Predictive regression model for crop biomass in clayey soil amended with bamboo biochar

By employing multiple linear regression (MLR) and random forest (RF) algorithms, the intricate interactions of clay soil amended with bamboo biochar were explored, aiming to introduce innovative strategies in precision crop management. The performance of two machine learning-based statistical models were assessed to predict maize biomass following a single application of biochar to clay soil at doses of 25 and 50 tons per hectare, respectively, either alone or in combination with vermicompost, across two cropping cycles. A total of 34 soil input variables, comprising chemical, physical, and hydraulic soil parameters, including pore size distribution, infiltration rate, and hydraulic conductivity, and one target crop biomass, were subjected to correlation analysis. The process of refining model parameters involved assessing feature importance and redundancy to optimize selection. Three models were chosen based on a dataset encompassing the 100^th, 50^th, and 75^th percentiles. Among them, the RF model using the 50^th percentile data demonstrated the best fit explaining 60% of the variability of the target crop biomass variable while yielding the smallest root mean square error (RMSE). Notably, all RF models identified potassium (K), phosphorus (P), and the magnesium-to-potassium (Mg/K) ratio as the most influential soil properties for biomass prediction. The RF models represent a valuable tool for predicting crop biomass yield in clay soils amended with biochar.