Waterlogformer: A Multimodal Model for Waterlogging Prediction

[WSDM 2026] This paper has been accepted by the 19th ACM International Conference on Web Search and Data Mining (WSDM 2026).

Introduction

Waterlogformer is a multimodal model for WD prediction.

To model hydrological mechanisms and effectively fuse multimodal data, a dual-branch multimodal architecture is developed for WD prediction, comprising three key components:

• Rainfall Branch — employs a Terrain-aware Rainfall Accumulation Unit which simulates rainfall accumulation over time and across locations under specific terrain conditions, embedding hydrodynamic knowledge of how rain propagates over landscapes.
• Waterlogging Branch — leverages historical WD time series together with static geographical information to capture spatio-temporal waterlogging patterns while respecting geographic constraints.
• Multimodal Fusion Prediction Module — integrates rainfall and historical WD representations and incorporates a distance- and terrain-similarity–based contrastive learning mechanism to enhance sensitivity to critical geographical factors during multimodal fusion.

Experiment results on a real-world dataset demonstrate the superior performance of Waterlogformer.

Prerequisites

Before proceeding, ensure Python 3.9 is installed. Install the required dependencies with the following command:

pip install -r requirements.txt

Acknowledgements

Our gratitude extends to the authors of the following repositories for their foundational model implementations:

• Time-Series-Library
• Autoformer
• PatchTST
• Crossformer