Modeling Soil Water Retention Under Different Pressures Using Adaptive Neuro-Fuzzy Inference System

Abstract

Soil Water Retention (SWR) is an important process in drainage, surface, and groundwater partitioning, hydrological modeling, water supply for irrigation, etc. Assessment of SWR characteristics is complex and difficult to conduct spatially in varied locations. Therefore, Pedotransfer Functions (PTF) which are empirical relations with easily available physical properties are commonly used. In the present study, the evaluation of soil moisture at different suction pressure using the adaptive neuro-fuzzy inference systems (ANFIS) approach based on soil texture (percentage of gravel, sand, silt, and clay) and compare with the PTF approach. The analysis was conducted for a total of eleven sites of two adjoining commands in India. The pressure plate apparatus along with coarse and fine sieve analysis, titration, and other tests were carried out to determine SWR, texture, organic carbon, and bulk density. The comparative analysis of Nash–Sutcliffe efficiencies of the best-fitted PTF models and ANFIS model confirmed that the ANFIS model can capture all variations of soil texture across all sites with Nash–Sutcliffe efficiency of nearly 1.0 indicative of an exact match, while no single PTF-based model can be used for all the sites. Therefore, the ANFIS model can be used to model soil water retention for the central India region using easily available texture properties of soils.