Extreme Gradient Boosting Regression Model for Soil Available Boron

摘要

Soil formation processes and agricultural practices determine the amount of plant-available boron (B) concentration in soils. In this study, the relationships between soil characteristics and plant-available B concentrations of 54 soil samples collected from Gelendost and Egirdir districts of Isparta province were investigated using the Spearman correlation and eXtreme gradient boosting regression (XGBoost) model. Plant-available B concentration was significantly correlated with the soils' phosphorus, potassium, copper, and electrical conductivity. The XGBoost model explained 63 of the variation in five components defining soil behavior, and one of these components showed the variance resulting from the plant-available B. The effects of explanatory variables on B concentration determined in the XGBoost model were the parameters that were also significant in the correlation analysis. The results indicated that the model could successfully estimate B availability from the routinely analyzed soil properties (Fig. 1).