Assessment through Machine Learning of Groundwater Vulnerability after Seismic Damage to Fuel Pipeline

Abstract

This study assessed the vulnerability of groundwater resources to the failure of the urban fuel distribution network under an earthquake. A case study of the Tehran, Iran, gas distribution network and the Tehran–Karaj Plain aquifer was conducted. To assess the seismic vulnerability of buried fuel pipelines in Tehran based on the fuel distribution network components, three possible earthquake scenarios were studied. To assess damage to the pipeline, a comprehensive model was developed using machine learning (ML). This model can assess and predict damage to a fuel pipeline and its type (i.e., leakage or full breakage). Moreover, aquifer contamination was assessed using the DRASTIC model. It was found that the ML-based pipeline seismic vulnerability assessment model had good performance in predicting seismic damage to the fuel distribution network, with a RMS error (RMSE) and a correlation coefficient (R) of 0.004 and 0.99, respectively. The results showed that the presented model had an acceptable efficiency in assessing the probability of seismic vulnerability of the buried pipeline and analyzing the pollution of the aquifer based on different earthquake scenarios. The developed groundwater seismic vulnerability assessment model can be used for further analysis in future research. Earthquakes are one of the most important natural disasters, and have caused widespread financial, human, and environmental losses in different regions of the world, especially in seismic areas. The existence of faults and the possible deterioration of buried pipes makes earthquake crisis and its serious damage to humans and the environment more severe. One of the most important threats in this situation is the contamination of underground water with hydrocarbon substances due to leakage from the fuel transmission network. In this research, to evaluate the pollution of the aquifer due to the damage to the fuel transmission network, we developed a model using the machine learning method to analyze the vulnerability of the buried pipeline. The DRASTIC model also was used to evaluate aquifer pollution. To evaluate the presented model, the fuel transmission network and aquifer of Tehran, Iran, were studied. The results indicated acceptable performance of the proposed model for assessing the seismic vulnerability of groundwater. The presented model can be used for other areas.