Social media and social networks are nowadays frequented by many millions of people providing a huge amount of various kinds of information. Processing of data coming from social platforms has attracted attention during last years due to the widespread availability of data and the ease to access to them. These data can be extracted and analysed to detect relevant information in different application contexts. Among social networks, Twitter is one of the most popular and fastest-growing blogging service able to provide relevant information for situation awareness and decision making support. This paper presents a novel approach for detecting events analysing real-time data incoming from the Twitter stream exploiting: tweets semantic analysis, rule based classification, and time-space detection models. The proposed approach has been implemented and applied in the detection of hazardous natural events and related consequences. Results of the application of the implemented software tool to a real scenario are reported.