An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model

Accurate initial conditions are crucial for improving numerical weather prediction (NWP). Variational data assimilation relies on a static background error covariance matrix (<b>B</b>), yet its variance estimation is often inaccurate, affecting assimilation and forecast performance. This...

Full description

Saved in:
Bibliographic Details
Main Authors: Lilan Huang, Hongze Leng, Junqiang Song, Dongzi Wang, Wuxin Wang, Ruisheng Hu, Hang Cao
Format: Article
Language:English
Published: MDPI AG 2025-06-01
Series:Applied Sciences
Subjects:
Online Access:https://www.mdpi.com/2076-3417/15/12/6399
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849433823626919936
author Lilan Huang
Hongze Leng
Junqiang Song
Dongzi Wang
Wuxin Wang
Ruisheng Hu
Hang Cao
author_facet Lilan Huang
Hongze Leng
Junqiang Song
Dongzi Wang
Wuxin Wang
Ruisheng Hu
Hang Cao
author_sort Lilan Huang
collection DOAJ
description Accurate initial conditions are crucial for improving numerical weather prediction (NWP). Variational data assimilation relies on a static background error covariance matrix (<b>B</b>), yet its variance estimation is often inaccurate, affecting assimilation and forecast performance. This study introduces DRL-AST, a deep reinforcement learning-based adaptive variance rescaling strategy that dynamically adjusts the variances of <b>B</b> to optimize forecast skill through improved assimilation performance. By formulating variance rescaling as a Markov Decision Process and employing an actor–critic framework with Proximal Policy Optimization, DRL-AST autonomously selects spatio-temporal rescaling factors, enhancing assimilation and forecast accuracy without additional computational cost. As a new paradigm for adaptive variance tuning, DRL-AST demonstrates competitive improvements in forecast skill in experiments with the Lorenz-96 model by generating initial states that better conform to model dynamical consistency. Given its adaptability and efficiency, DRL-AST holds great potential for application in high-dimensional NWP models, where deep learning-based dimensionality reduction and reinforcement learning techniques could further enhance its feasibility and effectiveness in complex assimilation frameworks.
format Article
id doaj-art-43877a3d4a2f426d8b621e71e085c480
institution Kabale University
issn 2076-3417
language English
publishDate 2025-06-01
publisher MDPI AG
record_format Article
series Applied Sciences
spelling doaj-art-43877a3d4a2f426d8b621e71e085c4802025-08-20T03:26:53ZengMDPI AGApplied Sciences2076-34172025-06-011512639910.3390/app15126399An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 ModelLilan Huang0Hongze Leng1Junqiang Song2Dongzi Wang3Wuxin Wang4Ruisheng Hu5Hang Cao6College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, ChinaCollege of Meteorology and Oceanography, National University of Defense Technology, Changsha 410073, ChinaCollege of Meteorology and Oceanography, National University of Defense Technology, Changsha 410073, ChinaCollege of Computer Science and Technology, National University of Defense Technology, Changsha 410073, ChinaCollege of Computer Science and Technology, National University of Defense Technology, Changsha 410073, ChinaCollege of Meteorology and Oceanography, National University of Defense Technology, Changsha 410073, ChinaCollege of Meteorology and Oceanography, National University of Defense Technology, Changsha 410073, ChinaAccurate initial conditions are crucial for improving numerical weather prediction (NWP). Variational data assimilation relies on a static background error covariance matrix (<b>B</b>), yet its variance estimation is often inaccurate, affecting assimilation and forecast performance. This study introduces DRL-AST, a deep reinforcement learning-based adaptive variance rescaling strategy that dynamically adjusts the variances of <b>B</b> to optimize forecast skill through improved assimilation performance. By formulating variance rescaling as a Markov Decision Process and employing an actor–critic framework with Proximal Policy Optimization, DRL-AST autonomously selects spatio-temporal rescaling factors, enhancing assimilation and forecast accuracy without additional computational cost. As a new paradigm for adaptive variance tuning, DRL-AST demonstrates competitive improvements in forecast skill in experiments with the Lorenz-96 model by generating initial states that better conform to model dynamical consistency. Given its adaptability and efficiency, DRL-AST holds great potential for application in high-dimensional NWP models, where deep learning-based dimensionality reduction and reinforcement learning techniques could further enhance its feasibility and effectiveness in complex assimilation frameworks.https://www.mdpi.com/2076-3417/15/12/6399variational data assimilationspatio-temporal optimizationdeep reinforcement learningadaptive variance rescalingnumerical weather prediction
spellingShingle Lilan Huang
Hongze Leng
Junqiang Song
Dongzi Wang
Wuxin Wang
Ruisheng Hu
Hang Cao
An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
Applied Sciences
variational data assimilation
spatio-temporal optimization
deep reinforcement learning
adaptive variance rescaling
numerical weather prediction
title An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
title_full An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
title_fullStr An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
title_full_unstemmed An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
title_short An Adaptive Variance Adjustment Strategy for a Static Background Error Covariance Matrix—Part I: Verification in the Lorenz-96 Model
title_sort adaptive variance adjustment strategy for a static background error covariance matrix part i verification in the lorenz 96 model
topic variational data assimilation
spatio-temporal optimization
deep reinforcement learning
adaptive variance rescaling
numerical weather prediction
url https://www.mdpi.com/2076-3417/15/12/6399
work_keys_str_mv AT lilanhuang anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT hongzeleng anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT junqiangsong anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT dongziwang anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT wuxinwang anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT ruishenghu anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT hangcao anadaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT lilanhuang adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT hongzeleng adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT junqiangsong adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT dongziwang adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT wuxinwang adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT ruishenghu adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model
AT hangcao adaptivevarianceadjustmentstrategyforastaticbackgrounderrorcovariancematrixpartiverificationinthelorenz96model