Generative Modeling for Atmospheric Convection
This work addresses the computational bottleneck in climate modeling by providing a cheaper alternative to simulate small-scale atmospheric convection, though it represents an incremental application of existing generative methods to a new domain.
The researchers tackled the problem of computationally expensive small-scale storm simulation in climate models by developing a Variational Autoencoder (VAE) trained on ~6 million global samples. The VAE successfully reconstructed convective spatial structures, performed unsupervised clustering of organization regimes, and identified anomalous storm activity, demonstrating its potential for stochastic parameterizations in climate models.
While cloud-resolving models can explicitly simulate the details of small-scale storm formation and morphology, these details are often ignored by climate models for lack of computational resources. Here, we explore the potential of generative modeling to cheaply recreate small-scale storms by designing and implementing a Variational Autoencoder (VAE) that performs structural replication, dimensionality reduction, and clustering of high-resolution vertical velocity fields. Trained on ~6*10^6 samples spanning the globe, the VAE successfully reconstructs the spatial structure of convection, performs unsupervised clustering of convective organization regimes, and identifies anomalous storm activity, confirming the potential of generative modeling to power stochastic parameterizations of convection in climate models.