HRD Seminar – Dr. Jonathan Poterjoy, University of Maryland – 16 October 2018

Dr. Poterjoy presented a seminar titled “Regional Weather Prediction Using an ‘Iterative’ Local Particle Filter”.


Particle filters (PFs) are sequential Monte Carlo methods that can solve data assimilation problems characterized by non-Gaussian error distributions for prior model variables or measurements. From the perspective of a geoscientist, PFs contain several theoretical properties that make them attractive for research and environmental prediction. Namely, they preserve dynamical balances during data assimilation update steps; they require no special treatment for nonlinear measurement operators or non-Gaussian errors; and they provide an elegant solution to the underlying Bayesian filtering problem. Recent efforts applying PFs for geophysical models have resulted in “localized” PFs, which approximate a given data assimilation application as a large set of loosely coupled problems that can be solved independently using relatively small ensembles – an approach long used for ensemble Kalman filters (EnKFs). While localization delivers a potentially transformative strategy for implementing PFs for high-dimensional systems, its use for real geophysical applications has been limited to small-scale, easily localized fluid motions such as moist convection in atmospheric models.

This seminar reveals findings from the first ever tests of PF for synoptic-scale weather forecasting. It also introduces a new “iterative resampling” method for improving filter stability for dynamical systems containing large spatial correlation length scales – commensurate with the Rossby radius of deformation in the current application. The experiments use a modified version of the Poterjoy (2016) local PF in the community GSI data assimilation package for its assessment in a cycling HWRF modeling framework. To measure potential benefits of the new system, medium-range HWRF ensemble forecasts generated from local PF members are scrutinized alongside forecasts generated using EnKF members. These forecasts occur over a four-week period that features the formation and evolution of several major hurricanes from the 2017 season; i.e., Irma, Jose, Katia, Lee, and Maria. The experiment poses a challenging geophysical data assimilation problem, owing to strong nonlinearity in the system dynamics and the extensive use of indirect remotely-sensed measurements from satellites. This research identifies several advantages of the local PF for an application known to pose challenges for Gaussian filters and smoothers, and describes broader implications of PFs for environmental prediction.

Poterjoy, J., 2016: A localized particle filter for high-dimensional nonlinear systems. Mon. Wea. Rev., 144, 59 – 76.

A copy of the presentation is available on the anonymous ftp site: