Using Artificial Neural Networks to Predict Soil Bulk Density.

Soil bulk density (BD) is a fundamental soil property in soil physics and hydrology and is usually predicted using PedoTransfer Functions (PTFs). Because of the increasing use of these PTFs in many soil and water simulation models, this study developed a set of PTFs for predicting soil bulk density using Artificial Neural Network (ANN) technique. The functions were elaborated with more than a thousand measured BD values from a wide range of soils from Chile. In addition, the estimates were compared to those obtained using a series of already existing PTFs. The results showed that among the functions developed, the best estimates were obtained when using sand, silt, clay, organic carbon content, soil depth, and water content at wilting point as inputs (r²=0.72). Additionally, the new set of PTFs enhanced the quality of estimates compared with the existing PTFs by improving the r² between 0.02 and 0.14, and reducing the overall error between 0.01 and 0.04 Mg m^-3.