Geography Reference
In-Depth Information
precipitation, and cloud-radiation interactions are parameterized using specified
cloud climatologies.
13.7
DATA ASSIMILATION
The capability of a numerical forecasting model to produce useful forecasts depends
not only on the resolution of the model and the accuracy with which dynam-
ical and physical processes are represented; it is also critically dependent on
the initial conditions employed for integrating the model. As Richardson's early
attempt to forecast weather numerically showed, observations cannot be used
directly to initialize a numerical forecast. It turns out that observational data
must be modified in a dynamically consistent fashion in order to obtain a suit-
able set of data for model initialization. This process is usually referred to as
data
assimilation
.
Traditionally, data assimilation was typically divided into two processes:
objec-
tive analysis
of the observations and
data initialization
. In the objective analysis
step, all data acquired for a given time (generally 00Z or 12Z) from the irregularly
spaced observational network of surface and upper air stations are checked for
accuracy and interpolated to points on a regular latitude-longitude grid at standard
pressure levels. A
background
or “first guess” estimate derived from a short-range
forecast is used to fill in grid points in data-sparse regions such as the oceans.
Such objectively analyzed data still contain noise that is likely to be interpreted as
spuriously large gravity waves when data are used as initial data in a numerical
model. In the initialization step, these objectively analyzed data are modified in
order to minimize the gravity wave noise, and hence reduce the magnitude of initial
velocity and pressure tendencies. More recently the objective analysis and initial-
ization steps have been combined in an approach called
four-dimensional data
assimilation
, which makes optimal use of all available observational and forecast
information.
13.7.1
The Initialization Problem
The importance of the initialization step in preparing initial data for forecasting
with primitive equation models can be illustrated by considering the relative mag-
nitudes of the various terms in the horizontal momentum equation in pressure
coordinates:
∂
V
∂t
+
ω
∂
V
(
V
·∇
)
V
+
∂p
=−
f
k
×
V
−
∇
(13.54)
For synoptic-scale motions in extratropical latitudes the wind and pressure
fields are in approximate geostrophic balance. Thus, the acceleration following
