README.md

metadata

task_categories:
  - image-segmentation
license: mit
tags:
  - climate
  - satellite imagery

ClimateBench-M: A Multi-modal Climate Data Benchmark

📖 Paper

ClimateBench-M is the first multi-modal climate benchmark designed to support the development of artificial general intelligence (AGI) in climate applications. It aligns data across three critical modalities at a unified spatio-temporal resolution:

1. Time-series climate variables from ERA5.
2. Extreme weather event records from NOAA.
3. Satellite imagery from NASA HLS.

This dataset is used in the paper: https://huggingface.co/papers/2504.07394.

To facilitate broad accessibility and reproducibility, ClimateBench-M is hosted on 🤗 Hugging Face Datasets. Due to its size and multi-modal nature, the dataset is split into two parts:

• ClimateBench-M-TS: Climate time series with extreme events aligned and labeled.
• ClimateBench-M-IMG: Satellite imagery data