Data utilization plays a vital role in the development of human societies. Today, the data are stored in different and heterogeneous sources. Providing an integrated view by fusion of the data is essential to enhance data utilization. Data quality is a challenging issue in the data fusion. One of the main challenges that affects the quality is data inconsistency, which has been addressed in this study. Some inconsistencies occur within a dataset. Some other inconsistencies occur between different datasets. This type of inconsistency is more likely because each dataset may have been generated by different people and updated at different times. People with different purposes, motives, and background knowledge may describe concepts or entities in diverse resources and as a result, inconsistency arises. Real inconsistencies between two sources can be handled with common techniques, such as voting. In this research, we worked on apparent inconsistencies that are caused by the difference in data representation. The purpose of this study is to manage this type of inconsistency to improve the entity recognition and data fusion quality. Entity resolution is an essential task in data fusion. Intelligent entity recognition improves the quality of data fusion and data veracity. Different criteria are defined for different data types in entity resolution process. One of the weaknesses of the criteria is the lack of attention to the level of ion and granularity of the data in different sources. The weakness causes conflicts and inconsistencies that if not managed properly, the process of identifying the entities will be erroneous and as a result the data fusion will not be sufficiently accurate. In this research, first, a framework is introduced for managing inconsistencies and data fusion. In the framework, using data granulation and knowledge bases, increases data veracity that is one of the new features of big data. The proposed algorithms belong to the category of online algorithms that is able to provide results by observing new records. It does not require the evaluation of the entire dataset from the beginning and just considers the necessary comparisons. In this work, granule and data granulation are key elements. Identification of spatial entities, formation of spatial granules, identification of the relationship between granules and clustering of them help to manage differences in data granulation. Creating a new semantic structure for spatial concepts and introducing a new criteria for measuring the quality of data based on data granulation are other achievements of this research. Blocking techniques significantly reduce the number of comparisons because only block members are compared with each other accurately in entity matching process. For this purpose, a new blocking method based on geographical features is introduced. An indispensable type of data is spatial data, with diverse application. This paper focuses on spatial data granulation. The proposed algorithms are based on spatial relationships between entities and have been tested on real air accident datasets. We developed and used an aviation accident datasets, but the proposed approach is not limited to this field and can be applied to almost any type of data that have hierarchy of concepts and values. By applying data granulation approach on several methods in the field of spatial data, the quality of data fusion based on the F-Score index has been improved.