Distributions of individual species, patterns of variability in species richness and abundance, and locations of sensitive or vulnerable habitats are essential inputs into marine spatial planning and risk assessment processes. Chatham Rise is an important deep-sea fishing region in New Zealand. Lying at the convergence of Sub-Tropical and Sub-Antarctic water masses, it has a highly diverse and dynamic physical environment, supporting high levels of biological production and encompassing a broad range of benthic habitats and fauna. Existing knowledge about seabed faunal distributions on Chatham Rise comes from records of museum specimens, fisheries and research trawl bycatch, and increasing from photographic surveys. Data from museum and trawl databases have been used to build models that predict species and community distributions in unsampled space but because the models are based on presence-only data from disparate sources and do not incorporate population density data, their predictions are considered uncertain.

To reduce uncertainty in predictions, we developed a new, spatially extensive, fully quantitative, and taxonomically consistent dataset of benthic invertebrate occurrence by merging data from five seabed photographic surveys, including an extensive dedicated survey (TAN1701) as part of this project (Bowden et al., 2019a). We then used this dataset to inform development of improved predictive models for Chatham Rise at both single-taxon and community levels, yielding maps of predicted population densities, beta-diversity (rate of change of community composition), and community classifications (Bowden et al., 2019b). Two independent modelling methods were used for each level: Boosted Regression Trees (BRT, De’ath 2007) and Random Forests (RF, Breiman 2001) for single-taxa, and Regions of Common Profile (RCP, Foster et al. 2013) and Gradient Forests (GF, Ellis et al. 2012) for communities, enabling ensemble model predictions for single taxa and comparison between classification methods for communities. For single-taxon models, the ‘hurdle’ model technique was used, combining predictions from presence-absence and abundance models to reduce bias associated with zero-inflated data. Sets of explanatory environmental variables (12 for single-taxon models, 18 for GF, and 9 for RCP) were selected from an initial set of 58 candidate layers and the 354 invertebrate taxa identified from the seabed image surveys were condensed into a set of 69 taxa by aggregation to higher taxonomic levels and exclusion of rarer and non-benthic taxa. Single-taxon models were produced for 20 taxa, selected according to their sensitivity or vulnerability to human-induced environmental impacts, while all 69 taxa were included in community models.

Outputs from the single-taxon models are presented as maps showing predicted occurrences as densities (individuals 1000 m-2) with associated estimates of model precision (CV) and cross-validation metrics. All models performed well by these criteria but a comparison using invertebrate bycatch data from the trawl database was inconclusive for most taxa modelled because of inadequate abundance information in the test data. While predictions for most of the taxa modelled have clear similarities with those of previous models, they also show differences, often driven by inclusion of density data. Outputs from the community models are presented as spatial classifications of the study area, analogous to existing spatial classifications such as the Marine Environments Classification (MEC, Snelder et al. 2007, Leathwick et al. 2012) and derivatives. RCP divided the area into 7 classes, whereas a hierarchical clustering method allowed GF results to be assessed at class levels from 7 to 50 classes and compared visually against existing classifications.

These predictions are the best-informed representations of seabed distributions at regional scales in the New Zealand Exclusive Economic Zone to date and provide a resource that will have applications in marine environmental management and ecosystem research. Potential applications include quantification of benthic impacts from bottom-contact fishing gear and other anthropogenic agencies, informing spatial management of biodiversity through, for example, the design of marine protected areas, and informing research into ecosystem linkages between water-column and seabed processes. A further obvious application and test of the predictions will be to use the modelled relationships developed here to predict faunal distributions across seabed areas beyond Chatham Rise.

This study is funded by Fisheries New Zealand (FNZ) under projects ZBD2016-11 and ZBD2019-01, with governance at FNS by Mary Livingston. The Principal investigator is David Bowden (David.Bowden@niwa.co.nz ) and the full team includes: Owen Anderson; Caroline Chin; Malcolm Clark; Niki Davey; Alan Hart; Andrea Mari; Andrew Miller; Ashley Rowden and Brent Wood.

References:

Bowden, D.A.; Rowden, A.A.; Anderson, O.F.; Clark, M.R.; Hart, A.; Davey, N., . . . Chin, C. (2019a). Quantifying Benthic Biodiversity: developing a dataset of benthic invertebrate faunal distributions from seabed photographic surveys of Chatham Rise. Aquatic Environment and Biodiversity Report No. 221. 35 p.

Bowden, D.; Anderson, O.; Escobar-Flores, P.; Rowden, A.; Clark, M. (2019b). Quantifying benthic biodiversity: using seafloor image data to build single-taxon and community distribution models for Chatham Rise, New Zealand. Aquatic Environment and Biodiversity Report No. 235. 67 p.

Breiman, L. (2001). Random forests. Machine Learning 45(1): 5-32

De'ath, G. (2007). Boosted trees for ecological modelling and prediction. Ecology 88(1): 243-251.

Ellis, N.; Smith, S.J.; Pitcher, C.R. (2012). Gradient forests: calculating importance gradients on physical predictors. Ecology 93(1): 156-168.

Foster, S.D.; Givens, G.H.; Dornan, G.J.; Dunstan, P.K.; Darnell, R. (2013). Modelling biological regions from multi-species and environmental data. Environmetrics 24(7): 489-499.

Leathwick, J.; Rowden, A.; Nodder, S.D.; Gorman, A.R.; Bardsley, S.; Pinkerton, M., . . . Goh, A. (2012). A Benthic-Optimised Marine Environment Classification (BOMEC) for New Zealand waters. New Zealand Aquatic Environment and Biodiversity Report No. 88. 54 p.

Snelder, T.H.; Leathwick, J.R.; Dey, K.L.; Rowden, A.A.; Weatherhead, M.A.; Fenwick, G.D., . . . Zeldis, J.R. (2007). Development of an ecologic marine classification in the New Zealand region. Environmental Management 39(1): 12-29.

NIWA

See:

Bowden, D.A.; Rowden, A.A.; Anderson, O.F.; Clark, M.R.; Hart, A.; Davey, N., . . . Chin, C. (2019a). Quantifying Benthic Biodiversity: developing a dataset of benthic invertebrate faunal distributions from seabed photographic surveys of Chatham Rise. Aquatic Environment and Biodiversity Report No. 221. 35 p.

Bowden, D.; Anderson, O.; Escobar-Flores, P.; Rowden, A.; Clark, M. (2019b). Quantifying benthic biodiversity: using seafloor image data to build single-taxon and community distribution models for Chatham Rise, New Zealand. Aquatic Environment and Biodiversity Report No. 235. 67 p.