Curated, machine-learning-ready dataset accompanying the paper "Ensembling geophysical models with Bayesian Neural Networks". For a detailed description of the dataset, please refer to the paper. If you find this resource useful, you're encouraged to cite our paper. The data is formed of the monthly total column ozone output from 15 CCMI refC1SD models, regridded to a common grid of 64x128. Experiment design for CCMI is described in this overview paper(www.geosci-model-dev.net/10/639/2017/). The model data was dowloaded from the Centrer of Environmental Data Analysis (CEDA) and from the Earth System Grid for the CESM models (information here http://blogs.reading.ac.uk/ccmi/badc-data-access/). Observational data was taken from v3.4 of the NIWA-BS total column ozone (https://doi.org/10.5281/zenodo.1346424).