EarthScape: A Multimodal Dataset for Surficial Geologic Mapping and Earth Surface Analysis

ICLR 2026 Conference Submission13767 Authors

Keywords: Datasets and Benchmarks, Multimodal Learning, Multimodal Fusion, Remote Sensing, Geospatial Analysis, Earth Science, Surficial Geology

Abstract: Surficial geologic (SG) maps are critical for understanding Earth surface processes, supporting infrastructure planning, and addressing challenges related to climate change and natural hazards. Current workflows are labor-intensive, subjective, and difficult to scale. We introduce EarthScape, an AI-ready multimodal dataset for advancing SG mapping and surface-aware geospatial learning. EarthScape integrates digital elevation models, aerial imagery, multi-scale terrain derivatives, and vector data for hydrologic and infrastructure features. We provide an end-to-end processing pipeline for reproducibility and report baseline benchmarks across single-modality, multi-scale, and multimodal configurations. Results show that terrain-derived features are highly predictive and that generalization across geologically diverse regions remains a key open challenge, positioning EarthScape as a benchmark for multimodal fusion and domain adaptation.

Primary Area: datasets and benchmarks

Submission Number: 13767