Transcript N C E P

N
C
E
P
Mesoscale Modeling Branch:
Where We Are and
Where We’re Going
Geoff DiMego
[email protected]
301-763-8000 ext7221
9 December 2008
1
Where the Nation’s climate and weather services begin
Who We Are
• Government Scientists
– Tom Black
– Jun Du
– Dennis Keyser
– Ying Lin
– Geoff Manikin
– Ken Mitchell* – LSM Lead
– Jeff McQueen
– Dave Parrish
– Matt Pyle
– Eric Rogers
– Wan-Shu Wu
<- Stacie Bender* (GWCMB)
<- Hui-Ya Chuang* (GWCMB)
• Visiting Scientists
–
–
–
–
–
Mike Ek
Zavisa Janjic
Fedor Mesinger*
Yoshiaki Sato* - JMA
Ratko Vasic
• Contractor Scientists
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
Ed Colon*
Brad Ferrier
George Gayno*
Ho-Chun Huang*
Dusan Jovic*
Dongchul Kim*
Pius Lee
Shun Liu
Guang-Ping Lou
Manuel Pondeca*
Jim Purser*
Perry Shafran
Youhua Tang
Marina Tsidulko
Caterina Tassone
Weiguo Wang
Jeff Whiting*
Vince Wong
Binbin Zhou
Julia Zhu*
Yanqiu Zhu
*personnel who are less than 100% working on MMB projects
2
WHAT – i.e. T O P I C S
•
•
•
•
Production Suite
Observations Processing
Marine Connections
Legacy Stuff
• HiResWindow Upgrades & Plans
• NAM / NDAS / DGEX March &
December Updates & Plans
• RTMA and DNG (aka smartinit) & Plans
• SREF: 2008 Package  2009 & Plans
• Air Quality: Ozone & Smoke Guidance
• Other Projects
3
NOAA’s NWS Model Production Suite
All the “stuff” in BROWN is my worry
Climate
CFS
Oceans
Hurricane
GFDL
HWRF
MOM3
HYCOM
WaveWatch III
~2B Obs / Day
Mostly
Satellite
+Radar
Global Data
Assimilation
Global
Forecast
System
North American
Mesoscale
WRF NMM
GFS, Canadian Global Model
ARL’s HYSPLIT
Severe Weather
Regional Data
Assimilation
North American Ensemble
Forecast System
Dispersion
Short-Range
Ensemble Forecast
WRF: NMM+ARW
ETA, RSM
WRF NMM
WRF ARW
Air Quality
NAM+CMAQ
Rapid Update
for Aviation
NOAH Land Surface Model
F
o
r
e
c
as
t
4
Observational Data Ingest
Mostly Satellite & Radar
Daily Satellite & Radar
Observation Receipt Counts
Daily Percentage of Data
Ingested into Models
(Not Counting Radar)
Level 2 Radar
2008 Data
125 M obs
2008
2015
Five Order of Magnitude Increase in Satellite
Data Over Fifteen Years (2000-2015)
Assimilated Data
2000
7%
2%
Selected Data
100 M obs
1990
1.7 B
100%
Received Data
Count (Millions)
1.7 B obs
210 M obs
17.3M
6.6M
Received = All observations received operationally from providers
Selected = Observations selected as suitable for use
5
Assimilated = Observations actually used by models
DOPPLER
RADAR
DATA
Will it be an 800 lb gorilla?
• Now Processing Level II Archive-Quality NEXRAD-88D Data
from ~130 Radar Sites
• Still Receiving Level II.5 super-obs (made onsite and 1st backup)
• Still Receiving Level III (aka NIDS 2nd backup)
• Still using VAD wind profiles
6
Radar Processing in 2008
• NSSL Level II Radar QC package (Shun Liu)
– Combined radial wind and reflectivity QC
– Not able to deal with new higher-res Build 10, so required
NCO to make Build 8 lookalike files
– Now includes generation of national 3D reflectivity mosaics
every hour (used by RUC’s diabatic digital filter)
• FY2009 plans
– Include estimate of mixed layer depth (and transport wind) for
NASA ROSES & RTMA
– NSSL has moved all its efforts over to WDSS-II which might
mean our efforts so far are effectively orphaned – we’ll see
7
Sample of the Level II National
Reflectivity Mosaic (Composite)
8
Observation Processing in 2008
• Dennis Keyser had
133 RFC’s
– Prepared for
observation
subtype [e.g.
aircraft airframe or
mesonet provider]
– To be used in
stratifying future
bias correction
work
Global Rawinsondes
Aircraft Wind/Temp Reports
Marine Obs -- 12 Hour Total
DMSP Imager – Sfc winds/PW
9
Polar Satellite Radiances (just 2 sat)
Satellite Winds from Geostationary Satellites
Aircraft Observation Processing
• Adapted NRL’s aircraft QC package (S. Bender)
– Process conventional and ACARS data
– Includes improved track-checking
– Ascent/descent reports generated as profiles
•
•
•
•
Associated with nearest METAR for surface baseline
Will expand fcst-vs-obs verification beyond just RAOBs
Will expand validation of model boundary layer
For use in RTMA of mixing height & transport wind (NASA ROSES
project to blend sat sources (COSMIC) with in situ estimates from
RAOB, ACARS, NEXRAD)
– Looking for an implementation in Q3 or Q4 FY2009
10
Comparison of Diagnosed PBL
Heights from 3 ACARS, NAM
Analysis and NAM 30 hr Forecast
Marina Tsidulko
11
Connections to Marine
Analysis and Modeling Branch
• NAM fields used to drive Great Lakes Wave Model when
not being driven by NDFD
• Downscaling technique from Netherlands being tested to
improve adjustment of winds as you move from land to
lake or ocean
• Next NAM will use RTG_SST_HR (1/12th degree)
• Overland lakes to be addressed: sst, ice or snow covered
(some invariably need to be “filled in” even in 4 km nests)
• Marine observation station dictionary to improve quality
control, specification of height / type of instrumentation
and to ensure brand new observations are kept out of
operations until they can be validated
• Integrating marine verification into EMC-wide unified
12
verification efforts
LEGACY: Eta MOS & NGM
• MDL has generated MOS coefficients from ~1-year
of NMM-based NAM
TIN08-89: REPLACEMENT OF ETA-BASED MOS
GUIDANCE WITH NAM-BASED
MOS GUIDANCE: EFFECTIVE DECEMBER 9, 2008
• This will replace the interim Eta-32k-based Eta MOS
• NGM and its MOS will be terminated 3 March 2009
TIN08-90: TERMINATION OF THE NGM AND NGMBASED PRODUCTS: EFFECTIVE MARCH 3, 2009
• NCEP will be able to turn off NGM, NGM-MOS,
Eta-32K run and Eta MOS
• NCEP & MDL will avoid having to port these to the13
new computer in FY2009
HiResWindow Fixed-Domain Nested Runs
Expanded 4-5 km Configuration as of 11 Sep 2007
• FOUR routine runs made at
the same time every day
• 00Z : ECentral & Hawaii
• 06Z : WCentral & Puerto
Rico
• 12Z : ECentral & Hawaii
• 18Z : Alaska & Puerto Rico
• Everyone gets daily high
resolution runs if & only if
hurricane runs are not
needed
Approved OSIP Gate 3 for inclusion in OB8.3 … then OB9 … then
14
dropped – but REINSTATED!! Data started flowing June 2008
Frequency of HiResWindow Runs:
No runs at all 50% of time in July & August
1
0.8
0.7
ALL HIRESW runs
NO large ARW
NO large NMM/ARW
NO HIRESW
0.6
0.5
0.4
0.3
0.2
0.1
N
r
ov
em
be
r
ob
e
O
ct
m
be
r
t
pt
e
gu
s
Se
Au
Ju
ly
Ju
ne
ay
0
M
Ratio of total model cycles
0.9
15
HiResWindow Updates in CY2008
• Feb 12
– Fix WRF Post to get frozen precip included in computation of simulated
reflectivity for ARW runs. (NWS Central Region)
• Apr 15
– 1) Fix a bug in the boundary smoothing along the eastern boundary of NMM
domains;
– 2) Change compilation options for NMM and task geometry for both NMM &
ARW to make codes run faster;
– 3) Trigger prelim job from 48 h GFS forecast rather than 84 h GFS forecast;
– 4) Change the NMM namelists to produce hourly model output (added to
NAWIPS 4/18;
– 5) Add updraft helicity to NMM output. (NCEP SPC)
• Aug 6
– 1) Change the generating process code (PDS octet 6) from 84 to WRF core
specific values of 112 (WRF-NMM) and 116 (WRF-ARW) (for AWIPS NCF);
– 2) Add vertical motion at 200, 250, and 1000 hPa isobaric levels (for DTRA);
– 3) Fix WRF Post to properly generate precipitation type and total cloud field
percentage for the WRF-ARW.
• Sep 15
– Generation of GRIB2 is changed so distinct GRIB records are made for wind
components (AWIPS NCF).
• Nov 4
– WRF post input code is corrected to eliminate risk of seg fault failure.
16
Matt Pyle Developmental 4.5 km NMM Run is Made Daily
Backup for SPC http://www.emc.ncep.noaa.gov/mmb/mmbpll/cent4km/v2/
17
Plans for HiResWindow
• Improve initial conditions (Liu & Parrish)
– Begin tests with Matt Pyle’s daily run
– Apply GSI analysis (as 3rd outer-loop)
• Using radial winds only
• Sharpen background error covariances
– Apply Diabatic Digital Filter (ala RUC)
• Force latent heating from reflectivity mosaic
– Expand daily Pyle run - 4 km NMM twice per day for all
of CONUS starting in Q2 FY2009 (for SPC & AWC)
• Upgrade resolution to ~2 km in 2011
– 1st Priority is to get NEMS & nesting into NAM
18
31 Dec 2007 Table of Differences Between
NAM/NDAS & GFS/GDAS Possibly Leading
to Reduced NAM Performance Relative to GFS
• Table 1. Analysis/Assimilation Differences
–
–
–
–
Prediction Model (2) & Domain Characteristics (6)
Assimilation Cycle Characteristics (7)
GSI Analysis Characteristics (4)
Observational Differences (17)
• Table 2. Prediction Model Differences (13)
• Next two NAM upgrades:
– Mar 08: GFS Gravity Wave Drag& Mt Bocking
19
– Dec 08: Partial Cycling
N
C
E
P
Decision Brief:
Upgrade to NAM/DGEX
Mesoscale Modeling Branch
Eric Rogers, Geoff DiMego, Tom Black, Mike Ek,
Brad Ferrier, George Gayno, Zavisa Janjic, Dennis Keyser,
Ying Lin, Matthew Pyle, Wan-Shu Wu
14 March 2008
20
where the nation’s climate and weather services begin
Mar 2008 NAM Bundle
18% Increase in NAM domain
§ Changes
to the WRF-NMM model physics
§GFS Gravity Wave Drag
§ Fix bug in ozone treatment
§ Improved computation of surface
longwave radiation
§ Unified land-surface physics
• Enlarge the computational domain of the
NAM by ~ 18%.
• Upgrade WRF-NMM code to IJK (faster)
version and keep pace with changes to the
public version distributed by DTC.
• New GSI code and recomputed NMM
background error covariances
• Assimilation of new/better observation
types (AIRS, MODIS wind, Mesonet obs,
SFOV GOES).
• Use 12-36 h forecast precipitation from the
00Z operational NAM as driver for NDAS
soil moisture in regions outside of the
CONUS.
• New terrain after 3 passes of smoothdesmooth, fixed oversized GSL and
21
waterfalls
GFS’ Gravity Wave Drag (GWD) &
Mountain Blocking (MB)
• “Mountain blocking” (Lott & Miller, 1997 ECMWF)
– Wind flow around subgrid orography
– Low-level flow is blocked below a dividing streamline
(air flows around, not over barrier)
• Gravity wave drag (Alpert et al., 1988, 1996; Kim &
Arakawa,1995)
– Mountain wave stress, pressure drag
– Vertical distribution of the wave stress,
changes winds aloft (momentum deposition)
– Extensive testing showed NAM did not need subgrid orography
factor for this component of GFS GWD
22
Example of GWD impact: Features More Progressive
48-72 h QPF valid 12Z 23 Dec 2006
Control
(in)
GWD
GFS
(mm)
RFC Verif
(in)
23
Verification : CONUS RMS Height Error :
Solid=Ops Dashed = Parallel
Retrospective Parallel Test Periods
March 2007
August 2007
NAM vs Parallel NAM : March 2007NAM vs Parallel NAM: August 2007
24
Verification : CONUS RMS Height Error :
Solid=Ops Dashed = Parallel
Real-time Parallel Dec 07 – Feb 08
25
Verification : CONUS RMS Vector Wind Error :
Solid=Ops Dashed = Parallel
Retrospective Parallel Periods
NAM vs Parallel NAM : March 2007NAM vs Parallel NAM: August 2007
26
Verification : CONUS RMS Vector Wind Error
Solid=Ops Dashed = Parallel
Real-Time Parallel Dec 07 – Feb 08
27
Time series of 36-h forecast 500 mb
Height errors: NAM vs Pll NAM
Ops NAM
RMS
Error
Parallel NAM
Bias
28
Q1 FY2009 (16 Dec I hope)
NAM / NDAS Change Package
• Analysis / Assimilation Changes
– Partial cycling: use GDAS forecast for atmospheric first
guess at start (tm12) of NDAS
– Assimilation of TAMDAR, Canadian AMDAR and
METOP2 data
– Updated GSI analysis code w/improved CRTM
• Model Changes (also in DGEX)
– Vertically mix/diffuse each hydrometeor species separately
– Radiation: double absorption coefficients for cloud
water/ice
– LSM: upgrades related to snow & frozen conditions
• New hi-res (23km) AFWA snow depth analysis
29
NAM Parallel Testing
• 1 August 2008 – present : EMC Real-time
• 26 Feb 2007 – 31 Mar 2007 : Retrospective
• 5 Aug 2008 – 28 Sept 2008 : Retrospective
• 28 Oct 2008 – present : NCO Real-time providing
gridded product for subjective evaluation
30
Equitable Threat (top) and Bias (bottom) QPF Scores:
Left=March 2007; Right=Aug-Sept 2008.
Red=Ops NAM, Blue=Pll NAM
March 2007
Aug-Sept 2008
31
24/48/72-h CONUS RMS Height Error :
Left=March 2007; Right=Aug-Sept 2008.
Solid=Ops NAM, Dashed=Parallel NAM
March 2007
Aug-Sept 2008
32
Real-Time
QPF Stats
for Nov.
CONUS
33
Real-Time Stats for Nov. CONUS
HEIGHT
Temperature
WIND
Rel. Humidity
34
Real-Time Stats for Nov. Alaska
HEIGHT
Temperature
WIND
Rel. Humidity
35
Q1 FY2009 NAM Upgrade
Performance Summary
• Upper air stats are much better
• Precip is slightly better with somewhat
lower biases (helps if NAM bias is high and
hurts if it is low – regime dependent)
• Surface stats generally neutral/slightly
better except for Alaska warm season bu
36
NAM Plans for FY2010
On New Computer
• Next machine (providing ~2-3x) August 2009
• Add CONUS & Alaska nests to NAM
– Nests run at ~NDFD spacing: CONUS 4.5km & Alaska 5.5km
– Nests run at least to 48 hours
– Nested guidance similar to HiResWindow except:
• Available every NAM run – no hurricane preemption
• Available 4 times per day – no intermittent availability
• Available ~3 hours earlier
– Nested output additional to existing NAM 12 km guidance
• No more NAM changes in CY2009
– Porting to new computer & NCO moratorium
– Move to new building
– Move to new ESMF-based NEMS
(NCEP Environmental Modeling Framework)
37
Future Nests Imbedded in 12 km NAM
38
12 km Terrain
4 km Terrain
GFS ~35km
39
Dots represent water points Domain is San Francisco Bay
DiMego’s Preferred Approach to
Reinstatement of FWIS
• Reinstate FWIS after NAM implementation on
new computer in FY2010 when NEMS oneway nesting capability can be used.
• Reinstating FWIS any sooner would mean:
– Squeezing it into already very busy 2009
– Having to do it within the WRF-Common
Modeling Infrastructure and with its nesting
– WRF solution would be very short lived, being
replaced within a year or so by a NEMS solution
– Efforts would certainly delay next NAM with
NEMS into late 2010 and even into 2011.
40
RTMA & DNG – OSIP
• Analysis of Record (AoR) – OSIP 05-009
• Real Time Mesoscale Analysis (RTMA) – OSIP 06-069
– Phase 1 of AoR
– RTMA for CONUS since August 2006
– New capability for OCONUS
• Requires downscaled 3-8 hr NAM fcsts for 1st guess for AK, HI and PR
• Requires downscaled 3-8 hr GFS fcsts for 1st guess for Guam
• Downscaled Numerical Guidance (DNG) – OSIP 06-041
– Run AWIPS process “SmartInit” at NCEP
– Use full resolution model fields instead of the degraded versions available
locally on AWIPS
– First do NAM to 84 hours CONUS, Alaska, Hawaii & Puerto Rico
– GFS to follow
• Will provide first guess for RTMA – Guam
• Will allow eventual termination of DGEX
– Possible improvements are many, but EMC lacks resources to pursue
41
RTMA & DNG
Accomplishments
•
•
•
•
•
•
CONUS RTMA Q3 FY2007
Alaska RTMA Q1 FY2008
DNG for NAM Q1-Q3 FY2008
Hawaii RTMA Q4 FY2008
Puerto Rico RTMA Q4 FY2008
Unified & Upgraded RTMA Q1 FY2009
42
PUERTO RICO NDFD DOMAIN
2.5 km RESOLUTION GRID
177 x 129 GRID POINTS on a Mercator Projection
TERRAIN FOLLOWING COVARIANCES
FIRST GUESS  SMARTINIT ON NAM FCST
Shaded Contours: Topography in meters
43
HAWAII NDFD DOMAIN
2.5 km RESOLUTION GRID
321 x 225 GRID POINTS on a Mercator Proj.
TERRAIN FOLLOWING COVARIANCES
FIRST GUESS  SMARTINIT ON NAM FCST
Shaded Contours: Topography in meters
44
Unify & Upgrade RTMA
Q1 FY09 [Today!]
•
•
•
•
•
•
Unify RTMA codes (CONUS & AK same as HI & PR)
Add analyzed surface pressure & analysis uncertainty (not mslp)
Add RTMA terrain (comes from RUC not AWIPS)
Faster code (2.5X faster via better use of multi-tasking)
Improved estimate of analysis uncertainty using Lanczos method
Analyze sensible T instead of Tv (decouples from moisture) and
give background more weight (both help to eliminate overfitting)
• Recompute, recalibrate & rescale (if necessary) Background
Error Covariance using terrain with ~no smoothing
• Broaden correlation lengths over water
• Improved obs quality control
–
–
–
–
Moisture checks inside 2D-Var (for dry-lines)
MADIS reject list (for CONUS but comes from ER)
Reject lists provided by NWS Regions
Tuned dynamic reject list
45
Estimated Analysis Error (K)
Observations Used
HAWAII RTMA Estimated
Temperature Analysis Uncertainty
Valid 12Z 2 September 2008
Note: Analysis Error is Computed via
the Lanczos Algorithm for Solving
Eigenvalue Problems in Connection
with the Conjugate-Gradient Method
of the GSI minimization. By-products
of the GSI minimization (gradient
vectors and step sizes) allow us to
reconstruct a good representation of
the Hessian Matrix. The Analysis
error covariance matrix is the inverse
of the Hessian.
GREEN DOTS : MESONETS
RED DOTS: NON-MESONETS
Analysis error is smaller where
there are obs and is equal to the
background error in data void
locations.
46
Unified RTMA Evaluations
• Josh Watson, “Eastern Region
wholeheartedly endorses the
suggested changes to be incorporated
into the RTMA.”
– The following (4) individuals and (15)
ER WFOs have evaluated the parallel
RTMA over the last 8 months and
provided positive comments and
feedback regarding the changes and
modifications:
– Joshua Watson, David Radell,
David Novak, Richard Watling
– WFO CAE, WFO AKQ, WFO CTP
– WFO CAR, WFO BOX, WFO PBZ
– WFO RAH, WFO RNK, WFO ILM
– WFO GSP, WFO BTV, WFO GYX
– WFO ILN, WFO CHS, WFO PBZ
• Greg Mann, Detroit WFO DTX
• Florida Institute of Technology –
Prof. Stephen Lazarus
• Dave Myrick (WR SSD), “The
following WR folks have participated
(some more than others) in the
evaluation of the parallel RTMA over
the past 8 months:
– WR HQ / SSD – David Myrick
– WFO Seattle - Kirby Cook and Brad
Colman
– WFO San Diego - Greg Martin
– WFO Reno - Shane Snyder
– WFO Spokane - Ron Miller
– WFO Hanford - Larry Greiss
– WFO Missoula - Gene Petrescu
– WFO Great Falls - Dave Bernhardt
– WFO Billings - Don Moore
– WFO Medford - Dennis Gettman
– WFO Pocatello - Dawn Harmon and
Dean Hazen
– WFO Flagstaff - Mike Staudenmaier
– WFO Elko - Ryan Knustvig
47
TIN08-95: ADD NEW EXPERIMENTAL NCEP RTMA GRIDS TO NOAAPORT
AND NDGD: EFFECTIVE JANUARY 28 2009
D E D I CATI O N
• We’re all dedicating this
RTMA implementation to
the memory of
Lee Anderson
48
DNG Includes a New Method for Generating PoP
PoP
Old smartinit method
totally based on model
QPF and RH such that
high model QPF=high PoP
With higher resolution
we get mesoscale bands
which lead to narrow bands
of high PoP surrounded by
large areas of low or 0 PoP
NAM
precip
New method combines
old method with SREF
pops
49
NAM DNG Distribution
• On 1 December, 2008 Brian Gockel wrote: NCEP's
NAM-Based Downscaled Numerical Guidance grids (for
the CONUS area) are scheduled for addition to the
SBN/NOAAPort ... into AWIPS. The activation is
scheduled for tomorrow (Tuesday Dec 2), effective with
the 12UTC distribution.
– This activation is for the CONUS [NAM] DNG only
[TIN08-87: ADDITION OF NCEP NAM-BASED CONUS-AREA DNG GRIDS TO NOAAPORT: EFFECTIVE
DECEMBER 2, 2008 ]
– The activation for the OCONUS NAM DNG is scheduled for late
January 2009.
[TIN08-92: ADDITION OF NCEP NAM-BASED OCONUS-AREA DNG GRIDS
TO NOAAPORT: EFFECTIVE JANUARY 28, 2009 ]
– … Note that the collective additional data volume is considerable
(139MB for each six-hourly batch, or 556MB/Day). For this
reason, staff at the NCF are included and should be aware (Jon
Beach and I spoke this evening, about this activation). …
• AWIPS users have found that local storage space needed
after conversion from GRIB2 to local NETCDF is
considerably larger.
50
NAM DNG
Local Storage Issue
Could Be
Reduced With
More Grid Clipping
to the Local Domain
51
RTMA & DNG Future Plans
• Funding = $10K (NWS/OST) + $20K (NCEP/OD)
–
–
–
–
–
–
–
–
–
–
–
–
–
Adapt the NAM smartinit to GFS
Generate DNG from GFS out to 7-8 days
Expand RTMA to Guam
Declare RTMA Operational
Generate Weather Type for DNG NAM & GFS
Upgrade to 2.5 km over CONUS
Upgrade to 3 km over Alaska
Upgrade to 1.25 km over Puerto Rico
Upgrade to 1.25 km over Hawaii and Guam
Apply DNG to RUC
Expand RTMA variables (visibility, wind gust etc)
Bias correct 1st guess prior to applying smartinit
Retire DGEX
52
NCEP
Provided
10 WRF runs
RDP
降水预报(国家体育场)
(0-36h), Initial:08080712
daily run at ~15 km resolution
CAMS
NMC
JMA
During the Opening Ceremony in Beijing 8/8/08, little
precipitation was predicted at National Stadium by
five RDP participants except Fr. & Aust.
Fr.&Aus
Canada
NCEP
53
Accumulated Precipitation
( 04pm Aug.8—06am Aug. 9)
National Stadium
54
Precipitation Forecast for National Stadium for
Opening Ceremony by Deterministic Models
WRF:
MM5:
17:00—24:00 无雨
17:00—20:00 无雨
20:00—23:00: 4mm
23:00--9日02:00:6mm
GRAPES:
T639:
17:00—23:00 无雨
17:00—9日02:00 零星小雨,
23:00—9日2:00,小雨,<1mm
T量-0.2mm
55
Q1 FY2009 SREF Changes - 1
• Upgrade model versions for NMM (2.2),
ARW (2.2) and RSM (2008)
• Increase horizontal resolution for non-Eta
members (Eta members run at 32km):
– To 32km from 40km for NMM (2x)
– To 32km from 45km for RSM (2.8x)
– To 35km from 45km for ARW (2.1x)
• Keep membership at 21 members, but:
– Add 4 WRF members from 6 to 10 (5 NMM + 5 ARW)
– Remove 4 Eta members from 10 to 6 (3 BMJ + 3 KF)
• Add more initial condition diversity using
ET method instead of breeding
– WRF members use 10 different GEFS initial conditions
– GEFS uses ET method instead of breeding
56
Q1 FY2009 SREF Changes - 2
• More physics diversity:
– Replace Zhou with Ferrier in 3 SAS members for RSM
– Replace Ferrier with WSM3 in all members for ARW
• SREF Output:
– Fix serious bug in ARW’s BUFR skin-temperature calculation.
– Breakout single bufr file into individual station time-series
used by forecaster’s
– Upgrade the WRFBUFR code to use MPI I/O
– Add aviation fields for AWC
– Add wind-variance fields for DTRA
– Add hourly output of each member: a
late requirement from SPC and AWC to
support aviation in Northeast corridor
during 2009+ convective season
57
Performance During EMC Testing
• Large scale features are similar between old
and new versions
• New versions show:
– More detailed spatial structures (especially in the
mountainous West)
– More intense in precipitation amount (expected
due to resolution increases
– Detailed structures are considered an improvement
over the current system but traditional scores such
as rmse and threat scores may suffer slightly
58
– Reduced 2m Temperature bias
Cold Season Retro Runs
• While NCO ran Real-Time Parallel in warm
season, EMC ran cold season retrospective cases
• HPC requested 3 cases:
– Dec. 14-16, 2007
– Jan. 1-6, 2008
– Feb. 9-14, 2008
• SPC requested one case:
– Feb. 2-5, 2008 (Super Tuesday Tornado)
59
SPC Requested Rerun of 3 Feb 2008
63-hr Forecasts Valid 00 UTC 6 Feb 2008
SREF
SREFp
Verf
SREF Forecasts of Probability of
Sig. Tornado Parameter >3
SREFp indicates higher probability
over lower MS Valley (also note
mean STP = 3 contour in SREFp
but not in SREF)
60
HPC: ZR forecast verifying at 00 UTC on 17 Dec 2007: SREFP
shows better areal coverage and amount
F48hr
F24hr
OPS SREF
F24hr
F48hr
PARA SREF
61
Precip Scores 4 Cold-Season Retro Runs (Ying Lin)
ETS
ETS
bias
bias
(1) Feb. 9-14, 2008, HPC
(3) Jan. 1-6, 2008, HPC
(2) Feb. 2-6, 2008, SPC (Super Tue Tornado)
ETS
ETS
bias
bias
(4) Dec. 14-16, 2007, HPC
62
Precip Verification from NCO’s parallel SREF: both
ETS and Bias scores improved (from Ying)
63
SREF 09z Calibrated Lightning
Verification of Lightning Fcsts Valid 00 UTC 5 Nov 2008
SREF 3 Day Sample
SREF 6 Month
Sample
SREFp 3 Day Sample
SREFp Thunderstorm Calibration Statistics:
• SREFp Brier score slightly improved
(13% vs 9% improvement over climo)
• SREFp ROC area slightly improved
(0.83 vs 0.78)
• Expect results to continue to improve
64 as
calibration adjusts to SREFp
Summary of SREF Evaluations
• Strong low-level temperature cold bias is gone for
NMM members
• Both individual members and ensemble mean perform
better
• Ensemble spread quality improves
• Probabilistic forecast improves
• More detailed features are seen such as precip and T
fields due to model resolution increase
• More frequent forecast outputs (hourly) and new
aviation ensemble products available to forecasters
• More balanced in membership among four models used
(about 5 each)
• SPC, HPC, AWC and Rich Grumm gave “Thumbs Up”
65
Chronology & Current Status
• July Code handed over to NCO
• August Major changes to produce hourly output
• August/September EMC ran cold season retro runs for HPC and SPC and
NCO & EMC worked to construct the NCO parallel system
• Sept 20 NCO parallel starts to run regularly
• Oct. 14 – Nov. 14 Official forecaster evaluation period
• Nov. 17 Fix to turn on precipitation rate field for ARW members (discovered
and requested by SPC/David Bright)
• Nov. 19 Failure due to script error (shared working directories)
• Nov. 21 ARW member fails due to bad input data
• Nov. 22-23 Three 21z runs fail with fix for shared directory problem
• Nov. 25 New SREF implementation is postponed
• EMC will establish a SREF parallel on new Cirrus machine
– Produce hourly output for SPC calibrations
– Re-submit clean SREF package for implementation later in 2009
66
FY2010 SREF Upgrade
• Expand bias correction to full domain (grids #221
for all of No. Am., #216 for AK and #243 for
Pacific Region)
• Downscale all members to at least ~12km with
NAM hi-res control
• Downscale all members to 5 km for those
variables provided by RTMA
• Add ensemble product based on BUFR
• Bias correct precipitation
• Replace breeding with ET method
• Add 10-50-90 precentile output
67
FY2008 Experimental
NAM-driven HYSPLIT
Alaskan Smoke Product
Now Run by NCO in
NWPROD
68
Operational FY08 Forecast Domain
48 h forecasts of Sfc Ozone at 06 and 12Z
259
grid
cells
CONUS “5x” Domain
(clean, static BC)
1. Operational NAM-CMAQ
2. Exp: NAM-CMAQ w/ PM
442 grid cells
69
2008 Real-time Runs
WRF-CMAQ 48 h forecasts
System
Domain
Chemistry
Emissions
PBL
Operational
Prod: 06/12Z
65p, 35 min
CONUS
22 L common
hybrid vertical
coordinate
CB-04
gas-phase
2005
EGU: 2008 DOE
projected
Area:2005 NEI
Mobile: OTAG &
2005 NEI
Asymmetric
Convective
Model
(ACM-2)
Experimental Para:
06/12Z
127p, 120 min
CONUS 22 L
“ “
CB05 w/
harmonized
aerosols
Same as Ops but for
expanded CB05
species
ACM-2
Develop
Dev2: 06Z only
127p, 120 min
CONUS 22 L
“ “
CB05 w/ heterog.
aerosols & smoke
fire sources
Same as EXP w/
HMS/BlueSky wild
fire sources
ACM-2 w/ NAM
Kz & mixed layer
depth
Results with CB05 were slightly worse
so no SMAQ upgrade this year.
70
8hr Max Ozone: July 10, 2008
Experimental Run :
71
• Better over SJV, SAC valleys
•Stronger Overprediction over LA basin by as much as 40 ppb comp. to obs
Air Quality FY09 Milestones
• Add experimental Ozone runs for Alaska and Hawaii
domains
• Declare Alaska HYSPLIT smoke run operational
• Add experimental Hawaii HYSPLIT smoke run and
CONUS dust product
• Upgrade HYSPLIT physics and test a CONUS dust
capability
• Continue exp & dev Particulate Matter runs with CMAQ
72
Proposed O-CONUS Domains
in FY09
Alaska 2.5X
Hawaii 0.5X
73
Other Projects
• DTRA / dispersion (McQueen)
• Verification implementation & unification
http://www.emc.ncep.noaa.gov/mmb/gplou/emchurr/nwprod/
• Aviation products for ICAO & Transition of
FAA AWRP algorithms from AWC to
NCEP’s CCS.
74
75
■ Cyclone tracking & verification system
NCEP/EMC Cyclone Tracking and
Verification System
J-Jobs
4 times daily: GFS, NAM, GEFS, SREF
2 times daily: ECMWF, CMC, NOGAPS,
UKMET, CENS (00Z, 12Z)
1 time daily: EENS (12Z)
SMS
21 CPU for GEFS, SREF
17 CPU for CENS
50 CPU for EENS
Single CPU for GFS, NAM,
ECMWF, CMC, UKMET,
NOGAPS
POE
Scripts
Model data
Search
Output forecast tracks
(ATCF, TIGGE)
Storage
Climate pairs each month
(ATCF, MySQL)
Climate verification
month, season, year
Forecast
tracks
Match
pairs
Analysis
tracks
Verify
F vs A
Visualization:
Forecast tracks
Analysis tracks
Position errors
X-Y biases, etc
Web display
76