds Detrainment Cirrus
– Cumulus momentum transport• Reduced spurious spinup of tropical systems• Cyclogenesis mostly confined to growth of systems later observed
• Testing involved– 835 days of retrospective data assimilation & model forecasts
• Summer (tropical cyclones)• Spring (severe weather)• Winter (temperature bias)
– Unable to test GFDL initialization thoroughly
21
What can we do for an encore? (cont)
2. Increase the rate of development for possible operational implementation
HOW?
22
What can we do for an encore? (cont)
HOW?Focus efforts on improving operational systems
– Improved project management• Operations needs to have increased influence on scientific direction of
applied research• Mutual discussion and execution of highest priority development projects• More rapid transition of development to operations
– Increase computing resources beyond Moore’s Law (constant $)– Consolidate software & forecast systems
• GSI (global and regional analysis system)
– Continue to exploit Test Bed concept• Enhanced Visiting Scientist exchange program • Increase the (potential) workforce capabilities
– Student education and training
• Support for non-operational users of operational systems
– Examples follow
Strong management support at NOAA level & above
23
NASA-NOAA-DOD Joint Center for Satellite Data Assimilation
(JCSDA)– NOAA, NASA, DOD partnership– Mission
• Accelerate and improve the quantitative use of research and operational satellite data in weather and climate prediction models
– Current generation data
– Prepare for next-generation (NPOESS, METOP, research) instruments
– Supports applied research• Partners
• University, Government and Commercial Labs
24
JCSDA Prioritized Applied Research Areas
• Advanced radiative transfer
• Clouds and precipitation
• Assess impacts of current instruments
• Improve sea surface temperature data and use of altimeter data
• Enhance land surface data sets (surface emissivity model)
• AIRS data implementation
25
Improve Sea Surface Temperature Data [X. Li & Derber]
SST Difference 29-28 October 2003 - Experiment
SST Difference 29-28 October 2003 - Control• New physical retrieval from AVHRR data, cast as variational problem• OPTRAN RTM & Linear Tangent Model• Eventual direct use of AVHRR (and other) radiance data
RMS and Bias Fits to Independent Buoy SST Data
NOAA-16 AVHRR data only
Northern Hemisphere Ex. Tropics
26
Improved Surface Emissivity Model for Snow [Yan, Okamoto and Weng)
Annual Mean RMS TB Difference (Obs – Simulated)
SnowEM
Operational
The Path to Operational Implementation
Code or Algorithm Development & RefinementRepeated case studies, proof of concept, eliminate bugs
Interface with Operational Codes & Data StructuresConnect input/output to BUFR/GRIB, develop backup version,Make code robust & efficient to fit time/cpu/memory window
Preliminary TestingLow resolution case studies, static initialization, relevant diagnostics,
warm & cool cases, assess short-term model climate (30 days)
Low Resolution Parallel Testing Connect to fully cycled data assimilation, run for all seasons,
accumulate verification statistics, identify&solve problems:e.g. biases, amplification through cycling, spin-up/down etc.
Pre-Implementation TestingOperational resolution fully cycled real-time parallel, more
comprehensive verification, documentation and user notification, real-time forecaster exposure/evaluation
EMCDTC/OTC
1-12+ 3-12
1-12 + 3-12
3-6 1-3
3-9
3-9
1-3
13-39
1-6
1-6
2-44-182-92-4
EMC total: 13-39 mo DTC-OTC-EMC total: 8-31 moWRF DTC
28
Shortfalls (or what do we need to do better)?
• Resources have not kept pace with the rapidly increasing complexity of today’s forecasts– Project management and technical support for
• Maintaining and developing complex operational Data Assimilation & Modeling codes and supporting code infrastructure*
• Interacting with external community (data, ideas, code transition, cultural education)*
• Example Hurricane WRF*
– Basic infrastructure (computing, testing & implementation capability)
– Timeliness of data delivery to operational centers and efficient product dissemination are marginal
– Recent additions to NOAA’s computing will help but…
29
30
Maximum Significant Wave Heights: Model vs. JASON
Direct hits: Altimeter through eye and maximum wavesWNA (green), NAH (red): Good track of build-up, set-down and maximumStorm’s eye (lower panel) well captured by both modelsEarly stages missed by WNA (green): weak GFS winds, small hurricanes
31
Genesis Summary NCEP GFS 2003 Atlantic Hurricane Season
Selected Storms (Isabel, Juan, Kate, Nicholas, Odette, Peter)
Total or Percent
Total Genesis Opportunities 92
Early or on time 42%
Late 27%
Failed to generate 31%
Timing Within
+- 6 h 13%
+- 12 h 41%
+- 24 h 67%
> 24 h 33%
Location Within
300 km 47%
500 km 77%
>500 km 23%
TimMarchokQingfu
Liu
32
GFDL(WRF)Hurricane Model
NCEPWAVEWATCH III
Flux
Wind &Air Temp.
Atmosphere
Ocean Waves
Currents,
Wave Boundary Model
SST
Wave Spectra
Flux
OceanModel
Wind &Air Temp.
Flux
GFDLHurricane Model
POM
SST &Current
SST
Flux
Atmosphere
Ocean
Elevations, & SST
Operational GFDL ModelOperational GFDL Model Future Coupled Hurricane-Wave-Ocean ModelFuture Coupled Hurricane-Wave-Ocean Model
URI & U. Miami partnerships
33
Shortfalls (or what do we need to do better)? (cont)
• A solid scientific strategy and policy are needed to guide an expanded and improved observing system– Evaluation of today’s system (e.g. current JCSDA
assessment)*– Ways to assess potential impact of new instruments
(e.g. OSSEs)*– Examples follow*– Recent activities are encouraging
• Ocean observing system for climate is step in right direction• Support for above projects is helping• Still a long way to go
34
Tropics 850 mb Vector Difference 20N - 20S (F-A) RMS 15 Jan - 15 Feb '03
0
1
2
3
4
5
6
7
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Forecast [days]
RM
S [
m/s
] '
control
no amsu
Tropics 850 mb Vector Difference 20N - 20S (F-A) RMS 15 Jan - 15 Feb '03
0
1
2
3
4
5
6
7
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Forecast [days]
RM
S [
m/s
] '
control
no hirs
Tropics850 mb Vector (F-A)RMS
The REAL problem is Day 1
Jung and Zapotocny
JCSDAFunded by
NPOESS IPO
35
Impact of Removing AMSU and HIRS Data on Hurricane Track Forecasts in East Pacific Basin
-25.0
-20.0
-15.0
-10.0
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
30.0
12 24 36 48 72
% I
mp
rov
em
en
t
NOAMSU
NOHIRS
Impact of Removing AMSU and HIRS Dataon Hurricane Track Forecasts in Atlantic Basin
-20.0
-15.0
-10.0
-5.0
0.0
5.0
10.0
15.0
12 24 36 48 72 96 120
% I
mp
rov
em
en
t
NOAMSU
NOHIRS
Satellite data ~ 10% impact
Jung and Zapotocny
JCSDAFunded by
NPOESS IPO
36
RMS Brightness Temperature Differences Between Observed Radiances and NCEP 6 Hour Global Forecast and Analysis
Moisture & Surface Channels
Temperature Channels
37
Doppler Wind Lidar (DWL) ImpactConv Only
Conv. + TOVS
Conv + TOVS + DWL(best)
Conv + DWL(non-scan)
Conv + DWL(PBL )
Conv + TOVS + DWL(non-scan)
Conv +DWL(Best)
Conv + DWL(Upper)
V at 200 hPa
V at 850 hPa
4
-4
4
-4
8
8
00
0
Time averaged anomaly correlations between forecast and NR for meridional wind (V) fields at 200 hPa and 850 hPa. Anomaly correlation are computed for zonal wave number from 10 to 20 components. Differences from anomaly correlation for the control run (conventional data only) are plotted.
Forecast hour
OSSE Results – Masutani et al
38
Opinionated Summary(Any opinion can be debated and “proven” wrong)
• Options for product improvement– More observations (needs more focus on anticipated
product improvement)– Field experiments (need more focus on understanding the
forecast system and correcting its errors)– Computational techniques (high priority)– Scientific development (high priority, more innovative
approaches, e.g. “resolvable scale modeling”*)– Testing, engineering and tuning (always necessary)
* Unfeasible unless huge increases in computing
39
Opinionated Summary (cont)
• Product enhancement– Increased societal impact– Broader spectrum of applications
• Research Strategy– Some course corrections needed
• Focus efforts more on improving operational systems• Involve more scientists without detracting from current rate of
development
• Obviously more infrastructure and computing needed• Strong management support at NOAA level & above
40
Opinionated Summary (cont)
The last word:
• Everyone still has a role to play• Sociology and our past may be greater enemies than
the science yet to be conquered*
*On runway, Birmingham AL airport, 10:15 PM, 12% laptop juice“Thunderstorms in Baltimore area have created a traffic jam”
top related