Dynamic simulation of soil water-salt using BP neural network model and grey correlation analysis

Soil water-salt dynamic under natural-artificial-biological conditions was studied with measured data of Luohui trench irrigation district in Shaanxi Province based on application of backpropagation(BP) networks of three layers, and then the additional momentum method and self adaptive tactic for training were adopted to feed forward BP neural networks. On the basis of the condition above, a sensitivity analysis about soil salt content and soil alkalinity was conducted according to each input factor by using default factor method, and the grey correlation analysis method was applied to certify the results. The results showed that the artificial neural networks model could express quantitatively the response relationship between groundwater dynamic and various factors with sufficient high accuracy. Soil water content, salt concentration of groundwater, and evaporation capacity were the main sensitive factors for soil water-salt dynamic in this irrigation district, the interaction amongst various factors formed coupling relationship under the complicated condition. The grey correlation analysis method could further verify the sensitivity degree amongst various factors. The combination of the above methods provides feasible method for analyzing the rules of soil water-salt dynamic under the condition of shallow groundwater depth during crop growing season, and it is complement and perfection for the traditional research methods of groundwater water-salt dynamic.