Abstract: [Objective] This paper puts forward a new method for identifying burned areas and fire intensity by using Landsat TM images and RS indices to construct the decision tree classification model. In combination with topographic factors such as slope, aspect and elevation the spatial distribution of fire severity was scientifically and systematically analyzed in this study to provide theoretical basis and data support for forest fire prevention and management in Daxing'anling Mountains.[Method] In this paper, Huzhong region of the Daxing'anling Mountains was targeted. TM images of post-fires in September 2010 and September 2007 were taken as the basic data, and DEM images and forest type maps were used as the auxiliary data. The NDVI, NDSWIR, MNDWI, dNBR and other RS indices were employed to build a decision tree classification model which then was used to identify ten burned areas of Huzhong in 2010. Fire severity was divided into four classes according to the threshold value of dNBR, and the Arcgis software was used to do an overlaying analysis on the fire severity map with slope, aspect, elevation.[Result] The overall accuracy and Kappa coefficient of the decision tree classification were 97.97% and 0.943 2. Compared with the Parallelepiped method and ISODATA method, the total classification accuracy was increased by 7.56% and 17.32%, respectively. The Kappa coefficient was also increased. In the decision tree method, the producer's accuracy and user's accuracy were 97.51% and 97.54%, the Parallelepiped method were 90.43% and 96.52%, and the ISODATA method were 94.35% and 95.68%. Fire severity was divided into four classes according to the threshold of dNBR:unburned, low, moderate and high. Moderate severity burned area accounted for 46.6% of the total, and high severity burned area was 33.2%. After overlaying analysis, 64.4% (4 177 hm²) of burned area located at the elevations from 1 000 m to 1 500 m, and 45.9% of burned area located at level Ⅲ slope (6°-15°). The burned area at the southern slope occupied 21.4% (1 391 hm²) of the total.[Conclusion] The decision tree classification model presented in this paper could identify burned areas accurately and the total classification accuracy was higher than the parallelepiped method and ISODATA method, and the burned area is closer to the method of visual interpretation. Moderate and high severity burned areas occupied most of the total burned areas, and there were some relations between the burn severity and slope, aspect, elevation.

Key words: burned areas, decision tree classification, fire severity, fire size, dNBR