The application of artificial intelligence in soil properties prediction has been progressing and developing in recent years. Determination of aggregate stability properties as indices against soil erodibility is time consuming and difficult. The predictability of three multivariate linear regression, neural network and neuro-fuzzy models efficiency in splash erosion prediction has been tested in this study. Due to low correlation of some properties of the soil, only four input parameters of Sodium Absorption Ratio (SAR), Porosity, Geometric Mean Diameter (GMD) of aggregates and runoff height have been analyzed as input variables in splash erosion prediction. The results indicated the priority of neuro-fuzzy method compared to others. Coefficient of determination of 0.798 in gauss2mf membership function with 3 membership functions and Hybrid Learning Algorithm have been obtained in adaptive neuro-fuzzy inference method. The small number of available data, in addition to samples distribution and spatial changes of samples led to low accuracy of multivariate linear regression method in splash erosion prediction.
Keywords: Aggregate Stability, Artificial Intelligence, Geometric Mean Diameter (GMD), Soil Erodibility
