Physics-guided machine learning for evaporation risk assessment during gasoline spill accidents

Gasoline spills pose significant environmental and public safety hazards due to the rapid evaporation of volatile organic compounds (VOCs), which contribute to atmospheric pollution and substantially increase explosion risk. Despite their importance, predictive models that quantify gasoline evaporation under diverse environmental conditions remain limited. This study proposed a novel Physics-Guided Gasoline Evaporation Risk Prediction (PG-GERP) framework that integrates physical modeling with machine learning to assess evaporation dynamics across different land-use types. The physical model incorporated key environmental variables, including wind velocity, ambient temperature, pool geometry, and turbulence intensity, accounting for heterogeneity in rural, suburban, and urban settings. Simulation results revealed critical patterns: while higher wind speeds and ambient temperatures enhanced evaporation, larger pool dimensions may suppress it due to vapor saturation. Notably, evaporation rates were approximately 10 % and 60 % higher respectively in suburban and urban environments, relative to rural areas, primarily driven by enhanced turbulence and vapor dispersion. A stacking machine learning model, trained on 3405 physically simulated scenarios, achieved exceptional predictive performance with coefficients of determination (R²) exceeding 0.99 across all environments. The PG-GERP framework provides a robust tool for assessing evaporation-driven risks of hazardous liquid fuels, enabling timely response and mitigation in real-world spill scenarios. SYNOPSIS: The study integrated numerical simulations and machine learning to predict gasoline evaporation under various environmental conditions, providing a robust framework for risk assessment and safety management in fuel storage and transportation.