The objective of this project is to create and implement new methods for improving the performance of filtering, or data assimilation techniques, applied to non-linear, non-Gaussian systems. Many problems in computational statistics, artificial intelligence, and geophysics involve such systems, and utilize specialized Monte Carlo sampling methods, called particle filters, for data analysis and forecasting, and for better understanding of the underlying phenomena. The principal investigator studies variational data assimilation methods to demonstrate that targeted ensemble generation using those methods delivers more effective filter performance, and that dynamic adaptation of the size of the particle ensemble improves the computational efficiency of the filter.
Statistical computations are an essential tool for the solution of problems in such diverse areas as artificial intelligence, industrial and consumer electronics, robotics, weather systems, climate change, ocean ecosystems, and land surface processes. The proposed research improves statistical computations required for the analysis, estimation, and forecasting of information that arrives over time, and contributes to a better combination of theoretical models with observational data. In addition, the project involves the training of undergraduate students in applied scientific research.
