Evaluating soil reinforcement by plant roots using artificial neural networks

Soil reinforcement by plant roots and its response to influencing factors are very important for bank stability evaluation and control. Models with improved accuracy are urgently needed for evaluating soil reinforcement. Using a back‐propagation (BP) learning algorithm, an artificial neural network (ANN) model with five input variables, including the number of roots, root area ratio, root tensile strength, soil shear strength, and soil moisture content, was developed to simulate the response of soil reinforcement to these factors. A connection weight approach was used to understand the relative importance of each factor. Using a data set published in 2003 and collected in Australia, soil reinforcement of four trees, Casuarina glauca, Eucalyptus amplifolia, Eucalyptus elata and Acacia floribunda, was simulated using three models: BP‐ANN, one described by Wu et al. in 1979 and the 2005 fibre bundle model (FBM) of Pollen and Simon. Comparisons of results from these models showed that the BP‐ANN model most accurately estimated the soil reinforcement. The simulated results indicated that only the effect of soil moisture content on soil reinforcement was negative. The influence of the other four factors was positive, and the relative importance was in the order: root area ratio > root tensile strength > the number of roots > soil shear strength. This study provides a new approach to soil reinforcement estimation and improves our understanding of soil resistance and bank stability.