ETA MODEL FORECAST SENSITIVITY TO INITIAL CONDITIONS
         FOR THE 22/23 FEB 98 FLORIDA TORNADOES CASE

                    Michael E. Baldwin
     NOAA/NCEP/EMC/General Sciences Corp., Camp Springs, MD

        Operational forecasts from the Early (32km) and Meso (29km) Eta
Models provided significantly different guidance for the tornadic
thunderstorms that hit central Florida around 0300-0600 UTC 23 Feb 98.
The Meso Eta, beginning at 0300 UTC 22 Feb 98, predicted strong upward
vertical velocities and heavy precipitation in the vicinity of these
thunderstorms at the time they were observed.  However, the Early Eta,
which began at 0000 UTC 22 Feb 98, forecasted weaker vertical motions
and significantly lighter precipitation.  Both models started out with a
strong disturbance off of the Texas coast, with a coherent region of
strong upward vertical motion associated with elevated convection.
Evidence of this elevated convection can be found in the 00h Early Eta
sounding near Houston, TX, which showed a nearly-saturated layer between
850-700mb.  The disturbance appeared as a line of precipitation that
moved across the Gulf of Mexico during the forecast.  Although the
forecast profiles were convectively unstable, the bulk of the heaviest
precipitation was explicitly predicted by the model's cloud physics
routines, rather than the model's convective parameterization.  The
convective scheme is designed to allow grid-scale (explicit)
precipitation to take over once the vertical motion is strong enough to
force grid-scale saturation.  The Meso Eta maintained this region of
intense upward motion as it moved across the Gulf of Mexico, while the
Early Eta weakened it considerably after ~15h into the forecast.

        The model systems contained differences in horizontal
resolution, domain size, boundary conditions, data assimilation
techniques, and forecast model code.  Experiments were performed to
examine the impact of these differences on the forecasts for this case.
The two models had slightly different formulations at the time of these
forecasts.  An experiment was performed with the Early Eta model code
running at 29km with Meso Eta initial conditions.  This produced nearly
identical results as the operational Meso Eta forecast, therefore, these
model differences do not appear to be the cause of the forecast
differences.  It is also possible that differences in domain size and/or
boundary conditions were important in this case.  The southern boundary
for the Meso Eta was approximately 500km from the Gulf of Mexico, while
in the Early Eta that boundary was approximately 1000km away from the
Gulf.  Boundary conditions for the Meso Eta were obtained by
interpolating the 3-h old Aviation run of the Global Spectral Model
(AVN), while in the Early Eta, a 12-h old AVN forecast provided the
boundary conditions.  The impact of these differences has not been
explicitly tested.

        The initial conditions for the Early Eta were provided by an
updated Eta Data Assimilation System (EDAS) which included
3D-variational (3DVAR) analyses, rather than the optimal interpolation
(OI) analysis scheme which was used in the Meso Eta EDAS.  One of the
important advances in the variational analysis system is the ability to
assimilate "non-traditional" data sources, such as GOES precipitable
water estimates.  These data were included in the Early Eta EDAS but not
in the Meso OI analyses, other such data excluded from the Meso Eta
include aircraft temperatures, VAD wind profiles from WSR-88D, SSMI
sea-surface wind speeds, and surface winds over land.

        The initial precipitable water fields indicate the Early Eta was
considerably drier over the southern Gulf of Mexico than the Meso Eta,
most likely due to the inclusion of GOES precipitable water data in the
Early Eta.  This dry air advected northward during the forecast and
interacted with the line of precipitation at the time it weakened in the
Early Eta.  An experimental rerun of the Early Eta EDAS without GOES-PW
data confirmed this result, the ensuing forecast maintained the line of
heavy grid-scale precipitation as it moved eastward across the Gulf,
reaching Florida around 0300 UTC 23 Feb 98.

        Due to the complexity of NWP systems, it is often very difficult
to link forecast discrepancies to specific model components or
particular characteristics of initial conditions.  Therefore, these
results should not be considered as a suggestion that GOES-PW data be
removed from the operational EDAS.  The drier air found in the GOES-PW
observations may have helped to maintain the severe convection over
Florida.  Typically, mid-level dry air is a prerequisite for strong
downdrafts and the creation of low-level baroclinicity, which may be an
important factor in the development of tornadoes.  Forecast fields
besides QPF and omega were not examined in detail and may prove to have
verified better in the Early Eta than in the Meso.  However, these
results do show the sensitivity of Eta Model forecasts to different
initial moisutre analyses and demonstrate that great care must be taken
when assimilating new sources of data.