Winter Weather Refresher

1
Winter Weather Refresher

• Stephen Jascourt and Bill Bua
• COMET NWP resources
• Stephen.Jascourt_at_noaa.gov
• Bill.Bua_at_noaa.gov at NCEP

2
OUTLINE 1. Global Forecast System what did we
learn last winter? 2. Global Forecast System
whats new for 2003-2004? 3. Eta Model what did
we learn last two winters? 4. Eta Model whats
new? 5. Nonhydrostatic Mesoscale Model (NMM) 6.
Short range ensembles (SREF) 7. RUC-20 8.
COMET Slides with this blue background indicate
transition to next item in above outline
3

• Global Forecast
• System
• What did we learn last winter?

4
October 2002 and August 2003 Implementations

• Increased resolution from T170L42 to T254L64 for
  first 84 hours of forecast (10/02)
• Replaced long wave radiation scheme 8/02 (now
  called RRTM)
• Results in warmer troposphere, colder
  stratosphere
• Should mitigate the GFS cold bias in troposphere
• Colder stratosphere may impact data assimilation
  of satellite radiances

5
Anecdotal evidence about the GFS

• GFS deepens mid-tropospheric troughs too much in
  eastern North America, particularly past day 3-4
• Evidence that the GFS prefers positive PNA
  pattern (ridge west, trough east North America)
• GFS often has better storm tracks along the Gulf
  and Atlantic coasts than the Eta-12
• Reasons for above are unknown and may be
  regime-dependent
• Likely will continue in winter 2003-04
• Forecasters should assess anew this winter

6
Contrasting Winter Regimes 01-02 vs 02-03
Negative PNA in winter 2001-02 (positive height
anomalies over eastern US)
Positive PNA in winter 2002-03 (positive height
anomalies over Alaska/Yukon, negative over east
coast)
7
Comparison of 5 day 500-hPa height error DJF
01-02 to DJF 02-03
Too much troughing over eastern US, not enough
across north Pacific and Hudson Bay
Too much ridging over northern oceans, too much
trough over NH mid-latitude continents
8
GFS lower tropospheric cold bias in cold season
00 UTC GFS analysis and forecast biases for
850-hPa temperatures at 12-h intervals for
January 2002
0ºC
24-h
analysis
Cold bias over west, spreads across northern
plains during forecast.
36-h
Magnitude may depend on flow regime
12-h
48-h
9
Compare January 2002 to 2003 850-hPa for 1st 48
hrs of forecast
00 UTC GFS analysis and forecast biases for
850-hPa temperatures at 12-h intervals for
January 2003
0ºC
24-h
analysis
Overall forecast bias still gets colder with time
but pattern of bias different under different
regime
PNA regime changes biases in upper Midwest by
about 1-2ºC?
36-h
12-h
48-h
10
Cold air damming and propagation along barrier
bad due to sigma coordinate 2001-02 at T170
Model topo. (m)
Contourstemperature error (surface, deg C)
Colorsmodel terrain height
24-hour GFS fcst of 2-m temperature was 8C too
warm over High Plains during start of arctic
outbreak (T170)!
11
Cold air damming and propagation along barrier
bad due to sigma coordinate 2002-03 at T254
Model topo. (m)
Contourstemperature error (surface, deg C)
Colorsmodel terrain height
At T254, 10ºC too warm at 2-m in GFS 48-hr
forecast near DEN!
12
AVN generally too wet too large precip area and
with a generally wet bias
AVN 36-h forecast of 24-h precipitation verifying
12z 3 March 2002
24-h gage analysis of precipitation verifying 12z
3 March 2002
13
T254 still has same precipitation bias
Grid 211 24, 48, 72-hr fcsts of 24-hr accum prec
ECMWF
ECMWF
0.4
0.4
GFS
UKMET
UKMET
GFS
0.2
0.2
Equitable Threat Score
Equitable Threat Score
JAN 2003 Dry month
FEB 2003 Wet month
UKMET
GFS
ECMWF
UKMET
ECMWF
GFS
1.0
1.0
BIAS
BIAS
14
GFS tends not to be able to remove enough
elevated CAPE

• Why do we care about this in the cold season?
• Can result in overdevelopment of frontal waves
• Waves move too far into the cold air
• Overdevelopment results in problems with amount
  (too much), location, and type of precipitation
• Example follows from winter 2001-2002
• Note This problem is not expected to improve
  with resolution increase on October 29, 2002
• The source of the problem is physics, rather than
  dynamics

15
GFS tends not to be able to remove enough
elevated CAPE - example
FORECAST versus ANALYSIS low positions and
pressures at 6 hour intervals from 00z7Dec01
through 12z9Dec01from GFS run of 00 UTC 7 Dec 2001
996
1003
1006!
1008
1012!
1011
12z09Dec01
12z08Dec01 (lows collocated)
16
GFS tends not to be able to remove enough
elevated CAPE - example
Verification, 12z8Dec01
Verification, 12z9Dec01
Low tracks Forecast Analysis
17
Reasons for overdevelopment in this GFS forecast
GFS tends not to be able to remove enough
elevated CAPE

• Deep, moist, conditionally unstable elevated
  layer
• Convective scheme cannot remove this instability
  (or enough of it)
• Grid-scale scheme convects instead, which results
  in too much
• Latent heating in 850-500 hPa layer
• Vorticity spin-up at low- and mid-levels
• Frontal wave intensification
• Moisture convergence in lower troposphere
  (results in even more moisture entering the grid
  column!)

18
Ensembles more consistent run to run than
operational higher-resolution GFS
Yellow operational MRF,
same valid times Initial00 UTC 8 April 2002
Initial00 UTC 9 April 2002
19
3-day forecast from 00 UTC 11/2/01, spaghetti
diagram for ensemble
global
Ensembles help assess forecast confidence and
range of scenarios
Uncertain location of incoming western trough
Uncertain amplitude of eastern trough
From CDC web site http//www.cdc.noaa.gov/map/i
mages/ens/ens.html
20
Relative Measure of Predictability (RMOP)measure
of how likely the ensemble mean is
http//wwwt.emc.ncep.noaa.gov/gmb/ens/index.html
(note wwwt may become www)

• Based on last 30 days of ensemble performance to
  take into account regime predictability and
  general model performance
• Ensemble mean and each ensemble member placed in
  equally likely climatological bins (bins vary
  seasonally and geographically to account for
  typical variability)
• RMOP colors with percentage below color bar show
  the percentile rank of todays forecast compared
  to the last 30 days for number of ensemble
  members agreeing with their ensemble means
  (agreeing with in the same bin)
• For example, red (90) means the ensemble
  distribution has more members in the same bin as
  the mean than 90 of the cases in the past 30
  days, suggesting this is among the most
  predictable forecasts in the last month
• RMOP probability numbers (above the color bar)
• Calibrated probability that ensemble mean will
  verify based on how often the ensemble mean
  verified when the same number of ensemble members
  were in the bin containing the ensemble mean
  during the past 30 days

21
Unpredictable heights in Strong gradient
Ridge/Trough Highly predictable
22

• Global Forecast
• System
• Whats new for 2003-2004?

23
Resolution through boreal winter 2003-04
GFS 00, 06, 12, 18 UTC MRF are same as fields
labeled AVN. MRF fields to be discontinued
T254 L64
T170 L42
T126 L28
84h 180h 384h
3½d 7½d 16d
2003 until . Planned change (may only be
resolution)
by 12/6/03
T126 L28
T126 L28
Ensembles
T62 L28
T62 L28
84h 180h 384h
3½d 7½d 16d
84h 180h 384h
3½d 7½d 16d
11 members (1 control, 10 perturbations)
11 members (1 cont., 10 pert) 00 UTC,
12 UTC 00 UTC,
06 UTC, 12 UTC, 18 UTC
24
Topography comparisonT254 topography (0-3.5 days)
25
Resolution of topography affects precip forecast
Verification
T170
T254
Sharper precipitation maxima, slightly better
placement of precipitation as a consequence of
increased horizontal resolution (first 3 1/2 days
only!)
26
Topography comparisonT170 topography (3.5-7.5
days)
27
Topography comparisonT126 topography (7.5-16
days), also ensembles 0-3.5 days until 12/06/03,
then 0-7.5 days
28
Topography comparisonT62 topography(ensembles
3.5-16 days until 12/06/03, then 7.5-16 days)
29
New Long Wave Radiation Scheme and Changes to
Cloud-Long Wave Radiation Interaction

• More efficient (runs twice as fast)
• More accurate (by a factor of 5 to 10!)
• Decreased lower tropospheric cold bias and upper
  tropospheric/lower stratospheric warm bias in
  parallel experiments
• High stratosphere cold bias occurs (may affect
  data assimilation of radiances?)

Details at http//meted.ucar.edu/nwp/pcu2/avclra
d2c.htm and http//meted.ucar.edu/nwp/pcu2/avradtr
4.htm
30
GFS warmer with new long wave scheme
Skin T
Skin and 2-meter temperatures with RRTM long wave
are higher than old GFS LW radiation early in the
forecast at high latitudes
2-meter T
31
Near-surface temperature increase in RRTM over
old GFS LW radiation increases through 5 days
(average difference around 1oC)
GFS warmer with new long wave scheme
32

• Eta ModelWhat did we learn last two winters?
• Model has been stable (no major changes, several
  minor fixes) from December 2001 through June 2003

33

• Examine the analysis!
• Compare against satellite, radar, surface data,
  etc.!
• Large scale features set the forecast scenario
• Model details and high resolution topography and
  coastlines will not help forecast accuracy if the
  large-scale winds are not well forecast or the
  cyclone track or intensity is off.
• Look off the coasts is the Atlantic ridge too
  weak in the model? Is the trough off the west
  coast sharp enough? Is the jet core, where
  parcels are peeling anticyclonically into,
  through, and out of, in the right position? How
  do you expect errors in such features will affect
  the strength of a cold surface high or the
  amplitude and timing of a major wave in the model
  forecast?

34
Sensitivity to multiple factors
B, F (in)
A,C, D, E, G, H (mm)
Initial conditions

• GFS vs. Eta initial state in same model
• compare B vs. F
• compare D vs. G

B
C
D
A
Physics

• Different convective parameterizations in same
  model
• compare C vs. D vs. H

Resolution
F
H
E
G

• Different resolutions in same model
• compare A vs. C vs. B

• Also, different models with same initial
  condition
• compare E vs. F

Look at different models and ensembles! Check
for bad init. cond. and unphysical behavior in
forecast
35
Remember, saturation for ice occurs at much lower
RH with respect to water

• Affects your interpretation of cloud base/top
  and cloud coverage

(new 32 km SREF is between brown and pink curves)
Model cloud top of overcast deck
Forecast sounding
Ice saturation threshold in 12 km Eta model
(new 32 km SREF is between brown and pink curves)
Model is saturated with respect to ice
Model cloud base
36

• Drying trend during forecast
• precipitable water becomes steadily drier during
  forecast compared to verification
• QPF also dries up at increasing forecast range
  compared to verification

• Monthly total 24-h forecast minus observed
  precipitation for Feb 2003
• lower (or more negative) values at later times
  into forecast period

Valid at 36 h
Valid at 60 h
Valid at 84 h
37
Watch for moisture stream getting intercepted by
convection

• Convective scheme drops too little precip,
  leaving moisture stream free to reach area where
  dynamics are causing grid-scale lift in colder
  air.
• In reality, convection intercepts moisture stream.

Moist inflow
7inches
5inches
1inch
Moist inflow
Moist inflow
38
Snow drifting downwind while falling Most falls
on upwind side but some advects downwind of ridges
39
Snow drifting downwind while falling Most falls
on upwind side but some advects downwind of ridges
40
Lake Effect

• Lake-effect band placement excellent
• precip intensity too weak by factor of at least 3
  overall and 10 for peak local amounts
• cant resolve multiple bands and waves

41
Isothermal layers form at 0 oC
Hourly BUFR sounding
0oC
42
Precipitation Type microphysics vs. diagnostic
output
Hourly BUFR sounding
now in BUFKIT
43
Precip type grids are not from model microphysics
Baldwin-Schichtel diagnostic algorithm

• Tends to have bias against SN in Eta
  overforecasts ZR
• Purpose is to alert forecaster to potential
  hazardous weather (ZR is most hazardous) so that
  forecaster inspects situation carefully and
  determines for him/herself the precip type

44
Precipitation Type microphysics vs. diagnostic
output
Overrunning case at LEX (Kentucky)
Hourly BUFR sounding
Both all liquid
0oC
45
Precipitation Type microphysics vs. diagnostic
output
Overrunning case at LEX (Kentucky)
Hourly BUFR sounding
Microphysics72 frozen precip Baldwinrain
0oC
46
Precipitation Type microphysics vs. diagnostic
output
Overrunning case at LEX (Kentucky)
Hourly BUFR sounding
Microphysicsmix of 21 frozen Baldwinrain
0oC
47
Precipitation Type microphysics vs. diagnostic
output
Overrunning case at MSL (Alabama)
Hourly BUFR sounding
Microphysics94 frozen Baldwinrain
0oC
48
Precipitation Type microphysics vs. diagnostic
output
Overrunning case at MSL (Alabama)
Hourly BUFR sounding
Microphysicsmix with only 13 frozen Baldwinrain
0oC
49
Patchy snow cover with bare ground spots changed
26 Feb 2002

• Before model fix (as in soundings to the left)
• 2-meter temperatures too cold over snow
• 850 temperatures too warm over Canada
• arctic boundary layer poorly handled
• After model fix (as in schematic below)
• 2-meter temps warmer, 850 temps cooler so
  verifies better
• arctic boundary layer structure still poor,
  seldom makes very stable even when it should

Too warm (before fix)
Too cold
50
Land surface upgrade summer 2001

• Cold season processes (Koren et al 1999)
• Patchy snow cover
• Frozen soil (new state variable)
• Variable snow pack density (new state variable)
• Soil heat flux under snowpack (Lunardini 1981)
• New maximum snow albedo database (Robinson
  Kukla 1985)
• Takes into account observed effect of vegetation
  on the albedo of grid box

51
Effect on high temperature with thin snow cover
NEW snowcover - ground gt freezing
OLD snowcover groundfreezing
Ground holds at freezing
SKIN TEMP
OBS
OBS
Model ? 0 C
First day better but still too cool Second day
worse because not all snow melted yet
2 m AIR TEMP
North Platt, Neb.

• previous model formulation (until snow completely
  melts)gt all incoming energy melts/sublimates
  snow gt skin temp held at freezing
• gt 2-m air temp held near freezing
• Current formulationgt patchy snow cover for
  snow depth less than threshold depth (veg-type
  dependent) gt reduces surface albedo gt
  accelerates melting gt more available energy at
  sfc
• gt skin temp can exceed 0 C gt 2-m air temp rises
  further above freezing.

0 C
18Z
52
Problems with light, fluffy snowcover

• If daily satellite snow analysis (from Satellite
  Analysis Branch) has snowcover where model first
  guess has none, then snow pack is added to the
  Eta analysis
• Depth assumed to be 1.5, 51 snowwater ratio
  (yields water equivalent of 0.3)
• If actual snow cover has less average water
  content than 0.3, it will melt sooner and the
  ground will heat faster in reality than in the
  model forecast

9-hr 850-hPa temp fcst
9-hr 2-m temp fcst
Forecast is lt4ºC
6º-12ºC
Verification is 8º-12ºC!
53

• Model outputs
• FREEZING LEVEL
• Extended below model topography using standard
  atmosphere lapse rate, but this will miss cold
  air trapped below valley inversion
• 12-km grids getting into AWIPS
• NCEP sends the following fields on 12-km grid for
  distribution over SBN
• precipitation and convective precipitation (3 h
  accumulation)
• T, RH at 2 meters
• U, V at 10 meters
• MSLP and EMSL (Shuell and Mesinger reductions to
  sea level)
• station pressure on model terrain
• precipitable water
• CAPE/CIN based on parcel from lowest model layer
• LI based on most unstable of the 6 bottom
  30-hPa-thick average parcels
• helicity (0-3 km using storm motion from Bunkers
  method)

54

• Eta Model
• Whats new for 2003-2004?

55

• Summer 2003 change bundle (5 items)
• Forecast impact mostly from item 3, especially
  better holding in low clouds. Other items combine
  for slight improvement in winds and moisture
  early in forecast (up to 12 hours)
• Convection and fundamentals of data assimilation
  unchanged, therefore overall forecast character
  will be same as before
• NEXRAD WINDS
• Superobs (combining individual data points) to 1
  km radial x 6o azimuth
• Not used where VAD quality flag says bad data
  (including birds) or where radar beam runs into
  model topography
• 32 km test shows no forecast impact
• 10 km test but without assimilation cycling shows
  improved fits to raob initially but no overall
  forecast impact
• 2) Radiance processing
• Upgrade to global model processing methods,
  including new emissivity model over land and use
  NOAA-17 polar orbiter data (not previously used,
  should improve analysis over Pacific though test
  error stats show no change)
• Allows channels used previously only over water
  to be used over land.
• Hardly any overall forecast impact (32 km test)

56

• Summer 2003 change bundle (5 items)
• 3) Output and Radiation made consistent with the
  new (Nov 2001) cloud microphysics
• Microphysics fields cycled
• Longwave effects of clouds updated hourly instead
  of every 2 hours
• Convective cloud fraction increased shallow
  convection ? 10 cloudcover
• Formulation for cloud fraction changes to Xu and
  Randall (1996), instead of Randall (1994)
• RH for cloud fraction calculated consistently
  with Nov 2001 microphysics
• Cloud optical properties changed
• Output changed (modified or new variables)
• Precipitable water field now includes only vapor
  (had included condensate, resulting in excessive
  values in regions of very heavy precipitation)
• Cloud water, cloud ice, rain, snow, and total
  condensate separately output
• Output cloud-base and cloud-top pressures from
  shallow nonprecipitating convection, deep
  convection, and grid-scale clouds separately
• Visibility calculations have been changed to use
  the new cloud fields, responding to mixing ratios
  of cloud water, cloud ice, rain, and snow
• Low-level clouds in the lowest 100 mb and
  upper-level clouds above the tropopause are no
  longer ignored

57

• Summer 2003 change bundle (5 items)
• 3) Output and Radiation made consistent with the
  new (Nov 2001) cloud microphysics
• Forecast impact in 32 km test
• lt0.5 precip improved, gt0.5 even drier
• Improved slightly all forecast hours fit to
  raobs and 2-meter temperatures
• General forecast impact usually seen in
  individual forecasts
• Increase in partly cloudy and overcast
  conditions (was too little)
• Smaller diurnal range when low clouds are
  present (was too large)
• Cooler daytime temperatures where shallow
  boundary-layer cumulus form
• Fog/low clouds burn off slower during the
  morning (occurred much too early still somewhat
  too early)
• Consistently better positioning of warm fronts
  during the daytime in moist situations when low
  clouds keep the surface cool, preventing mixing
  from advancing the surface warm front

58

• Summer 2003 change bundle (5 items)
• 4) GOES cloud-top assimilation
• Remove moisture above observed cloud tops
• Add moisture at observed cloud levels
• Use as top anchor point for precip assimilation
  if have to add precipitating cloud
• Forecast impact on overall error statistics
  mixed, but viewing individual cases shows
  improvement in structure and pattern of
  precipitation early in the forecast (32 km test)
• 5) Stage IV precipitation assimilation
• Upgraded from Stage II to Stage IV adds quality
  control
• Forecast impact small but improvement measured in
  precipitation during EDAS cycle, which improves
  soil moisture a little (32 km test)
• Also, many additional output variables in grib
  files on NCEP server
• Hourly output to 36 hours, and 6, 18 UTC runs out
  to 84 hours
• New land surface variables including snow depth
  and percentage of snow cover
• More discussion on the model change bundle at
• http//deved.meted.ucar.edu/nwp/pcu2/EtaMay2003upd
  ate.htm
• More technical details in the Technical
  Procedures Bulletin at
• http//wwwt.emc.ncep.noaa.gov/mmb/tpb.spring03/tpb
  .htm

59
No METAR surface temperatures in Eta analysis
12 UTC analysis 00 UTC EDAS analysis (includes
late data) plus 3
h forecast and assimilate data for new analysis
at 03, 06, 09, 12 UTC
Why METAR temperatures removed? Creates bad
analysis soundings!
Starts with good 00 UTC EDAS sounding
3h forecast from 9 UTC analysis much too cold
above nocturnal inversion
12 UTC raob
00 UTC raob
How? By spreading influence upward
Times are valid times of 1, 2, and 3 hour
forecasts
Arrows show when data added

• Hourly soundings starting with 00 UTC EDAS.
• Small radiative cooling in bottom 75 hPa every
  hour
• Large cooling through lowest 150 hPa after data
  added at 3, 6, and 9 UTC.

Sounding jumps when data added
60
No METAR surface temperatures in Eta analysis

• RUC still uses METAR temperatures. RUCs
  terrain-following coordinate system allows it to
  take advantage of good ways to handle surface
  data.
• Eta still uses other parts of METARs (winds,
  dewpoints, pressure)
• Eta stopped using all surface temperature
  observations over land on Sep 10, 2003
• Eta uses only surface observations within 6
  minutes of analyses times as of Sep 10, 2003
• GFS never used METAR temperatures, only uses
  METAR pressures

Whats the forecast impact?
Precip 3 weeks summer improved over east,
especially 24 h amounts gt 0.50, westno
change Surface T, Td, wind nearly same as with
METAR temperatures Aloft (heights, T, wind)
nearly same, slight improvement in
mid-troposphere Overall large improvement in
32-km reanalysis fits to obs, but brief 12-km
test smaller changes
RMS differences from raob temperature profiles
No T from METAR
With T from METAR
Analyses
12 h forecast 60 h
forecast
60 h forecast fits raob T same except worse near
ground
12 h forecast fits raob T same except worse near
ground
Fits raob T profile better except near ground
RMS temperature difference from raobs
32-km reanalysis, winter (1 month) 12-km
test, summer (3 weeks) 12-km test, summer (3
weeks)
61

• Changes testing for possible mid-winter
  implementation (late January or February)
• Follow the comet.eta newsgroup for updates in
  case things dont go as planned
• 1) Overhaul of short wave radiation but may not
  get in
• Will be like GFS except will ignore short-wave
  absorption by oxygen and carbon dioxide
• GFS solar radiation described at
  http//meted.ucar.edu/nwp/pcu2/avradtr1.htm
• Was the oldest component of the model and needed
  catching up
• Forecast impact will reduce the present high
  bias in incoming solar radiation, both clear sky
  and through clouds. Forecast is consistently
  colder preliminary tests show increased errors
  because too cold will not be included in change
  package unless can be corrected
• 2) Land surface changes
• Will reduce snow depth threshold for patchy snow
  (presently assumes bare patches if depth less
  than 5 over grass up to 16 over forest)
• Albedo will vary with solar zenith angle, so more
  solar energy is reflected and less available for
  heating at low sun angles (e.g. morning, evening
  and all day during winter at high latitudes)
• Precip adds to snowpack if model microphysics
  indicates frozen fraction gt 50 (presently, adds
  to snowpack if lowest layer air temperature is
  below freezing)
• Numerous other technical and numerical
  refinements
• Forecast impact expected to reduce diurnal range
  (which has been too large) and probably cause a
  net overall cooling compared to present model

62

• Changes testing for possible midwinter
  implementation (late January or February)
• 3) Bias adjustment of precipitation assimilation
• Multisensor precipitation analyses are currently
  assimilated into the EDAS
• Assimilated starting 12 hours before the model
  initial time
• Has affect of stimulating precipitation during
  12-hour assimilation cycle where observations
  show it should occur and inhibiting it elsewhere
• Affects soil moisture and thus later evaporation
• Helps spin up microphysics and vertical motion in
  right places based on observed precip
• NO PRECIP ASSIMILATION WHERE SNOW only where
  rain
• Currently the precip amounts assimilated have a
  low bias
• Bias as determined by large-scale several-week
  comparison of gauge and multisensor analyses will
  be removed this is the only change
• Forecast impact primarily warm season, not
  winter. Should slightly reduce the tendency of
  the model to dry out during the forecast
• 4) GOES 12 radiances added
• These will substitute for GOES-8 radiances.
  Eastern North America and offshore waters have
  had no GOES radiance coverage in the Eta model
  since GOES-8 was replaced by GOES-12 in April,
  2003
• OVERALL temperatures cooler and matches diurnal
  curve better

63

• NONHYDROSTATIC
• MESOSCALE
• MODEL
• New hybrid vertical coordinate
• Replaced the Eta model for high-resolution
  windows runs in 2002
• May replace Eta completely during 2005
• Provides operational support for fire weather and
  dispersion/emergency
• Similar physics to Eta
• No assimilation cycling yet, starts with Eta-12
  analysis
• BEGINS THE TRANSITION TOWARD WRF
  (Weather Research and Forecasting Model)
• Modular design different model options plug in
• WRF will allow collaborators outside NCEP to run
  the same model and to supply parameterizations
  that will more readily work in the NCEP
  operational environment (a great difficulty now)
• NMM will be one version of WRF
• WRF ensemble will run in high-resolution window
  slot Fall 2004

64
High-resolution Window Runs

• Eta used for initial and boundary conditions
• Once per day 10 km over Alaska, 8 km over west,
  central, east CONUS
• Twice per day 8 km over Hawaii, Puerto Rico
• Not operational system reliability pretty good
  but not 100

65

• Small-domain specialty runs
• 26 small domains
• One operational 8-km run each at 00, 06, 12, 18
  UTC
• Supports fire weather operations during fire
  season
• SPC selects domain if severe threat high
• HPC can select domain for winter weather event
• 4-km runs on call operational service to feed
  dispersion model in emergency

66
(No Transcript)
67
Hybrid and Eta Coordinates
Ptop (constant pressure)
Ptop (constant pressure)
? 0
Pressure domain
400 hPa
? 0
Sigma domain
? 1
MSL
? 1
68
Vertical resolution and vertical coordinate
Layers slope with terrain 60 flat
layers, 60 everywhere
fewer above high terrain

• Tick marks are actual model levels from BUFR
  sounding files
• Same stations used on left (NMM) and right (Eta)
• Note difference in station elevation. NMM
  usually lower (stations usually in valleys)
• Note difference in vertical resolution with
  increasing station elevation

NMM Eta
s coordinate up to 400 hPa p coordinate above 400
hPa
69

• NMM Forecast Characteristics (compared to Eta)
• Large-scale conditions same as Eta
• Nocturnal cold bias, esp. clear/calm
• Cold bias with old snow because treated same as
  fresh snow in NMM
• 10-m winds slightly too weak over western states
• More flow through valleys
• Not as good for valley inversion trapping
• Better for downslope (and diurnal upslope too)
• More flow over ridges instead of around ridges
• Better gravity waves affects winds and lee-wave
  temperatures
• More detail
• Catches more of the spotty mountain convection
  missed by Eta
• Too much orographic precip in heavy precip
  episodes
• Features closer to true amplitude
• Small displacement error gives bigger (worse) RMS
  error stats than too-smooth Eta model yet more
  physically descriptive forecast

70
Topography NMM vs. Eta

• 30 data (approx 1 km)
• Eta 12 km silhouette topo
• Valleys must be at least 2 grid boxes 24 km
  wide because step coordinate requires zero wind
  at valley walls
• NMM 8 km mean topo
• Peaks about same, valleys deeper and sharper

71
Topography NMM vs. Eta

• Peaks not generally higher in NMM
• Valleys deeper in NMM
• More detail in NMM
• Even more detail over high plains in NMM

Zoomed view
72

• Topographic influence on winds
• NMM Eta

Flow into/through valley
Flow over valley - good if well-mixed
- good for
trapping/damming
Downslope follows terrain
Downslope doesnt reach bottom

or corners
Overhead view, Topo contours
Flow over
Flow around
73
Dissemination

• GRIB and BUFR data available for ftp
• Fields do not go out over SBN
• Model will not run if hurricane model needs to
  run (they share same time slot on computer)
• How do I get it?
• The model grib files are accessed on
  ftp//ftpprd.ncep.noaa.gov/pub/emc/mmb/mmbpll/
• In these directories
• alaska10.t00z
• central08.t12z
• east08.t18z
• hi08.t00z
• hi08.t12z
• pr08.t06z
• pr08.t18z
• west08.t06z

74
Short-Range Ensemble Forecasts (SREF)

• What? (caution changes planned by December,
  details a few slides later)
• 5 Eta 48 km (control 2 perturbation pairs)
• 5 Regional Spectral Model 48 km (control 2
  perturbation pairs, based on GFS analysis) RSM
  currently has old AVN/MRF physics
• 5 Eta members using Kain-Fritsch convective
  parameterization
• When?
• 21, 09 UTC in time for your use with 00, 12 UTC
  Eta
• Status?
• Officially operational (24x7 computer
  support/reliability)
• Output might get into AWIPS in fall 2004 (OB-4)
• New user-friendly web interface linked from SREF
  home page,
• which is http//wwwt.emc.ncep.noaa.gov/mmb/SREF/
  SREF.html

75
http//wwwt.emc.ncep.noaa.gov/mmb/SREF/SREF.html N
ote wwwt address may change to www
Go here, and it brings up this
76
http//wwwt.emc.ncep.noaa.gov/mmb/SREF/SREF.html N
CEP Short range ensembles on the web. Note wwwt
address may change to www
77
Mean and Spread charts Dominant Precip Type

• Shows the precipitation type diagnosed in the
  largest number of ensemble members

Precipitation type determined by
Baldwin-Schichtel algorithm
78
Spaghetti Diagrams Range of uncertainty
79
Probability Charts percentage of members
exceeding threshold
SREF
Percentage of members with QPF gt .25/24h
010519/0000V63 SREFX-CMB 24HR PQPF OF .25
80
Probability Charts threshold exceeded by
specified percentage
SREF
Highest QPF at each point exceeded by 60 of the
ensemble members
010519/0000V63 SREFX-CMB SHADED, IN AT LEAST
60 OF MEMBES
81
Individual Station Plots MeteogramsEnsemble
mean and all members. Experimental, available by
fall 2003linked on http//www.emc.ncep.noaa.gov/m
mb/research/meso.products.html
82
Precip-type algorithmsensemble of ptype from
operational Eta. Available during fall/winter
linked on http//www.emc.ncep.noaa.gov/mmb/researc
h/meso.products.html
83
Tentative New SREF Configuration (by 12/03)

• 15 members (same number as now)
• More physics diversity (convection scheme and
  microphysics)
• Less initial condition diversity one
  positive-negative perturbation pair (2 now)
• 32-38km or equivalent (RSM) horizontal resolution
  with 60 levels
• Upgrade RSM physics to current GFS physics
• BUFR sounding and surface data to be available
  for all members

 
84
Reasons for proposed change

• Insufficient spread (verifying analyses falling
  outside range of solutions provided by the SREF)
• Evidence that physics diversity increases spread
  in solutions (even in cold season, when dynamics
  typically more important)

Contours around surface low
85
Distribution verifies better with physics
diversity

• Analyze distribution with Talagrand diagram
• Rank each ensemble member at each grid point from
  lowest to highest forecast value. 15 ensemble
  members means rankings are 1, 2, 3, , 15
• Identify verification as less than 1, between 1
  and 2, , between 14 and 15, higher than 15
• Make climatology of how many times verification
  falls into each position

Ensemble spread too small verification less than
smallest forecast or bigger than largest forecast
value 37 of the time
Ensemble spread slightly better verification
less than smallest forecast or bigger than
largest forecast value 25 of the time
SREF with 15 members 5 Eta using BMJ
convection, 5 Eta using KF convection, 5 RSM
SREF with 10 members 5 Eta using BMJ
convection, 5 RSM
Percentage with verification value gt largest
forecast value
Percentage with verification value lt smallest
forecast value

• New SREF, 15 members
• More physics diversity
• fewer initial condition perturbations

Ensemble spread nearly perfect! Verification
almost equally likely in every slot.
86
Winter Weather Experiment III Participants Map
87
When will WWE-III occur?

• Begins Oct. 1, 2003 (Intermountain WFOs)
• Test and evaluation period conducted Sept. 15 -
  Sept. 30, 2003
• Nov. 1, 2003 (Remaining WFOs)
• Test and evaluation period conducted Oct. 15 -
  Oct. 31, 2003
• Occurs twice daily
• HPC routinely produces enhanced graphics in
  support of WWE
• WFOs participate only when impacted by an event
• View graphics and participate in a 15 minute
  (maximum) collaboration call
• Collaboration occurs after arrival of updated
  operational and ensemble guidance, but still
  early enough in the forecast process to foster
  different opinions among participants
• Ends April 1, 2004
• May be extended to no later than May 1, 2004
  depending on how active the winter has been

88
WWE-III HPC will provide..

• Accumulation Graphics
• Loops of precipitation in 6 hour increments out
  to 72 hours (CONUS)
• Melted QPF, Snow, Freezing Rain, and the precip
  type grid used to convert melted to winter precip
• HPC forecasters chooses a QPF source (HPC, Eta,
  GFS, or SREF) and a precip type grid to convert
  the QPF to accumulations (from the Eta, GFS, or
  SREFs)
• EVENT Total Accumulation Graphic (ETAG)
• Manually edited accumulations over an event
• Separate graphics for Combined Snow/Sleet and
  Freezing Rain
• Valid from issuance time through end of event or
  72 hours max
• For non Intermountain Region WFOs only
• Watch/Warning Exceedance Potential Graphics
• Shows by how much the ETAG exceeds watch/warning
  criteria
• Separate graphics of both combined snow/sleet and
  freezing rain for both 12 and 24 hour thresholds
• Not viable in the Intermountain region
• Criteria by elevation does not allow this
  strategy to work
• Low Tracks Graphic
• HPC forecast of surface low position and track
  over the CONUS
• Forecast of central pressure of low in 12 hour
  increments out through 72 hours
• Clustering of available model solutions displayed
  on same map
• 500 mb Heights Graphic

89
HPC Winter Weather Experiment Products

• Products are available on WWE Web Site
• http//hpc.ncep.noaa.gov/wwe
• Password-protected to prevent general viewing of
  not-ready-for-prime-time products
• Contact Pete Manousos (Peter.Manousos_at_noaa.gov)
  to obtain password to site

90
Winter Weather Products directly fromSREF
(graphical)

• (Products on NAWIPS at HPC)
• Probability of freezing rain for each 3, 6, 12
  and 24 hour period
• Joint probability of freezing rain and PQPF
  exceeding specified criteria
• Mean, maximum and minimum snow amounts and
  freezing rain for 3, 6, 12 and 24 hour periods

91
SREF case examples 31 Jan 2002
Probability of Snow
Probability of Freezing Rain
27 hour forecasts valid 12 UTC January 31
Dominant Precipitation Type
9AM Radar Jan. 31, 2002
92
SREF case examples 31 Jan 2002
9AM Satellite January 31, 2002
Wheres the rain-snow line?
93
SREF case examples 30 Dec 2000 (Millenium
Storm)
24 h accumulated precipitation Eta 36 h forecast
from 12 UTC 29 Dec 2000
SREF spaghetti diagram 24 h accum. precip gt 0.5
in 24 h fcst from 12 UTC 29 Dec 2000

• Operational Eta forecast heavy snow across
• Washington, D. C. and Baltimore metro areas and
  
• southeast PA.
• Most SREF members kept the heavy precipitation
  offshore
• Official forecast Winter Storm Warning, 3-6
  DC, 5-10 Baltimore
• Verification clear skies, no precipitation
  across DC/Baltimore/northern Maryland but
• heavy snow fell from the northeast end of
  Philadelphia northeastward

94
6-7 JAN 2002, observed snowfall
SREF case examples 6-7 Jan 2002
95
SREF case examples 6-7 Jan 2002
3 h QPF 0.2
Sea-level pressure 1002 hPa

• Spaghetti diagrams from SREF run just hours
• before snowstorm began
• Forecast heavy precip and forecast snow area do
  not intersect - no forecast of heavy snow!
• Previous SREF runs had precip further south,
  hardly any where heavy snow verified
• Why was SREF forecast so bad? Continue

Contours outline precip type snow
96
SREF case examples 6-7 Jan 2002
Raob wind, raob hght, analysis wind, hght
Notice winds in trough axis, especially sharpness
of observed vs. analysis wind shift and 100 knots
observed at Atlanta. Eta analysis just as bad
97
SREF case examples 6-7 Jan 2002
Every individual ensemble member initial
condition which includes perturbations, such as
the member shown here (labeled rsmp1), still did
not come close to matching the observed strength
and sharpness of the winds in the trough
axis When the analysis has large errors, even
ensemble perturbations wont include reality, and
forecast verification will lie outside the
ensemble envelope! Always look out for bad
analyses!
98
SREF case examples 6-7 Jan 2002
Convection over Gulf in sharp trough often
trouble for analysis!
99
RUC-20

• 3D-VAR analysis implemented (yes it finally
  happened in May 2003)
• See info on COMET NWP matrix page at
• http//meted.ucar.edu/nwp/pcu2/index.htm and
  the
• RUC page at http//maps.fsl.noaa.gov/ (esp.
  see online TPB)
• Large improvement in error stats (precip, raobs,
  METARS, sensible weather, visibility, cloud
  patterns, everything!)
• 2-level snow model
• GOES cloud top assimilation, boundary layer
  profilers
• many many model changes.
• Still uses theta-sigma hybrid coordinate.

100
Expected Effects of New RUC 3-D Var

• Slight improvement or about equal skill overall
  in 3-h and 12-h forecasts compared to those from
  the previous RUC OI analysis as verified against
  rawinsondes.
• Closer fit to observations than the RUC OI
  analysis.
• A smoother analysis increment (correction to 1-h
  forecast field) resulting in less noise in
  short-range forecasts.
• Capability for assimilation of indirect
  observations such as radial winds, satellite
  radiances and wind speeds.

101
Precipitation Type in RUC

• RUC Precip type is
• not derived from the Baldwin-Schichtel algorithm
  used for Eta p-type grids
• not always exactly the same as in the models
  microphysics, though its close
• RUC Precip type output can be mixed (includes
  multiple types)
• define surface T based on minimum orography,
  same as used for 2-meter temperatures fits METAR
  elevations more closely than topography used for
  dynamics and has deeper valleys
• Includes rain when microphysics rain rate at
  ground is not too small and surface T gt 0oC
• Includes freezing rain same as rain but surface
  T lt 0oC and at any higher level T gt 0oC
• Includes sleet when microphysics graupel rate at
  ground is big enough and bigger than the snowfall
  rate at the ground and there is enough rainwater
  at some level and surface T lt 0oC
• Includes snow when
• Microphysics snowfall rate at ground is not
  exceedingly small, or
• Microphysics graupel rate at ground is big
  enough but smaller than snowfall rate, or
• Microphysics graupel rate at ground is big
  enough and 0oC lt surface T lt 2oC, or
• Microphysics graupel rate at ground is big
  enough and there isnt much rainwater at any
  level
• For details on RUC output variables, see
  http//ruc.fsl.noaa.gov/vartxt.html

102
RUC Land-surface Process Parameterization

• Updated in 20 km for fall 2002 (last year)
• change in thermal conductivity better diurnal
  cycle
• frozen soil physics, 2-layer snow model

Purpose Improve near-sfc, precip, cloud
fcsts Ongoing cycle of soil moisture, soil temp,
snow cover/depth/temp)
2-layer snow model
103
RUC 2-layer snow model update
Problem Too cold temps at night (with clear
skies, low winds) over thin snow layer. Similar
to Eta patchy snow problem solved with crisis fix
Feb 02 Solution couple thin layer of soil
with thin layer of snow cover (implemented
October 2002)
7.5 cm
S n o w
2-layer snow model
4 cm
5 cm
S o i l
1-layer snow model
coupled snow-soil layer
104
TEMPERATURE OVER SNOW COVER comparison between
operational RUC experimental RUC w/thin snow fix
RUC 2-layer snow model update
Valid 1200 UTC 5 March 2002
Difference big here!
Control (21-h fcst)
Experimental Control difference
Observed
Control - 19 C Experiment - 10 C Observed
- 11 C (12 F)
Experimental (21-h fcst)
105
After thin snow fix, area of snow cover better
matches the NESDIS snow coverthough still not
perfect.
RUC 2-layer snow model update
RUC-40, old
RUC-20, new
10 February 2001 1500 UTC
More realistic snow cover with fix
106
Oct 2002 thin snow fix improves surface
temperature forecasts in areas with shallow snow
cover
RUC 2-layer snow model update
Bias results from 4-14 February 2001
with thin snow fix, reduced cold bias
Avg surface temp biases from 3-h fcst for
stations with snow depth lt 10 cm
Feb 4 5 6 7
8 9 10 11
12 13 day
0 12 0 12 0 12 0
12 0 12 0 12 0 12
0 12 0 12 0 12 0 hour
with thin snow fix, reduced cold bias
Avg surface temp biases from 3-h fcst for
stations for all stations with snow cover
0 12 0 12 0 12 0
12 0 12 0 12 0 12
0 12 0 12 0 12 0 time
107
RUC good for light, fluffy snow cover
Remember the case shown in the Eta section when
thin snow melted early in the forecast period,
but not in the model, so Eta forecast 2-meter
temperatures were much too cold? Heres the same
case, RUC vs. Eta comparison. RUC did well.
RUC
Eta
3º-6ºC
10º-12ºC
108

• COMET
• Keeps you informed through the Eta and AVN
  newsgroups.
• Read it like email, but it doesnt get mixed up
  with your
• regular email
• Post questions, get answers
• Read questions from other forecasters
• Traffic light, not too much to read
• Discuss/explain important model error
  characteristics when
• they are happening, timely information!
• Start at http//meted.ucar.edu/nwp/newsgroups/inde
  x.htm

109
COMET ensemble training
Ensemble Module at http//meted.ucar.edu/nwp/pcu1
/ensemble/ Coming soon, might be ready by the
time you read this Ensemble Forecasting
Powerpoint From WDTB Winter Weather Workshop,
July 2003 Two versions Condensed version
http//www.wdtb.noaa.gov/workshop/WinterWxIV/pres
entations/shortversionWintWxWkshp2003SJ.ppt Full
version http//www.wdtb.noaa.gov/workshop/Winter
WxIV/presentations/WintWxWkshp2003SJ.ppt
110
The COMET NWP Model Matrix
http//meted.ucar.edu/nwp/pcu2/
111
http//meted.ucar.edu/nwp/pcu3/cases/
More cases being added, some under development now