Challenges in Adaptive Sampling

Sharan Majumdar

Rosenstiel School of Marine and Atmospheric Sciences
University of Miami

In recent years, objective adaptive sampling techniques have been designed to identify optimal locations and times for the deployment of mobile observational resources, in order to improve a given forecast. The two most widely used techniques to date have been Singular Vectors (SVs) and the Ensemble Transform Kalman Filter (ETKF). The latter technique is one of a family of square-root filters used in data assimilation. Theoretical similarities and differences between the two techniques will be discussed along with their strengths and limitations. These will be supported by illustrations of guidance issued by the SVs and ETKF for winter storm and tropical cyclone forecasts. Ideally, the analysis error covariance matrix estimated by the adaptive sampling technique would be consistent with that assumed by the data assimilation scheme. Unfortunately, this is not the case in practice, and hence the ability of the adaptive sampling technique to predict the reduction in forecast error variance produced by observations is compromised. This difficulty and other issues related to numerical weather prediction and data assimilation will be discussed.

Last Update: April 29, 2005