MODELING BULK DENSITY OF SUBSOIL UNDER NATIVE VEGETATION IN MINAS GERAIS, BRAZIL

Bulk density (rho(b)) is an important indicator of soil physical quality, but it is seldom assessed at greater depths due to sampling difficulties. Pedotransfer functions have been relatively successful in estimating rho(b), but specific functions for most Brazilian biomes are not available. The aim of this study was to develop mathematical functions to describe rho(b) to a depth of 1 m in areas of native vegetation in central and southern Minas Gerais, Brazil. For determination of rho(b), soil core samples were taken of 53 profiles of different soil orders at six depth intervals (0-5, 5-10, 10-20, 30-40, 50-60 and 90-100 cm). The soil rho(b) ranged from 0.66 to 1.74 kg dm(-3), with a mean of 1.25 kg dm(-3), and was generally lower in the 0-5 and 5-10 cm layers. Models based on particle size analysis and standard chemical properties were created by multiple stepwise linear regression and were able to describe rho(b) to a depth of 1 m. Soil organic carbon and clay, silt, and sand contents, as well as cation exchange capacity at pH 7.0, were the most relevant variables in the models, which achieved higher accuracy for Oxisols (R-ajust(2) = 0.85), followed by Inceptisols (R-ajust(2) = 0.69), nitic Alfisols (R-ajust(2) = 0.67), and Ultisols (R-ajust(2) = 0.51). Since modeling for the whole database resulted in R-ajust(2) = 0.50, we conclude that stratification by soil order was useful for improving the equations, with the exception of the Ultisol order.