Species diversity is a measure of the complexity of community structure and function, and the spatial distribution pattern of species diversity is affected by many environmental factors. In this paper, the relationships between species diversity and community type, habitat and community structure of forests in the Songshan National Nature Reserve were analyzed by diversity indices and multi-layer perceptron network. The dataset included 68 quadrats, 291 species, and 6 environmental variables (elevation, slope, aspect, litter layer thickness, soil depth, and soil solidity). The results showed that species richness, species diversity and evenness values of Ulmus macrocarpa + Populus davidiana mixed forest and Pinus tabulaeformis+ Populus cathayana mixed forest were higher, but these values of Ulmus macrocarpa forest and Larix principis-rupprechtii forest were lower. The Patrick and Shannon-Weiner indices varied in order of herb layer > shrub layer > tree layer; the Pielou value of Ulmus pumila forest varied in order of herb layer > tree layer >shrub layer, and other community types varied in order of shrub layer >herb layer > tree layer.
Species diversity of functional layers changed differently along elevational gradient. In tree layer, the species richness, diversity and evenness decreased with increasing elevation. In shrub layer, the species richness, diversity and evenness were significantly peaked at the intermediate elevation. In herb layer, the species richness and diversity decreased with increasing elevation, but the evenness value changed little.
Applying multi-perceptron network to predict the diversity of functional layers, we found that aspect affected species diversity of tree and shrub layers greatly, but elevation affected species diversity greatly in herb layer.