Real-Time, Crowdsourcing-Enhanced Forecasting of Building Functionality During Urban Floods

TL;DR

This paper introduces CRAF, a physics-informed, crowdsourcing-enhanced framework for real-time urban flood impact forecasting that improves accuracy and reliability by integrating sparse observations into dynamic impact predictions.

Contribution

The study presents a novel closed-loop, impact-based forecasting system that combines physics simulation with crowdsourced data, enabling real-time, multi-step predictions without online retraining.

Findings

CRAF reduces forecast errors by up to 95% during typhoon events.

The system operates with low latency, requiring only 10 minutes per update cycle.

Offline evaluation shows stable generalization across diverse storm scenarios.

Abstract

Urban flood emergency response increasingly relies on infrastructure impact forecasts rather than hazard variables alone. However, real-time predictions are unreliable due to biased rainfall, incomplete flood knowledge, and sparse observations. Conventional open-loop forecasting propagates impacts without adjusting the system state, causing errors during critical decisions. This study presents CRAF (Crowdsourcing-Enhanced Real-Time Awareness and Forecasting), a physics-informed, closed-loop framework that converts sparse human-sensed evidence into rolling, decision-grade impact forecasts. By coupling physics-based simulation learning with crowdsourced observations, CRAF infers system conditions from incomplete data and propagates them forward to produce multi-step, real-time predictions of zone-level building functionality loss without online retraining. This closed-loop design supports continuous state correction and forward prediction under weakly structured data with low-latency operation. Offline evaluation demonstrates stable generalization across diverse storm scenarios. In operational deployment during Typhoon Haikui (2023) in Fuzhou, China, CRAF reduces 1-3 hour-ahead forecast errors by 84-95% relative to fixed rainfall-driven forecasting and by 73-80% relative to updated rainfall-driven forecasting, while limiting computation to 10 minutes per update cycle. These results show that impact-state alignment-rather than hazard refinement alone-is essential for reliable real-time decision support, providing a pathway toward operational digital twins for resilient urban infrastructure systems.