Defining the relationship between bulk density and organic carbon content in forest soils using generalised linear mixed-effect models

Abstract

Background

In this study, we used a generalised linear mixed-effects model (GLMER) to establish a predictive pedotransfer function defining the relationship between forest soil bulk density and total organic carbon. More than 950 soil samples were obtained from four forested areas with a wide range of bedrock (limestone, loess, crystalline volcanic, sandstone, alluvial loam, polygenic loam and transported materials rich in organic carbon) and soil types (Leptosols, Cambisols, Fluvisols, Podzols and Technosols). Model validation was performed by testing against 10% of the data randomly selected from the original dataset (10% dataset) and an independent dataset from the Czech national forest inventory (NFI2 dataset).

Result

The GLMER including sample origin locality as random effect displayed a highly accurate predictive capacity. Subsequent analysis avoided model simplification by excluding sample origin and retaining the global GLMER only. For all samples, the final model covered a range from 0.16 to 27.70% for total organic carbon and from 0.27 to 1.94 g cm^− 3 for bulk density. Model residuals based on laboratory values were symmetrical with a median value just 0.09 g cm^− 3 higher. While validation with the 10% dataset confirmed model parameter validity with high accuracy, validation using the NFI2 dataset indicated slight discrepancies, possibly due to differences in sampling method used. Individual GLMs fitted both validation datasets better than the global GLMER; however, Wilcoxon tests showed better consistency in the original model on the 10% validation data. Consequently, we suggest the global GLMER may prove more suitable for direct use in expressing bulk density from total organic carbon.

Conclusion

The pedotransfer functions produced, particularly that based on global GLMER, can be used to express bulk density via total organic carbon content, or vice versa, with high accuracy. While based on a wide range of bedrock/soil types, further studies may be needed in other regions to validate the model for general application.