Development of ternary diagrams for estimating water retention properties using geostatistical approaches

Abstract Most pedotransfer functions (PTFs) have adopted soil texture information as the main predictor to estimate soil hydraulic properties, whether inputs are defined in terms of the relative proportion of different grain size particles or texture-based classifications. The objective of this study was to develop ternary diagrams for estimating soil water retention (θ) at − 33 and − 1500 kPa matric potentials, corresponding to the field capacity and wilting point, respectively, from particle size distribution using two geostatistical approaches. The texture triangle was divided into a 1% grid of soil texture composition resulting in 4332 different soil textures. Measured soil water retention values determined in 742 soil horizons/layers located in Portugal were then used to develop and validate the hydraulic ternary diagrams. The development subset included two-thirds of the data, and the validation subset the remaining samples. The measured soil water content values were displayed in the ternary diagram according to the coordinates given by the particles size distribution determined in the same soil samples. The volumetric water content values were then predicted for the entire ternary diagram using two different geostatistical interpolation algorithms (ordinary kriging and the empirical best linear unbiased predictor). Uncertainty analysis resulted in a root mean square error below 0.040 and 0.034 cm 3  cm − 3 when comparing the interpolated water contents at − 33 and − 1500 kPa matric potential values, respectively, with the measured ones included in the validation dataset. The estimation variance calculated with both methods was also considered to access the uncertainty of the predictions. The available water content of Portuguese soils was then derived from θ − 33 kPa and θ − 1500 kPa ternary diagrams developed with both approaches. The hydraulic ternary diagrams may thus serve as simplified tools for estimating water retention properties from particle size distribution and eventually serve as an alternative to the traditional statistical regression and data mining techniques used to derive PTFs.