THE NCEP ETA DATA ASSIMILATION

1
THE NCEP ETA DATA ASSIMILATION SYSTEM (EDAS)
AN OVERVIEW
COMET Faculty Course on Numerical Weather
Prediction
10 June 1999
Presented by Eric Rogers
2
OUTLINE

• Philosophy

• General Description / History

• Current Operational 32-km EDAS

- Impact of 3DVAR analysis and cycled EDAS
- Observations
- Recent Performance The Good, the Bad, and
the Ugly

• Future Plans

3
GOAL PRODUCE BEST POSSIBLE INITIAL CONDITIONS
FOR ETA MODEL FORECAST

• KEY COMPONENTS

- State of the art analysis (variational)
- Consistency between assimilating and forecast
model (resolution, physics, dynamics)
- Intelligent selection and use of observations
NOT necessarily the same as fitting all the
observations exactly (more to come on this)
4
EDAS PAST CONFIGURATION MILESTONES

• June 1993 Eta-80 (Early Eta) replaces LFM
  initialized with single Eta Optimum Interpolation
  (OI) analysis at 00/12Z using GDAS first guess

• March 1995 Eta-29 (Meso Eta) 3-h EDAS with
  OI analysis used as first guess for 03/15Z
  forecasts

• October 1995 Eta-48 replaces Eta-80 in Early
  Eta slot, initialized by a 12-h intermittent EDAS
  cycle with 3-h OI analysis updates

5
EDAS Original Configuration Eta-48 forecast
00Z/12Z, Eta-29 forecast from 03Z/15Z
6
48KM EDAS PROBLEMS AND CONCERNS

• Forecasters wanted higher resolution run earlier

• Starting the EDAS using a GDAS first guess still
  caused inconsistencies (e.g., soil moisture)

• Eta OI analysis not easily adaptable for use of
  non-traditional data sets (satellite radiances,
  NEXRAD radial velocity)

• Eta OI analyses heights, not temperatures, so it
  cant use aircraft mass information

7
Eta-48 Surface Meteogram at Riyadh, Saudi Arabia
from 12Z 6/25/97 Solid
Observations Dashed Forecast
2-m T
2-m Td
12Z 6/25 00Z 6/26
12Z 00Z
6/27 12Z

• Ran operational EDAS/Eta-48 system over Saudi
  Arabia
• - Forecast surface temperature 4-10oC too
  cold
• - 00-h analysis Td 19oC, observed Td
  -6oC!

8
(No Transcript)
9
(No Transcript)
10
SOLUTION ETA-32 UPGRADE

• Stage 1 Implemented 9 February 1998

- Resolution of 00Z/12Z EDAS and Early Eta
changed from 48km/38levs to 32km/45levs
- Eta OI analysis replaced by Eta 3-d
variational analysis (3DVAR)
- Partially cycled EDAS Basic
atmospheric variables (T,q,u,v,sfc p) at the
start of the 12-h EDAS obtained from GDAS soil
moisture / temperature, cloud water, turbulent
kinetic energy from the previous EDAS cycle
- Meso Eta-29 unchanged
http//www.nws.noaa.gov/om/tpb/447.htm
11
ETA 3DVAR ANALYSIS (Parrish et.al 1996 NWP
Preprint Volume)

• Loosely patterned after NCEP global SSI analysis

• Analysis variables

- Stream function
- Temperature
- Potential function
- Specific humidity
- Surface pressure
- Geopotential height

• More adaptable for use of new data types than OI
  (e.g., NEXRAD radial velocities used in Eta-10
  runs during 1996 Olympics)

12
(No Transcript)
13
3DVAR vs. OI 00Z/12Z First Guess Temperature vs
RAOBS
14
3DVAR vs. OI 00Z/12Z first guess vector wind
RMS vs. RAOBS
15
3DVAR vs. OI
16
WHY DO CYCLING?

• Initial conditions more consistent with
  forecast model

• Less spinup of divergence, cloud, precipitation,
  and TKE

• More accurate representation of soil moisture

17
IMPACT OF CYCLING ON SOIL
MOISTURE
48km EDAS 12-h cycle
32km EDAS 3 week cycle
18
48-h Eta forecast 2-m temperature - sfc T
obs Model shows no disquieting drift from
cycled EDAS
Bias (deg C)
19
ETA-32 UPGRADE

• Stage 2 Implemented 3 June 1998

- EDAS converted to full cycling mode with
atmospheric variables cycled from the previous
EDAS as well as soil/cloud/TKE
- Meso Eta-29 converted to 32km
1) 03Z Eta-29 ----gt 03Z Eta-32 (33-h fcst)
2) 15Z Eta-29 ----gt 18Z Eta-32 (30-h fcst)
3) Both off-time runs connected to fully cycled
EDAS
- Robust backup system in place to ensure that
EDAS soil moisture is never reinitialized from
the GDAS
20
(No Transcript)
21
DATA DUMP TIMES
EDAS
Early Eta
22
OBSERVATIONS USED BY ETA 3DVAR

• Upper air data

- Rawinsonde height/temperature/wind/moisture -
Dropwindsondes - Wind Profilers - NESDIS
thickness retrievals from polar orbiting
satellites (oceans only) - VAD winds from
NEXRAD - Aircraft (conventional and ACARS)
winds/temps - Satellite cloud drift winds - SSM/I
and GOES precipitable water retrievals -
Synthetic tropical cyclone data

• Surface data

- Surface land wind/temperature/moisture - Ships
and buoys - SSM/I oceanic surface winds
23
DATA QUALITY CONTROL

• CQC Complex QC of raob height/temps (baseline,
  hydrostatic, lapse rate, radiation correction,
  etc.)

• ACQC Quality control of conventional aircraft
  data (remove duplicates, track checks, create
  superobs)

• SDMEDIT NCEP Senior Duty Meteorologist can
  flag all or parts of suspect raobs

• 3DVAR Analysis performs gross check vs. first
  guess

- Temperature /- 15oC - Wind /- 25 ms-1 - RH
/- 90 - Precipitable water /- 12 g/kg -
Height /- 100 m
24
Use of Surface Data Eta OI vs. Eta 3DVAR
Eta OI Analysis
Eta 3DVAR Analysis
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
RECENT ETA/EDAS PERFORMANCE

• Original 3DVAR in Eta-32 degraded surface
  analysis (especially moisture)

29
EXAMPLE 00Z 7/13/98 at Rapid City, SD
Solid RAOB Dashed 80 km OPNL 3DVAR 5oC
analysis error in Td at 850 mb !
Problems 1) 3DVAR tuned to give too much weight
to first guess 2) code error excluded all
surface data!
Solid RAOB Dashed 80 km Modified 3DVAR
30

• New 3DVAR tested in July 1998 and showed
  improved fit to surface and raobs (especially
  moisture)

• Re-tuned 3DVAR implemented on 3 November 1998

• We thought everything was OK..

BUT..
31
24-H ACCUMULATED PRECIPITATION
EQUITABLE THREAT SCORES ALL FCSTS
Solid Eta Short Dash NGM
Long Dash AVN/MRF
12/1/97 - 2/28/98
12/1/98 - 2/28/99
10-15 drop in Eta skill between 1997-98 and
1998-99
32
Persistent synoptic error in Eta-32 during winter
of 98-99 weaker and faster Eastern Pacific
troughs/cyclones than observed
Example 48-h Forecasts valid 1200 UTC 17 March
1999
33
AVN had much better SLP forecast
then Eta
34
PROBLEM 1 November 98 change degraded
mass/wind balance in 3DVAR

• If mass / wind balance well-behaved, positive
  height correction is coincident with center of
  anticyclonic wind correction

850 mb ANL-GUESS height/wind 80KM EDAS valid 00Z
3/15/99

• Note 10 degree longitude displacement between
  centers of wind and height correction

• Problem is most severe in regions and at
  analysis times without widespread raob data but
  with large amounts of wind or mass only data
  (e.g., satellite winds)

35
SOLUTION Improve geostrophic coupling of
mass/wind analysis corrections in 3DVAR
Operational 3DVAR analysis
Modified 3DVAR analysis
Note Improved height/wind coupling near
Aleutians
36
PROBLEM 2 Horizontal/vertical correlations too
narrow observation had VERY limited impact on
analysis away from its level

• One observation test Insert one height
  observation 10 m greater than first guess at 200,
  500, 900 mb and measure impact in
  horizontal/vertical

37
Operational 32-km 3DVAR
Modified 32-km 3DVAR
200 mb
900 mb
38
Performance of new 3DVAR 3 December 1998 to 16
January 1999 test at 80 km resolution
24-h accumulated precipitation threat
scores All forecasts Dashed
Modified 3DVAR Solid Operational 3DVAR
Equitable Threat Score
Threshold (in)
39
00,24,48-h forecast vector RMS wind error vs RAOBS
Dashed Modified 3DVAR Solid Operational
3DVAR
48-h
24-h
Vector RMS error (m/s)
00-h
Pressure (mb)
NOTE 200 mb 24-h modified 3DVAR error 00-h
error in operational 3DVAR!
40
NEW 3DVAR IMPLEMENTED 13 MAY 1999 IMPACT
200 mb Vector Wind RMS error vs RAOBS
3/1/99-5/31/99 Blue 48-h Eta-32 fcst
Purple 48-h AVN fcst
Old 3DVAR
New 3DVAR
RMS Vector Wind Error (m/s)
Rogers on leave software breaks
Further documentation at http//www.nws.noaa.gov
/om/tpb/3d-eta.htm
41
FUTURE EDAS DEVELOPMENT

• 0-6 months (Day 1 on IBM)

- Move 03Z Eta to 06Z run 12-h EDAS for 00/12Z
cycle 06-h EDAS for 06/18Z cycle - Use of
GOES/TOVS radiances in Eta 3DVAR

• 6-12 months

- Increase resolution to 22km/50levs -
Assimilation of observed precipitation (Lin
et.al, 1998 NWP Conference, 1999 Mesoscale
Conference)

• 12-18 Months

- Initialization of cloud water (Zhao et.al 1998
NWP Conference) - Use of NEXRAD radial velocity
and surface mesonet data in Eta 3DVAR switch
from 3-h to 1-h analysis cycle in EDAS
42
Impact of precip assimilation on Eta-80 24-h
forecast valid 12Z 7/5/98
Control forecast
Forecast w/precip assimilation
24-h observed precip (RFC analysis)
43
Rainfall Data Assimilation (Lin et.al, 1998 NWP
Conference, Phoenix, AZ)

• During the 12h pre-forecast assimilation period
  at each timestep compare the model predicted
  rainfall to observed
• Adjust the models latent heating profile
  accordingly (Carr and Baldwin, 1991)

44
Cloud Data Assimilation(Zhao et al, 1998, 12th
NWP, Phoenix, AZ)

• Data sources
• real-time Neph Analyses (USAFGWC)
• hourly radar/gauge observations