Abstract:
|
The Ensemble Kalman Filter (EnKF) and its relatives provide a powerful family of methods for optimal Bayesian tracking of spatial dynamic processes, even when the input data stream is seriously compromised with missing values, irregular arrival of data, corrupt data, and changing spatial aggregation boundaries. The data model of the Kalman Filter, when properly expressed in linear algebra, is surprisingly flexible and elegant in its formulation. In this paper we explore the full range of data corruption that are expressible in the data model, and show the effectiveness of the tracking algorithm within what is quite possibly one of the worst data environments in all of international public health and epidemiology. The data modeling methods in this paper exploit techniques in linear algebra that are not well understood and employed in the spatial epidemiology literature.
|