Building a Seamless Data Assimilation and Ensemble Prediction System: What Should It Simulate? (Invited Presentation)

In recent years, we have witnessed an integration of ensemble-based
data assimilation systems and of ensemble prediction systems.
For example, an ensemble Kalman filter or an ensemble data
assimilation system can be used to provide the ensemble of
initial conditions for both a short-range and a medium-range
ensemble prediction system. This seamless use of ensembles
is desirable from a theoretical point of view but comes with
practical challenges.

A seamless data assimilation system must necessarily have a
global domain. Ideally, the initial ensemble will serve for
both global medium-range and regional short-range ensemble forecasts.
To obtain high-quality short-range forecasts, it is desirable
to have balanced initial conditions that reasonably match
observations from high-resolution observational networks. It
is also desirable to have frequent analyses from a short-window data
assimilation cycle. To have sufficient error growth in the
early forecast ranges, before errors project mostly on
baroclinic modes, errors with realistic amplitudes must be generated
for the observations, the model and the lower- and upper-boundary conditions.

Whereas the Monte Carlo frame work permits accounting for many
sources of error, it is still not clear how to arrive at a reasonable
and comprehensive understanding and description of error sources in data
assimilation and forecasting.