A localized particle filter for geophysical data assimilation

TL;DR

This paper presents Localized Particle Filters that partition high-dimensional geophysical systems into smaller regions, reducing computational costs and improving stability in data assimilation tasks like weather and ocean modeling.

Contribution

Introduction of Localized Particle Filters that exploit spatial localization to enhance efficiency and accuracy in high-dimensional geophysical data assimilation.

Findings

Achieved computational efficiency in large-scale systems.

Demonstrated improved stability and error performance.

Validated on rotating shallow water system.

Abstract

Particle filters are computational techniques for estimating the state of dynamical systems by integrating observational data with model predictions. This work introduces a class of Localized Particle Filters (LPFs) that exploit spatial localization to reduce computational costs and mitigate particle degeneracy in high-dimensional systems. By partitioning the state space into smaller regions and performing particle weight updates and resampling separately within each region, these filters leverage assumptions of limited spatial correlation to achieve substantial computational gains. This approach proves particularly valuable for geophysical data assimilation applications, including weather forecasting and ocean modeling, where system dimensions are vast, and complex interactions and nonlinearities demand efficient yet accurate state estimation methods. We demonstrate the methodology on a partially observed rotating shallow water system, achieving favourable performance in terms of algorithm stability and error estimates.