Transcript PPT

Global Climate and Weather
Modeling
NCEP Production Suite Review
John H. Ward
Chief, Global Climate & Weather Modeling Branch
December 7, 2011
Outline
 Personnel
 FY11 Upgrades
 FY12 and Beyond
 Global Model Roadmap
2
Personnel

Global Modeling



















Modeling Team Lead
S. Moorthi
Glenn White
Jordan Alpert
Mike Young

Yutai Hou
Huiya Chuang
DaNa Carlis
Henry Juang
Diane Stokes
Mark Rozwodoski
Qingfu Liu
Suru Saha
Fanglin Yang
Sajal Kar
Yali Ma
Kate Howard
Jia-Fong Fan
Feds - Blue
Climate Modeling




Climate Team Lead
Dave Behringer
Xingren Wu
Ensembles




Infrastructure










Mark Iredell – Lead
Edward Colon
Ratko Vasic
Dusan Jovic
Dmitry Sheinin
Weiyu Yang
Jun Wang
Nicole McKee
Patrick Tripp
Eugene Mirvis
Visiting Scientists – Red







Yuejian Zhu – Lead
Dinchen Hou
Jun Du
Richard Wobus
Bo Cui
M. Pena-Mendez
Jesse Ma
Jiayi Peng
Yan Lou
Bo Yang
Physics



Hualu Pan
Jongil Han
Ruiyu Sun
3
Contractor - Black
Personnel


Data Assimilation
















Feds - Blue
John Derber – Lead

Russ Treadon

Daryl Kleist

George Gayno

Andrew Collard

David Groff

Greg Krasowski
Xu Li
Haixa Liu
Shun Liu
Mike Lueken
Michiko Masutani
Dave Parrish
Jim Purser
Miodrag Rancic
Xiujuan Su
MingJing Tong
Paul Van Delst
Wan-Shu Wu
Yanqiu Zhu
Fedor Mesinger
Lidia Cucurull - Boulder
Visiting Scientists – Red
Land Surface









Mike Ek – Lead
Jiarui Dong
Jesse Meng
Helin Wei
Vince Wong
Yihua Wu
Youlong Xia
Ringqian Yang
Weizhong Zheng
4
Contractor - Black
Personnel

Dynamics






S. Moorthi
Henry Juang
Sajal Kar
Jia-Fong Fan
Mike Young
Jordan Alpert

Ocean/SST



Dave Behringer
Diane Stokes
Ensembles




Programming/User Support




Kate Howard
Mark Rozwodoski



Post Processing/Product




Huiya Chuang
DaNa Carlis
Yali Ma


Climate Modeling


Xingren Wu


Tropical Storms

Qingfu Liu

Infrastructure
Radiation/Physics

Yutai Hou
Hualu Pan
Jongil Han
Ruiyu Sun
Suru Saha












Diagnostics/Verification


Glenn White
Fanglin Yang
Yuejian Zhu
Dinchen Hou
Jun Du
Richard Wobus
Bo Cui
M. Pena-Mendez
Jesse Ma
Jiayi Peng
Yan Lou
Bo Yang



Mark Iredell
Edward Colon
Ratko Vasic
Dusan Jovic
Dmitry Sheinin
Weiyu Yang
Jun Wang
Nicole McKee
Patrick Tripp
Eugene Mirvis
5
FY11 Changes
 NAEFS: Downscaling for Alaska – 12/7/2010
 NAEFS: Inclusion of FNMOC Ens – 3/1/2011
 CFSv2.0.0 – 3/30/2011
 CFSv1 will continue to be run until further notice
 WAFS: Blended US/UK products
 Parallel for evaluation – 1/25/2011
 Initially evaluation complete
 Changes implemented based on feedback – 6/15/2011
6
FY11 Changes
 GDAS/GFS Bundle – 5/9/2011

Improvement in 10m winds in Southwest U.S. has
resulted in unrealistic GFS MOS winds

MDL is testing a fix – no implementation date as yet
7
FY12 & Beyond
 NAEFS: GEFS Upgrade – 2QFY12
 Global Aerosol System (GOCART) – 2QFY12
 Hybrid EnKF-3DVar GSI Data Assimilation –
3QFY12
 GFS Resolution Increase – 1QFY13
 Transition to NEMS – 3QFY13
8
Global Modeling
Roadmap
9
Basic Strategy
 All candidate models will be ported to the NOAA
R&D system at Fairmont Site-B
 Operational GDAS & GFS will be ported to the
NOAA R&D system


It will become the baseline for all tests
The baseline version will evolve as the operational
GDAS & GFS are upgraded
 Models will eventually be transitioned into the
NEMS framework
10
Basic Strategy
 Establish verification package

Combined UKMET, ECMWF analysis will be used for verification


If all forecast hours and/or variables are not available to verify a full
16-day forecast, the verification may need to be augmented with
analyses from one or more of the candidate models.
Current NCEP verification package will be used, with some
modifications





Options to verify a single forecast starting at a specific initial time
rather than forecasts valid at a specific time
Guarantee a homogeneous set of dates for all models being
evaluated – missing dates from one model will be eliminated for all
models
Include near surface, sensible weather elements into routine package
Both Grid-to-Grid and Fit-to-obs will be verified
Develop a UKMET-like index to aid in determining the overall impact
of proposed models
 Specific variables and weights will be determined prior to any model
testing
11
Basic Strategy
 Candidate Models





GFS
NMMB
FIM
Cubed-Sphere
MPAS
 Model configuration
 Each modeling may have one or more configurations with various
physics, dynamic, radiation, resolution, etc.
 All models will output pressure GRIB files on a standard lat-lon
grid
 Limiting factor is the operational resource window (CPU x Wallclock)

Since codes may not scale the same on different architectures, they
must be ported to the operational system to ensure each
configuration will fit within the operational resource window
12
Basic Strategy
 Model configurations will be tested as members of a multi-
model ensemble

Two Step Process
 Test at full resolution against the operational GFS
 Configurations that perform within a given delta of the GFS will
then be considered for an ensemble member at lower resolution
Evaluation of value-added to the ensemble will be
determined using the techniques developed for NAEFS
Configurations which add value will be implemented into an
operational multi-model ensemble, replacing the current
GFS-based GEFS
 Until the 2014/15 time frame, when the Operational CCS is
expected to have its next resource increase, the total
number of ensemble members will remain 20 and the model
resolution of each member will be similar to the then
13
Operational GEFS.


Basic Strategy
 All members of an Operational Multi-model Ensemble
must be run on the NOAA Operational computer system



Ensures all members will be available to produce ensemble
products
NAEFS does produce Ensemble Products from members
provided by CMC & FNMOC, but all of those are supported
operationally by their Center
Test or experimental products can be produced from Site-B
runs or from members provided from other systems, but
those should never be disseminated via Operational
channels.
14
Basic Strategy
 Specific model configurations from the ensemble
membership will then be tested for use as the
Operational deterministic model, which will
support Weather, Climate, and Data Assimilation


This is the opposite of current NCEP Operations,
where the ensemble system is usually the n-1
deterministic model at lower resolution
With a multi-model ensemble, however, the n-l
paradigm isn’t possible
15
Next Steps
 Infrastructure
 Code developers at each of the NOAA development sites
need easy access (on-site) to program analyst support for
each of the NOAA R&D system architectures
 A centralized group needs to be established to maintain
common libraries and utilities at all NOAA R&D sites
 A centralized group needs to be established to maintain
data flow between the NOAA operational and R&D sites
 The lack of development resources at NCEP will require
both retrospective and near real-time runs at the NOAA
R&D sites
 Subversion repository will be established to hold all
candidate codes
16
Next Steps
 Begin comparison of NNMB & FIM with current Operational GFS

Routine FIM runs from ESRL are being verified with the standard
NCEP verification package



10 days only
Initial results look very promising
NMMB with GFS physics in NEMS are running routinely for 00Z


6 days only
Need to begin verifications
17
Questions
18
NAEFS: Downscaling for Alaska
Statistical Down-Scaling Techniques for Alaska
♦
Variable: surface pressure, 2-m temperature, 10-meter wind component

♦
Variable: Tmax and Tmin

♦
work well using current operational technique for CONUS
Using latest define Tmax and Tmin defined period for Alaska
Variable: wind speed and direction
Using equal weight for wind direction distribution
Possible future improvement for wind direction
Alaska Verification


♦
♦
Stats for Tmax, Tmin, wind direction/speed
First a few key points:




Statistical down-scaling data adds value
Bias correction alone is of value
Bias correction with downscaling adds significant value to the forecasts
NAEFS is better than lone GEFS
More members are better
 Present (latest) to HPC (Alaska desk)


Dec. 7th 2009
19
Statistical downscaling for NAEFS forecast
 Proxy for truth
RTMA at 5km resolution
 Variables (surface pressure, 2-m temperature, and 10-meter wind)
 Downscaling vector
 Interpolate GDAS analysis to 6km resolution
 Compare difference between interpolated GDAS and RTMA
 Apply decaying weight to accumulate this difference – downscaling
vector
 Downscaled forecast
 Interpolate bias corrected 1*1 degree NAEFS to 5km resolution
 Add the downscaling vector to interpolated NAEFS forecast
 Application
 Ensemble mean, mode, 10%, 50%(median) and 90% forecasts

20
Back
21
NAEFS: Current Configuration
NCEP
CMC
Model
GFS
GEM
Initial uncertainty
ETR
EnKF
Model uncertainty
Stochastic
Yes (STTP)
Yes (multi-physics)
Tropical storm
Relocation
None
Daily frequency
00,06,12 and 18UTC
00 and 12UTC
Resolution
T190L28 ~70km
1.0 degree
Control
Yes
Yes
Ensemble members
20 for each cycle
20 for each cycle
Forecast length
16 days (384 hours)
16 days (384 hours)
Post-process
Bias correction for
ensemble mean
Bias correction for each
member
Last implementation
February 23rd 2010
July 10th 2007
22
NAEFS: NUOPC IOC
NCEP
CMC
FNMOC
Model
GFS
GEM
Global Spectrum
Initial uncertainty
ETR
EnKF
Banded ET
Model uncertainty
Stochastic
Yes (STTP)
Yes (multi-physics)
None
Tropical storm
Relocation
None
None
Daily frequency
00,06,12 and 18UTC
00 and 12UTC
00 and 12UTC
Resolution
T190L28 ~70km
1.0 degree
T119L30 ~1.0degree
Control
Yes
Yes
Yes
Ensemble members
20 for each cycle
20 for each cycle
20 for each cycle
Forecast length
16 days (384 hours)
16 days (384 hours)
16 days (384 hours)
Post-process
Bias correction for
ensemble mean
Bias correction for each
member
Bias correction for
member mean
Last implementation
February 23rd 2010
July 10th 2007
May 2010
Back
23
CFS2.0.0
 Forecast component frozen (T126L64)

Assimilation component can evolve with GFS and
GSI
 Current CFS Data Assimilation is fully coupled
version of T574 GFS/GSI

Will not change when the Hybrid data assimilation
is implemented
Back
24
WAFS
 Harmonization of UKMET and US WAFS products
 UKMET & US utilize different forecast models and different
algorithms to produce WAFS products (Turbulence, Cb,
Icing)
 Produces inconsistencies between the products from both
Centers
 UKMET & US have agreed to move toward consistent
algorithms to reduce differences
 Both Centers are testing blended products
 Mean products are averages of Center’s products
 Maximum products are values from either Center
Back
25
GSI/GFS Bundle
 Analysis Changes














Improved OMI QC
Removal of redundant SBUV/2 total ozone
Retune SBUV/2 ozone ob errors
Relax AMSU-A Channel 5 QC
New version of CRTM 2.0.2
Inclusion of Field of View Size/Shape/Power for Radiative transfer
Remove down weighting of collocated radiances
Limit moisture >= 1.e-10 in each outer iteration and at end of analysis
Inclusion of uniform (higher resolution) thinning for satellite radiances
Improve location of Buoys in vertical (move from 20 to 10m)
Improved GSI code with optimization and additional options
Recomputed background errors
Inclusion of SBUV from NOAA-19
Ambiguous vector quality control for ASCAT (type 290) data
 Model Changes



New Thermal Roughness Length
Set minimum moisture Value in Stratosphere to
Reduce background diffusion in the Stratosphere
26
Wind Speed Bias
27
Wind Speed Bias
28
Temperature Bias
29
500 MB Anomaly Correlation
Northern Hemisphere
Southern Hemisphere
30
Precipitation
Back
31
NAEFS: GEFS Upgrade
 Major Improvements
 Resolution Increase
T190L28  T254L42 (0 - 192 hrs)
 T190L28  T190L42 (192 - 384 hrs)
 Improved initial perturbations
 Improved stochastic total tendency perturbations
 Product Delivery Delays
 Raw GFS GRIB data will be delayed ~20 minutes


Delays will gradually increase with forecast length
Bias corrected GRIB will be delayed ~ 20 minutes

Probabilistic Products will be ON TIME

32
FCST
+4:35 --- +5:15
Current
FCST
+4:35 --- +5:35
NO IMPACT ON NAEFS
FNMOC_ENS_DEBIAS
+7:20 --- +7:30
20m late finish
NCEP_POST
+4:37 --- +5:17
Future
NAEFS products start
NCEP_POST (PGB)
+4:37 --- +5:37
Delays grow to
20 minutes at 16 days
+5:00
+6:00
+7:35
+7:00
+9:00
CMC_ENS_PREP
CMC_ENS_POST
+7:20 --- +7:22
ENS_DEBIAS
+4:40 --- +5:24
ENS_DEBIAS
+5:00 --- +5:44
CMC_ENS_DEBIAS
+7:22 --- +7:26
PROB_PRODUCTS
+5:24 --- +6:45
20m late start
+8:00
NAEFS_PROB_PROD (1)
+7:33 -- +8:08
PROB_PRODUCTS
+5:44 --- +6:40
5m early finish
GEFS/NAEFS 6-hr window flow chart
NAEFS products (2)
+8:08 --- +9:08
Back
33
GSI Hybrid EnKF-3DVAR Upgrade

Possible components (NSST and Combination with CDAS removed)












GPS RO bending angle rather than refractivity
Inclusion of compressibility factors for atmosphere
Retune SBUV ob errors – fix bug at top
Update radiance usage flags
Prepare for monitoring NPP and Metop-B
Add GOES-13 data
Add Severi CSBT radiance product
Satellite monitoring stats code included in Ops.
New Sat wind data and QC
EnKF hybrid system – modify inflation factors
Update to current version of trunk
New version of Forecast model


Restructured to include options for Semi-Lagrangian & NSST Model, and
corrected a bug in lake elevation.
Updated postprocessor



CAPE, CIN, & Lifted Index calculated from virtual temperature
Ability to output GRIB2 directly
10 new variables for Fire Wx & 6 for Wind Energy
Back
34
NRT NGAC configuration
Experimental (non-operational)
 120-hr dust-only forecast once per day (00Z)
 ICs: Aerosols from previous day forecast and meteorology from
operational GDAS
 3-hourly products: 3d distribution of dust aerosols (5 bins from 0.1 –
10 µm)
 Automatic output archive, post-processing and web update since
June 11, 2011
 Same physics and dynamics as operational GFS with the following
exceptions:



Lower resolution (T126 L64)
Use Relaxed Arakawa-Schubert scheme [Moorthi and Suarez, 1999]
with convective transport and tracer scavenging
Aerosol-radiation feedback turned off
35
Sample Forecast
Back
36
GFS 1QFY13 Upgrade
 T1178L64 Semi-Lagrangian
 Resource Nuetral
Back
37
FIM purpose and configurations
Purpose: A next-generation global model for NOAA
(candidate for NCEP ops, coupled model research)
Resolution
 Real-time testing at 60km, 30km, 15km resolution icosahedral horizontal grid
 64 vertical levels – hybrid θ-σ
 Ptop = 0.5 hPa, -top = 2200K
Physics
 Currently GFS physics suite (2010 version)
Initial conditions
•GFS/GSI spectral data to FIM icos hybrid θ-σ vertical coordinate
•Ensemble Kalman using GFS T254 – Jeff Whitaker, ESRL
Application at NCEP
Likely application in GEFS, candidate for future global model
38
Versions of FIM running – Fall 2011
Resolution
Init
conds
Physics
Stream
1.5
30km
GSI
GFS (July 2010)
No
30km
EnKF
GFS
Yes
FIM9
FIMX
15km
EnKF
GFS
No
60km
GSI/cyc GFS + WRF-chem,
chem
testing of Grell cu
No
FIM7
60km
GSI
GFS
No
FIMens
60km,
10mem
EnKF
GFS
No
FIMens
27km, 4mem EnKF
GFS
No
FIM
FIMY
- Stream 1.5
39
FIM Development at NCEP
Strong progress toward NCEP operational status
Preliminary tests of FIM done on NCEP CCS
MOA with NCEP to run FIM as part of Global Ensemble Forecast
System (GEFS) under NEMS.
GSD’s Advanced Computing Group worked closely with NEMS
developers at NCEP to work out basic NEMS design issues as well as to
get FIM under NEMS
Testing of FIM to start within NEMS GEFS framework at NCEP in next
few months
NCEP doing their own verification now of FIM
38-level version of FIM tested – equal performance with L64 –
readiness for GEFS. (Fewer levels should fare well with isentropic-hybrid
coordinate …. and apparently do)
40
HFIP Global Model/Physics Team plans for FY12
 FIM - Testing in GEFS under NEMS
 Model enhancements toward efficiency
 Testing of WRF physics (Grell cumulus), other physics

Critical for improving tropical cyclone diversity
 Physics – diversity between GFS and FIM should improve multi-model
TC ensemble



Grell 3-d convective scheme – now in testing
FIM-chem with more complex microphysics (2-moment cloud at least) for
coupling with aerosol (Saharan dust, sea salt, etc.) – Georg Grell
Stochastic physics, other devices to increase ensemble diversity – Jian-Wen
Bao, Georg
 Continued development of CFIM (coupled atmosphere-ocean FIM with
icos HYCOM)
 Use of long runs for evaluation of FIM TC climate and diagnosis of
physics problems
 Continue work toward FIM-based EnKF-hybrid (Phil Pegion, Jeff W)
41
7-day 500 hPa anomaly correlation
1 Aug – 31 Dec 2010, 135 cases
N. Hemis S. Hemis Arctic Antarctic
20-80N
20-80S
70-90N 70-90S
New version of FIM –
Nov 2011
73.46
65.66
66.66
46.97
GFSDC
72.90
65.20
65.49
44.46
Verification against GFS analyses, GFS used for initial conditions
NOTE: FIM physics matches that used for GFS in fall 2010, not using the May11
GFS physics modifications
Back
42
NMM-B Dynamical Core
 Nonhydrostatic Multiscale Model on B grid (NMM-B) (Janjic,
2005; Janjic and Black, 2007)



Further evolution of WRF NMM (Nonhydrostatic Mesoscale Model)
Intended for wide range of spatial and temporal scales, from meso
to global, and from weather to climate
Evolutionary approach, built on NWP and regional climate study
experience by relaxing hydrostatic approximation (instead of
extending cloud models to large scales; Janjic et al., 2001, MWR;
Janjic, 2003, MAP)



The nonhydrostatic option as an add–on nonhydrostatic module



Applicability of the model extended to nonhydrostatic motions
Favorable features of the hydrostatic formulation preserved
Reduced cost at lower resolutions
Easy comparison of hydrostatic and nonhydrostatic solutions
Pressure based vertical coordinate


Nondivergent flow on coordinate surfaces (often forgotten)
No problems with weak static stability on meso scales
43
Zavisa Janjic
Global
NMMB
~1 year
500 hPa
Height
Anomaly
Correlation
Coefficient
vs forecast
time
NMMB
initialized
and
verified
using GFS
analyses
Global
North Hemisphere
South Hemisphere
Tropics
44
Zavisa Janjic
Back