Expanded Rationale for the IPO/NOAA Bracketing OSSEs
G. D. Emmitt
10/20/99
The bracketing OSSEs are meant to explore the bounds of the potential impacts of a space-based
wind sounder on today's operational forecast models for several "technology neutral" observation
coverage and measurement error characterizations. By "technology neutral" we mean that the
instrument details are ignored but the general sense of how a lidar measurement is made is
retained. Thus, the following aspects of making wind measurements with space-based Doppler
wind lidars are captured:
Distributed vs. cluster sampling of the wind field within the TRV.
There are several reasons for conducting a series of bracketing OSSEs prior to performing
future OSSEs for very specific instrument concepts. First, we want to know the "ultimate"
sensitivity of our OSSE to the atmospheric parameter in questionin this case, winds. In other
words, the question is how much of an impact would be reported if perfect wind observations
from the Nature Run were available to the operational model. If the answer were "hardly
detectable" then there would be no reason to continue with the rest of the planned OSSEs. The
second reason for the bracketing OSSEs is to provide some measure of the relative return on
investment for several general DWL data products. In this case, some general form of
"cost/benefit" analyses can be achieved. A third reason is to develop the tools to evaluate the
more specific concepts that will be proposed to meet some stated requirements. A fourth
reason is to develop the understanding and experience of assimilating DWL data products long
before the instrument is launched. In this case, such a long lead-time increases the likelihood
that the instrument design may also benefit from the OSSE results.
The bracketing OSSEs listed in TABLE 1 are designed to establish the range of impacts that
could be expected from a range of DWL data product coverages and accuracies as they
compare with a "reference impact". The "reference impact" is that associated with the use of
perfect wind observations from the Nature Run. The perfect observations are constrained to
the temporal and spatial coverage of a space-based observing system. Otherwise, no cloud or
subgrid scale wind variance effects are simulated. The data product simulated for the
"reference" OSSE (R) is a wind profile from a Nature Run grid point closest to the center of a
200 km x 200 km data grid. This "reference" OSSE does not map to any real DWL. This data
set can be used with several assigned RMSEs to test for basic accuracy sensitivities of the
OSSE system.
The next four OSSEs in Table 1 have been defined to explore selected or limited data product
coverages that may, in a very general sense, be mapped to DWLs of differing coverage
potentials (referred to as the Coverage Series). The "Accuracy Series" of experiments uses the
same coverage scenarios as the Coverage Series but varies the instrument measurement
accuracy (e.g. m = 2 m/s or 7 m/s). Cloud and wind variance effects are invoked in all of the
OSSEs except for the "reference" OSSE (R).
The Experiment 1 data product is meant to represent that achieved with a very sensitive DWL
that is only prevented from making an observation by optically thick clouds. In some parlance
this data product would be referred to as the Holy Grail (200 km x 200 km version). The
coverage and accuracy (o << 1.0 mps) imply DWL systems which are well beyond the
current state of the art. In the "Accuracy Series" of OSSEs, the cases where m 1.0 mps
would map to all coherent systems and the increasing values of the measurement error would
map to less and less capable direct detection systems. (See discussion below regarding the
accuracy experiments.) Note that the coverage series data assumes perfect pointing knowledge
while the RMSE added for the accuracy series should account for all random measurement
errors.
Experiment 2 is designed to evaluate the relative impact of wind data that is obtained only
from clouds or the planetary boundary layer. The resulting data product might be similar to
that from a very modest sized coherent lidar. i.e. very accurate measurements from single
shots. As various amounts of measurement error are added to the base data product, the data
product begins to map to a direct detection aerosol lidar.
Experiment 3 represents a data product that would be obtained with an instrument that would
provide useful data only when there was a cloud free scene. Since a totally cloud free scene is a
very rare event for a 200km x 200km target area, we have used 50% cloud cover as the cutoff
for useful data. This product may map to the data product of a system that relies solely on
molecular returns.
Experiment 4 represents a bounding extreme in horizontal coverage. Whereas the swaths of
data in OSSE Experiments R,1,2 and 3 were all ~ 2000 km wide, the data in this case is
obtained from a non-scanning instrument. The resulting data pattern is a single LOS profile
provided every 200 km along the satellite ground track. The data product coverage in the
vertical is consistent with the same rules for Experiment 1, except that the shots within a 200
km x 200 km area are assumed to be clustered within a very small area of a few tens km
dimension.
(Note that an Experiment 5 OSSE has been added which represents a distinctly unique
combination of Experiments 2 and 3. While not considered a "bracketing OSSE" this
experiment should give some insight relative to the impact of data from high accuracy and
resolution in the PBL combined with lower accuracy and resolution in the mid- and upper-troposphere.)
There is an issue of how individual lidar shots are combined to yield a single wind observation
that will be assimilated into NWS models. Those DWL concepts that provide cross-track
coverage by scanning the lidar beam usually employ either a constant rate conical scan at a
fixed nadir angle (~30 45 degrees) or a step-stare conical scan, dwelling at prescribed azimuth
angles long enough to get sufficient signal to make a useful wind measurement. The following
can be taken as generally true for envisioned DWLs: Direct detection systems require
averaging multiple shots to obtain measurements with useful accuracy the more photons
collected, the lower the RMSE. Coherent detection systems have a basic accuracy of ~1.0 m/s
but may use shot accumulation to improve sensitivity in regions of lowest backscatter. Thus,
scanning and averaging imply including data taken over some extended areas to obtain a single
wind estimate for the models to use. The simulated data is used as if it pertained only to the
lat/long/height accompanying the wind speed. A data product based upon areal averaging is
treated with caution by the assimilation routines. Therefore, we have generated simulated
DWL data sets that include both the products resulting from large spatial averaging and an
observation that would result from a system that could concentrate all of its shots (that would
have been into the 200 km x 200 km column) into a very small area (~ 10 km x 10 km). The
OSSEs can then be run that should reveal the merits of the two extreme sampling strategies.
The following discussion addresses how various levels of RMSE can be added to the base
simulated data products provided for "Coverage Series" OSSEs.
The simulated DWL data sets are based upon a continuous conical scan and a prf of which
produces ~ 30 shots (15 forward and 15 aft) into a 200km x 200km area at the top of the
atmosphere. As the shots are propagated down through the Nature Run atmosphere, the
number of shots reaching a given level can be reduced by clouds in levels above. Thus N
(number of LOS samples) is not constant throughout a column.
In order to bracket the impacts due to instrument accuracy, errors of measurement and
representativeness need to be added to the simulated data used in the coverage OSSEs. That
additional error needs to be added in a RMSS manner and be a function of the number of
samples taken. There are two basic forms of the observation error (or weighting function used
in the assimilation cost function) that apply. Both forms derive from the following general
expression for o (observation error):
m = single shot measurement RMSE (illumination volume is ~ 100-1000 meters long and 10
meters diameter)
s = standard deviation of the wind field variability over the TRV
Nm = number of shots used to produce a single LOS wind observation where each shot has a
RMSE of m .
Ns = "effective" number of shots that reduce the potential sampling RMSE of s .
(Note: Nm and Ns are obtained by dividing the N in the simulated data records by the numbers
provided in Table 2.
In the Coverage Series of OSSEs the first term in this expression is assumed to be very close to
zero. Since we don't know s a priori, we might use the assigned o values for rawinsondes
(r) as substitutes for s (see Table 2).
Thus the weighting function that should be applied is
Ns = N/X for distributed shot case where X is found in Table 2; Ns = 1 for clusters
In the Accuracy Series of OSSEs, a Gaussian random error is added to represent various
instrument accuracies. It is taken that the accuracy applied is that which the instrument would
claim at the top of the atmosphere after accumulating or averaging all the returns from a
homogeneous, non-turbulent flow in a 200 km x 200 km x 1 km volume. As an example, for a
direct detection DWL system that is designed for a 3 m/s RMSE for a cloud free LOS
measurement using 25 shots, this would be interpreted to mean that
For the Accuracy Series of OSSEs the assimilation weighting function should be:
Ns = N/X for distributed shot case where X is found in Table 2; Ns = 1 for clusters
Table 1
Experiment Name
(Coverage Series)
Reference | 1 | 2 | 3 | 4 | 5 | |
Description of data product without regard to specific DWL technology | Perfect u,v observations from an orbiting instrument at single points within the TRV. No cloud or sub-grid wind variability effects accounted for. | Ultimate DWL that provides full tropospheric soundings, clouds permitting. | An instrument that provides only wind observations from clouds and the PBL | An instrument that provides mid and upper tropospheric winds only down to the levels of significant cloud coverage. | A non-scanning instrument that provides full tropospheric soundings, clouds permitting, along a single line that parallels the ground track | An instrument that provides all the data of experiment 2 plus some low resolution, lower accuracy data within the mid and upper cloud-free troposphere |
Vertical domain (km) | 0-20 | 0-20 | 0-20 | 3-20 | 0-20 | 0-20 |
Target Volume
(km x km x km)
(z> 2km) (z<2km) |
200 x 200 x 1
200 x 200 x.25 |
200 x 200 x 1
200 x 200 x.25 |
200 x 200 x.25 | 200 x 200 x 1 | 200 x 200 x 1
200 x 200 x.2 |
500 x 500 x 2
200 x 200 x.25 |
Swath width (km) | 2000 | 2000 | 2000 | 2000 | <200 | 2000 |
C: clustered shots
D: distributed shots |
C | C&D | C&D | C&D | C&D | C&D |
Table 1. DWL data height assignments and Observational errors Height (km) X Standard deviation .125 2 1.4 .375 1 1.6 .625 1 1.8 .875 1 1.9 1.125 2 2.0 1.375 1 2.2 1.625 1 2.3 1.875 1 2.4 2.5 5 2.4 3.5 5 2.6 4.5 5 2.8 5.5 5 3.0 6.5 5 3.2 7.5 5 3.4 8.5 5 3.4 9.5 5 3.4 10.5 5 3.2 11.5 5 3.1 12.5 5 3.0 13.5 5 2.7 14.5 5 2.6 15.5 5 2.5 16.5 5 2.5 17.5 5 2.5 18.5 5 2.5 19.5 5 2.7