Title: A%20Brief%20Overview%20of%20NCEP%20SREF:
1 A Brief Overview of NCEP SREF Where are we
and where are we heading? Jun Du, Geoff
DiMego, Zoltan Toth, Jeff McQueen
and Binbin Zhou Acknowledgements Eric Rogers,
Tom Black, Geoff Manikin, Brad Ferrier, Dusan
Jovic, Matt Pyle, Hui-Ya Chuang, Henry Juang,
Jongil Han, Zavisa Janjic, George Gayno, Kenneth
Mitchell, NCO stuff as well as Stephen Lord and
Louis Uccellini
http//www.emc.ncep.noaa.gov/mmb/SREF/SREF.html (N
CEP Ensemble User Workshop, Oct. 31 Nov. 2,
2006)
21.How to capture uncertainties (ensemble
system design) 2. How to convey forecast
uncertainty (ensemble products and data) 3.
Downstream applications of the SREF (driving
other prediction systems) 4. Verification
3(1) To Capture Uncertainty
4- Atmospheric IC uncertainty
- (1) Multi-analysis (GFS and NAM)
- (2) Perturbed analysis bred vector (current), ET
(ongoing/2007), and ET with DA-based rescaling
(future, work with Mozheng Wei) -
- Atmospheric LBC uncertainty
- (1) different LBCs for different members simply
provided by various NCEP global ensemble members - (2) plan to work on how to better couple LBC and
IC in future - Lower boundary/Earth surface forcing
uncertainty - (1) Land surface currently not perturbed but
soil moisture perturbation design is underway --
first a simple approach (Du, 2007) and then might
be an ET-like approach in near future - (2) Water/sea surface temperature, flux etc.
(currently not perturbed) - Upper boundary uncertainty from space????
5Small domain 500H spread, May 19, 1998, Eta32, 5
members
f00
f12
f24
f36
Perturb both IC LBC
perturb LBC only perturb IC only
63h-apcp forecast with WRF_ARW model
with NAM soil moisture
Diff between the two forecasts (nam -
gfs)
with GFS soil moisture
7Within WRF_NMM model Impact on T2m is
significant!
With nam soil moisture (NMM)
T2m diff (namSM gfsSM, NMM)
With gfs soil moisture (NMM)
8T2m -- impact comparison (NMM) Soil moisture
impact on T2m is bigger Than that of IC and of
model difference!
different IC (NMM namSM)
Diff soil moisture (nam gfs NMM)
different models (nmm vs. arw namSM)
9Model uncertainty (both dynamics and physics)
(1) multi-model Eta, RSM, WRF-NMM, WRF-ARW
(current), plan to replace Eta and RSM members
with all WRF members in 2008 (issue how
important will the diversity of model dynamics
be?) (disadvantage difficult to maintain
different model has different bias) (2)
multi-physics GFS, Eta, MM5, BMJ/SAT, KF/DET,
SAS/RAS (current), diversity in LSM, cloud, PBL,
radiation as well as WRF-GFS, WRF-RUC in near
future disadvantage dont produce enough spread
in larger scale (Du) and cannot maintain extra
spread after a certain time length (according to
Peter H.) (3) stochastic physics currently an
ongoing research on NCEP global ensemble
(Dingchen Hou) and could be adopted to SREF once
its ready. (issue can stochastic physics alone
perform as well as or exceed what multi-model and
multi-physics ensemble can currently provide in a
consistent manner?)
10(No Transcript)
11(No Transcript)
12 RSM Error
Corr. Coeff.
13Accuracy of probabilistic forecast measured by
RPSS and Reliability
14Before introducing physics diversity
After introducing physics diversity
15Cold season case 1 (09z, March 5, 2004) spgt
charts comparison at 39h fcst
Before physics diversity (988mb)
After physics diversity (988mb)
16Warm season June and July, 2004
PAR SREF
OPS SREF
17SLP
500H
Par - - - - - Opl --------
850T
850U
250U
850RH
41-case averaged outlier rate at all fcst lead
time
18Cold season case 1(09z, March 5, 2004) ens
spread comparison at 63h fcst similar
PAR/12h-apcp
OPS/12h-apcp
PAR/SLP
OPS/SLP
19 Sensitivity Experiment on IC vs Physics
(Experiment I)
IC-ensemble
PHY-ensemble Eta_BMJ_ctl
Eta_BMJ_ctl Eta_BMJ_n1
Eta_KFc_ctl Eta_BMJ_p1
Eta_SAT_ctl Eta_BMJ_n2
Eta_DET_ctl Eta_BMJ_p2
Eta_CON_ctl Eta_BMJ_n3
Eta_Fer_ctl Eta_BMJ_p3
Eta_RAS_ctl
20IC-ensemble spread (SLP at 63h)
PHY-ensemble spread (SLP at 63h)
More sensitive to IC pert than PHY Diversity in
general except for the Cyclone region where is
more convective
PHY-ensemble mean 63h fcst (SLP)
21IC-ensemble spread (12h-apcp at 63h)
PHY-ensemble spread (12h-apcp at 63h)
Similar sensitivity to both IC pert and to PHY
Diversity for precip in general. PHY diversity
enhances sensitivity in Convective areas.
PHY-ensemble mean 63h fcst (12h-apcp)
22PHY-ensemble spread (CAPE at 63h)
IC-ensemble spread (CAPE at 63h)
Equally sensitive to IC pert and PHY Diversity
for CAPE (focusing different Sub-regions though)
PHY-ensemble mean 63h fcst (CAPE)
23Besides dealing with the uncertainties in initial
state, external forcing, governing laws and
numerical methods as discussed above
Residual Part (either uncounted or over
estimated uncertainties) post processing to
calibrate ensemble system to have a reliable
probabilistic forecast. Currently, work is going
on only for the 1st-moment (statistical decaying
average by Cui and dynamical/flow-dependent
dual-resolution approach by Du (2007). BMA-type
approach is planned to be tested in future for
the 2nd- and higher moments (Toth) Huiling Yuan
of GSD (former FSL) is also working on a bias
correction method for the SREF.
24 Post-processing with Hybrid-Ensembling
appraoch (Du, 2004) fcst error of 21-mem ens
mean of 500mb H at 3.5 days
Raw SREF fcst error
Hybrid SREF fcst error
25Without extra observation within uncertainty
source area
f00
F51 targeted time
With Extra observation within uncertainty
source area (uncertainty reduced)
f00
F51 targeted time
21z, May 11, 2005 SREF SLP spread forecast
(current cycle)
26Without extra observation within uncertainty
source area
With extra observation within uncertainty source
area
rms error of SLP by ensemble mean forecast (error
reduced in targeted region)
27SREF Configuration
- 21 members (5 Eta_bmj, 5 Eta_kf, 5 RSM, 3
WRF_nmm, 3 WRF_arw ? about 10 WRF_nmm 10
WRF_arw in 2008) - 32-45km ? 32km for all members (2007) ? 20-25km
(2009) - Four cycles per day (03,09,15,21z) ? 00,06,12,18z
in future - 3hrly output to 87hr each cycle (possibly hourly
output in future) - Three output domains CONUS, Alaska and
Pacific/Hawaii
28(2) To Convey Uncertainty
29SPC first-ever Day-2 High Risk outlook for
April 7, 2006s over 60 tornadoes event
SPCs Significant Tornado Probability forecast
Derived from SREF (David Bright)
30Subject SREF and the sig tornado event last
FridayFrom David Bright David.Bright_at_noaa.govDa
te Mon, 10 Apr 2006 094412 -0500To Jun Du
Jun.Du_at_noaa.govHi Jun,I wanted to show you the
image SREF attached. This is an experimental
parameter one of the forecasters is testing and
it's definitely not ready for prime time (i.e.,
I'm not convinced its formulation is entirely
defensible at this point, and its false alarm
ratio has yet to be determined!). But, this F039
hour SREF forecast from last Thursday (09 UTC
SREF 6 April 2006) shows the primary significant
tornado threat centered over nrn AL, nrn MS, and
parts of TN (with SREF mean 850 mb height
overlaid).The SREF played a large part in the
first ever Day 2 "HIGH RISK" outlook ever issued
by the SPC (http//www.spc.noaa.gov/products/outlo
ok/archive/2006/day2otlk_20060406_1730.html). Of
course, other deterministic models, climatology,
and forecaster experience were involved in the
decision process, but the SREF played a key role.
Again, this product only in the evaluation
stage, but I thought you'd be interested
nonetheless.DavidLouis Uccellini
wroteJoe/Russ Between this SREF result, the
first time Day 2 High Risk and the 1.0 POD for
over 60 tornadoes, I think it is time to
celebrate in a big way. I recommend that you guys
need to put together a one pager and work with
Dennis and Carmeyia to highlight this historic
accomplishment(s). Louis
31SubjectSREF done good!FromRichard Grumm
Richard.Grumm_at_noaa.govDateWed, 25 Jan 2006
102426 -0500ToJun Du Jun.Du_at_noaa.govJun,Hel
lo, hope you are doing well. The recent event of
22-23 January was another good example of the
SREF doing quite well with PTYPE issues. Humans,
like the 2-3 April 2005 event did not heed the
guidance.FYIhttp//nws.met.psu.edu/severe/2006/
23Jan2006.pdfRich
32 Ensemble Product type in general 1. single
outcome mean, median, mode, extremes 2.
uncertainty spread, confidence factor 3.
distribution probability, spaghetti, meteogram,
postage stamps, clustering
33(No Transcript)
34 SREF Aviation product List (Binbin)
Variables (14) Products Levels/Regions Algorithms
Icing Prob of occurrence FL240,180,150,120 090,060,030,FL000 T-RH-Vertical drift
Clear air turbulence (CAT) Prob of occ (LGT,MDT,SVR) FL420,390,360,330 300,270,240,FL210 Ellrod 1992
Cloud Mean/spread Prob of ovc,sct,bkn,clr Ceiling, cloud base Model-output
Visibility Mean/spread Surface Stoelinga 1999
Ceiling Mean/spread Ceiling, cloud base Conditional
Flight-Restriction Prob of LIFR,IFR,MVFR,VFR Vis and/or Ceiling Visibility-ceiling
Jet-stream Prob of gt60,80,100 kt 4500, 18000, 34000 ft. Model-output
10m-wind speed Mean/spread Prob of gt10,20,30 kt 10m Model-output
Convective cloud Mean/spread/directions Grid convective cloud
Low level wind shear (LLWS) Mean/spread Prob of severe (20 kt/2000 ft) Wind vector difference
Tropopause Mean/spread Model-output
Freezing-level Mean/spread Sfc, 4000, 8000, 12000, 16000 Model-output
Fog Prob of occurrence Surface Cloud base/top threshold
Mountain obscuration to be added Prob of occurrence Mountains RH 80, 90 Cloud water/ice/snow
35Transition?
- Currently, all those ensemble products are
generated using an ensemble product generator
code - In future, they should be generated using NAWIPS
ensemble functionality package
36Field Forecasters like SREF web products
- Date Mon, 13 Feb 2006 154015 0000
- From David Eastlack ltDavid.Eastlack_at_noaa.govgt
- Subject SREF web page
- To Jun.Du_at_noaa.gov, Jeff.Mcqueen_at_noaa.gov
- Greetings, I felt compelled this morning
to send a message of appreciation for your
outstanding SREF web site! We peruse the SREF
products regularly, and find them a beneficial
part of the forecast process. Once again, great
work! I hope that it will be possible, sometime
in the near future, to have products such as
these considered operational. We do realize
though that since this is an experimental page,
it is not guaranteed to be always updated. Again,
thanks and great work! - Dave Eastlack MET WFO - OAX (Omaha, NE)
37(No Transcript)
38The problem is that
- Its not operational but run at personal account
(not reliable) - Solution is that it needs to be transferred to
NCO (a user survey strongly supports this)
39Survey announcement (09/08/2006)
List
Since the current SREF web page
(http//www.emc.ncep.noaa.gov/mmb/SREF/SREF.html)
is a developement display tool in nature and not
operational, it's not 100 reliable to users and
not friendly enough. As you may also know that
NCO has developed an official operational SREF
web (http//www.nco.ncep.noaa.gov/pmb/nwprod/analy
sis/) but not enough products on it currently.
However, the former should be eventually replaced
by the later for everybody's good. There will be
an ensemble user workshop at NCEP from Oct. 31
-Nov. 2 this year. Zoltan suggested that we
gether a list of products which field forecasters
use from SREF before the workshop. As a first
step, you might compare my current personal SREF
page and the NCEP official page to let us know
what products need to be transferred to the
official page. Surely, you can also request
whatever you might want to see.
Although I received
emails from forecasters from time to time about
the SREF web products, I didn't keep a track of
it (such as a forecaster from Texas area
requested the Post Stamp charts, but I couldn't
recall his email address). So, please
circulate this servey to field forecasters where
you see proper. Remember that this servey should
be done before Oct. 27. So, please act as soon as
you can. Thanks a lot. Jun
40Subject Re Fwd SREF web products
serveyFrom Jon Zeitler Jon.Zeitler_at_noaa.govDat
e Sun, 10 Sep 2006 113600 -0500To Peter
Manousos ltPeter.Manousos_at_noaa.govgt,
Jun.Du_at_noaa.gov
Jun/Pete, FYI, I was the one
requesting the postage stamp graphics.
Transferring Jun's
personal page to the operational versionis
exactly what I would like to see (except of
courseanything still in development). The
postage stamps,ability to select thresholds for
the spaghetti charts, andthe probability.
Eventually, it would be nice to getthese into
AWIPS like the GFSEnsemble data we
currentlyreceive. Many thanks,_________________
____________________________Jon W. Zeitler
Science and Operations OfficerNational Weather
ServiceAustin/San Antonio Weather Forecast
Office2090 Airport RoadNew Braunfels TX 78130
USA
41From John.Kwiatkowski_at_noaa.govJohn,Sally
forwarded your email to me since she knows I look
at the SREF on a consistent basis. When I am
trying to forecast QPF/probabilities/timingin
addition to looking at forcing and RH, I use the
QPF probabilitiesprovided by the model and IMHO
have found these to work better than MOSwhen
strictly looking at QPF POPs and timing for my
developing of aforecast. Also, I use the
instability and kinematic fields when tryingto
forecast whether or not the possibility exists
and/or the magnitudeof severe weather in the
12-60 hr period. One improvement I suggest isto
have more forcing/upward vertical velocity field
products. Thiswould help specifically in terms
of cap erosion forecasting andstratiform QPF,
but most importantly in the forecasting process
ingeneral. Bryan.Smith_at_noaa.gov
42Subject Re Fwd SREF web products
serveyFrom Seth Binau ltSeth.Binau_at_noaa.govgtDat
e Mon, 18 Sep 2006 132257 -0500To Jun Du
ltJun.Du_at_noaa.govgtCC Dan Baumgardt
ltDan.Baumgardt_at_noaa.govgt
Jun,
Thanks for the opportunity for
the field to provide some input on SREF. We use
SREF heavily in our office, and continue to
increase our use as more and more. While I use
the NCO SREF page a great deal, I do prefer your
page due to the wide array of products not
available on NCO's page.
If NCO takes this over
explicitly...there are some products we would
like them to make available as listed below...
As far as
PROBABILITY...1) 3/6/12/24 hour precipitation
and snowfall with the thresholds you have
(0.01"...0.10"...0.25"....0.50" etc)2)
precipitation type (rain/sleet/frozen/snow/)3)
CAPE/CINH
As far as Mean Spread1) MSLP2) 850T3)
500 Hght and Vorts4) Precipitation (3/6/12/24
hr) (mean)5) Snowfall (3/6/12/24 hr) (mean)6)
Precipitable Water7) Instability
As far as SPAGHETTI
(all thresholds you have)1) MSLP2) 850T3)
500 Heights4) Precipitation (3/6/12/24 hr)
As far as
Individual Members (stamp charts)1) MSLP2)
850T3) 500 Hght and Vorts4) Precipitation
(3/6/12/24 hr)5) Precipitable Water6) CAPE/CINH
Thanks
for providing such great support to the field!
Seth
Binau, NWS ARX
43Subject Re SREF web products serveyFrom Dan
Baumgardt Dan.Baumgardt_at_noaa.govDate Fri, 13
Oct 2006 153947 -0500To Jun Du
Jun.Du_at_noaa.govJun Du I STRONGLY agree that
your SREF maintained site should migrate to
operational!! The data is great that you have and
be AT LEAST the starting point for the ops
version!
We are using these data heavily! October 7 is
the last data - oh no! Thanks!
Dan Baumgardt -
KC9IWMScience and Operations OfficerNational
Weather ServiceN2788 County Road FA, La Crosse,
WI 54601Phone (608)784-8275 x766 Fax
(608)784-8238Email dan.baumgardt_at_noaa.govhttp/
/www.crh.noaa.gov/arx
44Subject Re Fwd SREF web products
serveyFrom David Zaff David.Zaff_at_noaa.govDate
Sun, 22 Oct 2006 085317 0000CC "jun.du gtgt
Jun Du" ltJun.Du_at_noaa.govgt
Jun,
I like the AniS java applet on
the official page. It'd be nice if you could add
a "framelabel" section with date/time stamps so
when one zooms in/out we'll have some clue as to
where we are in the forecast. The probability
graphics are a big hit in the BUF office,
particularly with precip. Nice job. These and
other ensemble charts really opened our eyes to
some of the utilities to ensemble forecasting.
Dave
45Subject SREF request followup from 8 Sept
2006From Walter Drag Walter.Drag_at_noaa.govDate
Mon, 23 Oct 2006 064353 -0400To Jun Du
Jun.Du_at_noaa.govCC David Vallee
David.Vallee_at_noaa.govGood morning Jun,
Surveying our
Taunton (KBOX) office, we would like to see some
of the following added to the NCO SREF page.
Probability of sustained winds greater than 33
knots (marine assistance) every 3 hrs out to 87
hours.
Probability of 24 hour snowfall (0.1, 2, 4, 8,
12, 16 inches).Walter Drag
cc Our KBOX Science and
Operations Officer - David Vallee
46Subject Re SREF web products serveyFrom James
A Nelson James.A.Nelson_at_noaa.govDate Fri, 27
Oct 2006 131045 -0800To Jun Du
Jun.Du_at_noaa.govJun,In Alaska, we would like
to see the same data you have on your web site to
be carried to the NCO web site. There is no SREF
data for AK on the NCO Site. As far as
additional probability fields, I would recommend
a CAPE value gt 200 as well as a wind speed
greater than 63 kt. The CAPE value seems to be a
good indicator of convection in AR. At least for
the Anchorage forecast area. The wind speed
greater than 63 kt speaks to the hurricane force
wind warnings for our marine areas.Jim (Alaska
Region)
47Subject Re SREF web products serveyFrom Jeff
Craven Jeffrey.Craven_at_noaa.govDate Fri, 27 Oct
2006 163135 -0500 To Jun Du
Jun.Du_at_noaa.govJun Du...thanks for
asking.Probability of exceedance charts are
extremely useful.Surface Visibility lt 1/2
mile10m Winds (sustained) gt15, gt20, gt25, gt34,
gt50 knots10m Gusts (mean wind in mixed layer
gt25, gt35, gt50, gt64 knots850 mb Temps lt0, lt-2,
lt-5Sfc Temp lt32, lt28, lt24, lt0Sfc Temp gt90,
gt100Sfc Td gt55, gt60, gt65, gt70, gt7512 hour
Precip gt0.01, gt0.10, gt0.25, gt0.50, gt1.00,
gt2.0024 hour Precip same as aboveCAPE gt0, gt250,
gt500, gt1000, gt2000, gt30000-6 km Shear gt20 knots,
gt30 knots, gt40 knots, gt50 knots0-1 km Shear gt15
knots, gt20 knots, gt25 knots, gt30
knotsCombination of CAPE gt500, 0-6 shear gt40
knots, and Conv Precip gt0.01Combination of
CAPEgt1000, same as aboveCombination of CAPE
gt2000, same as aboveCombination of CAPE gt1000,
0-6 shear gt30 knots, and Conv Precip
gt0.01Combination of CAPE gt2000, same as
aboveCombination of CAPE gt3000, same as
aboveCombination of CAPE gt1000, 0-6 shear gt20
knots, and Conv Precip gt0.01Combination of CAPE
gt2000, same as aboveCombination of CAPE gt3000,
same as aboveCombination of 2m RH lt30, 10m wind
gt15 mph, 2m temp gt60Probability of Precip
TypeSnow, Freezing Rain, Ice Pellets, RainMean
and Standard DeviationPrecipitable Water
1000-500 mb RH850 RH, 700 RH, 500 RH, 300 RH12
hour QPF12 hour SnowSurface PMSL850 mb
temp850 mb Tdsurface Td (2 m)surface temp (2
m)surface wind (10 m)700 mb temp500 mb
temp500 mb Height850 mb wind500 mb wind250 mb
windJeff Craven, SOO Sullivan/Milwaukee WI
48SREF data
- Raw data, 2-year archive at NCEP supercomputer
(accessible to NCEP service centers) - ftp site mean/spread/prob as well as individual
members of selected variables in grib1 format
(e.g. Penn State office to public) - NOMADS (e.g. Stony Brook etc. to public)
- NAWIPS 2007 (all WFOs)
49 SREF Grib Mean/Spread Products
Mean and spread Parameters Units Level
( not in Spread files) 2m
Temperature K Sfc 10m U,
Vwind m/s 10 m Total precipitation(3,6,12
,24hr) kg/m2 Sfc Convective Avail. Pot.
Energy J/kg Convective inhibition (CIN)
J/kg Storm RElative Helicity
(SREH) m2/s2 0-3000 m Lifted
Index K 0-30 mb abv grnd Sea Level
Pressure Pa Sfc Pressure Pa 1000
-50 mb (every 50 mb) ? Categorical rain
y/n Sfc Dominant precip type (over
3hr) 1-7 Sfc
12hrly Large scale Snow Fall kg/m2 Sfc 1
2hrly Snow Depth kg/m2 Sfc 12hrly
Accumulated Snow Fall kg/m2 Sfc Absolute
vorticity /s 1000-50 mb (every 50
mb)? Geopotential height gpm 1000-50 mb
(every 50 mb)? Relative humidity 1000-5
0 mb (every 50mb)? U, V- wind m/s 1000-50
mb (every 50 mb) Temperature K 1000-50
mb (every 50 mb)? Thickness gpm 1000-850,
1000-500, 850-700mb Total preciptable water
50 SREF probabilistic products
Probabilistic Parameters Units
Threshold____________ Convective Avail. Pot.
Energy J/kg 500, 1000, 2000, 3000,
4000 Convective Inhibition (CIN) J/kg
-50, -100, -200, -300, -400 Storm Relative
Helicity (SREH) m2/s2 100,
150, 200, 250, 300 Lifted Index K
0, -2, -4, -6, -8 Precipitation (3, 6,
12, 24 hr) Inches 0.1,
0.25,0.5, 1.0, 2.0 12hrly Accumulated
Snow Fall Inches 1, 2.5, 5,
10, 20 Prob precip type is rain
Prob precip type is freezing rain
Prob precip type is snow or sleet
51SREF Variables at NAWIPS/2007
- Means/Spreads
- Heights at 1000, 850, 700, 500, 250 mb
- UV at 1000, 850, 700, 500, 250 mb 10 m
- Temperature 850, 700, 500 mb 2 m
- Dew Point (RH) 850, 700, 500 mb 2 m
- QPF at 3, 6, 12 and 24 hour totals
- 12-hr Snowfall
- Sea Level Pressure
- Precipitable Water
- Probabilistic Fields
- 3-hr/6-hr QPF GE .01, .25, .50, 1.0
- 12-hr/24-hr QPF GE 01, .25, .50, 1.0, 2.0
- 12-hr Snowfall GE 1, 4, 8, 12 (have 2.5,
5, 10, 20) - Temperature at 2 m 850 mb LE 0oC
- 10 m Wind GE 25 kt, 34 kt, 50 kt
- CAPE GE 500, 1000, 2000, 3000, 4000
- Lifted Index LE 0, -4, -8
- Surface Visibility LE 1 mi, 3 mi
- Cloud Ceiling LE 500 ft, 1000 ft, 3000 ft
52(No Transcript)
53(No Transcript)
54(No Transcript)
55(No Transcript)
56(No Transcript)
573. Downstream Applications of SREF
- DTRA dispersion model (HPAC/SCIPUFF, Jeff
McQueen, Patrick Hayes, Steve Hanna and PSU/David
Stauffer) - SREF drives Air Quality Forecast system
(research, a student of Prof. Eugenia Kalnay from
UMD, Ms. Debra Baker, is working with us now) - Hydrology in future?
584. Verification
- Grid2grid package (against analysis, current)
- Grid2obs package (fvs, against station obs,
current) - Unified package in plan (lead by Toth and Geoff.
Yuejian will have a presentation on this from
global ensemble/NAEFS side)
59Thank You and Looking forward to work with you in
times to come!