Towards the operational use of tweets data in high impact weather scenarios: data mining and analytics in Basque Country.

Although social media industry is now a very congested Marketplace, Twitter continues to maintain its status as a popular social media platform. There are 330 million monthly active users and 145 million daily active users on Twitter sending more than 6,000 tweets every second in the world. In Spain case 85% population are social media users, with around 5 million tweeter profiles for a population around 47 million. In the autonomous community of Basque country (2.17 million inhabitants) around 20% of citizens use Twitter.

Twitter is a social tool that enables users to post messages (tweets) of up to 280 characters supporting a wide variety of social communication practices including photo and video attach. The Basque Meteorology Agency @Euskalmet with more than 115,3 K followers is one of the most popular accounts in Basque Country. Twitter is not only an opportunity to instantaneous spread messages to people without intermediaries, but also as a potential platform for valuable data acquisition using tweeter API capabilities. In this contribution, we present a study of different aspects related to the operational use of Twitter data in the context of high impact weather scenarios at local level.

The most important activity in Euskalmet are actions in severe weather events. Before the event, mainly centered in forecast and communication, during the event in nowcast, surveillance and impact monitoring and after the event in post-event analysis. During all these complex processes real time tweets posted by local users offer a huge amount of data that conveniently processed could be useful for different purposes. For operational staff, working at office during severe weather episodes, is critical to understand the local effects that an adverse phenomenon is causing and the correct perception of the extent of impact and social alarm. For this purposes, among others, different information associated with posted tweets can be extracted and exploited conveniently. In this work, we present some results that demonstrate how different data mining and advances analytics techniques can be used in order to include social media data information for different tasks and particularly during high impact weather events.

In this paper we summarize our experience during a proof of concept project for automatic real time tweeter analysis and the development of an operational tool for tweeter API data exploitation in the Basque Country. We present the main challenges and problems that we have had to face, including how to deal with the lack of geolocation information, since in the case of the Basque country, as in other parts of the world, tweets containing geotags are the exception, not the rule.