Underestimation of industrial contributions to VOC fluctuation in the Hangzhou Bay: Insights from simultaneous measurements and ML-SHAP analysis

As a major petrochemical hub, the Hangzhou Bay suffers from severe volatile organic compound (VOC) pollution. Here, we conducted parallel high-frequency VOC measurements at a coastal mainland site (Haiyan) and an offshore island site (Daishan) during April-May 2024. To disentangle average source contributions from dynamic peak drivers, we developed a novel integrated framework combining photochemical initial concentration reconstruction, adaptive dynamic-threshold event detection, Positive Matrix Factorization (PMF), and machine learning with SHapley Additive exPlanations (ML-SHAP). By reconstructing initial concentrations, we effectively mitigated the "masking effect" of photochemical degradation, revealing that conventional uncorrected observations significantly underestimate the true weights of highly reactive industrial emissions. The parallel PMF simulations confirmed that source categories remained consistent between observed and reconstructed frameworks, identifying eight primary VOC sources. However, the analysis revealed that industry-related emissions accounted for 30-40% of the average mass concentration in the observed model, which rose significantly after reconstruction. Although the offshore site exhibited lower average TVOC concentrations, our dynamic-threshold algorithm objectively isolated frequent, high-intensity short-term industrial plumes at this location, driving markedly higher skewness and kurtosis. Furthermore, while the PMF highlights baseline mass contributions, the ML-SHAP framework demonstrated that emission sources overwhelmingly dominate short-term VOC variability (>85%), with transient industrial pulses acting as the deterministic triggers (∼60%) for extreme concentration spikes. Trajectory analyses confirmed that coastal VOCs are shaped by regional transport and local emissions, whereas offshore fluctuations are predominantly driven by short-distance upwind industrial clusters. These findings highlight a fundamental paradigm shift in regional pollution control: moving from steady-state "total emission reduction" toward "pulse-oriented peak management," emphasizing the necessity of high-resolution early warning systems to mitigate episodic ozone formation in industrialized coastal environments.