Global Flood Prediction: a Multimodal Machine Learning Approach

TL;DR

This paper introduces a multimodal machine learning framework that combines geographical text data and historical disaster data to predict global flood risks over multiple years, demonstrating improved accuracy over single-modality models.

Contribution

The work presents a novel multimodal approach utilizing transfer learning and embeddings from text and time-series data for long-term flood prediction.

Findings

Achieves 75-77% ROCAUC in flood prediction

Multimodal approach outperforms single-modality models

Demonstrates potential for long-term disaster planning

Abstract

Flooding is one of the most destructive and costly natural disasters, and climate changes would further increase risks globally. This work presents a novel multimodal machine learning approach for multi-year global flood risk prediction, combining geographical information and historical natural disaster dataset. Our multimodal framework employs state-of-the-art processing techniques to extract embeddings from each data modality, including text-based geographical data and tabular-based time-series data. Experiments demonstrate that a multimodal approach, that is combining text and statistical data, outperforms a single-modality approach. Our most advanced architecture, employing embeddings extracted using transfer learning upon DistilBert model, achieves 75\%-77\% ROCAUC score in predicting the next 1-5 year flooding event in historically flooded locations. This work demonstrates the potentials of using machine learning for long-term planning in natural disaster management.