Evaluation of Volume Representation Networks for Meteorological Ensemble Compression

Abstract

Recent studies have shown that volume scene representation networks constitute powerful means to transform 3D scalar fields into extremely compact representations, from which the initial field samples can be randomly accessed. In this work, we evaluate the capabilities of such networks to compress meteorological ensemble data, which are comprised of many separate weather forecast simulations. We analyze whether these networks can effectively exploit similarities between the ensemble members, and how alternative classical compression approaches perform in comparison. Since meteorological ensembles contain different physical parameters with various statistical characteristics and variations on multiple scales of magnitude, we analyze the impact of data normalization schemes on learning quality. Along with an evaluation of the trade-offs between reconstruction quality and network model parameterization, we compare compression ratios and reconstruction quality for different model architectures and alternative compression schemes.