Upload
bartholomew-parrish
View
224
Download
1
Embed Size (px)
DESCRIPTION
3 Preparing for the Future (1) Improved forecast services –Greater focus on high-impact events –Additional environmental information service responsibilities –Provide more information to users and access to more info Support forecast offices –Efficient Grid Initialization (e.g. SmartInit) –Analysis of Record (and RTMA) –Probabilistic and ensemble methods Respond to external (NRC) reports –“Completing the Forecast” –“Fair Weather” Respond to NOAA Science Advisory Board reviews –Ocean modeling (National “backbone”) –Hurricane intensity (ensemble-based system)
Citation preview
1
The Proposed Next-Generation
NCEP Production SuiteEMC Senior Staff
January 2007
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
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
2
Overview
• Preparing for the future• Production Suite: Conceptual
prototype• Benefits• Summary
3
Preparing for the Future (1)
• Improved forecast services– Greater focus on high-impact events– Additional environmental information service responsibilities– Provide more information to users and access to more info
• Support forecast offices– Efficient Grid Initialization (e.g. SmartInit)– Analysis of Record (and RTMA)– Probabilistic and ensemble methods
• Respond to external (NRC) reports– “Completing the Forecast” – “Fair Weather”
• Respond to NOAA Science Advisory Board reviews– Ocean modeling (National “backbone”)– Hurricane intensity (ensemble-based system)
4
Preparing for the Future (2)
• Observations (number and availability)– Advanced Polar and Geostationary sounders (~100 X greater)
• NPOESS (<60 minutes globally) – 2012-2015 (or later)• METOP (1-4) – 2007 • NPP (90-120 minutes globally) – 2009• GOES-R – 2013 (or later)
– Next-generation Doppler radar• Advanced post-processing techniques for multi-model
ensemble (e.g. NAEFS project)– Bias correction– 2nd moment correction– CPC “consolidation” to quantify “value-added”
• Advanced dissemination strategies– E.g. NOMADS (“Fat server/Thin Client” technology)
• Next-Generation Air Traffic-control System (NGATS)– Geographically consistent solutions– Global to terminal scales– At least hourly updating globally
5
Preparing for the Future (3)• Three principals for moving forward
1. Maturing, ensemble-based, probabilistic systems offer the most potential benefits across wide spectrum of forecast services
“Model of the day” is not a scientifically supportable
solution for the future2. Ensemble composition
a. Managed component diversityb. Components must be institutionally supported (operational
or major research institution)3. Product delivery
a. Time is critical (perishable product)b. Information availability must be maximized
6
Production Suite: Conceptual Prototype
Products
• Three levels of information– Routinely delivered
1. Pointwise, single-valued, downscaled Most Likely Forecast from all available guidance on NDGD grid
2. Description of forecast uncertainty through probability density function (pdf)
– “On-demand” (via publicly accessible server)3. Individual ensemble member forecasts available• Prototype: NOMADS
7
Production Suite: Conceptual PrototypeApplication Areas
• Focus on high impact weather– Hurricane intensity (and track) and coastal impacts– Other “High impact” defined by
• Users• Type of event
– Goal: “warn on forecast” for highest resolution events (e.g. tornado)• Examples of other new applications
– Surface transportation (e.g. “winter weather”)– Environmental monitoring (AQ + Atmos. Constituents)– Ocean (HABs & ecosystems, fog & visibility, coastal inundation,
dynamic storm surge)– Hydrology (water quality, drought)
• Hourly updating– RTMA AOR (through Reanalysis/Reforecast)– Regional assimilation– Global assimilation+
+ if requirement and available computing and human resources
8
Production Suite: Schematic Overview
Model Region 1
Model Region 2
Global/Regional Model DomainAnalysis
• Concurrent execution of global and regional applications– More efficient execution of rapid updating
• In-core updating for analysis increments • Regional (CONUS, Alaska, Hawaii, Caribbean & Puerto Rico) • Global (if requirements and resources)
– All ensemble members may exchange information during execution• ESMF-based Common Modeling Infrastructure
9
Analysis--------------
OtherForecastSystems
Physics(1,2,3)
ESMF Utilities(clock, error handling, etc)
Post processor & Product GeneratorVerification
Resolution change
1-11-21-32-12-22-3
ESMF Superstructure(component definitions, “mpi” communications, etc)
Multi-component ensemble+
Stochastic forcing
Coupler
Dynamics(1,2)
Application Driver
ESMF* Compliant Component System
* Earth System Modeling Framework (NCAR/CISL, NASA/GMAO, Navy (NRL), NCEP/EMC)
2, 3 etc: institutionally (non-NCEP) supported
10
CFSMFS
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
WAV
CFS & MFS
GENS/NAEFSGFS
Next Generation PrototypePhase 4 - 2015
Regional
Rap Refresh
GlobalHUR
SREF
Reforecast
Hydro / NIDIS/FF
Hydro
NAM
GDAS
RDAS
RTOFS RTOFSAQ
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
AQ
Computing factor: 81
Concurrent• GFS*• NAM• SREF Hourly• GDAS• RDAS• Rapid Refresh Expanded• Hurricane capability (hires)• Hydro/NIDIS• Reforecast
* Earlier delivery of GFS concurrent combined products from NAM, GFS, SREF
11
CFS & MFS
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
CFSMFS
WAVGFSRegional
Rap Refresh
GlobalSREFReforecast
Hydro
NAM
GDAS
RDAS
RTOFS
RTO
FS
CFS & MFSAQ Hydro / NIDIS/FF AQ
GENS/NAEFS
>100% of 2015 computing
Next Generation PrototypeFinal – 2017+
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
GLOBAL NGATS
HU
R
Computing factor: > 240
ECOSYSTEMS
SPACE WEATHER
HENS
12
Summary of Benefits• GDAS and RDAS with advanced assimilation techniques
– Use high density time and space observations more completely– Prepare for NPOESS, METOP, next-generation radar obs– Provide potential for expanded Rapid Refresh capability to serve
NGATS• Concurrency and unified post-processing provide
– Earlier delivery of GFS products– Real time boundary conditions for regional systems– Global and regional ensembles giving complementary uncertainty
measures based on different physical mechanisms– Unified “D(prog)/Dt view from all guidance products
• Reanalysis/Reforecast capability– Enables maximum forecast skill and independent skill assessment– Provides
• Operationally supported probabilistic systems• Updated skill evaluation as forecast systems evolve
• Grouping of regional ocean applications allows– Hurricanes to use real-time ocean state– Coastal applications to run concurrently
13
Summary• Phased evolution of the NCEP Production Suite
– 2009-2015• Results in
– Improved services for high impact weather– Application of advanced data assimilation techniques for improved model
initial conditions– More efficient
• Use of computing• Incorporation of new product lines for improved services
– Earlier product delivery– More uniform and informative product stream
• Advanced ensemble suite including components supported outside NCEP• Improved statistical post-processing• Reforecast and Reanalysis become operationally supported
• Consistent with existing ESMF & global data assimilation development and interagency collaborations with– NASA– DOD– NCAR
14
ThanksQuestions?
15
Extras
16
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
GD
AS
GFS anal
NA
M anal
CFS
RTOFS
SREF NAM
AQ
GFSHUR
RD
AS
Current (2007)
GENS/NAEFS
Current - 2007
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
17
CFSMFS
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
WAV
CFS & MFS
GENS/NAEFSGFS
Next Generation PrototypePhase 4 - 2015
Regional
Rap Refresh
GlobalHUR
SREF
Reforecast
Hydro / NIDIS/FF
Hydro
NAM
GDAS
RDAS
RTOFS RTOFSAQ
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
AQ
18
Community-based Development• Strategy and roles:
– Focus on single component instead of entire model system
– Collaborative, not competitive– NCEP/EMC
• Maintains primary components for each part of Production Suite and for each application
• Supports ESMF applications in operations• In collaboration with community
– Integrates new ESMF-based components into operations– Performs final testing and preparation of upgrades of supported
components in operations– Collaborators
• Provide – Component upgrades to be tested in operational setting– Institutional support for their contributed components– Diversity and expertise complementary to operations
• Work through DTC, JCSDA, CTB, etc.
19
Conceptual Prototype:Numerical Forecast Guidance (1)
• Information should be optimally combined from all available sources– Domestic and international
models (global, e.g. NAEFS and regional)
– Same product format for all time scales (unified post-processing)
• Progress in numerical forecast system development should not be constrained by post-processing– Improved products come from
development of improved systems
Impact of Models on Day 1 Precipitation Scores
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Thre
at S
core
Human(HPC)
ETA
Linear(Human(HPC))Linear (ETA)
• Robust training and outreach program must– Accompany new probabilistic-
based system– Support NWS Field Operations,
commercial sector and international users
– Support advanced dissemination of forecast information on all time and space scales
20
Conceptual Prototype:Forecast System Development Areas
• Observations processing – JCSDA– Increased focus on quality control– Contribute to future observing system design
• Data assimilation– Coordinated development of advanced techniques
• Simplified 4-D Var (NCEP/EMC), annual updates thru 2008• “Classical” 4-D Var (with NASA/GMAO)• Ensemble Data Assimilation (with ESRL, UMD and others)• Development coordinated with NASA-NOAA-DOD JCSDA• 2008 decision 2010 implementation
– Better use of high time and space density, remotely-sensed data• Model accuracy (dynamics and physics)
– Hybrid vertical coordinate (sigma-pressure-theta)– Semi-lagrangian, Semi or Fully Implicit– Advanced radiation, shallow convection, deep convection– Stochastic forcing– Land surface tiling– Increased collaboration with community (Test Beds)
• Post-processing – Unified system across time and space scales and models– Includes bias correction and downscaling
21
NCEP’s Next Generation Operational Forecast System
YearsYearsWeeksWeeksMinutesMinutes DaysDaysHoursHours SeasonsSeasonsMonthsMonths
Type
of G
uida
nce
Warnings & Alert Warnings & Alert CoordinationCoordination
WatchesWatches
ForecastsForecasts
Threat Assessments
GuidanceGuidance
OutlookOutlook
Lead Time
Protection of Protection of Life/PropertyLife/Property
Flood mitigationFlood mitigationNavigationNavigation
TransportationTransportationFire weatherFire weather
HydropowerHydropowerAgricultureAgriculture
EcosystemEcosystemHealthHealth
CommerceCommerceEnergyEnergy
Initi
al C
ondi
tion
Sens
itivi
ty
Boundary Condition
SensitivityReservoir controlReservoir control
RecreationRecreation
Forecast Uncertainty
“Forecast Countdown for the Seamless Suite”
22
Concurrent execution of global and regional forecast models (2)
Model Region 1
Model Region 2
Global/Regional Model DomainAnalysis
Local Solution
• Real time boundary and initial conditions available hourly
– “On-demand” downscaling to local applications• Similar to current hurricane runs but run either
– Centrally at NCEP OR– Locally (B.C, I. C. retrieved from on-line data at NCEP)
• No boundary or initial conditions older than 1 hour – Flexibility for “over capacity” runs
• Using climate fraction must be planned• No impact on remainder of services
• Consistent solution from global to local with a single forecast system and ensembles providing estimate of uncertainty
23
ESMF Component Framework
Application
ChangeResolution
SurfaceCycling Atmosphere Post some other
Couplersome other Component
ATMDynamics
ATMPhysics Vertical Post Product
GeneratorOutput
GRIB/BUFR
24
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
GD
AS
GFS anal
NA
M anal
CFS
RTOFS
SREF NAM
AQ
GFSHUR
RD
AS
Data processingCurrent (2007)
GENS/NAEFS
Current - 2007
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
25
CFSMFS
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
Regional
Rap Refresh
GlobalRefcst
HydroHOURLY
RTOFSAQ
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
Ideal State
DATAASSIM
GARAGOROHU
RTOFS-ECOS
Hydro / NIDIS/FFAQ
HUR
SREF
GFS
GENS/NAEFS
WAV
NAM
MedRange
Refcst
Wk2 MFS & CFS
26
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
Reforecast
NA
M anal
CFS
SREF NAM
GFS
WAV
HUR
Next Generation PrototypePhase 1 - 2009
3-hourly GDAS (2)1-hourly RDAS (6)
GENS/NAEFS
RTOFSAQ
GFS A
nal
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
Added• 1-Hourly RDAS• 3-Hourly GDAS• Reanalysis/ Reforecast
Computing factor: 3
27
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
Reforecast
GFS A
nal
NA
M Anal
CFS & MFS
GFS
WAV
HUR
GENS/NAEFS
Next Generation PrototypePhase 2 - 2011
GDAS
SREF
RDAS
RTOFSHydro / NIDIS AQ
NAM
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
AQ
Added• Hydro/NIDIS products Moved• GFS ½ h earlier Expanded• Hurricane & wave products Incorporated• Multi-domain rapid updating
Computing factor: 9
28
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: Four Times/Day
Perc
ent U
sed
RUCFIREWXWAVESHUR/HRWGFSfcstGFSanalGFSensETAfcstETAanalSREFAir QualityOCEANMonthlySeasonal
Reforecast
GFS A
nal
NA
M Anal
CFS & MFS
GFS
WAV
HUR
GENS/NAEFS
Next Generation PrototypePhase 3 - 2013
GDAS
SREF
RDAS
RTOFS
NAM
NCEP Production SuiteWeather, Ocean, Land & Climate Forecast Systems
AQHydro / NIDIS/FFAQ
Computing factor: 27
Added• Flash flood products Moved• SREF concurrent to NAM Expanded• Reforecast capability
29
EVOLUTION of the Next-Generation NCEP Production Suite (1)
Phase Date ComputerPower*
Human Resources
Implementation of new services
1 2009 3 +8(2 - 2008 3- 20072 – 20081 - 2009)
Increased forecast accuracy with hourly RDAS/LDAS, 3 hourly GDAS/GLDAS (2), and Advanced Data
Assimilation (2010)Reanalysis-Reforecast (5)
HABs (1)
2 2011 9 +6(2 – 20071 – 20083 – 2009)
ESMF-based in-core system (5)Downscaled 4 Domain RR with Firewx
Global hourly Aviation productsLand-HYDRO-NIDIS seasonal products (1)
3 2013 27 +3(1- 20102 – 2011)
Concurrent NAM, SREFCoupled Land-Hydro & Flash Flood (FF) guidance (1)
Biogeochemical tracers (2)
* Relative to NCEP’s 2007 computer
30
EVOLUTION of the Next-Generation NCEP Production Suite (2)
Phase Date ComputerPower*
Human Resources
Implementation of new services
4 2015 81@ +5(2 – 2011)(3-2010)
Hourly GDAS+ Concurrent GFS+
Fully coupled global atmosphere-oceanAdvanced global ensemble system
Hurricane ensemble (1)@
Dynamic storm surge ensemble (1)@
NGATS support@
Final 2017(+)
240 +2(2 – 2015)
Concurrent GEFS/NAEFSHurricane ensemble
Dynamic storm surge ensembleFull ecosystem support (2)
NGATS support
* Relative to NCEP’s 2007 computer @ additional 3x computing upgrade in 2009required
+ If positive upgrade to services
31
Fig. 2. Schematic diagram illustrating the one-way flow of initial value related information in a traditional NWP forecast process.
Traditional NWP Process
32
Fig. 4. Schematic diagram illustrating the two-way flow of initial condition related information in the proposed new, integrated NWP forecast process.
Future NWP Process THORPEXPROGRAM
33
Primary & Secondary Components and Models
• Primary components– Observations ingest, processing and quality control– Data assimilation– Forecast model– Post-processing– Product delivery
• Primary model– Used in data assimilation cycle– Supported by EMC for NCEP’s operations
• Risk reduction• Optimum maintenance• Optimum enhancement
– Continued exposure to observations– Model improvements impact analysis and forecast
• Secondary model– Initialized from analysis– Not cycled– Must add value to operational system
• Skill• Diversity• Unique application
– Supported institutionally by external (to EMC) organization• E.g. Navy (through NUOPC)• First line of support at EMC (one person)
– Applications• Ensemble membership (managed diversity)• Not fully supportable by EMC (e.g. ecosystems)
34
The Environmental Forecast Process
Observations
Analysis
Model Forecast
Post-processed Model Data
Forecaster
User (public, industry…)
NumericalForecastSystem
Data Assimilation
35
EMP Model Strategy & ESMF• Concept of operations
– Single system for global and regional models– Performance permitting
• Migration to single model or• Multiple dynamics and physics options in single structure
– Single verification, observations data base obeying WMO standards
– Single analysis code– System perturbations from
• Model diversity• Stochastic physics (preferred)
• System supports both operational and research components– Dynamics– Physics
• Overall positive experience at Met Office• ECMWF maintains single model & data assimilation
system for global wx & short-term climate forecasting
36
Forecast System Development Strategy
• As far as possible, based on quantitative assessment:– A single physics– Applied across multiple scales– E. g. global weather and climate
• Possible extension to mesoscale• Successful for hurricane (GFS physics, NAM microphysics) GFDL)
• Excellent characteristics from all applications tested on other scales and implemented when ready
• Single model (and DA) structure makes this feasible• Ensemble application
– Requires approximately equal skill and system diversity– Secondary models and diverse components create manageable
system if ESMF compatible and institutional support• Disciplined competition fosters the best system
37
Overview of Future Jigsaw
• Enabled by ESMF• Hourly global data assimilation analyses• Hourly global rapid refresh
– Supports NGATS• Concurrent global and mesoscale and
mesoscale ensemble• Reforecast window• Target 2015+
38
Timing Summary (2015)Current (%) Future (%) Current Start
TimeFuture Start
TimeCurrent
End TimeFuture End
Time
GDAS 5.4 12.0 6:00 1:30 6:30 1:40
RDAS 3.2 7.2 5:30 1:30 6:00 1:40
GFSanal 4.4 0.8 2:55 1:30 3:20 1:40
NAManal 1.9 0.8 1:23 1:30 1:32 1:40
Rapid Refresh 2.2 4.8 1:25 1:30 1:47 1:47
NAMfcst 9.6 4.0 1:33 1:40 3:00 3:00
GFSfcst 12.8 6.5 3:21 1:40 4:40 3:30
GENS/NAEFS 7.1 7.6 3:23 3:30 5:30 6:00
SREF 7.8 7.6 0:30 1:30 1:22 3:40
HUR 7.7 7.8 4:30 3:40 4:30 6:00
RTOFS 7.7 7.2 xxxxx xxxxx xxxxx xxxxx
WAV 2.4 3.1 4:32 3:40 4:53 6:00
AQFS 2.4 2.1 2:00 2:30 5:00 5:00
CFS/MFS 25.4 17.1 xxxxxx xxxxx xxxxx xxxxx
Hydro-NIDIS 0 1.25 xxxxx xxxxx xxxxx xxxxx
Reforecast 0 11.5 None None
Bold=EarlierRed=Later
39
The ultimate target is a completed NOAA Framework of ESMF
Components within which NOAA scientists can work efficiently.
One solution is outlined in the next few slides.
A Project to Create the NOAA/NCEP Framework of
ESMF Components Mark Iredell
40
Proposed NOAA ESMF-based System Components (1)
• Change Resolution– Imports ATM state on one grid– Exports ATM state on another grid, possibly changing
variables and units• Surface Cycling
– Imports ATM boundary– Exports ATM boundary updated
• Atmosphere– Imports ATM state– Exports ATM state later in time
• Post– Imports ATM state– Exports selected products
41
• Dynamics– Imports ATM state– Exports ATM state later in time
• Physics– Imports ATM state– Exports ATM state adjusted
Proposed NOAA ESMF-based System Components (2)
42
• Vertical Post– Imports ATM state– Exports fields on selected levels but model grid
• Product generator– Imports fields on one grid– Exports fields on another grid as requested
• Output– Imports fields– Writes GRIB2 files (or BUFR or netCDF)
Proposed NOAA ESMF-based System Components (3)
43
• Independent validation for each component that anyone can theoretically run
• Examples– Dynamical core
• Held-Suarez, etc.– Physics
• Single column, etc.– Atmosphere
• NWP and climate verifications and diagnostics– GFS with GDAS
• Full parallel validation
Proposed NOAA ESMF-based System Components (3)
44
Other Components
Other components that could be coupled to the global atmospheric model using ESMF:
• Obs. Processing• Variational Analysis• Ensemble Members• Mesoscale Model• Storm Model
• Ocean Model• Ice Model• Hydrology Model• Chemistry Model• Space Model
45
Model FrameworkTime and People costs
• Estimated cost to NOAA for building:– 4 years– 17 man-years
• Maintenance cost unknown– Multiple major developers– Possible lack of coordination– Coordination may slow development and
system enhancement
46
Code/Algorithm Assessment and/or Development
Transition Steps (Modeling)Identification for Selection1
2
Interface with Operational Codes3
Level I: Preliminary Testing (Lower Resolution)4
Level II: Preliminary Testing (DA/Higher Resolution)5
EMC Pre-Implementation Testing (Packaging/Calibration)6
NCO Pre-Implementation Testing7
Implementation/Delivery8
47
EMCNCO
R&D Operations Delivery
Criteria
Transition from Research to Operations
Requirements
EMC
NCEP’s Role in the Model Transition Process
OPS Life cycleSupport
Service Centers
NOAAResearch
Concept of Operations
ServiceCenters
Test BedsJCSDA
CTBDTCJHT
User
Obs
erva
tion
Sy
stem
Launch List – Model Implementation Process
FieldOffices
Effort
EMC and NCO have critical roles in the transition from NOAA R&D to operationsOther Agencies
&International
Forecast benefits, Efficiency, IT Compatibility, Sustainability
48
Component Requirements• Code standards and documentation
– ESMF interfaces fully described– Lightweight so as to not impede research
• Some compile-time and run-time flexibility– Somewhat flexible processor and thread layout– Somewhat flexible memory indexing layout– Somewhat flexible import and export fields
• Standard metadata– CF convention– GRIB2, etc.
• Distributed grid support under ESMF to enable coupling• Standalone validation
49
Given this solution is acceptable, how can we manage to get there?
One framework project plan is given in the next few slides.
50
Steps in creating model frameworka) Planning and coordinationb) GFS I/O componentsc) GFS ATM gridded componentd) GFS physics componente) GFS dynamics componentf) UMO dynamics componentg) GFS post componenth) Change resolution componenti) Surface cycling componentj) Coupler componentsk) Land componentl) Ocean componentm) Ice componentn) Aerosol componento) Ionosphere componentp) Mesoscale componentsq) GSI componentr) Configuration management
51
Substeps in model framework (1)a) Planning and coordination (1.2)
1. Unified Model Infrastructure Group (UMIG)2. coordinate with EMC and other NOAA groups
b) GFS I/O components (0.2)1. I: input analysis, export distributed grid2. O: import distributed grid, export history file3. ESMF-ize
c) GFS ATM gridded component (0.2)1. import and export distributed reduced Gaussian grid2. internal state is both spectral and grid3. faster than current GFS4. keep dynamics and physics embedded at first5. ESMF-ize
d) GFS physics component (0.2)1. combine radiation and other physics2. standardize import and export states3. run on any set of points4. run with or without land model5. ESMF-ize
52
Substeps in model framework (2)e) GFS dynamics component (0.2)
1. import and export distributed reduced Gaussian grid2. all spectral data and transforms are internal3. may have two phases (two modes of invocation)4. ESMF-ize
f) UMO dynamics component (0.2)1. Import and export distributed UMO grid2. may have two phases3. ESMF-ize
g) GFS post component (1.0)1. adapt WRF post to GFS2. import distributed model grid3. write data in GRIB24. ESMF-ize
h) Change resolution coupler component (0.2)1. import one ATM model state and convert to another2. ESMF-ize
i) Surface cycling component (0.2)1. adapt current surface cycling component2. ESMF-ize
53
Substeps in model framework (3)j) Coupler components (4.0)
1. between ATM and ice and ocean for CFS2. adapted from Sheinin’s coupler3. ensemble coupler4. other couplers as well – generic coupler?5. ESMF-ize
k) Land component (1.0)1. ESMF-ize Noah land model
l) Ocean component (1.0)1. ESMF-ize MOM4 minimally2. EMSF-ize HYCOM
m) Ice component (1.0)1. ESMF-ize NCEP fast ice physics2. ESMF-ize GFDL slow ice physics
n) Aerosol component (1.0)1. ESMF-ize GOCART aerosol model
o) Ionosphere component (0.8)1. ESMF-ize IDEA ionosphere model
54
Substeps in model framework (4)p) Mesoscale components (2.6)
1. non-WRF NMM2. WRF and HWRF3. run nested or standalone4. ESMF-ize
q) GSI component (1.6)1. minimal subroutinization2. ESMF-ize
r) Configuration management (0.4)1. Install and maintain Subversion2. Subversion training and support3. Version control on scripts and code for CCS, R&D, and local servers
55
Dependence on ESMF software developed and maintained by NCAR
• Risk: e.g., ESMF did not compile on “mist” (turned out to be a compiler deficiency)
• Potential critical mass of users(NOAA, DOD, NASA)
• Potential support from IT contractors• Further layering (e.g., MAPL from NASA)
56
What is an ESMF Component?
• An ESMF Component has 3 methods: initialize, run, finalize
• Methods have a standard interface, including 1 import state and 1 export state
• A component also has an internal state, which contains whatever persistent data the component needs, and which is generally not visible to other components
57
75
80
85
90
95
100
anom
aly
corr
elat
ion
* 10
0
2 5 8 11starting month
Consolidation Ensemble AverageBest single model
Potential Benefits of Using 9 ModelsLead 5 Nino34 forecast 1981-2001
Gaussian Kernels
“Frequentist” methods
“Bayesian” methods
Construction of Optimum Forecast Guidance from Multi-Model Ensembles
1. Multiple independent realizations2. Historical “reforecast” data set3. Optimal postprocessing to produce “the best” forecast4. Compact information dissemination