Empirical assessment of the relationship between flood events, and land use change

In the region of Flanders in Belgium 5% of the territory was recorded to be subjected to at least one flood event between 1988 and 2016 (1) (2). Major flood events have occurred for example in 2010 and, most recently, in the spring of 2016, when exceptional weather events (with a record amount of precipitation in June 2016) led to sometimes repeated flooding of residential and agricultural areas (2). Experts warn that such extreme weather events will occur more frequently in coming decades due to climate change (3). Moreover, Flanders is one of the most urbanized regions in Europe. In 2009 12.9% of the total area in Flanders was sealed, which is notably higher than the European average of 1.8% and the Belgian average of 7.4% (3). A recent study has also shown that the area of built up parcels is currently increasing at a rate of 6 ha per day (4). There is a growing awareness (4) (5) that the high rate of surface sealing in- and outside flood prone areas affects the frequency and severity of flood events, since rainfall is increasingly unable to infiltrate in the soil, leading to ponding and to causing runoff to increase and accumulate downstream. The combined effects of climate change and rapid urbanization therefore mean that floods are likely to occur more often in Flanders and lead to increasingly severe economic and social damages (6) (3). Currently, the Flemish Environment Agency estimates that floods in Flanders cause a yearly, average damage of over 50 million euros (3). It is predicted that these damages will rise with 50% by 2050 if no extra measures are taken (7).

In this paper, we empirically test for the case of Flanders the now widely accepted but poorly scientifically documented hypothesis that the hydrological traits and spatial configuration of upstream land use systems affect the frequency and spatial characteristics of flood events. This is done by analyzing geodata about documented flood events, dating back to 1988, in a multivariate regression analysis on the one hand and a spatially explicit machine learning approach on the other hand through which the effects of land use distribution is highlighted against these of catchment characteristics and antecedent meteorological conditions.