Performance of the tangential model of soil water retention curves for various soil texture classes

The tangential model (TANMOD) is one of the few soil water retention curve (SWRC) models that can be applied in both unsaturated and saturated soils, from the positive suction range to the negative suction range, accounting for the effect of volume changes in the entrapped air in soil pores. The model has been successfully evaluated with relatively coarser soils. Its performance, however, has not been fully tested for various soil texture classes. In this study, we aim 1) to determine the TANMOD parameters for various soil texture classes, and 2) to assess the underlying relationship between the TANMOD parameters and the soil texture class. To address those objectives, the TANMOD was first fitted to 399 SWRC from 10 USDA soil texture classes in the UNSODA soil database. The model parameters consist of three coordinates (S-re, s(e)), (S-rm, S-m), and (S-rf, s(f)), three tangential slopes, c(e), c(m), and c(f), along the curve. Multivariate analysis and several machine learning algorithms were respectively used to evaluate model parameters for each soil texture class and reveal the relation between the model parameters and the soil texture classes. The results demonstrated that the TANMOD fitted well from coarser soils to finer soils. Unique sets of the model parameters and their uncertainties are proposed for 10 USDA soil texture classes. Unsupervised learning algorithms, hierarchical cluster analysis and k-means clustering, failed to classify the TANMOD parameters while one of the supervised machine learning techniques, random forest, adequately classified the TANMOD parameters to the USDA soil texture classes. The accuracy of the classification based on the random forest model is 62.6%. The maximum tangential slope, c(m), was the most important parameter in relation with the soil texture class. Consequently, the TANMOD parameters not only have their own physical meaning but also can be applied to various USDA soil texture classes.