A pedo-transfer function (PTF) for estimating soil bulk density from basic soil data and its comparison with existing PTFs

Abstract

Collection of non-destructive soil core samples for determination of bulk densities is costly, difficult, time- consuming, and often impractical. To overcome this difficulty, several attempts have been made in the past to estimate soil bulk densities through pedo-transfer functions (PTFs), requiring soil texture and organic carbon (OC) content data. Although many studies have shown that both organic carbon and texture predominantly determine soil bulk density, a majority of the PTFs developed so far are a function only of organic matter (OM)/OC. In addition, no attempts have been made to test and compare the applicability of these PTFs on an independent soil data set. Thus, through this study efforts have been made not only to develop a robust soil bulk density estimating PTF, based on both soil texture and organic carbon content data, but also to compare its predictive potential with the existing PTFs on an independent soil data set from 4 ecologically diverse micro-watersheds in Almora district of Uttaranchal State in India. Effects of varying levels of soil particle size distributions and/or OC/OM contents on the absolute relative errors associated with these PTFs were also analysed for assessing their applicability to the independent soil data set. Amongst the existing PTFs, Curtis and Post, Adams, Federer, and Huntington-A methods were found to be associated with positive bias or mean errors (ME) and root mean square prediction differences (RMSPD) ranging between 0.10 and 0.38, and between 0.23 and 0.45, respectively, whereas Alexander-A, Alexander-B, Manrique and Jones-A, Manrique and Jones-B, and Rawls methods were found to be associated with negative ME and RMSPD values ranging between -0.08 and -0.15, and 0.18 and 0.23, respectively. In contrast, Bernoux, Huntington-B, and Tomasella and Hodnett-PTFs, with RMSPD values ranging between 0.18 and 0.20, were the only methods associated with little or no bias. However, on comparing the predictive potential of the existing PTFs, in terms of their 1 : 1 relationships between the observed and predicted soil bulk densities and ME and RMSPD values, only Manrique and Jones-B (ME: -0.08; RMSPD: 0.18), Alexander-A (ME: -0.08; RMSPD: 0.19), and Rawls (ME: -0.11; RMSPD: 0.22) methods were observed to give somewhat more realistic soil bulk density estimations. The study revealed very limited predictive potential of the existing PTFs, due to their development on specific soils and/or ecosystems, use of an indirectly computed organic matter (instead of directly measured organic carbon) content as a predictor variable, poor predictive potential of developed regression model(s), and/or subjective errors. In contrast to this, the new soil bulk density estimating PTF was found to be associated with far better 1 : 1 relationship between the observed and predicted soil bulk densities and zero ME (or bias) and lowest (0.15 g/cm³) RMSPD values. The absolute relative errors associated with both the new and the existing soil OC/OM and texture-dependent PTFs were observed to be almost insensitive to the varying levels of silt and clay. However, compared with the existing PTFs, these errors associated with the new PTF were observed to be much more insensitive to the varying levels of OC/OM, thereby indicating the applicability of the new PTF to a wide range of soil types.