dataset · Global

TerraMesh

Hugging Face dataset ibm-esa-geospatial/TerraMesh

Claim this listing

Represent TerraMesh? Send a claim for review so we can discuss listing updates. Claims do not automatically grant ownership or verification.

Data offered docs-verified: multimodal Earth observation pre-training samples · sample_id and split (train/validation) · centre lon/lat, crs, bounds, geometry · S2_time and S1_time timestamps · cloud_cover
Coverage llm-inferred: Global (inferred from sample centre coordinates across several continents) · 2018–2023 (inferred from sample timestamps)
Formats docs-verified: parquet
Access docs-verified: Licence CC-BY-SA-4.0; no authentication listed in ENTRY (auth: null)

Use cases llm-inferred: Build a paired Sentinel-1/Sentinel-2 patch training set for a land-cover or habitat classifier prototype · Pre-train or fine-tune a multimodal EO foundation model with cloud-cover filtering · Benchmark cloud-cover and acquisition-date sampling strategies before applying a model to a site-specific habitat baseline
Audience llm-inferred: remote sensing and machine learning engineers · GIS analysts · ecologists building habitat-mapping prototypes

Constraints llm-inferred: Pixel resolution and update frequency are not stated in the docs · Preview shows only sample rows; the overall geographic and temporal coverage is not documented · Dataset card has no written description of the modalities or sensors · Pre-training dataset rather than an analysis-ready product; requires EO data-handling skills · CC-BY-SA-4.0 share-alike terms may restrict derived datasets or models · Validation split is 89.1k rows against 8.93M train rows; no test split is listed

Verified 2026-10-15 via HF API: exists; 19678 downloads/30d, 35 likes.

Metadata observed 14/10/2026 · research sweep: manual-2026-10-15 (Hugging Face datasets)