Prediction Models for Cation Exchange Capacity of Purplish Soils in Southwest China

Objective The aim was to establish an efficient predicting model for cation exchange capacity (CEC) in purplish soils.

Method The data from 171 purplish soil samples in the southwest region of China, including soil pH, organic matter content, and soil particle composition, have been employed a hybrid strategy of genetic algorithm (GA) and Piecewise mapping with Levy flight optimized Particle Swarm Optimization (PLPSO) to optimize the Back Propagation Neural Network (BPNN). Additionally, the classical statistical method was used to construct multiple models for predicting CEC in purplish soils, including MLR, BPNN, GA-BPNN, PLPSO-BPNN, PSO-GA-BPNN and PLPSO-GA-BPNN. The models were evaluated using the Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and Coefficient of Determination (R²).

Result The ranking of MAPE for the CEC prediction methods was as follows: PLPSO_GA-BPNN (0.6167) < PSO_GA-BPNN (0.7421) < PLPSO-BPNN (0.9871) < GA-BPNN (1.0821) < BPNN (1.2561) < MLR (1.7772). The RMSE ranking was: PLPSO_GA-BPNN (0.1112) < PSO_GA-BPNN (0.1254) < GA-BPNN (0.1349) < PLPSO-BPNN (0.1441) < BPNN (0.1894) < MLR (0.2557). The R² ranking was: MLR (0.6182) < BPNN (0.7313) < GA-BPNN (0.8637) < PLPSO-BPNN (0.8745) < PSO_GA-BPNN (0.8923) < PLPSO_GA-BPNN (0.9074). The PLPSO_GA-BPNN model performed the best in terms of RMSE and R², with values of 0.6167, 0.1112, and 0.9074, respectively.

Conclusion The PLPSO_GA-BPNN model demonstrated the best predictive performance, and the optimization effect of PLPSO_GA was the most effective, which not only providing a more direct and effective method for accurately predicting CEC in purplish soils, but also holding broad application prospects in the field of soil property prediction.