Analysis of historical flood events in Denmark with information from digital news media

Abstract

Reliable information on historical flood events is critical for flood risk analysis, climate change adaptation, verification of forecast models, etc. Unfortunately, such information is often difficult to find, due to e.g. lack of monitoring equipment at the location of a flood. In Denmark, management of water has traditionally been the responsibility of local authorities, which means there is a limited national overview of historical events and their consequences. Previous studies have employed different strategies for compiling a flood event inventory, including mining information from (1) insurance data, (2) social media data, and (3) newspaper archives. The aim of this study is to exploit a comprehensive digital news media archive to compile an inventory of Danish flood events in the period 2007-2020 with information on the time and location of the event, to classify the type of flood, and note any available information on local consequences and damages. We have gained access to the company Infomedia’s large digital media archive, which consists of digitized articles from news sources ranging from major national newspapers to small, local outlets. The archive contains more than 75 million news articles with the earliest articles dating back to 1990. The archive is searchable through calls to an API with a custom search language that combine user-specified keywords. A hydrologist has read all articles that match the keywords, noting all the relevant information. 1,118 distinct flooded locations where identified over the 14-year period of 2007-2020. Results show that there is large year-to-year variability in the different types of floods. Urban pluvial floods are experienced somewhere in Denmark every single year, while the number of both fluvial and storm surge floods are very low (or entirely missing) in some years. Urban pluvial floods occur throughout the year but are highly concentrated in the summer months with a mean date of occurrence in late July, while storm surges are observed only between September and March with a mean date in mid-December. Fluvial floods are the least concentrated type of floods and occur throughout the year with a slight overweight in winter months (mean date in early January). The spatial distribution of floods is uneven with four out the 10 municipalities that experience the highest number of floods being located in Eastern Jutland (Vejle, Horsens, Kolding, Aarhus) and another four located in the Northern half of Zealand (Copenhagen, Roskilde, Gribskov, Holbæk). Storm surge events occur over large geographical areas and we therefore speculate that they are more likely to be reported in news media than urban pluvial floods, which are often local events due to the small-scale nature of convective rainfall cells. Ongoing work is trying to quantify these aspects and validate the individual flood events in the inventory using additional data sources.