Rigorous accuracy and robustness analysis for two-scale reduced random Kalman filters in high dimensions

Andrew Majda, Xin T. Tong

Research output: Contribution to journalArticle

Abstract

Contemporary data assimilation often involves millions of prediction variables. The classical Kalman filter is no longer computationally feasible in such a high dimensional context. This problem can often be resolved by exploiting the underlying multiscale structure, applying the full Kalman filtering procedures only to the large scale variables, and estimating the small scale variables with proper statistical strategies, including multiplicative in ation, representation model error in the observations, and crude localization. The resulting two-scale reduced filters can have close to optimal numerical filtering skill based on previous numerical evidence. Yet, no rigorous explanation exists for this success, because these modifications create unavoidable bias and model error. This paper contributes to this issue by establishing a new error analysis framework for two different reduced random Kalman filters, valid independent of the large dimension. The first part of our results examines the fidelity of the covariance estimators, which is essential for accurate uncertainty quantification. In a simplified setting, this is demonstrated by showing the true error covariance is dominated by its estimators. In general settings, the Mahalanobis error and its intrinsic dissipation can be used as simplified quantification of the same property. The second part develops upper bounds for the covariance estimators by comparing with proper Kalman filters. Combining both results, the classical tools for Kalman filters can be used as a-priori performance criteria for the reduced filters. In applications, these criteria guarantee the reduced filters are robust, and accurate for small noise systems. They also shed light on how to tune the reduced filters for stochastic turbulence.

Original languageEnglish (US)
Pages (from-to)1095-1132
Number of pages38
JournalCommunications in Mathematical Sciences
Volume16
Issue number4
DOIs
StatePublished - Jan 1 2018

Fingerprint

Robustness Analysis
Robustness (control systems)
Kalman filters
Kalman Filter
Higher Dimensions
Filter
Model Error
Estimator
Uncertainty Quantification
Data Assimilation
Kalman Filtering
Error Analysis
Error analysis
Fidelity
Quantification
Dissipation
Turbulence
Multiplicative
High-dimensional
Filtering

Keywords

  • Filter accuracy
  • Filter robustness
  • Model reduction
  • Reduced Kalman filters

ASJC Scopus subject areas

  • Mathematics(all)
  • Applied Mathematics

Cite this

Rigorous accuracy and robustness analysis for two-scale reduced random Kalman filters in high dimensions. / Majda, Andrew; Tong, Xin T.

In: Communications in Mathematical Sciences, Vol. 16, No. 4, 01.01.2018, p. 1095-1132.

Research output: Contribution to journalArticle

@article{4b8594d798514899a7c1fc94ac14ac66,
title = "Rigorous accuracy and robustness analysis for two-scale reduced random Kalman filters in high dimensions",
abstract = "Contemporary data assimilation often involves millions of prediction variables. The classical Kalman filter is no longer computationally feasible in such a high dimensional context. This problem can often be resolved by exploiting the underlying multiscale structure, applying the full Kalman filtering procedures only to the large scale variables, and estimating the small scale variables with proper statistical strategies, including multiplicative in ation, representation model error in the observations, and crude localization. The resulting two-scale reduced filters can have close to optimal numerical filtering skill based on previous numerical evidence. Yet, no rigorous explanation exists for this success, because these modifications create unavoidable bias and model error. This paper contributes to this issue by establishing a new error analysis framework for two different reduced random Kalman filters, valid independent of the large dimension. The first part of our results examines the fidelity of the covariance estimators, which is essential for accurate uncertainty quantification. In a simplified setting, this is demonstrated by showing the true error covariance is dominated by its estimators. In general settings, the Mahalanobis error and its intrinsic dissipation can be used as simplified quantification of the same property. The second part develops upper bounds for the covariance estimators by comparing with proper Kalman filters. Combining both results, the classical tools for Kalman filters can be used as a-priori performance criteria for the reduced filters. In applications, these criteria guarantee the reduced filters are robust, and accurate for small noise systems. They also shed light on how to tune the reduced filters for stochastic turbulence.",
keywords = "Filter accuracy, Filter robustness, Model reduction, Reduced Kalman filters",
author = "Andrew Majda and Tong, {Xin T.}",
year = "2018",
month = "1",
day = "1",
doi = "10.4310/CMS.2018.V16.N4.A8",
language = "English (US)",
volume = "16",
pages = "1095--1132",
journal = "Communications in Mathematical Sciences",
issn = "1539-6746",
publisher = "International Press of Boston, Inc.",
number = "4",

}

TY - JOUR

T1 - Rigorous accuracy and robustness analysis for two-scale reduced random Kalman filters in high dimensions

AU - Majda, Andrew

AU - Tong, Xin T.

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Contemporary data assimilation often involves millions of prediction variables. The classical Kalman filter is no longer computationally feasible in such a high dimensional context. This problem can often be resolved by exploiting the underlying multiscale structure, applying the full Kalman filtering procedures only to the large scale variables, and estimating the small scale variables with proper statistical strategies, including multiplicative in ation, representation model error in the observations, and crude localization. The resulting two-scale reduced filters can have close to optimal numerical filtering skill based on previous numerical evidence. Yet, no rigorous explanation exists for this success, because these modifications create unavoidable bias and model error. This paper contributes to this issue by establishing a new error analysis framework for two different reduced random Kalman filters, valid independent of the large dimension. The first part of our results examines the fidelity of the covariance estimators, which is essential for accurate uncertainty quantification. In a simplified setting, this is demonstrated by showing the true error covariance is dominated by its estimators. In general settings, the Mahalanobis error and its intrinsic dissipation can be used as simplified quantification of the same property. The second part develops upper bounds for the covariance estimators by comparing with proper Kalman filters. Combining both results, the classical tools for Kalman filters can be used as a-priori performance criteria for the reduced filters. In applications, these criteria guarantee the reduced filters are robust, and accurate for small noise systems. They also shed light on how to tune the reduced filters for stochastic turbulence.

AB - Contemporary data assimilation often involves millions of prediction variables. The classical Kalman filter is no longer computationally feasible in such a high dimensional context. This problem can often be resolved by exploiting the underlying multiscale structure, applying the full Kalman filtering procedures only to the large scale variables, and estimating the small scale variables with proper statistical strategies, including multiplicative in ation, representation model error in the observations, and crude localization. The resulting two-scale reduced filters can have close to optimal numerical filtering skill based on previous numerical evidence. Yet, no rigorous explanation exists for this success, because these modifications create unavoidable bias and model error. This paper contributes to this issue by establishing a new error analysis framework for two different reduced random Kalman filters, valid independent of the large dimension. The first part of our results examines the fidelity of the covariance estimators, which is essential for accurate uncertainty quantification. In a simplified setting, this is demonstrated by showing the true error covariance is dominated by its estimators. In general settings, the Mahalanobis error and its intrinsic dissipation can be used as simplified quantification of the same property. The second part develops upper bounds for the covariance estimators by comparing with proper Kalman filters. Combining both results, the classical tools for Kalman filters can be used as a-priori performance criteria for the reduced filters. In applications, these criteria guarantee the reduced filters are robust, and accurate for small noise systems. They also shed light on how to tune the reduced filters for stochastic turbulence.

KW - Filter accuracy

KW - Filter robustness

KW - Model reduction

KW - Reduced Kalman filters

UR - http://www.scopus.com/inward/record.url?scp=85062148461&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85062148461&partnerID=8YFLogxK

U2 - 10.4310/CMS.2018.V16.N4.A8

DO - 10.4310/CMS.2018.V16.N4.A8

M3 - Article

AN - SCOPUS:85062148461

VL - 16

SP - 1095

EP - 1132

JO - Communications in Mathematical Sciences

JF - Communications in Mathematical Sciences

SN - 1539-6746

IS - 4

ER -