War landform mapping and classification on the Verdun battlefield (France) using airborne LiDAR and multivariate analysis

Acting as efficient earth-movers, soldiers can be viewed as significant geomorphological drivers of landscape change when replaced in the recent debates on Anthropocene Geomorphology. 'Polemoforms', generated by military activities, correspond with a set of human-made landforms of various sizes and geometries. They are particularly common on the World War One battlefield of Verdun (France) which ranks among the largest battles of attrition along the Western Front. The artillery bombardments and building of defensive positions in that battle significantly altered the landscape, resulting in thousands of shell craters, dugouts, and gun positions that have altered both the meso and microtopography. This paper proposes an innovative methodology to make an exhaustive inventory of these small-scale conflict-induced landforms (excluding linear features such as trenches) using a digital terrain model (DTM) acquired by airborne LiDAR on the whole battlefield. Morphometric analysis was conducted using Kohonen's self-organizing maps (SOMs) and hierarchical agglomerative clustering (HAC) in order to quantify and classify the high number of war landforms. This combined approach allowed for mapping more than one million landforms which can be classified into eight different shapes including shell craters and various soldier-made landforms (i.e. shelters, gun positions, etc.). Detection quality evaluation using field observations revealed the algorithm successfully classified 93% of shell craters and 74% of anthropologically constructed landforms. Finally, the iconographic database and map series produced will help archaeologists and foresters to better manage the historic site of Verdun, today covered by a large forest of similar to 10 000 ha. (c) 2019 John Wiley & Sons, Ltd.