The knowledge graph has attracted wide attention in recent years. At first, it is proposed to enhance the search function. Then with the progress of its research, the knowledge graph has become the foundation and bridge of semantic retrieval. There are still many challenges in the process of constructing a knowledge graph. Compared with the construction of a knowledge graph in the open areas, it has more domain characteristics in the restricted areas. For example, during the construction of a knowledge graph in the field of disaster news, the geographical name entities are very important and they have the characteristics of hierarchical nature, which brings great convenience to construct the graph. This paper proposes a method to construct a knowledge graph of disaster news based on an address tree. The address tree is corresponding to each news corpus and obtained by using geographical name characteristics. Then a knowledge graph is obtained step by step by fusing other kinds of knowledge based on the address tree. Based on this method, a prototype system of disaster knowledge graph is implemented and it illustrates the feasibility of the method.