Using pedo-transfer functions to estimate dry soil layers along an 860-km long transect on China’s Loess Plateau

Abstract

Dry soil layer (DSL) development as a result of imbalance in water input and output is a widespread pedo-hydrological phenomenon in arid/semi-arid regions such as the China’s Loess Plateau (CLP). To build sufficient data for large-scale DSL estimation, soil water data for the 0–5 m soil profile were collected for the period 2013–2016 along an 860-km long transect on CLP and analyzed for pedo-transfer functions (PTFs). The objective was to determine the effects of environmental factors on DSL variation and to develop an effective PTF for the estimation of DSLs on CLP. Three DSL evaluation indices were calculated — DSL thickness (DSL-T), DSL formation depth (DSL-F) and DSL mean soil desiccation index (DSL-SDI). The results showed that DSLs were mainly distributed in the northcentral part of the transect, with a mean thickness of 2.77 m. We compared the performances of PTFs developed by different approaches — multiple regression (MR), artificial neural network (ANN) and the indirect and direct methods. It showed that the ANN approach effectively predicted DSL formation and indices. The indirect method improved simulation accuracy of DSL indices. The combination of the ANN approach and the indirect method gave the best estimation accuracy for DSL indices. The application of the PTFs not only reduced labor and time needed for field survey of DSL, but also improved DSL research on CLP and beyond. The indirect method based on soil moisture and/or hydrological models was promising for the estimation of DSL indices.