Evaluation of pedotransfer functions for predicting particle density of soils with low organic matter contents

Abstract

Soil particle density (ρs) is a basic soil physical property which is essential for soil pore volume estimation and sediment source discrimination. The measurement of soil ρs is labor intensive and time-consuming. Pedotransfer functions (PTFs) have been developed for estimating ρs from readily available soil properties during the past decades. Existing PTFs are developed mainly based on soil datasets covering wide ranges of soil organic matter (SOM) contents. For soils from subtropical regions with low SOM values the accuracy of previous PTFs remains unknown. In this study we evaluated the performance of existing ρs PTFs by comparing model estimates to measurements of 175 soils from a subtropical region in China. The investigated soils had relatively low SOM contents (mostly<0.02 g g−1) and covered texture ranges from coarse sand to clayey soils. The Adams-based PTFs performed poorly, while the modified Adams PTFs and several simple linear or multiple regression PTFs produced relatively accurate ρs estimates for the 175 soils with root mean square errors (RMSEs) around 0.05 g cm−3 and mean biases within ± 0.03 g cm−3. Further multiple regressions indicated that the ρs of subtropical soils was simply a negative linear function of clay content. Despite conflicting several early reports, an independent dataset showed that the proposed linear model provided accurate ρs estimates for soils with low SOM contents from subtropical regions with a RMSE of 0.030 g cm−3 and mean bias of −0.013 g cm−3.