AI

EarthRecord: From Pixels to Land Change with Multimodal Agents over Four Decades of Satellite Imagery

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 11/11/2026, 05:03:23 UTC.

Briefings

EarthRecord is a new benchmark for evaluating multimodal agents in land-change analysis using four decades of satellite data. It features 56 sites with 238 questions across ten tasks, testing agents on raw multi-sensor archives. Twelve agents were evaluated, showing they complete analyses end-to-end but fail when mapping from imagery, as half of their answers follow satellite alterations rather than ground truth.

Key points

• EarthRecord uses 56 sites with four decades of Landsat, MODIS, VIIRS, and Sentinel-2 data.
• Agents fail to map from imagery, as half of their answers follow satellite alterations rather than ground truth.

Context

This benchmark addresses the gap where existing benchmarks use curated images or preprocessed inputs, leaving unmeasured the ability to analyze raw multi-sensor archives for reliable land-change analysis.

Limitations

Evaluation results are limited to twelve multimodal agents; the record moves answers from prior knowledge to pixels but not yet to real ground change. Some answers may follow satellite alterations rather than ground truth.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 11/11/2026, 05:04:08 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibc2jqoa6d3rza24vtxetn6tj473rk24zdn7aml4vsiug7tcmgaay