Land cover change is an important scientific issue for the land evaluation and eco-environmental change forecasting.It is the necessary means to study the eco-environmental changes in coal fire zone by acquiring high precise land cover map through accurate classification approaches.In this paper

Maximum likelihood classification（MLC）

Spectral Angle Mapping（SAM）

object-oriented classification（OOC） and the multi-level classification based on compound subregion（MCBCS） approaches are used to classify land cover in the Wuda coal fire area.The results show that the multi-level classification based on compound subregion method leads to the highest accuracy up to 92.97%and the Kappa Coefficient is 0.9155.This method segments the study area based on the thermal characteristic

thermal anomalies

landscape and the disturbing to ecosystems.It reduces the confusion among different landcover types

emphasizes the zonal and regularity of land cover of the coal fire area in Gobi

and increases the separability of land cover.The multi-level classification based on compound subregion increases the whole accuracy by the accuracy improvement of single land cover.