Continually Learning Neural-Operator Surrogate for Three-Dimensional Airborne Electromagnetic Bayesian Inversion

This paper addresses the computational bottleneck in three-dimensional probabilistic inversion of time-domain airborne electromagnetic (AEM) data. While a single forward simulation takes only tens of seconds, a Bayesian inversion of a survey containing millions of soundings requires approximately 10^10 forward evaluations. The authors develop a continually learning neural-operator surrogate of the 3D AEM forward operator that replaces the traditional solver within the Bayesian inversion framework, enabling practical probabilistic inference for large-scale geophysical surveys.