A New Look to an Old Problem: Using Gaussian Mixture Models and Topo-Bathymetric LiDAR Data for Morphological Classification of Large Lowland Rivers

River morphological classifications tend to be integrative of a large range of parameters summarizing the general setting of the system studied and information focused on hydraulics, sediment transport and morphology. However, existing approaches are often subjective and not adapted to the diversity of fluvial patterns.One way to improve river morphodynamics and classification is based on recent development of remote sensing technologies such as topo-bathymetric LiDAR (LTB). This tool captures a high density of data (> 10 pts.m-2) with a centimetric accuracy over large spatial scales and on small time-periods. A survey carried out on the largest river of France, the Loire River between Nevers and Nantes (ca. 450 km) allowed testing the statistical potential of the LTB to investigate the morphological classification of contrasted river sectors (Garcia-Lugo et al., 2015). As a working hypothesis, we assume that the morphology of the channels of a river reach can be synthetized by a statistical distribution of dimensionless (and detrended) elevations and slopes which, when simplified, allows the delimitation of fluvial geomorphological units.Five sites characterized by contrasted morphological settings were retained for this work: 1) anabranching, 2) sinuous single channel, 3) braided, 4) formerly trained by groynes and 5) trained by groynes. Density curves were calculated for elevation of dimensionless detrended Digital Elevation Models (DEMs) and slope data. Simplification of elevation statistical distribution was achieved using a Gaussian Mixture Model (GMM) that divided the initial signal into a set of Gaussian functions corresponding to geomorphological units. Slopes statistical distributions were described using classical statistical parameters (skewness, kurtosis, median).It was shown that statistical distribution of dimensionless elevations varied in terms of shape and location on the x-axis of the plot of density curves. The shape gives an indication of lateral connectivity of the reach considered, while the location on the x-axis transcribes the predominance of low or high elevations of the system regarding the elevation magnitude. Description of an anabranching site led to the identification of secondary channels network. Braided reach shows a narrow distribution indicating a system with homogeneous relative elevation. The density curve of the site constrained by groynes is clearly bimodal traducing 2 disconnected and clearly visible morphological units. The site with removed or modified groynes showed a staggered aspect. Slopes distribution is also varying following reach morphology. Anabranching is showing the widest distribution while braided reach shows the narrowest.Results of this study are in line with literature concerning both braided and regulated rivers (Ashmore et al., 2013; Campana et al., 2011). For the anabranching site, methodology seems capable to describe efficiently channel staggering and allows the calculation of a lateral connectivity index. Results underline GMM capacity to discretize the initial signal into several corresponding to morphological units. The methodology is promising for the understanding of river morphodynamics and opens perspectives for river management and classification.