Abstract
Accumulation of instream large wood (i.e., fallen trees, trunks, branches, and roots) at bridges during floods may exacerbate flooding, scour and cause structural failure. Yet, explaining and predicting the likelihood of a bridge trapping wood remains challenging. Quantitative data regarding wood accumulation at bridges are scarce, and most equations proposed to estimate the accumulation probability were derived from laboratory experiments , and include variables such as flow velocity, Froude number, and approaching wood volume or size which are difficult to obtain. Other evaluations based on technical reports and information regarding wood removal have been proposed but are mostly qualitative. Until now, a data-driven approach combining multiple quantitative accessible variables at the river reach and catchment scales remains lacking. As a result, the controlling parameters explaining whether a bridge is prone to trap wood are still unclear. This work aims to fill this gap by analysing a database of 49 bridges across the United Kingdom (UK) classified as prone and not prone to wood accumulation. The database contained information regarding the geometry of the bridge (i.e., number of piers and pier shape) and we added parameters describing the upstream river channel morphology, the riparian land-cover, and high-flow characteristics. We applied multivariate statistics and a machine learning approach to identify the variables that explained and predicted the predisposition of bridges to wood accumulation. Results showed that the number of bridge piers, the unit stream power, the pier shape, and the riparian forested area explained 87% of the total variability for the training dataset (0.87 training accuracy), and the selected model had a testing accuracy of 0.60 (60%). Although limited by the sample size, this study sheds light on the identification of bridges prone to wood accumulation and can inform bridge design and management to mitigate wood-related hazards.