Evaluation of parameter models for estimating loess soil particle-size distribution

The soil particle-size distribution (PSD) is a fundamental soil physical property. PSD curve is widely used as a basis for estimating soil hydraulic properties such as the water retention curve and saturated and unsaturated hydraulic conductivities. However, the disconnected experimental data of PSD sets often limit its uses, and the parameter models are needed to simulate PSD curves. Many models have been developed for simulating PSD curves, and the performances of these models varied with the soil types. To determine the optimal model to fit the PSD curves for loess soils, ten popular models (two 1-parameter models, six 2-parameter models, and two 3-parameter models) were used to fit 828-PSD data sets measured in the Loess Plateau of China. Three criteria, correlation coefficient（R）, root mean square error (RMSE), and Akaike’s information criterion (AIC) were used to evaluate the performances of ten models. The results show that the Fredlund model with three parameters presents the best performance for the majority of soils, and has little influence by the soil texture. The performance of Weibull model with two parameters is less than that of Fredlund model and better than that of others. The Jaky model with single parameter presents the worst performance in the whole models. The Fredlund model with three parameters is testified to be suitable for estimating the PSD curves of the loess soils.