In the field of spatial information sharing, with different cognitive styles of the real world, people will focus on different aspects when they observe and describe the same geographical object. It will lead to semantic heterogeneous. The semantic grid gives an ideal solution to this issue. This paper discusses the necessity of semantic-oriented metadata management in semantic grid firstly. Existing ontology-based spatial information sharing systems almost assume a centralized global ontology to access spatial information. In WAN, the hypothesis will not realize. Focus on the status of spatial information sharing, from the view of actual needs, this paper proposed a semantic-oriented metadata management model which can be extended. The model combines the features of ontology and encodes metadata semantic information and related managements elements of distributed, heterogeneous grid node computing pools. It has three layers, the grid node computing pool layer, the metadata management layer and the metadata recording layer. The most important part is the semantic metadata mapping service. The service transforms ontology metadata concepts between domain ontology metadata base and common ontology metadata knowledge base so as to all users of each field can access the spatial information from every grid node computing pool which has registered in metadata management layer and ensure that all operations are based on semantic. The carrier of semantic information-ontology plays a leading role in all operations. It guarantees the userspsila semantic can be transform to every aspect; avoids semantic distortion and realized semantic-oriented information integration and interoperability. At last, one example provides a detailed way on how to implement the semantic metadata mapping service in this model. In conclusion, the adoption of the model makes each grid node computing pool in grid only need safeguarding the information resources itself. By the collaboration, all grid no- des can discover, understand and use entire distributed, heterogeneous and constantly changing information resources in grid. It solves the differences in metadata definition and standards among different units and enables users to access spatial information systems freely from the impact of semantic differences.