Quantifying the effects of variable selection, spatial scale and spatial data quality in marine benthic habitat mapping

Mapping benthic habitats has become critical in many contexts like conservation and management. While marine habitat mapping methods strongly rely on tools and methods from geography and geomatics, habitat mapping practitioners with a background outside of these specialized areas do not always have a full understanding of the spatial concepts behind these tools and methods. This phenomenon is amplified when marine geomorphometry, the science used to quantify seafloor terrain characteristics, is integrated into the marine habitat mapping workflow. This dissertation reviews the use of spatial concepts in the field of marine benthic habitat mapping; many concepts are poorly understood or poorly implemented in the habitat mapping workflow, among which spatial scale and spatial data quality stand out as being of particular importance. While geomorphometry is commonly used in marine benthic habitat mapping, no framework existed to test which terrain attributes should be used as surrogates of species distribution, leading to an inability to compare results from different studies. This dissertation explores different options for terrain attribute selection and proposes an optimal combination that can be used as standard in all habitat mapping studies. This selection is then tested using two approaches to benthic habitat mapping and is shown to perform better than others. Bathymetric data, the primary input for marine geomorphometry analyses and one of the main data inputs for habitat mapping, are commonly impacted by data acquisition artefacts. Very little work has been done on trying to understand how these artefacts propagate throughout the habitat mapping workflow. The impact of artefacts on the bathymetry and its derived terrain attributes is described, and it is shown that artefacts modify the spatial and statistical distributions of depth and terrain attribute values. However, when these affected data are used in habitat mapping, their impact is not always predictable. Some artefacts were found to sometimes inflate measures of accuracy and performance and sometimes decrease them. Overall, habitat maps were shown to be very sensitive to the effects of variable selection, spatial scale and data quality, and as such have serious implications when they are used to inform decision-making, for instance in marine conservation and management. This dissertation raises awareness about these issues and highlights the need for careful integration of spatial data in habitat mapping practices