Source term estimation (STE) of hazardous gas leakages in chemical industrial parks (CIPs) is important for addressing environmental pollution and improving safety and reliability in engineering practice. To achieve real-time STE

least squares-based STE methods have recently been developed. However

these methods require the number and locations of potential hazardous gas leakage sources are known as a priori

which is difficult in many practical scenarios. To address this limitation

we propose a new data-driven STE approach

which enables the STE to be implemented in real time and applicable to complicated turbulent dispersion scenarios. The linear independent analysis in data science is applied to historically collected concentration data of a hazardous gas of concern from a network of sensors to extract the sensor data which represent independent hazardous gas leakage scenarios (IHGLSs). An appropriate Gaussian model approximation to a high-fidelity computational fluid dynamics (CFD) model that must be used to represent the hazardous gas leakage scenarios of concern is built

and the off-line STE of IHGLSs using the approximating Gaussian model is then performed to build the data-driven STE model. The performance of the proposed approach is evaluated by using data that are generated by simulating ethane leakage scenarios in a CIP using a CFD model. Results indicate that the leakage localization accuracy is 100% and the mean relative estimation error for the leakage strength is 6.76%. Moreover

the proposed approach is validated with real data in Prairie Grass field dispersion experiments

demonstrating the practical applicability of the proposed approach.