Fragmentation is a common but critical problem of big data analytics. Although big data analytics have the 5-v benefit, including volume, variety, veracity, velocity and value, fragmentation may make such benefit illusory. The benefit of big data analytics is based on the complete coverage of data collection and the sharing of data among multiple levels and multiple branches of the organization. A field investigation is conducted to test the big data analytic systems of an environmental protection agency of a province in China at the level of province, prefecture, and county. This field investigation reveals that the data collection coverage fails to cover some essential aspects, particularly when such data are generated by other governmental branches. The field investigation also reveals that the data are not shared among different levels or different branches of the government, which leads to the problems of fragmentation. Further investigation reveals that this fragmentation mainly lies in the insufficiency of prediction in the design of database framework and the fragmental purchase of big data analytic services. This paper recommends predictive designing in the initial design of the framework of big data analytic systems so that it may better accommodate changes in the future.