Skip to content
Research · Oct 31, 2026

Localized posterior calibration reduces predictive errors in ocean turbulence model

Researchers propose amortized localized posterior inference to address model discrepancy and observation uncertainty in calibration, demonstrating lower errors than global methods in twin experiments and ocean model applications.

Trust79
HypeLow hype

1 source · cross-referenced

ShareXLinkedInEmail
TL;DR
  • A new method for model calibration addresses discrepancies across context domains by constructing localized posteriors weighted by proximity to target contexts.
  • Localization reduces aggregate predictive errors compared to global posterior methods in twin experiments with the Branin-Hoo model.
  • Application to an ocean turbulence closure shows localized calibration lowers temperature and zonal-velocity errors relative to deterministic and global-posterior approaches.
  • The approach uses Gaussian localization weights, covariance inflation for context uncertainty, and a conditional normalizing flow for amortized variational inference.

Researchers propose a method called amortized localized posterior inference to address model discrepancy, where a single calibration parameter may not reproduce a physical response uniformly across a context domain. The approach constructs a family of localized posteriors by weighting observation likelihoods according to their proximity to a target context, incorporating context uncertainty through Gaussian localization weights with covariance inflation.

The method includes reflection at domain boundaries and an optional correction for uneven context sampling. A conditional normalizing flow approximates the family of posteriors via amortized variational inference, and integrating the localized posteriors under a prescribed context measure yields the marginalized parameter distribution.

In twin experiments using the Branin-Hoo model, the authors compare global and localized inference under both uniform and skewed context distributions. The results show that localization reduces reported aggregate predictive errors relative to the global posterior, with improvements observed across both context distributions.

The researchers apply the method to calibrate the k-ε turbulence closure in a single-column ocean model, using atmospheric surface forcing as context. Localized posterior calibration reduces temperature and zonal-velocity errors compared to both deterministic and global-posterior calibration approaches. Both twin experiments are designed to admit local compensation, ensuring the validity of the localized inference framework.

The work is authored by researchers from AIRSEA and Mines Saint-Étienne, and the paper is published as arXiv:2610.37731 in the cs.AI category.

Sources
  1. 01arXiv cs.AI — Amortized Localized Posterior Inference for Model Calibration with Application to an Ocean Turbulence Closure
Also on Research

Stories may contain errors. Dispatch is assembled with AI assistance and curated by human editors; despite the trust-score filter, mistakes happen. We correct publicly — every article links to its revision history. Nothing here is financial, legal, or medical advice. Verify before relying on any claim.

© 2026 Dispatch. No ads. No sponsorships. No paid placement. Reader-supported via Ko-fi.

Built by a person who cares about honest AI news.