With the advances in sensing, communication, and computation, there is an increasing need to track mobile events such as air pollutant diffusion, toxic gas leakage, or wildfire spreading using mobile sensors such as robots. Lots of existing work use control theory to plan the path of mobile sensors by assuming that the event evolution is known in advance. This assumption has severely limited the applicability of existing approaches. In this work, we aim to design a detection and tracking algorithm that is capable of identifying multiple events with dynamic event signatures and providing event evolution history that may include event merge, split, create and destroy. Simulation results show that our approach can identify events with low event count difference, high event membership similarity, and accurate event evolution decisions, while using a reasonable number of tracking robots.