Predicting loblolly pine site index from soil properties using partial least-squares regression

Abstract

Data from 15 sites with loblolly pine (Pinus taeda L.) plantations located in the southeastern United States were used to predict site index (SI) from soil physical and chemical properties from the top 15 cm of the mineral soil. Two modeling approaches were used to predict SI from soil properties. First, the ordinary least-squares method of multiple linear regression was used, which selected calcium, potassium, AND sand percentage as the significant predictor variables. Second, partial least-squares regression was used, which selected total nitrogen, carbon, calcium, magnesium, AND sand percentage as the significant predictor variables. The partial least-squares regression approach addressed multicollinearity in the data and produced a better model to predict SI. The partial least-squares regression model explained 77% of the variation in SI.