Facing the challenges with current tornado warning systems, we explore alternative approaches. Specifically, we present a database engineered system that integrates information from heterogeneous rich data sources, capturing both climatology data for tornadoes and those data just before a tornado warning. Such a system aids in predicting tornado occurrences by identifying the data points that form the basis of a tornado warning. Applications of this system to US data highlights the advantages of using a classification forecasting recurrent neural network (RNN) model in our system. The application results also highlight the effectiveness of our database engineered system for big data analytics on tornado climatology—especially, in accurately predicting tornado lead-time, magnitude, and location.