NWC REU 2014
May 21 - July 30



Photo of author

An Evaluation of Applying Ensemble Data Assimilation to an Antarctic Mesoscale Model

Lori Wachowicz, Steven Cavallo, and Dave Parsons


What is already known:

  • The lack of observations in the Antarctic make it difficult to observe the changing mass balance of the ice sheets.
  • The use of high resolution numerical models provide insight to finer-scale processes that may be affecting the Antarctic ice sheets.
  • Ensemble data assimilation is optimal for data-sparse regions by providing uncertainty estimates in order to improve the model.

What this study adds:

  • Using the analysis increment provides a diagnostic for determining model bias.
  • Surface observations may be biased and not well-represented by the ensemble.
  • An upper-level warm bias is leading to a potential circulation bias in the ensemble model.


Knowledge of Antarctic weather and climate processes relies heavily on models due to the lack of observations over the continent. The Antarctic Mesoscale Prediction System (AMPS) is a numerical model capable of resolving finer-scale weather phenomena. The Antarctic’s unique geography, with a large ocean surrounding a circular continent containing complex terrain makes fine-scale processes potentially very important features in poleward moisture transport and the mass balance of Antarctica’s ice sheets. AMPS currently uses the 3DVAR method to produce atmospheric analyses (AMPS-3DVAR), which may not be well-suited for data-sparse regions like the Antarctic and Southern Ocean. To optimally account for flow-dependence and data sparseness unique to this region, we test the application of an ensemble adjustment Kalman Filter (EAKF) within the framework of the Data Assimilation Research Testbed (DART) and AMPS model (A-DART). We test the hypothesis that the application of A-DART improves the AMPS-3DVAR estimate of the atmosphere. We perform a test using a one-month period from 21 September - 21 October 2010 and find comparable results to both AMPS-3DVAR and GFS. In particular, we find a strong cold model bias near the surface and a warm model bias at upper-levels. Investigation of the surface bias reveals strongly biased land-surface observations while the warm bias at upper-levels is likely a circulation bias from the model warming too rapidly aloft over the continent. Increasing quality control of surface observations and assimilating polar-orbiting satellite data are expected to alleviate these issues in future tests.

Full Paper [PDF]