In the snow melt period, snow depth variation has the features of high dimension, nonlinear and non-normal distribution. In this study, the multi index problem on snow depth change was transformed into the problem of single projection index by the projection pursuit method with optimized the projection direction by genetic algorithm. So the snow depth change dimention has been reduced with the method. The result shows that the factors effect sequence on snow depth dynamic change follows: the ground temperature > net radiation > air temperature > soil moisture > wind speed > relative humidity > total precipitation > vapor pressure. MLP neural network was used to build the snow depth model, which can simulate snow depth. The result shows that using projection pursuit method to get the snow depth of major impact factor is reliable, and major impact factor can simplify the snow depth model.