Transcript N C E P

N NCEP’s Central Computing System
EMC’s
Numerical
Modeling
C
Mesoscale Modeling
E
P
Where We Are and Where We’re Going:
Geoff DiMego
[email protected]
301-763-8000 ext7221
24 June 2003
Where the Nation’s climate and weather services begin
TOPICS
•
•
•
•
•
•
•
NCEP’s Central Computing System (CCS)
What we do with it (current model suite)
Stuff made for Alaska by Meso Eta, HRW etc
Latest Bundle of Changes for Meso Eta (8 July)
What we want to do with it (future plans)
Weather Research and Forecasting (WRF)
The North American Regional Reanalysis
Central Computer System (CCS)
• Initial 3 year base period followed by
two 3 year option periods
– Each three year period contains an upgrade
• Total of 6 major increases through 2010
providing guaranteed performance (initially
2.5x then 8x…) over our current computer
• 70% for use on weather + 30% for climate
– Previously 90% for weather + 10% for climate
• Installed Sept 02 at IBM Gaithersburg, MD
• Accepted Dec 02, OPS switchover May 03
Central Computer System (CCS)
Phase / Increase
Date
Processors
Clock Speed
Memory
Disk Space
Tape Storage
Current
2432 375MHz
1216 MB
30 TB
200 TB
Phase I / 2.5x
May 2003
Phase II / 8.0x
June 2004
1408
1408 MB
42 TB
1250 TB
1.3GHz
NCEP Weather Computing Capability Relative to
2752
2752 MB
84 TB
2500 TB
December 2002
1.8+1.3GHz
40
35
30
Six Increases in Weather Portion of CCS
25
20
15
10
5
0
NCEP Computing Ca
Bad News – Development has filled its half
Production
Development
Wx Production Suite Made Up of Four Uniform Cycles per Day
Proposed NCEP Production Suite
Weather Forecast Systems
Version 1.2 January 15, 2003
Percent Used
100
80
60
40
20
0
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
Hawaii
FIREWX
COFS
RUC
EDAS
Waves
GFSens
HUR/NWM
GFSfcst
GFSanal
ETAfcst
ETAanal
SREF
GDAS
NCEP Model Suite-Global
Run Slot
#/day
Mission & (Notes)
Domain Fcst
(h/v)
Range
Resolution
Global
Forecast
System (GFS)
4/day
Global general weather and aviation
guidance to 15 days (winds, temp, rainfall)
Boundary + initial conditions for Eta,Waves
Initial conditions for ensemble generation
Supports Model Output Statistics
Hurricane tracks
global
30 km
384
55 km/ 64l
75 km/ 42l
after day 3
Global Data
Assimilation
System
(GDAS) 4/day
Provides best guess for GFS analysis
Verification & validation
(3-D Variational 6-hr update frequency
with digital filter)
global
30 km
55 km/ 64l
9
6 hr
update
Global
Ensemble
2/day
Probabilistic rainfall (QPF) and general global
weather to 15 days
50 km
(10 members with initial condition
perturbations generated from bred
modes)
360
(h/v)
100km/28l
200km/28l
after day 7
NCEP Model Suite-Ocean/Waves
Run Slot
#/day
Mission & (Notes)
Domain Fcst
Resolution
(horiz) Range (horiz/v)
Global Wave
2/day
Marine safety
global
126
1.25x1.0
long/lat
Northwest
Atlantic
Waves 2/day
Marine safety
region
126
0.25x0.25
long/lat
Alaskan
Waves
2/day
Marine safety
region
126
0.5x0.25
long/lat
Ice Drift
1/day
Marine safety
bi-polar
384
190 km
ROFS
1/day
Marine safety
region
48
10 km
nearshore
to 20 km
offshore/
19L
NCEP Model Suite-Regional
Run Slot #/day Mission & (Notes)
Domain Fcst
(h/v)
Range
Resolution
(h/v)
NGM 2/day
Used for Model Output Statistics
No.America
20 km
48
83km/16l
Meso Eta
(aka early)
4/day
North America High Resolution early
North
84
guidance on precipitation, general weather America
& domestic aviation (strategic); provides
25 km
HiResWindow lateral boundaries and is
used for Model Output Statistics
12km/60l
Eta Data
Assimilation
System
(EDAS)
4/day
Provides best guess for Meso Eta analysis
Verification & validation
(3-D Variational 3-hr update frequency )
North
12
12km/60l
America 3hr
25 km
update
Short Range
Ensemble
Forecast
(SREF) system
2/day
Probabilistic rainfall (QPF), precip type
and general weather and its uncertainty
(15 members = 10 Eta + 5 RSM, initial
condition perturbations generated from
bred modes)
North
63
America
25 km
48km/60l
for Eta
48km/28l
for RSM
NCEP Model Suite-Regional
Run Slot
#/day
Mission & (Notes)
Domain
(h/v)
Fcst
Rng
Resolution
(h/v)
HiResWindow
High Resolution Window runs of
Nonhydrostatic Meso Model (NMM)
for daily high resolution test guidance
from next generation mesoscale system
for everyone in US when fewer than
two Hurricane runs are being made
Alaska 0z,
Western US 6z,
CentralUS 12z,
EasternUS 18z,
Hawaii 00+12z
Puerto Rico
06+18z
25 km
48
8 km/60l
10km/60l
in Alaska
IMET / Fire
Weather
4/day
Runs of Nonhydrostatic Meso Model
(NMM) to support IMETs and SPC etc
for Fire Weather
Selectable ¼
of large nest
48
8 km/60l
Homeland
Security
On demand!
Run of Nonhydrostatic Meso Model
(NMM) to support run of HYSPLIT
Selectable ¼
of large nest
48
4 km/60l
Hurricane
Hurricane track & intensity guidance
for TPC warnings (up to 4 storms)
75o x 75o
126
18/55km/ 42l
CONUS
3 & 12
(0,3…
21)
20 km / 50l
4 large nests/day
4 small nests/day
4/day
Rapid Update Cycle Tactical/hourly Aviation Guidance for FAA,
(RUC) 24/day
domestic aviation and NCEP’s AWC and
SPC
20 km
Meso Eta Alaskan Output Grid #216
45 km Polar-stereographic
Contents identical to 40 km CONUS grid #212
Meso Eta Alaskan Output Grid #217
22.5 km Polar-stereographic
Contents identical to 20 km CONUS grid #215
Meso Eta Alaskan Output Grid #242
11.25km Polar-stereographic
Contents identical to 12 km CONUS grid #218
Grids 216, 217 + 242 MUCH Better
Than Original AWIPS Grids 207&214
207, 214
216, 217 & 242
Except for its vertical longitude which screws up IPFS & NDFD!
Grids 221 + 104 Cover Alaska
32 km Lambert
90 km Polar-stereographic
Full complement
NGM-look-alike
Grids 243 Covers Eastern Pacific
0.4 deg by 0.4 deg lat-long with content
same as 40 km CONUS grid #212
Map of current BUFR sites in Alaska
49 original
38 new
Graphics Available from Web Site
http://www.emc.ncep.noaa.gov/mmb/meteograms/
Surface Meteogram
Vertical Time Section
Nonhydrostatic Mesoscale Model (NMM)
• Model used for runs at grid spacings less than 10 km:
Homeland Security, HiResWindow and Fire Weather
• See Janjic, Gerrity,and Nickovic, 2001 for model
equations, solution techniques & other test results
[MWR,Vol. 29, No. 5, 1164-1178]
• Highly refined version of nonhydrostatic option
released in May 2000 upgrade to NCEP’s workstation
Eta http://wwwt.emc.ncep.noaa.gov/mmb/wrkstn_eta/
• NMM retains full hydrostatic capability
– Incorporate nonhydrostatic effects through  where =(1/g) dw/dt
– Then split prognostic equations into:
• hydrostatic parts plus
• corrections due to vertical acceleration
– Set  to zero to run in hydrostatic mode
Nonhydrostatic Mesoscale Model
Feature Comparison With Meso Eta
Feature
Dynamics
Meso Eta Model
Hydrostatic
Horizontal 12 km E-grid
Nonhydrostatic Meso Model
Hydrostatic plus complete
nonhydrostatic corrections
8 or 4 km E-grid
grid spacing
Vertical
60 step-mountain 60 hybrid sigma-pressure
coordinate eta levels
levels
Terrain
Unsmoothed
Unsmoothed grid-cell mean
silhouette with
everywhere
lateral boundary set
to sea-level
Physics (BMJ convection, turbulence etc) have been tweaked in NMM.
Hybrid versus Step (Eta) Coordinates
Ptop
Ptop
=
0
Pressure domain
=
0
Sigma domain
ground
ground
=1
MSL
=1
NMM vertical domain compared to Eta
NMM 60 –Layer Distribution
18 pressure
layers, model
top still at 25hPa
18 layers
1st layer interface above
420 hPa is bottom of first
fixed pressure layer
420 hPa
42 sigma
layers
between
surface about
420 hPa
42 layers
HiResWindow Fixed-Domain Nested Runs
Alaska Nest = AWIPS grid #249
• Users want routine runs
they can count on at the
same time every day
• 00Z : Alaska-10 &
Hawaii-8
• 06Z : Western-8 &
Puerto Rico-8
• 12Z : Central-8 &
Hawaii-8
• 18Z : Eastern-8 &
Puerto Rico-8
• This gives everyone a
daily high resolution
run when fewer than 2
http://www.emc.ncep.noaa.gov/mmb/mmbpll/nestpage/
hurricane runs needed
Terrain Used in Eta-12 and Eta-10
Eta-12
Eta-10
NMM terrain would not be restricted to discrete values like the Eta’s step mountain terrain.
HRW NMM Alaskan Output
• Grid #249 10 km Polar-stereographic
• BUFR soundings for anonymous ftp from NCEP
ftp://ftpprd.ncep.noaa.gov/pub/emc/mmb/mmbpll/alaska10.t00z/bufrsnd/
• Output for anonymous ftp from NCEP server
ftp://ftpprd.ncep.noaa.gov/pub/emc/mmb/mmbpll/alaska10.t00z/
• Output for anonymous ftp from TOC server
ftp://tgftp.nws.noaa.gov/SL.us008001/ST.opnt/MT.meso_CY.00/RD.20030621/PT.grid_
DF.gr1_AR.alaska10/
• Web displays of Alaskan Nest (Meso Eta vs
HiResWindow vs NCAR WRF) On left of page, sweep
down through parameters, 2-m temperature and 10-m
wind fields have northern and southern regional display
options http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/hiresw.alaska10/
12 km Meso Eta vs 10 km NMM On Web
12 km Meso Eta vs 10 km NMM vs 10 km WRF On Web
On Call Emergency Response
• SDM receives request for run with specs of release
and initiates run into production where it may
preempt existing or future production runs
• 4 km NMM run produces hourly output
• Hourly output drives 4 km HYSPLIT run to
describe atmospheric spread of hazardous material
(designed for radiological accidents)
• HYSPLIT output sent to WFO & emergency
managers
• No output of the NMM meteorological fields (yet)
26 Selectable 4 km domains for
homeland security response
Fire Weather / IMET Support Run
• SDM receives request for run via coordination call
with Boise, WR, SPC etc.
• Runs are made within dedicated run slots at 00z, 06z,
12z and 18z running over the top of the Meso Eta
• 8 km NMM run produces 3 hourly output grids
• Output grids (look just like HiResWindow) picked up
by WR, clipped to relevant subregion and prepared
for transmission to the IMET laptops using same FXNET procedure developed for Olympics. SPC gets
grids directly.
26 Selectable 8 km Domains For Fire
Weather / IMET Support Identical To 4
km Homeland Security Domains
Fire Weather / IMET Run Output
• The firewx grids are available out to 48 hours on the
TOC ftp server (tgftp.nws.noaa.gov) under the
following format:
/SL.us008001/ST.opnl/MT.nmm_CY.{CC}/RD.{YY
YYMMDD}/PT.grid_DF.gr1_AR.nest{xx} where
CC = 00, 06, 12, or 18
YYYYMMDD = the current date
xx = 01 - 26 (geographic location)
Filenames follow the convention:
fh.{hhhh}_tl.press_gr.awpreg (hhhh = 0000, 0003,
0006, ... , 0048)
Sample GIF File Denoting Area
of Fire Wx Run
8 km versus 4 km Hybrid Terrain
12 km Meso Eta vs 8 km NMM Winds
12 km Meso Eta vs 8 km NMM Winds
Alaska Case Eta-12 vs NMM-4
17 March 2002
Alaska Case Eta-12 vs NMM-4
17 March 2002
http://wwwt.emc.ncep.noaa.gov/mmb/SREF/SREF.html
This page will
add Alaskan
products by
October 2003
N Implementation of
C Eta Upgrade Bundle
E
P
Geoff DiMego
[email protected]
301-763-8000 ext7221
18 June 2003
Where the Nation’s climate and weather services begin
Planned Changes to Eta-12
• Upgrades to Gridscale cloud & precipitation (Brad Ferrier)
– Begin proper cycling of total condensate (Eric Rogers)
– Upgrade microphysics and improve cloud - radiation interaction
• Upgrades to 3DVAR analysis (Dave Parrish)
– Add direct analysis of WSR-88D radial velocity from NWS Multicast
– Upgrade radiance processing, stop thinning, use NOAA-17 (20x increase)
• Upgrades to Precipitation assimilation (Ying Lin)
– Assimilation of GOES cloud top pressures (w/ Jim Jung)
– Assimilate hourly precip from Stage IV instead of Stage II
• Extend off-time (06z & 18z) runs to 84 hours (Eric Rogers)
• Increase output in both frequency and content
http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/etapllsup12.etax/
http://wwwt.emc.ncep.noaa.gov/mmb/tpb.spring03/tpb.htm
Sample Total Cloud Cover
Better due to reduced longwave cooling
Before Changes
After Changes
Sample Distribution of Processed
88D Radial Velocity Data
~ 5 km processing of an hour’s worth of scans
Sample Cloud Top Pressure Field
All NWS RFC’s Stage III are used in Stage IV
except AKRFC & NWRFC
Example of new fields postable from Eta
Cloud Water
Rain
Cloud Ice
Snow
Baldwin Type versus Precip Type
Direct from Model
Baldwin Diagnosed Precip Type
Percent Frozen Direct from Model’s
Gridscale Scheme
Bundle Verification Results 24hr QPF
ETS
BIAS
Surface Temperature Response
True for East and West for both 00z and 12z runs.
Bundle Verification Results Sfc RH
Bundle Verification Results Upper-Air
24 hr
T
RH
Z
V
Bundle Verification Results Upper-Air
60 hr
T
RH
Z
V
PLANS FOR THE FUTURE
For each of the possible six
upgrades/phases of the CCS
contract with IBM
Global Forecast System (GFS)
Prediction Model Data Assimilation
Phase/YR
T-254 / L64
3D-VAR, AMSU-B, Quikscat
Current
T-254 / L64 add
Grid point version, AIRS, GOES
I / 2004
2 passive tracers
imagery
45 km / L64
3-D Background error covariance,
II / 2005
cloud analysis, minimization
45 km / L64 +
Absorption / scattering in
III / 2006
improved microphysics
radiative transfer
40 km / L80
Aerosols in radiative transfer, GIFTS IV / 2008
40 km / L80
NPP, integrated SST analysis
V / 2009
35 km / L100
Advanced 4DDA, NPOESS, IASI +
air quality
VI / 2010
Ensemble Forecasts
Global Ensemble
Short Range Ensemble Forecast
Phase/YR
T126/L28 -10 members 48 km 15 members
Current
T126/L42 -15 members 28 km 15 members
I / 2004
90km/L42-30 members 20 km 20 members
II / 2005
90km/L42-50 members 18 km 20 WRF members III / 2006
80km/L64-50 members 16 km 20 WRF members IV / 2008
80km/L64-50 members 14 km 20 WRF members V / 2009
70km/L64-50 members 12 km 25 WRF members VI / 2010
North American Early Guidance System
Prediction Model
Data Assimilation
Phase/YR
12 km Meso Eta partial
12 km 3DVAR 88D radial velocity,
Current
hourly output
GOES sounder cloud top pressures
12 km Meso Eta, support
12 km 3DVAR improve background
air quality, new radiation
error covariances
I / 2004
10 km NMM move top to 10 km AIRS, GOES imagery & move top II / 2005
2mb, enlarge domain
to 2mb, hourly updates
10 km WRF improved
10 km absorption scattering in radiative
physics full hourly output
transfer
10 km WRF
10 km WRF 4DDA
IV / 2008
10 km WRF
10 km aerosols in radiative transfer &
V / 2009
4 member ensemble
reflectivity
8 km WRF 4 member
8 km NPP, advanced 4DDA, NPOESS,
ensemble improved physics
IASI & air quality (ozone, particulates)
III / 2006
2 member ensemble
VI / 2010
Planned 15% Expansion in 2005
Hi Res Window & OCER
On-Call Emergency Response
Hi Res Window
8 km nested NMM no analysis 4 km nested NMM
8 km nested NMM
4 km nested NMM
with 3DVAR and top at 2mb
with top at 2mb
8 km WRF
6 member ensemble
7 km WRF
6 member ensemble
7 km WRF
10 member ensemble
6 km WRF
10 member ensemble
6 km WRF
15 member ensemble
4 km nested WRF
Phase/YR
Current
I / 2004
II / 2005
with improved physics
3.5 km nested WRF
III / 2006
with improved physics
3.5km nested WRF
IV / 2008
with aerosols&reflectivity
3 km nested WRF
V / 2009
with explicit convection
3 km nested WRF
with explicit convection
VI / 2010
7000000
6000000
Model Plans
Linked To AWIPS
Total Generated by Models at NCEP
5000000
4000000
3000000
2000000
•First iteration complete
•Connect model upgrade
plans to a (conservative) set
of proposed AWIPS product
upgrades -ALASKA
INCLUDED
•Presented to
NWS/OCWWS folks at
meeting 10 September
•No second iteration to date.
1000000
0
2002200420052006200820092010
120000
100000
Total Transmitted on AWIPS (PROPOSED!)
80000
60000
40000
20000
0
0.03
2002 2004 2005 2006 2008 2009 2010
Ratio of Generated at NCEP to Transmitted on AWIPS
0.025
0.02
0.015
0.01
0.005
0
2002 2004 2005 2006 2008 2009 2010
Why Can’t We Do Better?
• Numbers just don't get us there - weather portion of
computer in 2010 is 36x of present capability. If all this
is used to increase horizontal resolution alone:
– Take the cube root (x,y,t) of that 36 which is 3.3
– For Meso Eta: 12km / 3.3 = 3.63km = highest resolution possible
– NO ALLOWANCE for ANY other upgrades (domain,
forecast range, vertical resolution, 3DVAR or cost due to
additional data sources like radar or satellite, physics
upgrades or implementing or increasing membership of
ensembles).
• Getting a bigger machine is unlikely since current talk
describes FY2004 budget as being extremely LEAN.
• Nesting degrades results.
Weather Research and Forecasting (WRF)
• End-to-end Common Modeling Infrastructure
–
–
–
–
Observations and analysis
Prediction model
Post-processing, product generation and display
Verification and archive
• For the community to perform research
• For Operations to perform NWP
• USWRP sponsorship - many partners: NCAR, NCEP,
FSL, OU/CAPS, AFWA, FAA, NSF and Navy
• Initial implementation in HiResWindow in 4QFY04
• Ensemble approach to be taken instead of single-run
deterministic approach
Accelerates NWS toward
WRF SREF end-state objective
Initial
Condition
Diversity
Global
ESMF
DA
Init
1
2
3
1
2
3
N
N
Prediction
Model
Diversity
Dynamic
Dynamic
Core
Core
1 1
+
Dynamic
Dynamic
Core
Core
2. 2
.
.
S
R
E
F
Dynamic
Dynamic
Core
NN
Core
NOAH Convective Boundary Radiation
Land-sfc Schemes
Layer
Schemes
NCEP WRF Ensemble Design:
•
In June 2004, computer increase will total 6x
• Therefore, establish 6-member ensemble run in
place of single HiResWindow run
–2
Control members
•NCEP NMM
core & NCEP physics, Dx = 8 km
•NCAR Mass core & NCAR physics, Dx = 10 km
–4
Additional members
•
•
alternative physics or bred mode initial condition perturbations
Qualify cores and evaluate potential ensemble
members according to the WRF Test Plan
Sea-Level Pressure RMSE
3.5
3
millibars
2.5
3 A+B+C
Model A
Model B
Model C
2
1.5
1
0.5
0
3
15
27
39
Forecast Hour
51
63
Sea-Level Pressure RMSE
3.5
3
millibars
2.5
15 3x5 Bred
3 A+B+C
Model A
Model B
Model C
2
1.5
1
0.5
0
3
15
27
39
Forecast Hour
51
63
850 mb Relative Humidity RMSE
2.5
25
2
20
15 3x5 Bred
3 A+B+C
Model A
Model B
Model C
1.5
1
percent
degrees C
850 mb Temperature RMSE
0.5
15 3x5 Bred
3 A+B+C
Model A
Model B
Model C
15
10
5
0
0
3
15
27
39
51
63
3
15
Forecast Hour
6
12
5
15 3x5 Bred
3 A+B+C
Model A
Model B
Model C
4
3
2
meters per second
meters per second
14
6
4
0
0
Forecast Hour
51
63
15 3x5 Bred
3 A+B+C
Model A
Model B
Model C
8
2
39
63
10
1
27
51
250 mb Wind Speed RMSE
7
15
39
Forecast Hour
850 mb Wind Speed RMSE
3
27
3
15
27
39
Forecast Hour
51
63
Example of Ensemble Probability Product
0-6km Shear >40kts Prob
Conv Precip >0.01” Prob
CAPE > 1000j/kg Prob
Severe Convection Prob
Courtesy NOAA-SPC
WRF Test Plan for Summer 2003
•
•
•
•
•
•
Participants: NCEP, NCAR, FSL, AFWA/NAVO-MSRC
Two Cores: NCAR mass core & NCEP NMM core
Two physics suites: NCAR suite & NCEP suite
IC=RUC and Eta – each with bred perturbations
BC=Eta with SREF-based anomalies
Retrospective Runs:2 nests for 30 days in 4 seasons
- Aug/02 Central & West - Oct/02 Alaska & East
- Feb/03 West & East
- May/03 Central & East
• Real-time Runs at NCEP: Human Forecaster Feedback
– 2 nests for 6 weeks each
– Jul/Aug West & East
– Aug/Sep Alaska & Central
Office of
Global
Programs
N
North American
C
Regional Reanalysis
E
P
Where the nation’s climate and weather services begin
Domain Coverage of NARR
190 km
32 km
For the Eta, a 32 km run takes 178 times the work of a 190 km run!
NARR GOAL: To Improve on
NCAR/NCEP Global Reanalysis
•
•
•
•
•
•
•
•
•
Higher resolution 32 km vs T-62 (~180 km)
More frequent updates & ouput 3 hr vs 6 hr
Add precipitation assimilation
Add satellite radiance assimilation
25 year period 1979-2003
To be perpetuated by NCEP/CPC like CDAS
Production ongoing on NCEP’s old IBM
Completion expected in Fall 2003
Could be source for 2.5 km downscaled climatology
for use in NDFD and IFPS techniques
Downscaling Strategy - A
• Correct model bias (on model grid)
– Today’s forecast vs current model history (1-2 months)
– Current ensemble mean vs Reanalysis climate mean
• Correct model spread (on model grid)
– Today’s forecast vs current model history
– Current ensemble spread vs Reanalysis climatological spread
• Apply corrections to all ensemble members
• Result: forecast anomaly on model grid, corrected for
climatology
• Calculate most probable anomaly from ensemble
(error weighted mean)
• Given high resolution, gridded climatology for each
forecast element:
–
–
–
–
Add most probable anomaly to climatology for downscaled forecast element
Not guaranteed to be physically consistent (like model grids)
Forecast anomaly on model grid needs to be transmitted
High resolution climatology resident at WFOs
Downscaling Strategy – B
• Bias correction directly on NDFD grid
– High resolution information still needed from local
climatology
• Can be done locally or centrally
– If locally, assumes WFOs receive all ensemble forecast
members
• Neural Network application
– Input: ensemble forecasts, lat, lon, elevation, climatology etc
– Output: bias corrected ensemble forecasts on NDFD grid
– Penalty function: probabilistic measure (e.g. Brier Skill
Score)