Comparison of estimated soil bulk density using proximal soil sensing and pedotransfer functions

Abstract

Due to disadvantages of the core sampling method for determining soil bulk density (BD), many soil surveys lack data of BD. To meet extensive demands for BD in hydrology, agriculture, environment, etc., an approach based on a pedotransfer function (PTF) is often established using available data of soil properties such as texture, organic matter content, pH and soil depth. Recently, an approach based on proximal soil sensing (PSS) was also proposed to fulfill this purpose, i.e., subtracting volumetric water content (θ, %) estimated using visible and near infrared spectroscopy (vis-NIRS) from BD of wet soils determined using gamma ray attenuation. With 120 samples collected from 20 profiles and 38 topsoils in a hilly subtropical area in China, this study compared accuracy of the two approaches based on leave-one-out cross-validation (LOOCV) and repeated cross-validations with a wide range of randomly split samples. Texture, organic matter content, pH and depth of the samples were measured, and they were used within multiple linear regression to establish PTFs. Results showed that the approach based on PSS was at least 12% more accurate than the other and its estimated BD were close to those measured using the core sampling method. For instance, in LOOCV, estimated BD based on PSS had root mean square error (RMSE) of 0.082 g cm−3, coefficient of determination (R2) of 0.92 and residual prediction deviation (RPD) of 2.07, while estimated BD based on the best PTF had RMSE of 0.094 g cm−3, R2 of 0.70 and PRD of 1.82. The results also showed that accuracy of the approach based on PSS was determined by accuracy of estimated θ using vis-NIRS. However, it is limited to improve accuracy of estimated BD using the approach based on PSS further by improving accuracy of estimated θ. This study demonstrated that the approach based on PSS is promising for estimating BD in future soil surveys.