Discovering variables and understanding their relations, which impacts emergency response, provide important knowledge to the development of decision models, e.g. Bayesian networks, in autonomous typhoon response systems (ATRS). Given the text inputs (containing natural language), learning the network structures still remains a challenge although learning Bayesian networks from data has been extensively investigated in various fields. In this demo, we develop a deep learning based framework for identifying typhoon relevant variables and build their causal relations from text. We use the CausalBank dataset and typhoon specific relation rules to refine the learned relations and allow users to further improve the models through their domain knowledge. We integrate the new learning tool into the existing ATRS and demonstrate the empirical results through real-world typhoon reports.