A framework for spatial correlations between industrial pollution sources and groundwater vulnerabilities based on machine learning and spatial cluster analysis: implications for risk control

The efficacy of groundwater pollution risk control is often limited by a lack of information on spatial correlations between industrial pollution sources and groundwater vulnerabilities. To overcome this limitation, a novel data-driven framework was established to reveal the spatial correlations between industrial pollution sources and groundwater vulnerabilities in Guangdong Province, China, mainly using a combination of genetic algorithm (GA), back propagation neural network (BPNN), kernel density estimation (KDE) and bivariate local Moran’s I (BLMI). The subjectivity in the indicator weighting of a DRASTICL model was successfully reduced by GA-BPNN, and then the groundwater vulnerability map was created by the GA-BPNN-DRASTICL model. A smoother surface of the industrial pollution source distribution was produced by KDE. The spatial clustering map between industrial pollution sources and groundwater vulnerabilities was generated by BLMI, explicitly showing their distribution characteristics and implying that the specific measures should be taken for controlling risks of groundwater pollution in the different parts of the study area.