This thesis compares the forecast performance of four strategies for coupling

global and limited area data assimilation: three strategies propagate information

from the global to the limited area process, while the fourth strategy feeds back

information from the limited area to the global process. All four strategies are

formulated in the Local Ensemble Transform Kalman Filter (LETKF) framework.

Numerical experiments are carried out with the model component of the National

Centers for Environmental Prediction (NCEP) Global Forecast System (GFS)

and the NCEP Regional Spectral Model (RSM). The limited area domain is an extended

North-America region that includes part of the north-east Pacific. The GFS

is integrated at horizontal resolution T62 (about 150 km in the middle latitudes),

while the RSM is integrated at horizontal resolution 48 km. Experiments are carried

out both under the perfect model hypothesis and in a realistic setting. The coupling

strategies are evaluated by comparing their deterministic forecast performance at

12-hr and 48-hr lead times.

The results suggest that the limited area data assimilation system has the

potential to enhance the forecasts at 12-hr lead time in the limited area domain at

the synoptic and sub-synoptic scales (in the global wave number range of about 10

to 40). There is a clear indication that between the forecast performance of the

different coupling strategies those that cycle the limited area assimilation process

produce the most accurate forecasts. In the realistic setting, at 12-hr forecast time

the limited area systems produce more modest improvements compared to the global

system than under the perfect model hypothesis, and at 48-hr forecast time the

global forecasts are more accurate than the limited area forecasts.