license: c-uda
pretty_name: MillionST
viewer: false
tags:
- image
- geospatial
- timeseries
- remote-sensing
- earth-observation
- satellite-image-time-series
- spatiotemporal
- self-supervised-learning
- masked-image-modeling
size_categories:
- 1M<n<10M
MillionST
MillionST is a large-scale satellite image time series dataset curated for pre-training spatiotemporal foundation models for Earth observation.
The dataset contains approximately 1 million satellite images from 100,000 geographic locations, with each location observed across 10 temporal phases over five years. It is designed to capture diverse geospatial changes, seasonal variations, and long-term land-surface dynamics.
MillionST was introduced in the paper:
TiMo: Spatiotemporal Foundation Model for Satellite Image Time Series
https://arxiv.org/abs/2505.08723
Code is available at: https://github.com/MiliLab/TiMo
Dataset Details
- Dataset name: MillionST
- Dataset type: Satellite image time series
- Domain: Remote sensing / Earth observation
- Scale: Approximately 1 million images
- Geographic locations: 100,000 locations
- Temporal phases: 10 phases
- Temporal span: Five years
- Associated model: TiMo
- Primary purpose: Self-supervised pre-training for spatiotemporal representation learning
Data Access
This dataset is released for computational research use. Users should follow the license and any access terms shown on this page.
If you use MillionST in your research, please cite the associated paper.
Citation
@article{qin2025timo,
title={TiMo: Spatiotemporal Foundation Model for Satellite Image Time Series},
author={Qin, Xiaolei and Wang, Di and Zhang, Jing and Wang, Fengxiang and Su, Xin and Du, Bo and Zhang, Liangpei},
journal={arXiv preprint arXiv:2505.08723},
year={2025}
}