Using a causal-based function to estimate soil bulk density in invaded coastal wetlands

Although pedotransfer functions (PTFs) have been routinely used for predicting soil bulk density (BD) in past decades, most of PTFs were developed based on empirical models that solely focused on statistical prediction. In this study, a causal-based PTF that focused on mechanistic explanation and prediction was developed by using partial least squares structural equation modelling (PLS-SEM). In an initial step, we identified key processes that theoretically affect BD variation, and assign measured variables for each process. This was achieved by investigating previous literature and researchers' experience based on characteristics of our study area. The proposed method was tested in an invaded coastal wetland in eastern China, with 45 samples. The causal-based model explained 71% of the variance in BD variation, indicating a permissible fit to the data. The model showed that the direct effect of nutrient cycling, plant invasion, and depth dependence on BD was significant. The results suggested that the soil processes and their interactions identified in the model were beneficial not only to improve predictive accuracy based on the cross-validation, but also to improve our understanding of BD variation from a system level. The findings highlight that a set of relative merits, such as no assumptions on the data and small sample size requirement, can improve the practical usefulness of PLS-SEM in BD prediction applications. This study also highlights that PLS-SEM may promote process-based soil-landscape modeling in terms of theory and methodology, especially for soil prediction or mapping.

Automated summary

This study successfully developed a predictive model for explaining and predicting soil bulk density variation using a causal-based partial least squares structural equation modeling approach. The model was validated in a coastal wetland in eastern China and showed significant direct effects of nutrient cycling, plant invasion, and depth dependence on bulk density. The findings highlighted the practical usefulness of the model in improving predictive accuracy and understanding soil variation at a system level.