ChronoEarth-492K: A Large Scale and Long Horizon Spatiotemporal Hyperspectral Earth Observation Dataset and Benchmark

2026-05-15

AI Summary

This work addresses the limited temporal coverage of existing hyperspectral self-supervised learning datasets, which hinders long-term spatiotemporal modeling. To this end, we present ChronoEarth-492K, the first large-scale, temporally aligned hyperspectral Earth observation dataset, integrating 17 years of continuous imagery from NASA’s EO-1 Hyperion mission across 185,000 global locations. The dataset comprises over 490,000 radiometrically corrected image patches and 28,000 multi-temporal sequences. Through rigorous radiometric consistency processing, fusion with multi-source geospatial products, and a standardized evaluation protocol, we establish a unified benchmark spanning static, short-sequence, and long-sequence tasks. ChronoEarth-492K thus provides the first open and systematic foundation for hyperspectral spatiotemporal representation learning.

Abstract

Hyperspectral imaging (HSI) provides dense spectral information for the Earth's surface, enabling material-level understanding of land cover and ecosystem dynamics. Despite recent progress in hyperspectral self-supervised learning (SSL), existing datasets remain temporally shallow, limiting the development of long-horizon spatiotemporal modeling. To address this gap, we introduce ChronoEarth-492K, the first large-scale, temporally calibrated hyperspectral SSL dataset built upon NASA's EO-1 Hyperion mission, the world's longest continuous hyperspectral archive up to date (2001-2017). ChronoEarth-492K comprises 492,354 radiometrically harmonized patches across 185,398 global locations over 17 years, with 28,786 sites containing multi-temporal sequences ($\geq 3$ observations) that enable both short- and long-horizon temporal analysis. Building on this foundation, we establish the ChronoEarth-Benchmark, a unified evaluation suite spanning static, short-horizon, and long-horizon temporal tasks, constructed from six open-source geospatial products covering land cover, crop type, forest dynamics, and soil properties. We further introduce a standardized evaluation protocol and report extensive baseline results across state-of-the-art hyperspectral foundation models. Together, ChronoEarth and benchmark provide the first large-scale, temporally grounded platform for systematic spatiotemporal hyperspectral representation learning.

Problem

• hyperspectral imaging
• spatiotemporal modeling
• self-supervised learning
• long-horizon temporal analysis
• Earth observation

Innovation

• hyperspectral imaging
• self-supervised learning
• spatiotemporal modeling
• long-horizon dataset
• Earth observation benchmark