Urban Fire Spatial–Temporal Prediction Based on Multi-Source Data Fusion

Abstract

Urban fire incidents pose significant risks to public safety and infrastructure, necessitating precise spatiotemporal prediction to enhance fire prevention and emergency response strategies. However, predicting fire occurrences remains a complex task due to the intricate interplay between spatial and temporal factors, including dynamic environmental conditions, historical dependencies, and inter-regional correlations. Temporal variables, such as past fire incidents and external influences like meteorological conditions, significantly impact fire risk, while spatial attributes, including regional characteristics and cross-regional interactions, further complicate predictive modeling. This study introduces UFSTP, an innovative framework for Urban Fire Spatial–Temporal Prediction that integrates multi-source data for enhanced predictive accuracy. UFSTP employs a neural region state representation to capture both intrinsic and extrinsic temporal dependencies, alongside a spatiotemporal propagation mechanism to model inter-regional correlations. Key environmental and historical features are extracted from real-world datasets to construct a comprehensive fire risk representation, facilitating the precise forecasting of fire occurrence in both time and space. Extensive evaluations on real-world datasets from Anci and Guangyang Districts demonstrate that UFSTP achieves a 16.2% average reduction in Mean Absolute Error (MAE) for time prediction and a 3.3% average improvement in top-1 hit rate for regional prediction over state-of-the-art baselines. The proposed framework offers a robust and interpretable approach to urban fire risk assessment, providing critical insights to optimize fire prevention measures and emergency resource allocation.

Keywords

• urban fire events
• multi-source data
• deep neural network
• spatial–temporal prediction