Geoscience Reference
In-Depth Information
Chapter 23
Ensemble Adaptive Data Assimilation
Techniques Applied to Land-Falling
North American Cyclones
Brian C. Ancell and Lynn A. McMurdie
23.1
Introduction
Adaptive data assimilation is becoming an increasingly important aspect of numer-
ical weather prediction. Traditional data assimilation involves combining a set
of routine observations with a first-guess field provided by a numerical weather
prediction model to produce an analysis of the atmospheric state. These analyses
subsequently serve as the initial conditions for extended forecasts. There are three
primary modern data assimilation methods that assimilate routine observations at
operational centers around the world and within a number of research applications:
(1) three-dimensional variational (3DVAR) systems, (2) four-dimensional varia-
tional (4DVAR) systems, and (3) ensemble Kalman filter (EnKF) systems. Each of
these techniques are based on the assumption that the errors of both the first-guess,
or background, variables and the observations are distributed normally, and aim to
identify the most likely atmospheric state within the statistical framework of Bayes'
Theorem (overview provided in Kalnay 2002 ).
Adaptive data assimilation allows the consideration of observational impact in
some way beyond the aggregate effects of a set of routine observations. There
are two primary types of adaptive data assimilation: (1) observation impact, and
(2) observation targeting. Observation impact methods estimate the relative impact
of each assimilated observation, or any subset of assimilated observations, on
a chosen forecast metric. In turn, these techniques are able to identify which
observations are important, and which are redundant, with regard to a number
Search WWH ::




Custom Search