Multi-source Composite Model for High-precision Prediction of Community VOC Emission

Volatile Organic Compounds (VOCs) significantly impact air quality and human health, yet their emissions are challenging to predict due to complex spatio-temporal variations in community environments. This study develops high-precision machine learning model to predict VOC emissions in an industrial park in Kaifeng, China, by integrating multi-modal data (low-cost VOC sensors, meteorological stations, and satellite-derived image features. Three machine learning algorithms—Random Forest, Gradient Boosting Machine, and eXtreme Gradient Boosting—were employed to capture non-linear relationships between emissions and environmental factors. The Optimal model achieved an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.742 at 100 × 110 m spatial and 1-hour temporal resolution, outperforming traditional methods. SHAP analysis identified industrial activity intensity, temperature, and wind speed as key emission drivers. This approach demonstrates the potential of multi-modal data fusion and machine learning for high-resolution air quality management in industrial communities.