Debris-flow is one of the major geological hazards in southwest China, which are a global threat and happen and results in thousands deaths and injuries and billions of dollars in damages globally. Automatic multi-temporal DEM coregistration for detecting terrain changes is an attractive but inherent very difficult research topic. Many methods have been proposed in recent years, but all of them can only deal with DEM with limited percentage of terrain changes. However, in landslide and debris-flow areas, the rate of terrain changes is very high. To solve such a problem, a new method for detecting terrain changes using local invariant patches is proposed in this paper. According to the character of the debris-flow activities, the peak and ridge are rarely affected. From where some invariant patches can be extracted associated by the feature extraction method. After co-registration these invariant patches, a coarse matching can be reached. Therefore, two DEM can be compared after applying this coarse matching. With an appropriate threshold, most of terrain changes can be eliminated, and then a fine matching can be anticipated reasonable. The accuracy terrain changes will be derived. The new method can estimate the terrain changes quantificationally and automatically, and verifies by a real application. The experimental results illustrate the proposed method is of robust, accuracy, and timeefficiency.