Data Assimilation

ePiSAT
FACT BOX

SOFTWARE TOOL   Data Assimilation system
COLLABORATORS   CSIRO, eMAST@Macquarie University, NCI, NCAR (USA)
APPLICATION   Modelling and data integration system for ecosystem science.

Data assimilation, specifically, ‘ensemble Kalman filtering’ (running multiple instances of a model to generate different states) is using real observations to nudge the model towards a state that is more consistent with the observations themselves.

One if eMAST’s key objectives has been to develop and deploy a continental scale data assimilation system on Australia’s National Computational Infrastructure (NCI) at the Australian National University (ANU).

Recent Australian-US collaboration has progressed the development of a research community focused ecosystem modelling and observation integration system for Australian on one of the nation’s top computing facilities.

Researchers from CSIRO, Macquarie University and the NCI teamed up with US collaborators to install and run the Data Assimilation Research Tested (DART) on NCI’s supercomputer (Raijin) and coupled to it Australia’s Community Atmosphere Biosphere Land Exchange (CABLE) land surface model. The endeavour marks significant progress toward the vision of eMAST to develop Australia’s first modelling and data integration system for ecosystem science and monitoring at unparalleled scales in space and time. The system will bring together a range of disparate ecological observations from ground- and space-based sensing networks into CABLE’s modelling framework.

The DART package provides a range of ensemble-based filtering techniques to assimilate both point and spatial observations into CABLE, including variants of the Kalman filter and particle filter. DART offers a convenient “package” for evaluating different assimilation techniques for CABLE, and will be used (along with NASA’s Land Information System) in a methods intercomparison planned for later this year and driven by the OzEWEX Working Group on Data Assimilation, but will also engage working groups on Model Evaluation and Benchmarking and Vegetation Processes. Outcomes of the intercomparison will be reporting on the accuracy and computational feasibility of the various data assimilation methods for CABLE, and importantly will provide indicative resourcing requirements for large-scale/continental data assimilation need for scoping operational systems. The team recognises the enormous computational challenges of assimilating the variety of observations into CABLE and the generated model estimates (and uncertainty) at daily and 1-km resolution across Australia. The endeavour will not be possible were it not for the NCI, not least of all because it remains the only computing environment accessible to researchers from government institutions and universities alike.