ZeroFlood: A Geospatial Foundation Model for Data-Efficient Flood Susceptibility Mapping

2025-10-27

AI Summary

Flood susceptibility mapping (FSM) in data-scarce regions is hindered by the reliance of hydrodynamic models on dense, high-resolution geophysical inputs—particularly accurate topography and hydrological parameters—which are often unavailable. Method: We propose Thinking-in-Modality, a cross-modal reasoning framework that jointly leverages Sentinel-1/2 remote sensing imagery and synthetic flood maps to enable few-shot, cross-regional transfer learning on Geospatial Foundation Models (e.g., TerraMind, Prithvi). Contribution/Results: This work establishes the first foundation-model-driven FSM framework operational without high-fidelity terrain or hydrological data, effectively bridging heterogeneity across multi-source geospatial datasets. Experiments demonstrate that TerraMind-Large achieves an F1-score of 67.21 on multi-region generalization tasks—substantially outperforming conventional methods—and confirms strong robustness, scalability, and practical deployability in real-world low-data settings.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Flood susceptibility mapping (FSM) is vital for disaster prevention but remains challenging in data-scarce regions where hydrodynamic models require dense geophysical inputs. This work introduces ZeroFlood, a geospatial foundation model framework for data-efficient FSM. The approach fine-tunes Geospatial Foundation Models (GFMs) with Thinking-in-Modality (TiM) reasoning, enabling flood prediction from basic Earth observation data such as Sentinel-1 or Sentinel-2 imagery. Using paired EO and simulated flood maps from data-rich regions, ZeroFlood bridges data availability gaps through cross-modal representation learning. Experiments with TerraMind and Prithvi GFMs show that TiM enhances model robustness, with the TerraMind-Large configuration achieving an F1 score of 67.21. The results demonstrate the feasibility of foundation-model-based FSM as a scalable and data-efficient solution for flood risk management.

Problem

• Addresses flood prediction in data-scarce regions
• Uses geospatial foundation models with basic satellite imagery
• Enables data-efficient flood susceptibility mapping through cross-modal learning

Innovation

• Fine-tunes Geospatial Foundation Models with TiM reasoning
• Uses cross-modal learning from Earth observation data
• Achieves scalable flood prediction with minimal data