Floods are the most common and among the most severe natural disasters in many countries around the world. As global warming continues to exacerbate sea level rise and extreme weather, governmental authorities and environmental agencies are facing the pressing need of timely and accurate evaluations and predictions of flood risks. Current flood forecasts are generally based on historical measurements of environmental variables at monitoring stations. In recent years, in addition to traditional data sources, large amounts of information related to floods have been made available via social media. Members of the public are constantly and promptly posting information and updates on local environmental phenomena on social media platforms. Despite the growing interest of scholars towards the usage of online data during natural disasters, the majority of studies focus exclusively on social media as a stand-alone data source, while its joint use with other type of information is still unexplored. In this paper we propose to fill this gap by integrating traditional historical information on floods with data extracted by Twitter and Google Trends. Our methodology is based on vine copulas, that allow us to capture the dependence structure among the marginals, which are modelled via appropriate time series methods, in a very flexible way. We apply our methodology to data related to three different coastal locations on the South coast of the United Kingdom (UK). The results show that our approach, based on the integration of social media data, outperforms traditional methods in terms of evaluation and prediction of flood events.
展开▼
