Adaptive neuro fuzzy inference system approach for developing soil water retention pedotransfer functions in flood spreading areas

Abstract

When using indirect approaches for accurate estimation of soil hydraulic properties, it is important to consider time and cost savings as part of water management. Models based on an adaptive neuro fuzzy inference system (ANFIS) outperform artificial neural networks (ANNs) in terms of forecasting error, computational speed, and estimation. The present study used ANFIS to develop pedotransfer functions (PTFs) to estimate soil hydraulic parameters (van Genuchten water retention curve parameters α, n, and residual water content [θ_r]), soil water retention at saturation, field capacity (FC), and permanent wilting point (PWP) using basic soil properties such as soil particle size distribution, bulk density (BD), medium porosity (0.2 to 30 μm [7.9 × 10⁻⁶ to 0.001 in]), and organic carbon (C). The ANN-based Rosetta model and ANFIS-based PTF were compared for accuracy of prediction of saturated soil water content (θ_s), FC, and PWP of soil in the flood spreading areas of Iran. The ANFIS-based models were able to estimate soil hydraulic properties with reasonable accuracy. It was concluded that adding medium porosity (0.2 to 30 μm) as an input variable to the ANFIS-based models improved model accuracy for FC, α, and n. Except for the SSCBDθ₃₃ model, prediction of water content at FC and PWP by Rosetta improved significantly over those obtained using the ANFIS-based PTFs.