Uncertainty Analysis of Fuzzy Inference System to Predict Saturated Soil Hydraulic Conductivity

Estimation of saturated hydraulic conductivity (Ks) is important as an essential parameter in soil physics. Up to now, a variety of different models have been developed for estimation of Ks which include artificial intelligence (AI) models recently developed for this purpose. Although ROSETTA as an old computer program could simulate soil hydraulic parameter, Fuzzy Inference System (FIS) might be useful to estimate this parameter since it is less time-consuming with lower complexity and cost. A new challenging issue which has received less attention is the uncertainty analysis of the results due to different dataset and different membership functions. In this study, dataset of 151 samples collected from arable land around Bojnourd City in north Khorasan province were analyzed based on stepwise regression and bulk and particle densities determined as the most important inputs to FIS. Monte Carlo simulations of 1000 different samples were achieved and results derived from training period with some available membership functions in Matlab. Totally, three performance criteria showed reliable estimates of hydraulic conductivity for low and high values and, specifically, the best result was obtained from Gaussian membership function.