Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Sun Han Neo, Sachith Seneviratne, Herath Mudiyanselage Viraj Vidura Herath, Abhishek Saha, Sanka Rasnayaka, Lucy Amanda Marshall

TL;DR

The paper tackles real-time, high-resolution flood mapping by upsampling coarse-grid flood maps with latent diffusion models conditioned on physics-informed inputs (coarse maps and DEM). The proposed LDM framework achieves fine-grid accuracy with substantially faster inference than traditional physics-based methods and CNN-based super-resolution, while demonstrating strong zero-shot generalization across unseen catchments and benefiting from transfer learning for rapid regional adaptation. Key findings include improved generalizability, notable inference-time reductions through timesteps and initialization optimizations, and improved flood-inundation metrics (POD, RFA, CSI) when compared to coarse-grid baselines, all grounded by the DEM and physical constraints for interpretability. The work signals a practical path to deploy diffusion-based, interpretable flood-mapping surrogates in real-time risk management and disaster response settings.

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.

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

TL;DR

The paper tackles real-time, high-resolution flood mapping by upsampling coarse-grid flood maps with latent diffusion models conditioned on physics-informed inputs (coarse maps and DEM). The proposed LDM framework achieves fine-grid accuracy with substantially faster inference than traditional physics-based methods and CNN-based super-resolution, while demonstrating strong zero-shot generalization across unseen catchments and benefiting from transfer learning for rapid regional adaptation. Key findings include improved generalizability, notable inference-time reductions through timesteps and initialization optimizations, and improved flood-inundation metrics (POD, RFA, CSI) when compared to coarse-grid baselines, all grounded by the DEM and physical constraints for interpretability. The work signals a practical path to deploy diffusion-based, interpretable flood-mapping surrogates in real-time risk management and disaster response settings.

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.

Paper Structure

This paper contains 15 sections, 7 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Flood Mapping Techniques
  4. Diffusion Models
  5. Methodology
  6. Data
  7. Model Architecture
  8. Experimental Results
  9. Model Comparison
  10. Inference Time Reduction
  11. Transfer Learning
  12. Flood Inundation Analysis
  13. DEM Ablation Study
  14. Conclusion
  15. Acknowledgements

Figures (3)

  • Figure 1: Overview of our proposed approach.
  • Figure 2: Comparison of fine- vs coarse-grid flood maps bomers_19. Coarse grids use fewer, larger cells, enabling faster computation but reducing spatial detail, whereas fine grids use more, smaller cells, providing greater accuracy at higher computational cost.
  • Figure 3: Comparison of coarse-grid (CG), fine-grid (FG), and super-resolution (SR) flood maps for a segment of Catchment 1. The purple circle marks a flooded area missed by the CG but successfully recovered by the SR model. The right panels show depth differences (FG – CG and FG – SR), where negative values in FG – CG indicate CG overestimation. The SR map reduces these errors, closely matching the FG at lower cost, while preserving flood contours and providing sharp predictions without distorting inundation boundaries.