Elsevier

Computers & Geosciences

Volume 145, December 2020, 104599
Computers & Geosciences

Research paper
A statistical analysis of lossily compressed climate model data

https://doi.org/10.1016/j.cageo.2020.104599Get rights and content
Under a Creative Commons license
open access

Highlights

  • Evaluating lossy compression of climate model output requires detailed analyses.

  • Two leading compression algorithms show promising fidelity at low error tolerances.

  • However, detectable artifacts are produced at higher error tolerances.

  • Analyses of artifacts must be tailored to features of the variable(s) of interest.

  • We advocate for collaboration between user communities and development teams.

Abstract

The data storage burden resulting from large climate model simulations continues to grow. While lossy data compression methods can alleviate this burden, they introduce the possibility that key climate variables could be altered to the point of affecting scientific conclusions. Therefore, developing a detailed understanding of how compressed model output differs from the original is important. Here, we evaluate the effects of two leading compression algorithms, sz and zfp, on daily surface temperature and precipitation rate data from a widely used climate model. While both algorithms show promising fidelity with the original output, detectable artifacts are introduced even at relatively tight error tolerances. This study highlights the need for evaluation methods that are sensitive to errors at different spatiotemporal scales and specific to the particular climate variable of interest.

Keywords

CESM
Climate variability
Earth system models
Lossy compression
sz
zfp

Data Availability

The original (uncompressed) CESM-LENS data used is available for download from the Climate Data Gateway at NCAR (formally known as the Earth System Grid) (Deser and Kay, 2020). In addition, we have made the reconstructed versions of the data that we analyzed with sz and zfp at various tolerances available (Baker and Hammerling, 2020).

Cited by (0)