A machine learning and geostatistical hybrid method to improve spatial prediction accuracy of soil potentially toxic elements

Effective environmental management and contamination remediation require accurate spatial distributions predictions of potentially toxic elements (PTEs) in the soil. However, no single method has been developed to predict soil PTE accurately. This study evaluated advanced geostatistical empirical Bayesian kriging regression prediction (EBKRP), machine learning algorithm random forest (RF), hybridized model (RF-EBKRP) map PTEs content greenspace soils. As identified by RF, organic carbon, matter, total (nitrogen, phosphorus, potassium), topographic features, urban functional types were used as significant covariates improve accuracy The performance was using root mean square error (RMSE), absolute percentage (MAPE), coefficient determination (R2). Results showed that RF performed much better than EBKRP predicting PTEs, with lower errors a higher R2. RMSE, MAPE, R2 values for 0.25–85.32 mg/kg, 3.86–25.40%, 0.77–0.90, respectively, while 0.51–99.03 5.42–32.13%, 0.40–0.66. Moreover, RF-EBKRP produced more individual models, improvements 122.5% 15.58% RF. performances are due its incorporation various ability handle complex nonlinear relationships between covariates. In end, hybrid is promising approach improving distribution PTEs.