betaFeedback

WorksMathematics

Laplace Bayesian transport maps for generative modeling of non-Gaussian climate fields

Abstract

Transport maps provide a powerful framework for the generative modeling of complex, non-Gaussian spatial fields, such as those encountered in climate science. By mapping a complicated target distribution to a standard Gaussian reference distribution, they can serve as efficient stochastic surrogates. However, current Bayesian transport maps, which are modeled using autoregressive Gaussian processes, typically rely on empirical Bayes estimates that ignore hyperparameter uncertainty. Here, we contrast this point-estimate approach with a fully Bayesian model that provides posterior uncertainty quantification in transport map hyperparameters. To avoid the high cost of Markov chain Monte Carlo (MCMC), we also propose a Laplace approximation to the hyperparameter posteriors, which can be computed at negligible additional computational cost relative to empirical Bayes. We provide numerical evidence and a real-data climate application demonstrating that these Laplace approximations can accurately capture most posterior summaries at a fraction of the cost of MCMC, while maintaining good predictive properties. Thus, our approach offers a scalable, uncertainty-aware framework for the generative modeling of non-Gaussian climate fields.

Paper details

vlnc:W_98d6kvty
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…