NPOESS Advisory Committee for Observing Systems Simulation Experiments
Report No. 3
September 1999
Introduction
The NPOESS Observing System Simulation Experiments (OSSE) Advisory
Committee provides technical oversight and scientific guidance to the investigators
involved in the NPOESS OSSE project. These investigators have requested input from
a committee representing the potential users of the polar-orbiting satellite data
forthcoming from the future NPOESS satellites' sensor suites. The Advisory
Committee has been asked to convene as necessary to review progress on development
and implementation of the OSSE system, and report on the progress to the NPOESS
Integrated Program Office. This offering constitutes the third in a series of such
reports.
Progress Summary
The third committee meeting was held at the National Centers for
Environmental Prediction, Environmental Modeling Center's office in the World
Weather Building, Camp Springs, MD on September 28, 1999. In attendance at that
meeting were the following committee members: Akira Kasahara, T. N. Krishnamurti,
Greg Mandt, Jan Paegle, Edwin Eloranta, and chairperson Donald Norquist. Prior to
the meeting, the committee received a set of documents from the NPOESS OSSE
project investigators (hereafter referred to as the OSSE Team) that described progress
on the project since the last committee meeting in May 1998. At the meeting, the
committee heard presentations from several of the OSSE Team members, and from a
representative of the NPOESS Integrated Program Office (IPO). The following
discussion summarizes the collective information provided in the document set and at
the meeting.
Steve Lord of NCEP/EMC, the OSSE Team leader, presented an overview of
the progress of the OSSE Team since the last advisory committee meeting. He also
distributed a document entitled "Observing System Simulation Experiments for
NPOESS" which was his presentation in written form. Steve listed the participants on
the OSSE Team, representing NOAA/NWS/NCEP/EMC, NASA/GSFC, Simpson
Weather Associates, and NOAA/NESDIS. He reminded the committee that the two
objectives of the NPOESS OSSEs are: to test the NWP impact of advanced instruments
using simulated data and make recommendations for configuration based on
quantitative NWP impact of NPOESS instruments, and to prepare for use of advanced
instrument suite as soon as possible after space deployment. Candidate instruments
being simulated in the OSSEs are Doppler Wind Lidar (DWL), Cross track infrared
counter (CrIS), Conically scanning Microwave Imager/Sounder (CMIS), and Advanced
Technology Microwave Sounder (ATMS).
Steve reported that the following new work had been completed since the last
committee meeting:
report on Nature Run cloud evaluation and adjustment (with input from ECMWF),
OSSE data base upgraded to accommodate DWL line-of-sight (LOS) winds,
NCEP global data assimilation model (SSI) upgraded to use LOS winds,
conventional and DWL observations simulated, TOVS data simulation underway,
real data sensitivity test for existing instruments repeated using current SSI and
TOVS 1B radiance data,
OSSE data assimilation environment setup continuing, including enabling of
software for application on massively parallel computer architecture,
evaluated simulated DWL winds scenario sensitivity results,
evaluated simulated conventional observations.
Steve next discussed the present status of computing resources available to
NCEP for the NPOESS OSSEs. First, the NCEP operational massively parallel
computer, the IBM SP, is not available to the project by NWS mandate. This and other
"reimbursable" projects (that is, not NWS core funded) will not receive support of any
kind from NWS operations. Next, computer resources that have been used will become
unavailable soon. An example is the NCEP Cray 5 (J916) computer, which will be
taken out of service in Spring 2000. Steve stated that a lack of computer resources has
greatly hindered progress on the OSSE project, and the situation will only worsen now
that the OSSE experiments themselves are imminent. To alleviate these shortfalls in
computing resources, the OSSE project has teamed with other NCEP-based
reimbursable projects to purchase an SGI Origin 2000 computer with 32 processors.
The OSSE Team is currently involved in porting all OSSE-related software (configured
for massively-parallel architectures) from the IBM SP. The NCEP-based reimbursable
projects will have to share the Origin 2000, and are expected to overtax the system at
times. This will likely present the need for the purchase of another "projects" computer
if funds from the projects are available.
In conclusion, Steve presented the following planned tasks for the OSSE Team
in fiscal year 2000:
complete sensitivity tests for real data (existing instruments): forecasts and analysis
of results,
complete sensitivity tests for simulated data (existing instruments) "baseline"
OSSE, including completion of simulated TOVS, calibration with real data
experiments, and any necessary adjustment of simulated data error levels,
continue analysis sensitivity tests for simulated DWL data, including DWL
technology scenarios,
consider non-NPOESS data that may be available in the NPOESS-operational era,
perform OSSEs with simulated DWL data,
develop and test adaptive observation strategies for DWL data,
develop and execute simulation strategies for other potential NPOESS instruments:
CrIS, CMIS, ATMS.
Michiko Masutani of NCEP/EMC next presented results from the real data
sensitivity tests for existing instruments, repeated using current SSI and TOVS 1B
radiance data. The resolution of the global forecast model (the NCEP MRF) used in the
data assimilation was triangular 62 wave truncation with 28 vertical levels. The
NCEP/NCAR Reanalysis data set was used for initial conditions and verification. The
tests were conducted over the period 1 Jan 28 Feb 1993. Three types of conventional
data withheld in the experiments were TOVS 1B (NOTOVS), rawinsonde winds
(NOWINS), and radiosonde temperature (NOTEMS). Evaluation of the impacts of
withholding combinations of the data types showed that: in the Northern Hemisphere,
NOWINS has more impact than NOTEMS, NOTOVS has slightly less impact than
NOTEMS; in the Southern Hemisphere, NOTOVS has much greater impact than
NOTEMS or NOWINS; in the tropics, NOTEMS had greater impact than NOWINS or
NOTOVS. In general, the results showed that wind observations are more important in
the wind analysis than are temperature observations, wind observations are at least as
important as temperature observations in the temperature/height analysis, TOVS
radiances are most important in the Southern Hemisphere, temperature is most
important in the tropical western Pacific. Initial adjustment (to the withholding of data)
takes about 20 days to complete. This may have implications for the minimum duration
of the nature run in the OSSEs. Forecasts from the tests' analyses are yet to be
executed and evaluated.
Michiko Masutani then presented results and conclusions of the evaluation of
the nature run (NR) cloud cover to be used in the OSSEs. A written report was
distributed to committee members before the meeting. NR clouds were compared with
cloud analyses from the Air Force (RTNEPH), International Satellite Cloud
Climatology Project (ISCCP), NOAA/NESDIS (CLAVR, currently just total cloud
cover), and Warren (surface observations only). Some of the major findings from the
study were as follows:
NR total cloud cover is in general agreement with the cloud analyses,
large differences between the cloud analyses preclude definitive conclusions about
NR cloud quality,
NR appears to have too much high cloud, but uncertainties in reference analyses
do not justify adjustments,
uncertainties in mid-level cloud do not justify adjustments in NR clouds,
NR had less low cloud than ISCCP and RTNEPH to an extent that adjustments
seemed necessary: excessive clouds over snow were reduced, insufficient marine
stratocumulus were enhanced.
Subsequent discussion centered on the implications of the reference analysis
uncertainties, particularly in the high cloud estimate. Steve Lord suggested that since
there is so much uncertainty in the specification of the real cloud cover state, it may be
necessary to run a range of OSSEs in which the cloud cover are set to a range of values
bracketed by possible bounds.
Sid Wood of Simpson Weather Associates (SWA) presented an overview of
their lidar simulation model (
""
Joe Terry of NASA/GLA gave a report on the simulation of the conventional
observations from the NR data set. BUFR-formatted real observations (that were used
in the NCEP/NCAR Reanalysis) and the NR data are combined in a process to extract
simulated "observations" from the NR data set. The next step is to introduce random
error in the simulated observations, using published observing error standard
deviations. This represents an attempt to account for instrument error and spatial
representativeness error. Then the rawinsonde drift simulator algorithm is used to apply
modifications to the rawinsonde simulations to account for spatial and temporal
displacement of the sonde during ascent. To date, the following types of conventional
data have been simulated for January and February 1993: rawinsonde and pilot
balloons, aircraft reports, buoy, PAOB, surface observations, and cloud motion winds.
Discussion following the report centered on two topics. First, how to account
for the increase in types of observing systems that are available globally since 1993?
How do we account for this in the OSSEs, so that we can realistically predict the
impact of the NPOESS instrument suite when it is fielded? Second, random error is
being added to the extracted observations, but not systematic error. We know there are
biases in the observations (notably, relative humidity). Is it important to account for
this in simulating the conventional observations, and if so, how?
Jack Woollen of NCEP/EMC-GSC gave a presentation on the Doppler Wind
Lidar database for use with the NCEP global data assimilation model (SSI). In addition
to time and location of the DWL LOS wind profile, the following parameters are
included for each vertical level in the profile: altitude, elevation angle, azimuth,
number of laser shots in product, backscatter source index, atmospheric depth of
product layer, horizontal LOS component, standard deviation of LOS estimates, DWL
quality mark. The simulated DWL LOS winds assumes a conical scan strategy, having
two LOS wind estimates in each 200 km grid cell. These could be used to compute the
(u,v) component wind profile, but the SSI can assimilate the LOS estimates directly.
Tom Kleespies of NESDIS gave a brief update of the progress on the simulation
of the satellite sounding data. The two primary goals of the effort are to make realistic
simulations of TOVS radiances from the NR data to test the OSSE sounding
infrastructure, and then to extend the simulation technology to the advanced sounders
planned for NPOESS. The necessary ingredients in the simulation process are: radiative
transfer model, specification of geotemporal locations for the soundings, the nature
run, specification of realistic instrument noise, and specification of scene noise. The
phenomena that are necessary for inclusion in the radiative transfer model are
transmittance, emissivity, cloud, cloud liquid water, and precipitation. For the OSSE,
the radiative transfer model (also known as the forward model) to apply to the NR data
to simulate radiances should be similar to but sufficiently different from that used in the
data assimilation process. The radiative transfer model known as OPTRAN has been
selected for use in the assimilation process, and the model known as RTTOV5 will be
used to simulate the radiance data. Tom presented tables comparing the features of
these two forward models. Specification of geotemporal locations should be realistic,
and ideally taken from orbit geometry and sampling strategies planned for the NPOESS
sensor suite. For simulated TOVS, these can be obtained from the real TOVS 1B data
corresponding to the OSSE period. For the NPOESS advanced instruments, this will
require a knowledge of the scan patterns and orbital parameters. Parameters used from
the NR data for satellite sounding simulations include profiles of temperature,
moisture, cloud liquid water, and cloud fraction, as well as surface temperature and soil
moisture. Ozone profiles for the period will be obtained from the Climate Prediction
Center. Realistic instrument noise can best be specified from the TOVS existing HIRS
and MSU instrument estimates by monitoring space and black body calibration targets.
Instrument noise for advanced sounders will have to be based on engineering "best
guesses." Scene noise specification should include inhomogeneous features in field of
view, unresolved clouds, precipitation, and other similar effects. The current status of
the satellite sounding simulation component of the OSSE project is: (1) the
infrastructure for simulating the cloud cleared, limb adjusted radiances was almost
completed when the contract expired, (2) NCEP data assimilation model is now
assimilating TOVS 1B radiance data, which are not cloud-cleared or limb adjusted, so
(3) the modified simulation infrastructure will have to account for the presence of
clouds in the simulated 1B radiances. One-time-only funds have been identified to
accomplish the simulation of the TOVS 1B radiances, and the work is to start October
1. Continued funding from the NPOESS IPO is required to accomplish the simulation
of the advanced sounders.
Michiko Masutani presented an evaluation of simulated DWL winds scenario
sensitivity results. Some preliminary OSSEs were conducted using the February 6 and 7
real and simulated observational data for the following experiments: real conventional
data (identical to NCEP analysis), simulated conventional data, simulated DWL data
and real conventional data, and simulated DWL data and simulated conventional data.
The latter experiment was conducted in three DWL technology scenarios: best case,
moderate technology, and worst case (see paragraph above on Sid Wood's
presentation). Starting from the NCEP analysis without TOVS radiances, each OSSE
involved the assimilation of the designated observational data over the two-day period.
Because it has no data simulated from (extracted from) the NR, the comparison of the
real conventional data analysis and the NR should yield the largest differences. Then
each OSSE involving simulated data should bring the resulting analysis closer to the
NR (truth). In the DWL OSSEs the simulated DWL winds were assimilated as LOS
estimates directly, which required the transformation of the analysis background fields
of (u,v) wind components to LOS estimates to compute the observation increment. A
six-hour window (± 3 hours) about the analysis time was used in the 3DVar
assimilation. Results showed that the best case DWL simulated data reduced differences
considerably from the real data-NR differences, while the worst case DWL simulated
data had little effect on the differences.
Dave Emmitt of Simpson Weather Associates made a presentation on progress
made in Doppler Wind Lidar (DWL) wind measurements. His talk focussed on
developments occurring across the DWL research and development community. Dave
gave a brief description of some initiatives in DWL measurement systems being
undertaken by agencies of the U.S. government. Also, at least two European agencies
are conducting research and development activities in DWL measurements and impact
studies. In summary, much activity is ongoing in agencies sponsoring DWL research
that is of relevance to NPOESS DWL project.
Dave next focused his remarks on the current NPOESS DWL simulation study.
He commented on the bracketing scenarios being considered for the OSSEs. These are
described on the SWA web site given above. He stated that what is not being accounted
for in the proposed bracketing assumptions includes realistic hardware, realistic
backscatter distribution, attenuation factors, and situation dependent cases. He feels that
not accounting for these factors puts them in a position to do trade studies that is,
how much gain can be gotten from how much extra cost of the DWL system.
Dave described the consensus agreements made at a meeting on metrics for
measuring impacts of the OSSEs, conducted in July 1999 and involving OSSE Team
members. The metrics that were agreed upon at that meeting were:
number of cyclones, positions, and intensities,
number of jets, their positions and core speeds,
occurrence and maximum and minimum temperature locations (including number of
extreme events)
anomaly correlation by spectral binning,
data rejection statistics,
flight path winds,
shear layers, location and strength.
Subsequent discussion centered on the need for a water vapor metric. Advisory
committee members expressed the need for the OSSE Team to consider such a metric,
for example water vapor anomaly. The Team is somewhat dubious at this time as to
how much information can be gained from the inclusion of a water vapor metric. One
reason is that the impacts on water vapor distribution can be very much dependent on
the characteristics of the prediction model used in the OSSEs.
Joe Terry briefly presented work being done at NASA/GLA on cyclone tracking
methods. The characteristics of individual cyclones are of interest in the OSSE
evaluations, and there is a need to associate cyclones simulated in the OSSEs with
corresponding cyclones in the NR. Joe described a method under study at NASA/GLA
that identifies individual cyclones from the global weather parameter grids using the sea
level pressure field to find the centroid of the innermost closed contour. The cyclone
tracking method he described estimates the forecast position of the cyclone center by
extrapolation or 500 hPa steering. A search is conducted in the resulting area of
estimate to determine the whether the cyclone's state is one of continuation, cyclolosis,
or cyclogenesis.
Steve Lord next informed the advisory committee of highlights from a
workshop on the evaluation of space-based lidar technology, hosted by the Office of
Oceanographic and Atmospheric Research on September 23, 1999. NESDIS is seeking
guidance from OAR on a possible future data buy. More information on this is
contained in the following paragraph. In addition, regional and global lidar wind
impact OSSEs may be conducted under the auspices of OAR. This represents a separate
OSSE study for which funding may be limited to one year.
Jim Ellickson of NOAA/NESDIS next gave a brief discussion of the proposed
NOAA/NESDIS commercial global data buy. This idea began in fiscal year 1998 when
NESDIS began consideration of the feasibility of commercializing satellite weather data
collection. Jim explained that in a data buy program, the government enters into a
contract with a private company. The contract is taken to a lending institution to
borrow capital needed to build, launch, and operate the observing platform and system.
The contract would obligate the government to purchase the data if contractual
agreements on data quality, timeliness, and coverage were met. Public meetings and
requests for information are used by the government to collect information from
potential contractors necessary to formulate the request for proposals. OSSEs and other
NPOESS-funded projects would supply information to the contract winners, as this
information is all public domain. The data buy approach is a possible alternative to a
government build and field approach. A cost study is being conducted at NASA-Stennis
to estimate the cost to the government of the latter approach.
Steve Mango of the NPOESS IPO presented some remarks on the NPOESS
program in general and the OSSE project in particular. The IPO wants to continue to
fund the OSSEs. The payloads depend on it, as they need the information in their
design and fabrication. Payload proposals have been evaluated and winners have been
announced. Now the instruments are in the engineering management and development
phase. By the year 2008, a constellation of three coordinated polar orbiters will be in
operation two from NPOESS, and one from Europe (the METOP satellite). Steve
showed a list of notional payloads necessary to satisfy IORD-1, the existing operational
requirements document. There are instruments planned for the NPOESS sensor suite
for which no OSSEs are planned. Steve said that the timelines of fielding these
instruments dictates that their development has to be done in parallel with the OSSEs.
The OSSE team has to make the hard choices to select which instruments to simulate in
their OSSEs the instruments likely to have the most impact on global weather
analysis and prediction. NPOESS currently faces an $80M reduction in FY2000 out of
a total $160M budget. The combined DOC and DoD contribution is $35M less than the
FY1999 appropriation. In spite of these projected budget shortfalls, the NPOESS IPO
remains committed to fund the OSSE effort.
Issues and Recommendations
Issue: Definition of "cloud" in the ECMWF Nature Run data set. In particular, the
need for a standard "threshold" measure of cloud on which cloud coverage is based,
such as a threshold level of optical depth, emissivity, or ice water content. It is
accepted that most conventional satellite analyses of high level clouds report too little
high cloud. Since the greatest impact of clouds on the conduct of the OSSE is likely to
be in the simulation of DWL wind observations, it may be necessary to use a
supplemental measure of cloud density, like threshold optical depth or ice/liquid water
content.
Recommendation: The OSSE Team should supplement their study of the nature run
cloud cover with a evaluation of the consistency between NR cloud cover and cloud
water content. If the two cloud indicators are consistent in time and space, then
threshold levels of cloud water content as dictated by the characteristics of the
bracketing lidar technologies should be used to indicate the essential presence or
absence of cloud insofar as the lidar beam interception or non-interception is
concerned. Cloud cover may be used as before to estimate the probability of
interception by cloud that is considered opaque.
Issue: "Bracketing experiments" have been proposed to test the sensitivity of weather
analyses and forecasts to a range of lidar power and sampling options. Do these
experiments give us confidence that a particular technology will be realistically
simulated for use in the OSSEs? Are these options relevant enough to feasible lidar
technologies?
Recommendation: The NPOESS OSSE Team should comprehensively survey all
proposed or operating lidar technologies and, with space deployment in mind, use these
to check the bracketing assumptions proposed for the OSSEs. Conservatively, all
assumptions should lie within the range of viable proposed or operating lidar
technologies. The survey should include both domestic and international lidar
researchers. The Oregon lidar meeting held earlier this year should provide substantial
information for such a survey.
Issue: The number of global observations routinely available is (1999) already greater
than in 1993 (date of nature run). How do we account for this in the OSSEs? Perhaps
more importantly, how can we project the observational coverage and its impact for
when the NPOESS sensor package is flown? How do we reconcile the need to
determine likely future impact with the need to simulate exactly the same observations
present in 1993 for OSSE calibrations? Does this suggest the need for a series of
sequentially more future OSSEs, in which incrementally more and newer observing
systems are simulated, culminating in the NPOESS sensor suite? Is there a possibility
that NPOESS instruments' impacts will be overtaken by progress in other sensor
systems?
Recommendation: The advisory committee does not expect the OSSE Team to foretell
the future of availability and accuracy of observing system in the next ten years.
However, we feel that it is crucial to estimate the impact of NPOESS sensors on global
weather analysis and prediction with the future and not the present constellation of
non-NPOESS observing systems in mind. It is possible that, with the advances in
observing systems and atmospheric modeling systems seen recently, it may be very
hard to show appreciable impact from the NPOESS sensors, particularly in the
Northern Hemisphere. One aspect of the projection of assimilation of data in global
weather analysis systems that must be factored in is the capacity and efficiency of
computer systems available in the future. One OSSE Team task planned for FY2000 is
entitled "Consider the 2008 data world." We recommend that a series of non-NPOESS
OSSEs be conducted that attempt to account for the evolution of observing systems and
prediction systems between now and 2008. Another suggestion is to rerun critical
OSSE experiments between now and 2008 whenever major changes occur in the
assimilation system or the globally available observations. Finally, the OSSEs might
profit by their expansion to the wider research community. This would require making
a community model, perhaps using the NCEP assimilation system as a template. Then
researchers (especially instrument scientists) could run their own OSSEs and estimate
their sensor's impacts on the community system. This allows for competition and
intercomparison of results, and distributes the labor and computational burden. It does
require significant documentation, management, and servicing.
Issue: Random error is being added to the simulated conventional observations, but not
systematic (mean) error. We know that there are biases in the observations (notably
relative humidity). Is it important to account for this in the simulation of the
observations? If so, how can we do this? Is there any such information that can be
used? Or can the (known) systematic error simply be removed from the observations
before they are assimilated?
Recommendations: The OSSE Team could review other recent OSSEs reported in the
literature to determine if systematic error of sensor systems was accounted for in the
simulations. Then the sensitivity of the OSSEs to imposed systematic error could be
tested by repeating a preliminary OSSE experiment in which the magnitude of
systematic error imposed on certain simulated conventional observations is varied
between experiments. Is the response linear with the imposed error level? If the error is
imposed on a moisture observation, what effect does it have on a mass or motion field
analysis or prediction, or vice versa? The impacts from these experiments may suggest
the benefit of removing systematic errors (if known) from observations before they are
assimilated.
Issue: Metrics for evaluating the impact of future observing systems were presented at
the meeting. The advisory panel suggested that a water vapor metric be included, such
as water vapor anomaly. Another was a wind impact, more spatially widespread than
the list presented at the meeting. That list was:
number of cyclones, positions, intensities
number of jets, positions, core speeds
number occurrence and locations of max. and min. temperatures (plus # of
extremes)
anomaly correlation by spectral binning
data rejection statistics
flight path winds
shear layers, location, and strength.
Recommendation: The OSSE Team should look at recent observing system experiments
conducted at NCEP and elsewhere to assess the metrics used in the assessment of
impact by candidate observing systems. With the increase emphasis on cloud and
precipitation prediction in the meteorological community, it seems likely that suitable
moisture impact metrics may be identified. On the other hand, it may be found through
such a study that attempts to measure impacts from or on the atmospheric moisture
distribution are frustrated by the sensitivity of the individual NWP model formulations.
Wind impact is expected to be high from high powered lidar technologies. Translating
the physical impact into an economic impact would be a highly beneficial metric.
Issue: There seems to be more instruments planned for NPOESS than there are OSSEs
planned for impact assessment. Unfortunately, the timelines of fielding these
instruments dictate that their development has to be done in parallel with the OSSEs.
The NPOESS IPO is asking the OSSE team to make the hard choices to select which
instruments to simulate in their experiments, and how to simulate them. Does this mean
to choose to include the instruments hypothesized to have the more marginal impact, to
guide the NPOESS IPO in their funding decisions as to which instruments actually get
deployed? Or does this mean to choose to include the instruments likely to have the
greatest impact, so that assimilation software systems can be fully developed, tested,
and in place to use these major-contributing sensors as soon as they begin sending real
data?
Recommendation: In cooperation with the NPOESS IPO, the OSSE Team should
re-examine the subset of sensors they have chosen for inclusion in the simulation
impact studies. The advisory panel has yet to see a rational justification for the choice
of sensors to be simulated in the OSSEs. What do these sensors potentially have to
offer that makes them a better choice for representation in the OSSEs than the sensors
not chosen? Do we run the risk of not being prepared to assimilate the data from
NPOESS sensors that may have significant but undiscovered potential? The OSSE
Team should provide an interim report on this topic to the advisory committee, with a
reasoned and comprehensive survey of each notional payload planned for IORD-I, and
why the payload is or is not represented in the planned OSSEs, and an explanation of
the simulation approach for the selected payloads.
Issue: A lack of affordable and accessible computing resources looms as a bottleneck in
the execution of the many OSSEs planned in the next phase of the project. This
shortfall could cause significant delays in the processing of the experiments, which in
turn could delay the information getting to the NPOESS mission planners. Creative
solutions to this problem must be found this fiscal year (FY2000). What are the
options, to avoid the costly purchase of additional computer hardware? It was
mentioned that special project money (like NPOESS) has been pooled from several
such projects to purchase some limited computer resources. Much more are needed
because these have to be shared among the projects. Are there accessible government
computer systems available? How do we find out, and can they be used for the OSSEs
with little or no cost to the project?
Recommendation: The OSSE Team should determine the availability, accessibility, and
feasibility of use of DoD computing resources, so that not all of the burden rests on
computers local to NCEP. The DoD High Performance Computing Modernization
Office http://www.hpcmo.hpc.mil/ would be a great place to start. While it may be
felt that the disadvantages of shipping files back and forth from remote computational
centers would outweigh their usefulness, it should be kept in mind that the OSSE
project does not require real-time turnaround. It may be that the government already
has adequate computational resources to handle the OSSE computations in a timely
manner, obviating the need for the purchase of more. This avenue should be seriously
explored before committing additional scarce resources to the purchase of new
computational hardware. Other possible solutions include: international cooperation
(for example, cooperation with Japanese scientists who will have enormous
supercomputing capabilities in two or
three years in connection with the development of the earth simulator at the Frontier
Research System for Global Change, Tokyo, Japan); asking the other sponsoring
organizations, NOAA and NASA, to help with computer time support (operational data
assimilation centers, like NOAA and NASA, will benefit from the development of
assimilation methodologies for the OSSEs).