Prediction of Soil pH Improvement Through Biochar: A Machine Learning Based Solution

Abstract

Biochar has achieved good results in improving soil properties. The rapid development of machine learning technology makes it possible to predict soil physicochemical properties. The objective of this study was to investigate the underlying mechanisms impacting soil pH in biochar‐improved soil using machine learning models. This study, based on the Lightweight Gradient Boosting Machine (LightGBM) and Deep Neural Network (DNN) algorithms, established machine learning models of soil pH after biochar addition and explored the influence of different input combinations of biochar information on the accuracy and performance of the model. The results show that biochar pH and biochar cation exchange capacity have a significant influence on model accuracy. Compared to the DNN model, the LightGBM model was more appropriate for predicting soil pH, and the LightGBM_a model performed the best, with R2 of 0.92, MAE of 0.291, and RMSE of 0.539. Shapley additive explanations (SHAP) value analysis, Partial Dependence Plot (PDP) analysis, and Individual Conditional Expectation (ICE) analysis further indicated that biochar electrical conductivity and biochar cation exchange capacity were important characteristics that have an extremely significant impact on model accuracy. The simultaneous citation of biochar pH, biochar cation exchange capacity, and biochar electrical conductivity has a synergistic effect. At the same time, it provides a reference for predicting other physical and chemical properties of soil after biochar is added.