Enhancing an Unsupervised Clustering Algorithm with a Spatial Contiguity Constraint for River Habitat Analysis

The spread of two-dimensional numerical hydrodynamic tools for ecohydraulic applications allowed for the development of automatic habitat detection methods, adopted as predicting tools for river habitat analysis. These automatic approaches differ for the employed identification rules, such as preference curves, fuzzy rules and clustering methods. Previous research has shown promising results in the automatic identification of mesoscale habitat patches by using clustering algorithms together with numerical hydrodynamic model results. These algorithms attempt to implement and simulate some of the expert-based requirements adopted in the field to delineate habitat patches. Spatial contiguity is one of such expert-based requirements that has not been enforced and exploited in automatic mesohabitat identification so far. In this work, we propose a novel tool (BASEmeso) based on an agglomerative hierarchical clustering algorithm where we enforced a spatial contiguity criteria. We compare our approach with a more established method without spatial constraints, considering a synthetic river reach where the composition of mesohabitat patches is known a priori, and on three experimental river reaches, to investigate the effects of different river morphologies. Our results show that when employing a contiguity constraint, a patch's extent is better captured, different patches can be distinguished better and the distribution of patch characteristics is smoother. This holds for all investigated morphologies. Together, it suggests that including a spatial contiguity constraint can improve the automatic delineation of river mesohabitat patches. The proposed methodology could positively contribute in the development of automatic, objective and predictive meso-scale habitat assessment workflows.