• Guidance

Filling gaps in marine data using Gradient Forest models

This describes how large numbers of different data sets can be effectively pooled to group marine species into 'community assemblages'. These assemblages can be used to identify biodiversity hotspots and explore trade-offs resulting from management decisions.

Aotearoa New Zealand’s waters cover a vast area, 4.1 million km2, most of which is deep sea.

View/download guidance

While many coastal areas and some offshore areas like the Chatham Rise have been well characterised, large parts of the EEZ (exclusive economic zone) remain unsampled as marine surveys in deep, offshore habitats are logistically difficult and expensive.

The lack of comprehensive information about what species are where, and in what numbers, makes it difficult to:

  • Understand the distribution of most species, particularly rare species or those found only in a few locations
  • Identify biodiversity ‘hotspots’
  • Make robust management decisions about resource use and conservation
Predicting marine biodiversity across Aotearoa New Zealand's vast marine area

We used a Gradient Forest model to accurately model how 253 species of bottom-feeding (demersal) fish species were grouped together over a range of environments within the New Zealand Continental Shelf Zone to depths of 2,000 m.

The model pooled a large number of existing data sets, combining environmental information with demersal fish occurrence records (data of where a species has/has not been found). It then grouped fish species according to the oceanographic and environmental conditions in which they live – ie species that tend to be found in the same habitats (eg cold, deep water) at the same time.

We used these groups, or ‘community assemblages’, as a proxy to estimate patterns of species diversity in inaccessible areas – using information from common species to ‘fill the gaps’ where data was limited for less common species.

We validated the model with independent data, which showed it was very robust at predicting patterns of species composition (species that make up the community assemblage) and turnover (the change in species composition over space).

We, jointly with a NIWA project, then used our 30 demersal fish community assemblages to determine optimal locations for biodiversity conservation, and to explore trade-offs between resource use and biodiversity conservation.

Summary of the research

  • 207,000 records of occurrence from 27,400 locations, showing presence/absence of 253 demersal fish species from research trawl surveys in the waters surrounding New Zealand
  • Detailed environmental data eg tidal current speeds, temperature gradients, salinity, seafloor, roughness, sediment type
  • 30 community assemblages with clear differences in environmental characteristics were used to
    • Determine biodiversity hot spots, and optimal locations for biodiversity characteristics
    • Explore trade-offs between resource use and biodiversity conservation

Advantages of this approach

  • It requires less data to run than considering 100s of species individually
  • It provides a limited number of groups of species (community assemblages) for decision makers to consider, which is:
    • Easier than an individual assessment for 100s of species
    • More holistic as these species interact and affect one another
  • It predicts assemblages that serve as proxies for rare species that cannot be modelled as they are poorly represented in available data
  • Filling gaps in marine data with GF models - best for print

    4 MB | Adobe Acrobat PDF file

  • gradient forest models

    430 KB | PNG image - good general-purpose format