Summary: We developed invertible neural networks to accelerate model calibration and simulation of computationally efficient earth system models.

Accomplishment: We proposed an invertible neural network (INN) to effectively and efficiently solve model calibration and simulation problems. INN provides bijective mappings between model inputs and outputs; thus, it solves pro babilistic inverse problems and forward approximations simultaneously. INN consists of a sequence of reversible blocks and is developed from a normalizing flow strategy. After training INN using paired samples of model inputs and outputs, for a given observation it produces parameter posterior distributions by inverse evaluation, and for a given parameter sample it generates corresponding model outputs by forward evaluation. The proposed INN has potential to fundamentally change how model calibration and simulation are made in traditional earth system modeling. INN is computationally efficient which solves both inverse and forward problems in seconds after training, and thus can be used for accurate parameter estimation and prediction in scenarios where rapid evaluations are required.

Reference: https://ai4earthscience.github.io/iclr-2022-workshop/accepted