Copula-based synthetic data augmentation for machine-learning emulators

TL;DR

This paper investigates the use of copula-based models to generate synthetic data for augmenting training datasets in weather and climate emulators, demonstrating significant improvements in prediction accuracy.

Contribution

It introduces a copula-based data augmentation method specifically for enhancing machine-learning emulators in weather and climate modeling.

Findings

Prediction error reduced by up to 62%.

Copula-augmented datasets outperform original data.

Method effective on toy physical models.

Abstract

Can we improve machine-learning (ML) emulators with synthetic data? If data are scarce or expensive to source and a physical model is available, statistically generated data may be useful for augmenting training sets cheaply. Here we explore the use of copula-based models for generating synthetically augmented datasets in weather and climate by testing the method on a toy physical model of downwelling longwave radiation and corresponding neural network emulator. Results show that for copula-augmented datasets, predictions are improved by up to 62 % for the mean absolute error (from 1.17 to 0.44 W m$^{-2}$).