Big Data are collections of data sets so large and complex to process using classical database management tools. Their main characteristics are volume, variety and velocity. Big Data integration is a new research area that faces new challenges due to these characteristics. Ontologies represent knowledge as a formal description of a domain of interest. They are widely used in data integration. This paper illustrates an approach for ontology based Big Data integration taking into account their characteristics. Our approach is based on a NOSQL database namely MongoDB and modular ontologies. It follows three steps: wrapping data sources to MongoDB databases, generating local ontologies, composing the local ontologies to get a global one. A tool implementing the generation of the local ontologies is also detailed.