DART: A community facility for ensemble data assimilation

The Data Assimilation Research Testbed (DART) is a mature community

software facility providing researchers access to state-of-the-art

ensemble data assimilation tools. The freely-available DART distribution

includes fully functional low-order and high-order models, support

for commonly available observations, hooks to easily add both new

models and observation types, diagnostic programs to interpret the

results, and a full tutorial suitable for self-study or teaching

data assimilation concepts, including exercises using the models

distributed with DART.

DART is used regularly with a number of geophysical models including

NCAR's WRF and CAM atmospheric models. DART/WRF is being used for

tropical storm analysis and prediction in the Pacific and Atlantic

and was used to produce real-time predictions during the 2009 Atlantic

hurricane season. DART/CAM has played an integral part in the development

of the new CAM version 4 that will be used for NCAR's contribution to

the next IPCC. DART/CAM has been run for many model configurations

to evaluate CAM systematic errors and parameterization options. DART

is also in use for chemical assimilation in the WRF-CHEM and CAM-CHEM

versions of these models.

New models, both small and large continue to be added to the set

compatible with DART. During 2009, DART assimilation was

developed for the POP (Parallel Ocean Program) ocean general circulation

model that is being used for decadal coupled model prediction at

NCAR. The newest version of the Planet WRF model, configured for

Martian data assimilation, is also now in use with DART.

Novel observation types also continue to be added to DART. For instance,

assimilation capabilities for radiance observations from the MOPITT

instrument on earth and from TES on Mars have been added in 2009.

New tools to improve and simplify the use of ensemble assimilation

continue to be added to the DART facility. Recent research has focused

on automated tools for reducing sampling errors in ensemble

assimilations.