Standard Talk (15 mins) Australian Society for Fish Biology Conference 2022

Habitat partitioning and machine learning help map species distributions in an iconic crustacean (#162)

Matthew McMillan 1 , Susannah Leahy 1 , James Daniell 2 , Nora Louw 2 , Eric Roberts 2 , Matthew Campbell 1
  1. Queensland Dept of Agriculture & Fisheries, Brisbane, QLD, Australia
  2. James Cook University, Townsville, Qld, Australia

Knowledge of species’ distributions is critical to many aspects of marine science and management. However, distribution data are often lacking, requiring modelled interpolations to ‘fill in the gaps’. In the case of benthic animals, relationships between habitat type and abundance are often strong and can be used to inform such models. Two species of Moreton Bay bugs, or bay lobsters, are not differentiated in historical fisheries records, complicating efforts to assess long-term trends in their abundance. We used Random Forest models to map sediment properties for the entire Queensland east coast and inform species habitat preferences based on existing and newly collected data. These preferences informed a Boosted Regression Tree model used to predict and map distributions of each species. We found strong habitat partitioning with smaller Thenus parindicus preferring shallower inshore areas and finer sediments and larger T. australiensis preferring deeper areas with coarser sediments. As a result, both species could be differentiated in the historical catch records, allowing the assessment of long-term trends in their abundance. Similar approaches may offer potential to map distributions of other benthic species with strong habitat preferences.