Investigation of landfill leakage points using geophysical exploration and machine learning methods

The compression and settlement of waste deposits in landfills may cause the deformation failure of leachate collection pipelines in municipal solid waste treatment facilities. This failure will further lead to the rupture of the impermeable layers and subsequent leakage of landfill leachates. Primary methods for monitoring damage to leachate impermeable layers include: laboratory chemical analysis of the pollutant levels of soil and water samples from landfills, and surface geophysical exploration. Building upon the above traditional environmental geological exploration methods, this study achieved effective prediction of landfill leachate leakage using geographic information system (GIS) technology combined with machine learning and influencing factor prediction. Specifically, the landfill area was divided into grid cells with resolution of 5 m × 5 m, which were then assigned corresponding spatial coordinates. A machine learning approach was employed to establish artificial neural network (ANN) models for predicting landfill leakage points. Based on a field investigation of a landfill in North China, 880 test samples and 220 validation samples were selected. The data obtained from these samples, including groundwater table depth, resistivity, polarizability, anomalies of seismic events in radar reflection waveforms, spatial relationships between seismic events and adjacent monitoring boreholes, and half-life information, were used as the input for the models. The output provided the leakage probabilities of all grid cells. Finally, the prediction results of two machine learning methods, the backpropagation (BP) and radial basis function (RBF) neural networks, were comparatively analyzed. The findings indicate that the BP neural network model showed higher prediction performance. The groundwater table depth, resistivity, and the relationship between resistivity and monitoring borehole positions served as the dominant factors influencing the seepage probability prediction of the impermeable layer. By superimposing the posterior probabilities of various influencing factors, the prediction model generated a classification map for leakage risk zones, achieving the objective of graded monitoring of leachate leakage risks within the landfill area.