A novel interpolation method to predict soil heavy metals based on a genetic algorithm and neural network model

• Published in: The Science of the total environment Vol. 825; p. 153948
• Main Authors: Yin, Guangcai, Chen, Xingling, Zhu, Hanghai, Chen, Zhiliang, Su, Chuanghong, He, Zechen, Qiu, Jinrong, Wang, Tieyu
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 15.06.2022
• Subjects: algorithms; China; environment; Environmental Monitoring - methods; Genetic algorithm; Interpolation; kriging; Metals, Heavy - analysis; Neural network model; neural networks; Neural Networks, Computer; prediction; Risk Assessment; Soil; Soil heavy metals; Soil Pollutants - analysis; Spatial Analysis; China; Interpolation; Soil heavy metals; Genetic algorithm; Neural network model
• ISSN: 0048-9697; 1879-1026; 1879-1026
• DOI: 10.1016/j.scitotenv.2022.153948

To improve the prediction accuracy of soil heavy metals (HMs) by spatial interpolation, a novel interpolation method based on genetic algorithm and neural network model (GANN model), which integrates soil properties and environmental factors, was proposed to predict the soil HM content. Eleven soil HMs (Cu, Pb, Zn, Cd, Ni, Cr, Hg, As, Co, V and Mn) were predicted using the GANN model. The results showed that the model had a good prediction performance with correlation coefficients (R2) varying from 0.7901 to 0.9776. Compared with other traditional interpolation methods, including inverse distance weighting (IDW), ordinary kriging (OK), universal kriging (UK), and spline with barriers interpolation (SBI) methods, the GANN model had a relatively lower root mean square error value, ranging from 0.0497 to 77.43, suggesting that the GANN model might be a more accurate spatial interpolation method and the soil properties together with the environmental geographical factors played key roles in prediction of soil HMs. [Display omitted] •A novel interpolation model was developed to predict soil HMs in a provincial scale.•Combining genetic algorithm and neural network model enhanced prediction accuracy.•GANN interpolation method performed advantages to 3 dominant traditional methods.•The optimized method showed lower root mean square error values and data anomalies.