MRB For SPU Data Errors

1
Validation of the HIRDLS Vertical Resolution
using GPS Radio Occultation Data
John Barnett, Scott Osprey, Chris Hepplewhite,
Oxford University John Gille, R. Khosravi,
University of Colorado and National Center for
Atmospheric Research plus the whole of the rest
of the HIRDLS Team
Aura Science Team Meeting, October 2008, Columbia
Maryland.
Abstract The Aura HIRDLS instrument was always
intended to provide measurements with a high
vertical resolution of order 1 km and horizontal
resolution that is commensurate for smaller scale
atmospheric phenomena such as gravity waves. The
launch problem which partially blocked the HIRDLS
optical path resulted in a different placement of
vertical profiles, but with the vertical
resolution potentially unaffected. Profiles are
now approximately 100 km apart along-track and
the orbit spacing cross track. Validation of the
vertical resolution is difficult because few
other measurements provide that resolution
sufficiently closely in time for atmospheric
features not to have changed. However, the COSMIC
suite of radio occultation satellites that
exploit the U.S. GPS transmitters to obtain high
resolution (1 km) temperature profiles in the
stratosphere does provide sufficient profiles
nearly coincident with those from HIRDLS.
Comparisons show a good degree inter correlation
between COSMIC and HIRDLS down to about 2 km
resolution, with similar amplitudes for each,
implying that HIRDLS and COSMIC are able to
measure the same small scale features. Hence, the
optical blockage within HIRDLS does not seem to
have affected this capability. This is of crucial
importance for gravity wave studies because
HIRDLS provides an excellent data set for their
quantitative study and measurement.
2
HIRDLS Vertical Resolution Objectives
One of the primary goals for HIRDLS has always
been to push the hard on the limits on vertical
resolution, but at the same time have
commensurate horizontal resolution. The aim is
to achieve a vertical resolution of order 1 to
1.5 km. Atmospheric features of this size
typically have a horizontal scale 100 to 200
times as large. Hence a profile spacing of a few
hundred km is needed to obtain a coherent
picture. The time scale may be as short as
minutes, so ideally the measurements should be
made nearly simultaneously. The optical blockage
within HIRDLS has modified the horizontal
placement of profiles originally there was to be
a cross track scan leading to a 2D mesh of
profiles at 500 km spacing. Now we have just a
line of profiles at about 100 km spacing almost
as many profiles but organised as a curtain
along the orbit. This situation may actually be
better for viewing some small scale structures
such as gravity waves.
3
FORMOSAT-3/COSMIC Data
The FORMOSAT-3 COSMIC system has 6 small
satellites each carrying a GPS receiver. As the
satellite set or rises relative to a GPS
transmitter, the path through the limb is
refracted by the atmosphere, and the extra time
of travel is used to deduce a combination of air
and water vapour density. Above about the
tropopause where there is almost no water this
yields a temperature profile. Vertical resolution
probably about 1 km above the tropopause (better
lower down). COSMIC data are generally said to
be accurate up to 30-35 km (4-5 scale heights).
Figure from http//www.cosmic.ucar.edu
See paper in March 2008 Bulletin of American
Meteorological Society.
4

• COSMIC profile locations for a typical day.

7 December 2007 - 1878 soundings at quasi random
times.
From a presentation by Bill Schreiner of the
COSMIC Project Office, HIRDLS Science Team
Meeting, January 2008
5
Data used
These comparisons used HIRDLS profiles within
0.75º great circle, i.e. 83.3 km and 500 sec of
time of the COSMIC profile. This is much
tighter than is normally possible with
validation, but is necessary because the small
scale features change quickly. Such a tight
window is possible because of the large number of
HIRDLS and COSMIC profiles each day (5500 and
2000). Sometimes 2 (possibly 3) HIRDLS profiles
matched this criterion, in which case they were
averaged together. Days 192 2006 to 365 2007
data were used. Very few were missing. COSMIC
data are from latest (late 2007) reprocessed
publicly available version. HIRDLS are Version
4 (2.04.19 internal to HIRDLS), as available
from the NASA DISC. NASA GSFC GEOS-5.01
meteorological analyses are used to make a
gradient correction in the HIRDLS retrieval and
are also compared.
6
Method of Comparison

• Two methods of comparison were used
• Subtract a smoothed profile (this is equivalent
  to a high pass filter) then inter-correlate
  HIRDLS, COSMIC and GMAO. Each used its own
  smoothed profile. Results are for a 0.5 pressure
  scale heights full width at half height filter.
• Fourier analysis over the range 2.2-5.7 scale
  heights, after fitting a parabola to remove the
  background, and apodizing.

Full Width Half Height is 0.5 scale heights
(approx 3.5 km)
Pressure Scale Height Distance over which
pressure changes by a factor of e. This is
typically 7 km.
7
Example comparison day 336 2006
Day 336 2006 COSMIC profile atmPrf_C002.2006.336
.14.42.G17 57.6 º S, 39.2 ºE, 144015 HIRDLS
57.0º S, 39.6 ºE, 144018
Black COSMIC profile Magenta nearest
HIRDLS Blue adjacent HIRDLS profiles to
south Gold adjacent HIRDLS profiles to
north Blue and gold are offset by 4 K per
profile.
8
Example comparison day 344 2007
Day 344 2007 COSMIC profile atmPrf_C006.2007.344
.01.15.G25 19.8 º N, 8.6 ºW, 011305 HIRDLS
19.8º N, 8.44 ºW, 011135
Black COSMIC profile Magenta nearest
HIRDLS Blue adjacent HIRDLS profiles to
south Gold adjacent HIRDLS profiles to
north Blue and gold are offset by 4 K per
profile.
9
Example Comparisons, before and after high pass
filtering
Initial profiles
After subtracting smoothed profile.
10
COSMIC vs HIRDLS correlation and standard
deviations from high pass filtered data
HIRDLS vs. COSMIC standard deviation of
temperature from smooth profiles over 2.2-5.4
pressure scale heights (16-40 km) for near
coincident profiles. Crosses are colour coded
with the correlation coefficient over this range.
Note that most profiles are positively
correlated.
11
COSMIC vs GMAO model correlation and standard
deviations
COSMIC standard deviations are much bigger than
those of GMAO but correlation still tends to be
positive
12
HIRDLS vs GMAO correlation and standard
deviations
13
Fourier analysis approach.
These viewgraphs give the results of
intercorrelating the Fourier analysed COSMIC,
HIRDLS and GMAO profiles. The range of 2.0 to
5.7 pressure scale heights was used (16-40 km).
Profiles had to be present over the whole of this
range. In most cases the COSMIC profile was the
cause for not filling the domain, in which none
was used. This range was a compromise between
maximising the range and maximising the number of
profile comparisons. The data were interpolated
to 1/48 intervals in log10(pressure), which is
half of the HIRDLS interval, giving 72 levels. A
classical Fourier analysis was used, i.e. each
sine and cosine component was evaluated
separately, rather than an FFT to avoid having to
pad out to a power of 2. The data were apodised
with a triangular function, but this gave
essentially the same result as a cosine bell
apodisation. A background profile was
subtracted from each. After much experimentation,
mainly with polynomials of different order, a
parabolic fit was used (a separate fit for each
of COSMIC, HIRDLS and GMAO). This would be
expected to attenuate the lowest wavenumber.
Higher order polynomials gave attenuation to
progressively higher frequencies but seemed to
leave the still higher ones about the same.
14
Fourier analysis results.
1436 profiles used (slightly less than for
previous figures because whole vertical range had
to be present)
15
How do sequences of profiles compare?
Successive profiles are offset by 5 K (-7.5,
-2.5, 2.5, 7.5 from absolute).
COSMIC 284 2006 Time in day 13694-14030 secs
16
How do sequences of profiles compare?
HIRDLS 284 2006 Time in day 24856-24949 secs
Successive profiles are offset by 5 K (centre
profile not offset).
COSMIC 284 2006 Time in day 13694-14030 secs
17
How do sequences of profiles compare?
HIRDLS 225 2006 Time in day 80260-80353 secs
COSMIC 225 2006 Time in day 39259-39685 secs
18
What do radiosondes observe?
Data are for 1 - 19 January 2008 in time order
from left to right. Temperature is correct for
first profile and offset 7 K for each subsequent
one.
Data source U.K. Met Office high resolution high
altitude radiosonde data, launched from Mount
Pleasant, Falkland Islands (52 deg S, 58 deg W) ,
provided at 2 second intervals during the ascent.
19
What do radiosondes observe?
As before but for a different time of the year.
Data are for 26 May to 13 June 2008 in time order
from left to right. Temperature is correct for
first profile and offset 7 K for each subsequent
one.
Data source U.K. Met Office high resolution high
altitude radiosonde data, launched from Mount
Pleasant, Falkland Islands (52 deg S, 58 deg W) ,
provided at 2 second intervals.
20
Conclusions
HIRDLS agrees well with COSMIC data down to 2 km
resolution (4 km wavelength) or better. COSMIC
shows similar amplitudes to HIRDLS suggesting
that either they have similar vertical
resolutions or they are both are adequate to
resolves the scales that are present Agreement
on finer vertical scales is difficult to verify
because of small amplitudes in both HIRDLS and
COSMIC data the atmospheric waves seem to have
a decreasing amplitude. This may be due to the
along-line-of-sight averaging inherent in both
HIRDLS and COSMIC measurements with smaller
vertical wavelengths being associated with
smaller horizontal wavelengths.