Landshift Sft V1: Satellite Image Pairs for Land Cover Change Detection

6388 records of temporal Sentinel-2 satellite image pairs over sampled locations, processed into tiles. The dataset, created by NuTonic and last updated on 2026-04-25, includes NDVI delta calculations to highlight change and emits captions and optional grounding for changed regions.

Use Cases

• Training change detection models based on temporal Sentinel-2 image pairs.
• Developing image captioning systems for satellite imagery based on emitted change captions.
• Segmenting connected change regions based on optional grounding data.
• Analyzing land cover dynamics based on NDVI delta calculations.

Strengths

• Contains 6388 total records, split into 5114 for training, 633 for validation, and 641 for testing.
• Processes 3798 satellite image tiles, providing a substantial base for analysis.
• Includes temporal pairs with a longer baseline, which may be suitable for detecting slower land cover changes.

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

Hugging Face, author NuTonic.

Collection Method

Temporal Sentinel-2 pairs processed over sampled locations.

Time Range

Temporal pairs suggest a time-series component, but specific dates are not provided.

Freshness

Last updated 2026-04-25 20:37:14; freshness should be verified.

Geography

Sampled locations, but specific geographic coverage is not detailed.