Abstract

Computational methods for Bayesian statistical inference in climate and environmental applications have advanced dramatically over the past 50 years, with the adoption of techniques such as Markov chain Monte Carlo or variational Bayes. The field is undergoing a new revolution as it embraces the representational capacity of neural networks, optimisation libraries, and graphics processing units to learn complex mappings between data and inferential targets. These new inferential tools are amortised, in the sense that they allow rapid inference through fast feedforward operations and have several compelling advantages over classical methods: they do not require knowledge of the likelihood function, they are relatively easy to implement, and they facilitate inference at a substantially reduced computational cost. In this lecture I present the decision-theoretic foundation of neural-inference methods. I then detail how they can be used for point estimation and for (approximate) Bayesian inference. The lecture showcases several environmental applications where we have used these methods for making rapid inference with spatial and spatio-temporal process models, and it concludes with a discussion of the benefits and perils of these new computational tools.