February 24, 2010, 2:00 pm - 3:00 pm

February 24, 2010, 2:00 pm - 3:00 pm

Particle filter for time series modeling and its application to geophysical data

Hiromichi Nagao (The Institute of Statistical Mathematics)

Various methods based on the Bayesian approach such as the Kalman filter algorithm have been used for time series modeling. They enable us a precise time series modeling, but needed some complicated artifices in algorithms and/or models because of long computation time especially in the case that non-Gaussian distribution functions are included in the model. Following recent dissemination of high performance computing environments such as PCs with multi-core processors have ability to give a solution to this problem, we develop software using the particle filter (PF) algorithm applicable to multivariate time series modeling in various fields of science. The state space model is directly programmable in the framework of the PF, e.g., a non-Gaussian probability distribution function can be expressed without any approximation. We report an overview of our software and show some examples of its application to geoscience data. We also introduce an oncoming web application “CloCK-TiME (Cloud Computing Kernel for Time-series Modeling Engine, tentative name)”, in which our software is implemented on a cloud computing system available via a user interface on internet.