MillionST: Satellite Image Time Series for Spatiotemporal Foundation Models

A large-scale satellite image time series dataset curated for pre-training spatiotemporal foundation models for Earth observation. It contains approximately 1 million satellite images from 100,000 geographic locations, with each location observed across 10 temporal phases over five years. The dataset was introduced in the TiMo paper and is hosted by Hillui on Hugging Face.

Use Cases

• Pre-training spatiotemporal foundation models based on the described satellite image time series.
• Analyzing long-term land-surface dynamics based on the five-year observation period.
• Studying seasonal variations in land cover based on the 10 temporal phases per location.
• Modeling diverse geospatial changes across 100,000 geographic locations.

Strengths

• Large scale with approximately 1 million satellite images.
• Spatiotemporal coverage across 100,000 geographic locations.
• Temporal depth with 10 observation phases per location over five years.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hillui on Hugging Face, introduced in the TiMo paper.

Collection Method

Curated for pre-training spatiotemporal foundation models.

Time Range

Observations span five years.

Freshness

Last updated 2026-07-03 01:35:30; freshness should be verified.

Geography

Covers 100,000 geographic locations.