Predicting of Soil Bulk Density Using Bulgarian Dataset

Abstract

The objective of this study was to determine the most appropriate pedotransfer functions for the soil bulk density prediction in depth for virgin and arable soils. The statistical analyses were performed on datasets containing 578 records for the soil bulk density, soil texture fractions, humus/SOC and carbonate contents at different soil depths from virgin and arable lands. We used data from our surveys as well published data by others. The pedotransfer functions were derived separately for virgin and arable lands and for datasets containing textural fractions determined according to the Katchinski method and according to the ISO 11277:2009. Mean absolute errors of the multiple regression equations were 0.07–0.10 g cm⁻³, and R ²_adj  = 69–71% for the virgin soil profiles. The prediction accuracy of Db of the arable soils was lower and corresponded in most case to Db of the equilibrium state. The advantage of local calibration and deriving of the most appropriate relationships stemmed from the available proxy variables and the methods used for their determination (e.g., soil particles size analyses).