A novel unsupervised method for assessing mesoscale river habitat structure and suitability from 2D hydraulic models in gravel-bed rivers

Application of mesoscale habitat models in gravel-bed rivers is increasingly common for a variety of purposes, from ecological flow design, impact assessment and conservation programmes. Integration with 2D hydraulic modelling offers the potential for broader applicability of mesoscale habitat models, extending applications to larger streams and nonwadable flow conditions, when on-the-ground and in-stream surveys are challenging or even prohibitive. In this work, a novel fully unsupervised procedure that allows the segmentation of the river channel area at a given flow condition at a scale that is consistent with the mesoscale is presented. Further, it defines an objective methodology to choose segmentation parameters and thus an optimal segmentation, based on intrinsic spatial properties of the resulting regions. Segmentation parameters are objectively selected by minimising a Global Score, which is based on three metrics representing intrasegment homogeneity, intersegment heterogeneity and an optimal range of segment numbers based on an empirically defined mesoscale. Applications of the model are tested on two reaches of the multithread Mareta and meandering Aurino Rivers in South Tyrol (NE, Italy). Model outcomes are compared with ground mesohabitat surveys, and habitat suitability is then assessed for three fish species (marble trout, grayling and European bullhead). A high level of agreement is found when comparing model- and survey-based habitat suitability estimates, with an overall value of R2=0.91$$ {R}<^>2=0.91 $$. The proposed approach shows potential for application of the mesohabitat concept for large gravel-bed rivers and nonwadable flow conditions. By allowing habitat estimates at flow ranges that could not be surveyed in-stream, the approach facilitates applicability of mesoscale habitat models to nonwadable conditions and large streams. The workflow is river-independent and fully unsupervised, as it does not require calibration or subjective choices of segmentation parameters. Significance statement Quantifying suitable habitat for riverine fauna is increasingly used for ecological flows assessment, with the use of mesoscale habitat modelling approaches becoming more common in the past few decades. Existing mesoscale habitat modelling approaches often rely on field surveys, which, however, become prohibitive at nonwadable flow conditions and in large streams. Here we develop and test against field data a fully unsupervised approach able to extract mesohabitats and their hydraulic characteristics from the outputs of 2D hydraulic models. Compared with existing approaches, our approach allows an automated segmentation of the wetted reach into mesoscale units through the implementation of an unsupervised optimality step in which optimal segmentation parameters are defined, and a final mesohabitat mosaic is selected. The methodology makes it easier to expand the applicability of mesoscale habitat modelling to a broader range of river sizes and conditions.