Application of artificial neural networks on extracting permafrost information

A back propagation artificial neural network was used to extract permafrost information in Huzhong National Nature Reserve, China. 400 field sites of permafrost information were investigated as training sites for the network and accuracy test for the classification results. Different combinations of input sources were used as training layers for the network, which included land cover, equivalent latitude, aspect, and radar image. Results of permafrost classification suggest that the neural network is available in the study area. The best combination to extract permafrost information includes land cover, equivalent latitude and radar image. But aspect may have little use in classifications of permafrost in the area.