Marine versus terrestrial predictive mapping: Geographic modelling constraints of working underwater

Maps of seafloor characteristics are in high demand for marine conservation and planning, including: marine park placement and zoning (Friedlander et al., 2003); marine resource management (Bax et al., 1999); human impact monitoring (Kendrick et al., 2002; Plummer et al., 2003); and integrative ecological research (Durako et al., 2002; Fonseca et al., 2002). Predictive habitat modelling is emerging as a useful approach to meet the requirements of these applications, showing promise for mapping stationary biological seafloor inhabitants with hydroacoustics in water too deep for optical remote sensing (Figure 1). In its most common form, predictive habitat or species modelling refers to the formal statistical modelling of relationships between the occurrence of an organism or class (typically presence / absence dependent variables) and quantifiable environmental drivers, which can then be applied to full coverage datasets to produce spatial predictions (Guisan & Zimmerman 2000). Techniques developed for terrestrial applications are being applied in marine settings, but fundamental differences in the ecology and sample collection lead to an uneasy marriage of terrestrial methods and marine data. This paper presents geographic analysis and modelling issues in marine predictive habitat modelling, which result from or are exacerbated by differences between terrestrial and marine systems. Examples are supplied from current marine mapping projects in southeast and western Australia.