Flood-LDM: Generalizable Latent Diffusion Models for rapid and accurate zero-shot High-Resolution Flood Mapping

2025-11-17

AI Summary

Traditional physics-based flood modeling is computationally expensive and unsuitable for real-time, high-resolution mapping, while existing CNN-based approaches suffer from poor generalizability and fail to enable zero-shot transfer to unseen regions. To address these limitations, this paper proposes a zero-shot flood super-resolution framework based on Latent Diffusion Models (LDMs). The method takes coarse-resolution flood maps and physics-informed constraints—such as topography and hydrological parameters—as inputs, and employs a physics-guided feature encoder coupled with a coarse-to-fine mapping network to reconstruct high-fidelity outputs efficiently in latent space. Crucially, it requires no training data from the target region and achieves cross-regional adaptation, generating ≥10 m resolution flood maps with high fidelity. Experiments demonstrate that our approach accelerates computation by two orders of magnitude over conventional physics models, significantly outperforms state-of-the-art methods in PSNR and SSIM, and maintains robust generalization on previously unseen watersheds—thereby enabling real-time flood risk assessment and emergency response.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• 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: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Flood prediction is critical for emergency planning and response to mitigate human and economic losses. Traditional physics-based hydrodynamic models generate high-resolution flood maps using numerical methods requiring fine-grid discretization; which are computationally intensive and impractical for real-time large-scale applications. While recent studies have applied convolutional neural networks for flood map super-resolution with good accuracy and speed, they suffer from limited generalizability to unseen areas. In this paper, we propose a novel approach that leverages latent diffusion models to perform super-resolution on coarse-grid flood maps, with the objective of achieving the accuracy of fine-grid flood maps while significantly reducing inference time. Experimental results demonstrate that latent diffusion models substantially decrease the computational time required to produce high-fidelity flood maps without compromising on accuracy, enabling their use in real-time flood risk management. Moreover, diffusion models exhibit superior generalizability across different physical locations, with transfer learning further accelerating adaptation to new geographic regions. Our approach also incorporates physics-informed inputs, addressing the common limitation of black-box behavior in machine learning, thereby enhancing interpretability. Code is available at https://github.com/neosunhan/flood-diff.

Problem

• Generating high-resolution flood maps rapidly for emergency planning
• Overcoming computational intensity of traditional physics-based hydrodynamic models
• Improving generalizability of flood mapping across unseen geographic regions

Innovation

• Latent diffusion models perform flood map super-resolution
• Physics-informed inputs enhance interpretability of predictions
• Transfer learning accelerates adaptation to new geographic regions