This article was recently accepted for
publication
in the preprint volume for the 9th Conference on Satellite Meteorology
and Oceanography, Amer. Meteor. Soc., Paris, France to be held in May
1998.
NOAA's Privacy Policy
A version of this article more suitable for
printing
is available as a
PDF
file.
This page is maintained by Dan Birkenheuer, birk@fsl.noaa.gov
P5.27A RADIANCE ASSIMILATION OF POLAR AND GEOSTATIONARY
SATELLITE
DATA IN LAPS
Daniel Birkenheuer1
NOAA Forecasts systems Laboratory
Boulder, Colorado
1. INTRODUCTION
The Local Analysis and Prediction System (LAPS) was
conceived
in the late 1980s at the Forecast Systems Laboratory (FSL; McGinley et
al. 1991). LAPS fills an anticipated need in the local forecast office.
It has been apparent that the large data volumes generated from current
and planned data acquisition systems overwhelm weather forecasters
because
it is impractical to review all available data soon enough to maintain
timely forecast generation. An analysis system that integrates
available
real-time data produces a comprehensive synopsis of the atmospheric
state
in a timely fashion. Going beyond subjective application, the
feasibility
of forecasting with local-scale numerical models initialized from LAPS
analyses has been demonstrated (Stamus and McGinley 1997).
FSL's operational LAPS analysis is typically
performed
on a 10-km mesh approximately centered over Denver, Colorado. The LAPS
grid has ranged from 600 km on each side to a newer larger domain 1,250
by 1,050 km. The vertical coordinate is pressure with 50-hPa increments
spanning 1100 hPa to 100 hPa; however, the pressure spacing is
adjustable.
LAPS background fields are interpolated from
large-scale
global or regional models; typical choices have been the Rapid Update
Cycle
(RUC) or Mesoscale Analysis and Prediction System (MAPS) 60-km and
40-km
national scale forecasts Benjamin (1991), the U.S. National Centers for
Environmental Prediction's Eta, nested grid model (NGM), and the U.S.
Navy
Operational Global Atmospheric Prediction System (NOGAPS) models. LAPS
routinely generates state variable analyses of t, p, z, u, v, w,
and q; and can produce special fields such as three-
dimensional
(3-D) cloud distribution, cloud type, precipitation; and derived fields
such as lifted index (LI), LI x w; and integrated precipitation
over time. In addition, LAPS can serve special needs of the user, e.g.,
aviation forecast problems.
This paper reviews recent progress in utilizing the
different
satellite data sources and forward models for LAPS analysis
applications.
2. RADIANCE DATA USED IN LAPS
At its inception, LAPS was designed to address the
local
forecast/nowcast problem, aimed at severe weather, rapid updates, and
the
merging of high frequency data sources (e.g., radar data) into a
frequently
updated analysis. For this reason, it was natural to develop LAPS with
a Geostationary Operational Environmental Satellite (GOES) interface.
Since then, demands have changed, and domains have
now
been analyzed outside of the range of GOES coverage in places such as
Bosnia,
China, and more domains are planned for other overseas applications
(i.e.,
Sydney 2000 Olympic Games). These activities, external to the United
States,
lack the GOES coverage that we have depended on for years. LAPS is
currently
being configured to utilize other geostationary platform data such as
Meteosat
and GMS, but even with these capabilities, a polar capability appears
desirable.
GOES data have many advantages, one of them being the
exceptional temporal continuity that is especially vital in local-scale
analysis and monitoring for severe weather. However, when using
analyses
for model initialization, polar satellite data can offer significant
benefits
in data-sparse areas, especially if the model runs are initialized at a
time coincident with a polar pass. Available at all locations on the
globe
including our laboratory, polar data are attractive for development
because
they can be checked out in-house and their performance could be
anywhere
in the world. Code maintenance also becomes standardized since one
satellite
(or type of satellite) is involved. Furthermore, we can test the polar
data alongside GOES and apply experience with GOES to the polar
infrared
(IR) and exploit the microwave sensing unique to polar craft.
The experience to date has been with IR data, which
is
the focus of this paper. Eventually, the logical progression of our
work
is to follow the Television and IR Operations Satellite (TIROS)
high-resolution
infrared sounder (HIRS) with TIROS microwave sounding unit (MSU) data
and
then explore Defense Meteorological Satellite Program (DMSP) data.
Thus,
we will eventually be utilizing microwave satellite data in LAPS. The
selection
of the forward models studied parallels this development. This work
provides
a foundation for combining satellite data of all types and the
algorithms
now being developed must handle navigation and integration of
heterogeneous
data (at least IR, visible, and microwave). This paper also addresses
the
progression from an iterative variational scheme to more contemporary
matrix
solutions with the ultimate objective to utilize one model to support
all
data types from most platforms.
GOES data can effectively support LAPS from both
direct
readout (10-bit) and low-precision radiances inferred from 8-bit
imagery.
Birkenheuer (1996) documents FSL's experiments with using satellite
broadcast
network (SBN) 8-bit image data to supply its algorithms with satellite
brightness temperatures. Imager data were initially used since they are
readily available for Advanced Weather Interactive Processing System
(AWIPS)
and are directly applicable to the local forecast office using current
data sources supporting AWIPS. With a better source of digital radiance
imager data (10 bit instead of 8 bit), it is conceivable that the
positive
impact could be extended further. This study demonstrates that the
13-bit
sounder data from GOES is superior to imager data in the GOES
variational
technique.
TIROS data used in LAPS originate from
high-resolution
picture transmission (HRPT) files. In particular, the TIROS ingest
software
processes the TIROS Operational Vertical Sounder (TOVS) information,
which
provides radiation measurements in 19 IR spectral regions from the
shortwave
(4.3 m) to the longwave regions, as well as the visible wavelength
(0.69
m).
An intermediate processing step exists between the
LAPS
ingest and the raw HRPT files. The primary function navigates the data
to earth locations (latitude and longitude) and computes the brightness
temperatures for each IR channel using the interactive TIROS processing
package (ITPP). The LAPS ingest software directly reads the
ITPP-produced
files and computes the mapping transform for each HRPT file. Because
the
TOVS data resolution is approximately 42 km along the satellite track,
this mapping is not time-consuming and requires about 2 minutes to
process
each intermediate file.
3. FORWARD MODELS AND ANALYSIS TECHNIQUES
To date, LAPS has used two primary approaches for
radiance
data assimilation. Both methods are 1-D variational techniques,
modifying
portions of the vertical column one grid point at a time. In the case
of
GOES data, the radiances are densely spaced enough for the 10-km
analysis
to allow an analysis at each grid point, using the choice of nearest
neighbor
or averaged clear radiances values.
In the case of TIROS data, the field-of-view (FOV) is
such that only the LAPS grid point closest to the center of the FOV is
analyzed. The results from the ensemble of analyses are spread
throughout
the LAPS grids by conventional analysis techniques such as Barnes
(1964).
3.1 University of Wisconsin Model (GOES Application)
The forward model used for GOES assimilation produces
a simulated radiance based on temperature, moisture, and ozone profiles
along with the temperature of the surface or cloud top, and the
pressure
of that radiating surface (i.e., surface pressure or cloud top pressure
whichever applies). Also needed are the zenith angle, used to determine
the airmass path and optical depth between the FOV and the satellite.
The
forward model used for this work was obtained from the University of
Wisconsin--Madison
and it is described in Hayden (1988). The forward model coefficients
used
for this study were vintage late 1995, furthermore, this model is not
configured
to produce a Jacobian for matrix solutions.
In order to apply the forward model variationally,
clear
and cloudy FOVs need to be determined. The LAPS cloud analysis is used
to identify clear and cloudy LAPS grid points (Albers et al. 1996). The
analysis presented here is only working from the FOVs classified as
clear.
The cloudy FOVs probably can be used; however, further research and
refinements
are needed to exploit cloudy and especially partly cloudy regions.
The functional evaluated at each grid point has the
form,
where the goal is to determine the optimum
coefficient
(c), where c is a scaling factor for the moisture corresponding
to the atmosphere between 500 and 100 hPa. No modification is made to
the
moisture profile anywhere else in the column. The forward model
radiance
(R) for a specific channel i is a function of LAPS
temperature
profile (t), ozone climatology profile (o), and the unmodified
LAPS
mixing ratio profile (w) and the scaling coefficient c.
The
moisture profile used in each run of the forward model is the modified
moisture profile denoted by cw, where each level of w
has
been scaled by the level-dependent coefficient c. The observed
radiance
derived from AWIPS image data is designated as Roi,
where subscript i indicates the imager channel number.
The first term in the functional maximizes agreement
between the forward model and observed radiance at the expense of only
modifying the water vapor profile. The second term adds stability and
gives
more weight to solutions in which the coefficient's departure from
unity
(no change to the initial profile) is minimized. The second stabilizing
term, helping constrain the solution to approximate unity, is more
important
when multiple layers are solved (a case not considered here). Weights
based
on error characteristics can eventually be added, but for now the two
terms
have equal weight. Error statistics become more important when the
functional
grows in scope to include other data sources (i.e., Radiosonde
Observation
(RAOB) data).
Note that differences in all three channels, not only
the moisture channel, are minimized in this technique. Thus, any
improvement
in the "dirty window," imager channel 5, will also contribute to the
solution.
The Powell (1962) method used to minimize this function typically
required
three to 10 iterations to converge. Here the maximum number of
iterations
was set at 50, and if this limit was reached, the coefficient for that
particular grid point was excluded from the algorithm.
Once the coefficients are determined, they are
applied
to the specific humidity field at each pressure level for which they
are
designated. The modified specific humidity field is then advanced to
the
final analysis step.
3.2 RTATOV (Polar Application)
To perform the polar experiments for LAPS, the
forward
model/variational solution was selected to be RTATOV, a radiative
transfer
model obtained from the European Centre for Medium Range Weather
Forecasts
(ECMWF). RTATOV solves both the forward model problem plus the Jacobian
"K matrix," adjoint, and tangent linear operators. Here the gradient
matrix
is used to solve a 1-D or 3-D variational solution to reconcile the
analyzed
thermal and moisture profiles with measured satellite radiances. This
model
was selected because it embraces a state-of-the-art approach to
radiance
assimilation and can effectively be formulated for use in a homogeneous
or heterogeneous analysis situation. Here it was only applied
homogeneously
to the moisture profile; future applications will be heterogeneous by
expanding
the functional to include other data sources and their respective
forward
models.
The functional used to minimize the radiance data
against
the background is:
The 1-D minimum variance solution is applied after
Eyre
(1989) using:
where xb is the background
vertical
profile vector containing thermal and moisture data from modified model
background data along with climatology above 100 mb. C is the
covariance
error matrix of the background, K is the gradient matrix from
RTATOV,
E is the covariance error matrix for the HIRS channel data, ym
is the measured radiance vector, and y(xb) is the
forward
modeled radiance vector computed from the background vertical profile
using
RTATOV.
The solution is only applied in clear areas even
though
methods exist to handle clouds in the model. We selected this approach
for convenience, since we are using the model for the first time. For
cloud
detection we rely on the LAPS cloud analysis that uses GOES data in its
current formulation.
4. RECENT VARIATIONAL ANALYSIS EXPERIENCES WITH GEOSTATIONARY AND POLAR
DATA
The positive results using GOES imager data to
enhance
both moisture analyses and local forecasts have been shown (Birkenheuer
1998a). Recently the iterative variational approach has been applied to
13-bit GOES sounder data, demonstrating additional improvement over
those
obtained using radiances derived from 8-bit image data. These solutions
used three comparable channels (10, 8, and 7) from the GOES 9 sounder
instrument
that correspond to the GOES 9 imager channels (3, 4, and 5). Therefore,
improvement is likely due to better data precision, signal-to-noise
ratio,
and better modeling of the sounder radiances.
Figure 1a plots the specific humidity bias error,
with
the most notable improvement at 400 hPa. The imager analysis bias is
shown
as dotted lines and the sounder bias error in solid lines. The reported
value at 350 hPa represents only 2 data points (since 350 is not a
mandatory
RAOB pressure level), and the remaining points represent statistics
from
64 comparisons made from mid-December 1997 through early February 1998,
pooling both 1200 and 0000 UTC validation times.
Fig.
1a. Plot of specific humidity bias error (g/g) with pressure over the
portion
of the LAPS analysis that is modified using GOES 9 data. Imager
and
sounder results errors are computed using Denver RAOB data. A
total
of 64 cases comprise the sample except for 350 hPa, which only had two
cases for the time interval studied.
Figure 1b contrasts the imager and sounder RMS errors.
Following 1a, the imager RMS error is indicated by the dotted line, and
the sounder-based result by the solid line. Here we see a very
consistent
reduction throughout the column except at 400 hPa, where there is
substantially
more improvement in RMS error.
Fig.
1b. Plot of specific humidity RMS error (g/g) for the same cases
as in Fig. 1a.
Overall, for the entire atmosphere between 400 and 100
hPa, the analyzed bias error using imager data was 1.287x10-5 compared
to 2.599x10-6 g/g from sounder-based moisture analyses, reduction in
moisture
bias was a striking 80%. The RMS error statistics are also improved
with
sounder RMS error of 3.592x10-5 compared to the imager RMS error of
4.826x10-5
g/g, an overall improvement of about 25%.
Work is ongoing in applying matrix solutions to our
analysis;
these methods are not yet mature enough to replace our iterative
scheme.
Initial experiences with RTATOV are discussed in Birkenheuer (1998b).
The
major finding was that mesoscale structure can be derived from the
method,
but consistent improvement in analysis quality has yet to be
demonstrated
from a large data sample. The RTATOV work is now on hold while LAPS is
configured for Optical Path Transmittance (OPTRAN; McMillin et al.,
1995).
5. SUMMARY AND PLANS
To date we have had the greatest success and most
experience
applying GOES radiance data to LAPS using iterative variational schemes
for moisture enhancement. We are steadily progressing to the more
versatile
variational approaches. In the summer of 1998 we hope to apply the
variational
minimization analysis not only to the moisture analysis but also to the
thermal analysis. The matrix methods currently being explored will
enable
us to bring together multiple data sources, and to solve for multiple
variables,
eventually producing the best solution for a wide set of data types.
One candidate for replacing the iterative GOES method
is RTATOV coupled with OPTRAN as its forward model component in its
more
natural coordinate system. OPTRAN is currently being configured to
interface
to LAPS. The expectation is to devise one model that will support the
large
variety of satellites and sensor data providing efficient maintenance
while
serving a host of satellite needs.
In the meantime, it is unfortunate that GOES sounder
radiances remain beyond the reach of the local forecast office. LAPS
will
remain configured to use GOES imager data for its source for
radiometric
satellite data in AWIPS; however, until sounder radiances can be
offered
to the local forecast office, LAPS will not realize its full potential
from GOES. When LAPS is operated in a location where it can receive
sounder
radiance, the techniques are in place to put these data to effective
use.
Future plans now include refining the LAPS GOES
analysis
techniques to utilize more of the sounder channels. Care is required to
offset the improvement that more data add to the analysis against the
longer
processing times incurred when processing more channels.
6. REFERENCES
Albers S., J. McGinley, D. Birkenheuer, and J. Smart
1996: The Local Analysis and Prediction System (LAPS): Analyses of
clouds,
precipitation, and temperature. Weat. and Forecasting, 11, 273-287.
Barnes, S. L., 1964: A technique for maximizing
details
in numerical weather map analysis. J. Appl. Meteor., 3, 396-409.
Benjamin, S.G., 1991: Short-range forecasts from a
3-h
isentropic-sigma assimilation system using ACARS data. Fourth
International
Conference on Aviation Weather Systems, Paris, France, Amer. Meteor.
Soc.,
329-334.
Birkenheuer D., 1996: Exploiting available satellite
data in AWIPS-era workstations. Eighth Conference on Satellite
Meteorology
and Oceanography, Atlanta, GA, Amer. Meteor. Soc., 46-49.
______, 1998a: The effect of GOES image data on RAMS
forecasts initialized with LAPS. 12th Conference on Numerical Weather
Prediction,
Phoenix, AZ, Amer. Meteo. Soc., 300-303.
______, 1998b: Analysis of polar satellite data in
LAPS
using RTTOV. 12th Conference on Numerical Weather Prediction, Phoenix,
AZ, Amer. Meteo. Soc., 304-307.
Eyre, J. R. 1989: Inversion of cloudy satellite
sounding
radiance by nonlinear optimal estimation. I: Theory and simulation for
TOVS. QJRMS, 115,1001-1025.
Hayden, C.M., 1988: GOES-VAS simultaneous
temperature-moisture
retrieval algorithm. J. Appl. Meteor., 27, 705-733.
McGinley, J.A., S.C. Albers, and P.A. Stamus, 1991:
Validation
of a composite convective index as defined by a real-time local
analysis
system. Wea. and Forecasting, 6, 337-356.
McMillin, L. M., L. Crone and T. J. Kleespies, 1995:
Atmospheric
transmittance of an absorbing gas. 5: Improvements to the OPTRAN
approach.
Appl. Opt. 34 (36) 8396.
Powell, M.J.D, 1962: An iterative method for finding
stationary
values of a function of several variables, Computer J., 5, 147-151.
Stamus P. A., and J. A. McGinley, 1997: The Local
Analysis
and Prediction System (LAPS): Providing Weather Support to the 1996
Summer
Olympic Games. 13th International Conf. on Interactive and Processing
Systems
(IIPS) for Meteorology, Oceanography, and Hydrology, Long Beach, CA,
Amer.
Meteor. Soc., 11-14.