NASA Simulation Experiment for Airborne Remote Observations completed Nov 2005

1
NASA Simulation Experiment for Airborne Remote
Observations (completed Nov 2005)
A global Observing System Simulation Experiment
(OSSE) was recently conducted at NASAs Goddard
Laboratory for Atmospheres for the Boeing
Company. The following slides describe the
experiment design, illustrate the results of the
experiment, and provide conclusions based on
these results. The purpose of this effort was to
design and execute an OSSE to evaluate the
potential impact of adding two airborne remote
observing systems to the current observing
network. The sensors, proposed by Boeing,
include a direct detection Doppler Wind Lidar
(DWL) and a Global Positioning System limb
occultation receiver (GPS). Both observing
systems will furnish remotely-sensed atmospheric
profiles from commercial aircraft platforms.
2
Basic Steps in the Performance of the OSSE

• Simulate DWL and GPS observations
• Execute global assimilations and forecasts using
  simulated observations from current observing
  network (Control)
• Execute global assimilations and forecasts using
  Control observations and observations from DWL
  and GPS (Control DWL GPS)
• Evaluate results using metrics designed to
  highlight the potential impact of these new
  observing systems on key large-scale and synoptic
  scale atmospheric analysis and forecast fields

3
Core Components of OSSE
Nature run
Model forecast data set from which all simulated
observations are extracted. Also used as the
verification or truth data set. Spatial
resolution 0.5 latitude x 0.625 longitude x
35 pressure levels Temporal resolution 6
hours Forecast range 06 UTC 11 Sep 1999 18
UTC 5 Nov 1999 Full set of diagnostics with
varying sea surface temperature field.


Data Assimilation System (DAS)

• FVSSI (hybrid system using NASA FVGCM model
  coupled with NCEP SSI analysis)
• FVGCM 1 horizontal resolution, 55 eta level
  model
• SSI T62/64 level Spectral Statistical
  Interpolation system

Forecast model
FVGCM 1 horizontal resolution, 55 eta level
model
4
Experiments with FVSSI
1. Control
Assimilation from 00 UTC 12 Sep 1999 18 UTC 31
Oct 1999 (50 days) Observations include all
conventional, satellite cloud motion wind,
Quikscat, and NOAA-12 and NOAA-14 HIRS and MSU
satellite temperature retrievals. 5-day free
forecasts initiated every 2 days starting 00 UTC
17 Sep 1999 (total of 20 forecasts).
Prognostic fields are saved every 6 hours.
2. Control DWL GPS
Assimilation from 00 UTC 12 Sep 1999 18 UTC Oct
1999 (50 days) Observations include all the
observations used in the Control and DWL and GPS
observations. 5-day free forecasts initiated
every 2 days starting 00 UTC 17 Sep 1999 (total
of 20 forecasts). Prognostic fields are saved
every 6 hours.
5
DWL and GPS Simulation Methodology

• The simulation of the observations was based on
  the following
• Personal communication with Boeing and HRL
  scientists
• Available literature on space-borne or airborne
  DWL and GPS sensors
• 3. Limitations of the DAS (limit to the number
  of observations, no GPS observation operator)

DWL simulation

• U and V components of the wind are simulated
• Existing ACARS in situ observation locations,
  times, and flight levels provide the point at
  which profiles are generated
• 50km maximum horizontal resolution
• 250m vertical resolution from the surface to 3km
  altitude, 1km resolution above 3km
• Full profile from aircraft to surface assumed in
  cloud-free locations

6
DWL and GPS Simulation Methodology (continued)

• Profile length in locations of cloud cover is
  determined by probability of beam attenuation
  through nature run cloud cover beneath aircraft.
  Standard FVGCM overlapping cloud algorithm used
  to compute total cloud fraction down to each
  potential observation level
• Random noise is applied to the observations with
  an estimated 1.0 meter per second error standard
  deviation
• Final profiles are formatted to resemble
  rawinsonde profiles for inclusion to DAS

GPS simulation

• Temperature is the only simulated quantity
  moisture is expected to be available from this
  observing system however it is NOT simulated in
  the current OSSE due to uncertainties regarding
  error levels and ability to properly utilize this
  quantity in current DAS (based on AIRS
  experiments).
• Existing ACARS in situ observation locations,
  times, and flight levels provide the point at
  which profiles are generated
• 85km maximum horizontal resolution

7
DWL and GPS Simulation Methodology (continued)

• An estimated along-beam resolution of 150m is
  adjusted to 6km to achieve a manageable number of
  observations for current DAS. This method of
  thinning is consistent with application to real
  data of this nature.
• Profiles are simulated only after aircraft has
  reached a relatively stable cruising altitude.
  400hPa level is chosen as a nominal lower cutoff.
• Assume no cloud attenuation
• Random noise is applied to the observations with
  an estimated 1.5 Kelvin error standard deviation

• Random selection of incoming radio beam direction
  is made for each profile to simulate the maximum
  possible reception angle lying 90 on either side
  of the aircrafts flight path
• Random selection of incoming radio beam elevation
  is made for each profile in the range of 1 above
  aircraft to 4 below aircraft. This is to
  simulate the possible reception of radio beam
  above and below aircraft relative to the flight
  path

8
DWL and GPS Simulation Methodology (continued)

• Maximum 200km profile length (10 profiles reach
  the ground)
• Final profiles are formatted to resemble ACARS in
  situ observations for inclusion to DAS

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
Conclusions

• RMS impacts (slides 11-14) indicate an
  improvement in the standard SLP and Z500 analysis
  fields with the inclusion of DWL and GPS
  observations in most areas where these new
  observations exist.
• The largest impact, especially in SLP, is over
  the North Atlantic (the maximum impact being
  about twice as large as that of the North
  Pacific). This North Atlantic impact is likely
  from two sources local effects of the
  concentrated aircraft flight paths over the North
  Atlantic along the typical storm track advection
  from upstream, where even greater sampling is
  being performed over the U.S. Primarily, local
  effects are likely responsible for the smaller
  positive impact in the North Pacific.
• Over the continental U.S. the impact is almost
  negligible in SLP but is maximum over the
  southeast U.S. in Z500. However, the average
  Z500 impact values are all quite small (maximum
  of only 6 meters over Florida in October). Even
  though the vast majority of DWL and GPS
  observations are located over the U.S., this
  already well-sampled region leaves very little
  room for improvement and therefore little or no
  impact would be expected there.

29
Conclusions (continued)

• Persistent negative impacts are very apparent
  over Canada and around the North Pole. Little or
  no DWL and GPS observations were simulated in
  these regions so some of this effect may not be
  local. Many of the fields that were examined
  (some of which are not included in this PPT file)
  show a very sharp boundary, near the
  U.S.-Canadian border, where impacts change sign.
  This very closely delineates between a data-rich
  region, the U.S., where there is an immense
  concentration of conventional and DWL and GPS
  observations, and a data-sparse region, Canada,
  where relatively few conventional and DWL and GPS
  observations exist. We believe that the negative
  impacts may be due to enhanced error growth of
  certain features due to artificial gradients
  created by the sharp data cutoff. The DAS would
  then propagate these errors zonally and poleward.
  Its unclear as to whether this is entirely as a
  result of the very asymmetrical distribution of
  such a large number of DWL and GPS observations
  or a manifestation of the DAS. In general,
  however, the addition of virtually any new
  observing system, real or simulated, results in
  both positive and negative impacts with all
  global data assimilation systems that have been
  used at NASA. Still, this will require further
  investigation.
• The anomaly correlation figures (slides 15-25)
  show the forecasts with DWL and GPS observations
  included in the initial conditions to be, on
  average, the same or slightly better than the
  control forecasts. As in the RMS analysis
  impact figures, the forecast improvement appears
  to be greatest over the North Atlantic. Small,
  incremental, forecast improvements from the
  addition of new and proposed observing systems to
  the existing observing network is typical, so the
  results of this OSSE are consistent with past
  real data and simulated data experiments for this
  metric.

30
Glossary
Nature run A long (generally one month or
longer), high resolution model forecast having
suitable outputs from which to create synthetic
observations. This forecast is used as
verification or truth when evaluating DAS
analyses and model free forecasts.
Data Assimilation System (DAS) A method of
blending large numbers of observations from a
diverse set of observing systems with model
background fields (ie model first guess) to
obtain the best analysis possible. The analysis
is used as initial conditions for model free
forecasts.
Model first guess (or background fields) A 6-hour
model forecast designed to extend an assimilation
6 hours forward in time prior to the introduction
of observations. Utilizing a model first guess
when generating an analysis provides continuity
between analysis times, fills gaps in data sparse
regions, and reduces field imbalances that may be
created by observations.
Model free forecast A 5-day model forecast
starting from an analysis. The term free is
used to distinguish this type of forecast from
the 6-hour first guess forecast that is run as
part of the DAS.
31
Glossary (continued)
DAS cycle One 6-hourly fixed length of
assimilation which involves generating the 6 hour
first guess, gathering observations in a 6 hour
block centered on the synoptic time (00,06,12, or
18 UTC), and using these observations to correct
the first guess for that synoptic time.
Example Start with an analysis from the
previous cycle at, say, 12 UTC generate a first
guess valid at 18 UTC introduce the observations
gathered from 15 UTC to 21 UTC correct the first
guess using these observations the result is an
analysis at 18 UTC.
Anomaly Correlation A commonly-used OSSE metric
that compares a forecast with the nature run
while using climatology as an anchor. Its also
known as a pattern correlation since it
emphasizes similarities in weather patterns, that
is, it rewards the forecast if its weather
patterns are similar to the nature run weather
patterns. Pattern consistency is an important
aspect when evaluating a forecast. Like the
standard correlation, values range from 1 to 1
anomaly correlation values below 0.6 indicate no
forecast skill.
FVSSI NASA hybrid DAS composed of FVGCM model
and NCEPs SSI analysis.
FVGCM Finite Volume General Circulation Model.
SSI NCEPs Spectral Statistical Interpolation
analysis system.
SLP Sea Level Pressure in units of millibars
(mb) or hectopascals (hPa).
Z500 500 mb geopotential height. This and SLP
are the most common fields examined.
32
(No Transcript)