Abstract: This paper presented the research on natural forest classification and mapping in China based on NDVI time series datasets derived from NOAA AVHHRR pathfinder data, and some ancillary geographical data. The authors first had a land cover classification by applying an ISODATA clustering method on some NDVI matrices indices from NDVI time series data, then detected the natural forest border through classes combination based on vegetation type map, TM image and field sampling point data. Secondly, in order to improve the classification accuracy, the authors performed a PCA transform on the NDVI time series data to remove noises. Finally, combined with other geographic data and field investigation data, the authors trained and constructed a neural network classifier to get the original forest classification map in China, then based on forest types, phenology character, China climatic division map and the national second and third classification system on vegetation, through classes combing, the forest type map in China was obtained. The final classification result showed that the classification method used in the paper can not only give a more detailed forest classification, but also have a better classification accuracy.

Key words: NDVI, forest mapping, neural network, time series