Exploring the impact of lime amendments: a machine learning analysis of influential factors

Introduction The effectiveness of liming materials in ameliorating soil acidity depends on their physical, chemical, and mineralogical properties. Although several factors influencing lime efficacy have been identified in previous studies, their relative importance and interactions remain insufficiently explored. This study therefore applies a machine learning (ML) approach to better understand the complex interactions among variables affecting soil pH change following lime application. Methods A dataset comprising 857 observations was compiled and analyzed within a supervised ML framework using a Gradient Boosting Model (GBM). The model was used to evaluate the relative importance of key factors, including particle size, lime reaction time, lime application rate, neutralizing value (NV), initial soil pH, and limestone type. Model performance was assessed using grouped cross-validation to ensure robust and unbiased evaluation. Results and discussion The GBM demonstrated strong predictive performance, achieving a coefficient of determination (R²) of 89.7%, along with low intercept and RMSE values, indicating minimal prediction bias and error. Particle size, reaction time, and lime application rate emerged as the most influential factors controlling lime efficacy, particularly through their interactions. The findings further highlight the limitations of relying solely on Effective Calcium Carbonate Equivalence (ECCE) and emphasize the need for integrated quality indices when evaluating liming materials. Optimal soil pH response was observed between 90 and 400 days after lime application, suggesting that liming should be done well in advance of cropping. Additionally, the optimal lime application rate ranged from 10 to 21 t ha⁻¹ for effective soil pH improvement. Overall, the study demonstrates that ML provides a robust, data-driven framework for evaluating liming efficacy and offers clear advantages over conventional reductionist approaches by capturing complex, multivariate interactions.