Spatial Distribution and Prediction of Cu and Mn in Helan Farmland Soil Based on GIS and Machine Learning

ABSTRACT Cu and Mn in soil play a crucial role in crop growth, but excessive amounts of Cu and Mn can pose a health threat to crops and humans. Therefore, accurately understanding Cu and Mn spatial distribution in soil is significant for protecting the ecological environment and human health. This study employed four machine learning (ML) methods, namely ordinary kriging (OK), back propagation neural network (BPNN), random forest (RF), and support vector machine (SVM), to predict the spatial distribution of Cu and Mn. Taking Helan County as study area, the predictive accuracy of four different models was compared by combining multiple auxiliary variable sources. The prediction accuracy of the four models was evaluated using root mean square error (RMSE) as the primary metric and mean relative error (MRE) as the secondary metric. The results of the spatial prediction model for Cu content were as follows: RF (MRE = 0.162, RMSE = 2.042) > BPNN (MRE = 0.154, RMSE = 2.160) > OK (MRE = 0.094, RMSE = 2.655) > SVM (MRE = 0.167, RMSE = 2.810), demonstrating that RF showed the best performance for Cu prediction in the study area. The spatial distribution trends for Cu predicted by the four models were similar, all showing a spatial pattern with higher concentrations in the central region and lower concentrations in the northern region. The prediction accuracies of the four models for Mn were as follows: RF (MRE = 0.107, RMSE = 3.629) > BPNN (MRE = 0.108, RMSE = 4.695) > SVM (MRE = 0.131, RMSE = 5.115) > OK (MRE = 0.077, RMSE = 6.165), indicating that RF was more suitable for predicting Mn in the study area. The Mn content distribution map showed a spatial pattern with higher concentrations in the southwestern and northwestern regions and lower concentrations in the southeastern region. The RF model effectively handled the nonlinear relationship between Cu and Mn content and multiple auxiliary variables. These results provided important theoretical basis and data support for exploring the spatial distribution trends of Cu and Mn content.