Quantitative inversion of soil heavy metal pollution using a GA-BP neural network model

With the rapid development of industrialization in China, significant economic benefits have been accompanied by varying degrees of threat to the soil environment, particularly from heavy metal pollution. The rapid quantitative inversion of heavy metal concentrations and assessment of pollution risks are urgent tasks. This study selected the affected area of a non-ferrous metal smelting slag yard in Gejiu City as the research subject. Farmland soil samples were collected from this area, and the basic physicochemical properties of the soil, along with the contents of Pb and Cd, were analyzed. Models for quantitatively inverting Pb and Cd contents based on soil physicochemical parameters were established using multiple linear regression (MLR), backpropagation neural network (BPNN), and genetic algorithm-optimized BPNN (GA-BPNN). Among the three inversion models, the GA-BPNN model demonstrated the highest accuracy, with inversion precision R² reaching 0.8980 and 0.9013 for Pb and Cd, respectively; the corresponding root mean square error (RMSE) values were 0.0001868 and 0.0001821. The inversion model established a nonlinear relationship between heavy metals and basic soil physicochemical properties, enabling further heavy metal pollution assessment based on the inversion results. This method provides a novel approach for inverting heavy metal content in soil and holds practical value for heavy metal remediation.