TAMDAR and its impact on RUC Forecasts

1
TAMDAR and its impact on RUC Forecasts

• Bill Moninger, Stan Benjamin, Brian Jamison,
  Tracy Lorraine Smith, Ed Szoke, Tom Schlatter
• NOAA / ESRL / GSD

2
AMDAR and TAMDAR

• AMDAR (Automated Meteorological Data and
  Recording) are automatically sent from
  commercial aircraft, mostly large jets
• TAMDAR (Tropospheric AMDAR) automatic reports
  from (currently) 50 turboprops flying regionally
  in the US Midwest

3
World-wide AMDAR coverage, Tuesday 5 June
2007 more than 250,000 observations in 24 h
4
Over CONUS, all altitudes more than 172,000
observations in 24 h
5
Coverage is limited to major hubs below 20 Kft,
(without TAMDAR)
6
Below 20 Kft, with TAMDAR better regional
coverage in the Midwest
7
TAMDAR Variables

• TAMDAR measures temperature and winds aloft, as
  does the rest of the AMDAR fleet
• In addition, TAMDAR measures
• water vapor
• turbulence (not discussed)
• icing (not discussed)

8
Parallel real-time RUC cycles

• dev2 includes TAMDAR and all other data
  typically assimilated by the RUC
• dev lacks only TAMDAR data
• Both cycles use NAM boundary conditions
• Both run at 20-km, but are otherwise similar to
  the operational 13-km runs
• Background fields are set equal every 48 h

9
Verification

• dev and dev2 forecasts are verified against RAOBs
• Usually at 00 UTC only (when TAMDAR are flying)
• Usually we look at 3-h forecasts
• We consider 3 regions

10
Great Lakes Region includes 13 RAOBs Eastern US
Region includes 38 RAOBs National Region is the
RUC domain (CONUS and adjacent)
Results today are from the Great Lakes region,
where most TAMDAR aircraft fly
11
RAOB Verification

• RUC and RAOB soundings are interpolated to 10 mb
  levels
• Comparisons of T, RH, W are made at every level
  from the surface upward
• No RAOB data are automatically eliminated
• A few RAOBs eliminated by hand
• (about a dozen in 18 months)

12
Temperature RMS error time series, 3-h
forecasts, Great Lakes Region, surface to 500 mb
Red dev RMS error Blue dev2 RMS error Black
difference
TAMDAR impact up to 0.2 K
13
Temperature RMS error profile, 3-h forecasts,
Great Lakes Region
TAMDAR impact 0.4K at 900 mb
For 4 months in early 2007
14
Maximum expected impact

• Because of RAOB instrument and representativeness
  errors, we expect the forecast error to never be
  less than the analysis error
• The analysis RMS T error is 0.5 K
• TAMDAR reduces the 3-h forecast error from 1.6 K
  to 1.2 K, or 0.4 K
• The maximum reduction would be from 1.6 K to 0.5
  K, or 1.1 K
• Thus TAMDAR impact represents about 35 of the
  maximum expected 3-h T forecast error reduction
  at 900 mb.

15
Wind error time series, 3-h forecasts, Great
Lakes Region, surface to 500 mb
TAMDAR impact small but consistently positive.
Limited by low quality aircraft heading
information.
16
Wind RMS error profile, 3-h forecasts, Great
Lakes Region
TAMDAR impact 0.25 m/s at 700 mb
17
Maximum expected impact

• Analysis fit to RAOBs is 2.2 m/s
• Thus, TAMDAR impact on 3-h wind forecasts
  represents a 15 reduction in 3-h wind forecast
  error at 700 mb

18
RH error time series, 3-h forecasts, Great Lakes
Region, surface to 500 mb
TAMDAR impact up to 2 RH
19
RH RMS error profile, 3-h forecasts, Great Lakes
Region
TAMDAR impact 2 RH below 550 mb (slightly
negative at 200 mb! of little importance)
20
RH RMS error profile, no TAMDAR (red), TAMDAR
(blue), and Analysis (black), Great Lakes Region
RMS errors for the dev and dev2 3-h forecasts
RMS error for the dev2 analysis
For 3-h RH forecasts, TAMDAR reduces the RMS
error by 15 to 25 of the way to the analysis
fit below 550 mb, thus represents a 15-25
reduction in 3-h RH forecast error in this region
21
Retrospective Runs

• All RUC data were saved for a 10-day period
• 12 UTC 26 November to 12 UTC 5 December 2006
• Includes a potent winter storm early, and more
  moderate weather later
• We have rerun the RUC over this period with a
  variety of
• assimilation parameters
• TAMDAR data variations

22
RH Assimilation parameters

• High temporal- and spatial-resolution upper-air
  RH data have not been available in the past
• What is the appropriate RH observation error
  (like a weight) to be used for TAMDAR?
• Weve tested several errors for impact.
• Too small an error gt overfitting
• Too large an error gt reduced impact

23
An example of overfitting. (This plot uses
verification data from mandatory levels only.)
Between 13 Sept 1 Dec 2005, RH error was set to
1. Note negative TAMDAR impact.
24
RH error tuning

• We have tried the following RH errors
• 1, 3, 12, 18, 25
• 1 yields overfitting and negative TAMDAR impact
  on 3-h RH forecasts
• 12 yields best 3-h RH forecast impact
• Higher RH errors cause a gradual dropoff in
  TAMDAR impact

25
Increased TAMDAR impact on 3-h RH forecasts with
12 RH error
some other changes as well, as detailed in the
preprint
Each curve shows the amount that TAMDAR reduces
the RMS error. New processing reduces the error
more gt more TAMDAR impact.
New processing causes TAMDAR to reduce 3-h RH
forecast errors (RMS) by 5 RH at 500 mb. Under
old processing (3 RH error). maximum reduction
in 3-h RH forecast error was 3.5 RH
26
A look back and ahead (1)

• TAMDAR has improved the skill of RUC T, W, and
  RH forecasts
• TAMDAR data from the Mesaba fleet is likely to
  become an operational data set (purchased by the
  NWS) in the near future
• The data will be restricted to the U. S.
  Government, but will be used in
  publicly-available operational models

27
A look ahead (2)

• AirDat will install TAMDAR on additional fleets
  over the next several months
• Covering Alaska and the Western US
• These fleets include some jet aircraft
• higher altitudes, speeds (implications??)
• better heading gt reduced wind errors
• GSD will evaluate the quality and impact of these
  data (with FAA funding)
• (Unfortunately, these new data will not be
  available beyond GSD, per AirDat)

28
TAMDAR in Alaska 1 Pen Air aircraft out of 8
planned
29
Anchorage TAMDAR sdg, compared with RAOB 1.5 h
later (agreement is pretty good, even for wind
which AirDat does not yet vouch for due to some
suspect magnetic declination data)
30
We expect to report at the AMS Annual Meeting
next January on initial results from these new
fleets and regions
31
Indirect RH impact

• Can TAMDAR temp. and wind. forecasts alone
  improve RH forecasts?
• by providing better RH advection forecasts
• We performed a retro run using no TAMDAR RH data

32
Answer no TAMDAR impact on 3-h RH forecasts
when TAMDAR RH data are excluded
Red all TAMDAR (and other) data Blue no TAMDAR
RH data Black no TAMDAR data at all
Removing TAMDAR RH data degrades forecast skill
at 3-h as much as removing all TAMDAR data.
33
Some TAMDAR impact on 9-h RH forecasts when
TAMDAR RH data are excluded
Red all TAMDAR (and other) data Blue no TAMDAR
RH data Black no TAMDAR data at all
TAMDAR W and T data alone reduce 9-h RH forecast
errors at 450-500 mb about half as much as the
whole TAMDAR dataset does.
34
Effect of vertical resolution on TAMDAR 3-h
Temperature forecast impact
Each curve shows the amount that TAMDAR reduces
the RMS error. Low-res reduces the error less gt
less TAMDAR impact.
Decreased vertical resolution decreases TAMDAR
impact on 3-h T forecasts by 30 at 750 mb 10
at 900 mb
35
Vertical Resolution

• Higher resolution data can improve forecasts, but
  increases communication costs
• We tested the impact of vertical resolution on
  forecast skill
• Full resolution
• every 10 mb in lowest 200 mb
• every 25 mb above that
• Reduced resolution
• every 10 mb in the lowest 100 mb
• every 50 mb above that
• yielded about 50 less TAMDAR data

36
Results

• Decreased vertical resolution degrades 3-h T
  forecast skill by 10 to 30
• Moreover, the maximum possible vertical
  resolution has provided critical information
  during some potentially-severe weather events.