Compressing multidimensional weather and climate data into neural networks

Langwen Huang, Torsten Hoefler

arXiv preprint · 22 October 2022 · ETH Zurich

In one sentence. Coordinate-based neural networks compress weather and climate grids by 300x to 3000x, beating SZ3 on error, and a 790x compressed WeatherBench loader raises RMSE by under 2%.

Abstract

Weather and climate simulations produce petabytes of high-resolution data that are later analyzed by researchers in order to understand climate change or severe weather. We propose a new method of compressing this multidimensional weather and climate data: a coordinate-based neural network is trained to overfit the data, and the resulting parameters are taken as a compact representation of the original grid-based data. While compression ratios range from 300x to more than 3,000x, our method outperforms the state-of-the-art compressor SZ3 in terms of weighted RMSE, MAE. It can faithfully preserve important large scale atmosphere structures and does not introduce artifacts. When using the resulting neural network as a 790x compressed dataloader to train the WeatherBench forecasting model, its RMSE increases by less than 2%. The three orders of magnitude compression democratizes access to high-resolution climate data and enables numerous new research directions.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.