An atlas of seabed biodiversity for Aotearoa New Zealand

Abstract

Abstract. The waters of Aotearoa New Zealand span over 4.2 million km2 of the South Pacific Ocean and harbour a rich diversity of seafloor associated taxa. Due to the immensity and remoteness of the area, there are significant gaps in the availability of data to quantify and map the distribution of seafloor and demersal biodiversity, limiting effective management. In this study, we describe the development and accessibility of an online atlas of seabed biodiversity that aims to fill these gaps. Species distribution models were developed for 579 taxa across four taxonomic groups: demersal fish, reef fish, subtidal invertebrates and macroalgae. Spatial layers for taxa distribution based on habitat suitability were statistically validated and then, as a further check, evaluated by taxonomic experts to provide measures of confidence to guide the future use of these layers. Spatially explicit uncertainty (SD) layers were also developed for each taxon distribution. We generated layer-specific metadata, including statistical and expert evaluation scores, which were uploaded alongside the accompanying spatial layers to the open access database Zonodo. This database provides the most comprehensive source of information on the distribution of seafloor taxa for Aotearoa New Zealand and is thus an invaluable resource for managers, researchers and the public that will guide the management and conservation of seafloor communities.

Modelling spatial distributions of biogenic habitat‐forming taxa to inform marine spatial planning

“…In total, 30 environmental variables were available to predict biogenic habitat‐forming taxa group distribution. The suite of variables represents a wide range of substrate, water chemistry and physical and topographical characteristics at the seafloor that have been used for previous modelling of seafloor invertebrates, fish and macroalgae distributions in New Zealand (Stephenson et al, 2020; Stephenson et al, 2023). Notable exclusions due to data availability include information of stressors (e.g.…”

“…Variables included bathymetry and its derivatives (topographic proxies), as well as variables related to light, current and food availability (productivity). An explanatory variable selection process was used to select variables for each respective biogenic habitat‐forming taxa group (Stephenson et al, 2023). First, an RF model was fitted using conditional permutation of variable importance in the extendedForest package (Liaw & Wiener, 2002) in R. This method accounts for any co‐linearity in explanatory variables when determining the relative importance of each variable in the model through the implementation of a conditional approach to calculation of variable importance (Ellis, Smith & Pitcher, 2012; Strobl et al, 2008).…”

“…Despite acceptable model fits, three models out of 20 were deemed inaccurate following an expert evaluation process (mussels, Brachiopoda and calcareous tubeworms). This showcases the value of obtaining expert assessment of the accuracy of predictions as this can guide whether modelled outputs are incorporated in future planning exercise analyses (Bennion et al, 2023;Stephenson et al, 2023).…”

“…The working group comprised environmental non-governmental organization, New Zealand government agency and fishing industry representatives. Spatial prioritisations were generated using the modelled outputs described herein but were complemented by other modelled layers of species distributions (Stephenson et al, 2023) and compositional turnover (Stephenson et al, 2020), fishing industry value and effort layers (MPI, unpublished data) and information of historical fishing effort (Rowden et al, n.d.). The New Zealand government has subsequently released proposed options for bottom fishing access zones (referred to as 'trawl corridors') where bottom fishing would continue to be permitted (FNZ, 2023).…”

“…Experts were also asked to provide a score of their level of knowledge (1 = Very high to 4 = Low) and a score for the distribution of occurrence records (1 = Very accurate to 4 = Inaccurate). This approach closely follows the expert evaluation process detailed byStephenson et al (2023); see supplement therein for more information.T A B L E 2 Model fit metrics for the random forest (RF) and boosted regression trees (BRT) models for 20 biogenic habitat groups.…”