The application of autonomous underwater vehicles to challenges in marine habitat mapping and predictive species distribution modelling

In a context of increasing anthropogenic pressure on deep-sea ecosystems with fishing, mining and pollution now reaching the largest environment on the planet, protection measures have to be implemented to preserve this fragile environment. Conservation of biodiversity relies on a sound and comprehensive understanding of species distribution and ecosystems composition and dynamics. Currently, our knowledge of many marine ecosystems, particularly in the deep sea, is insufficient to anticipate changes over time and design appropriate strategies. Among the many challenges that have long impeded the efficient conservation of deep-sea ecosystems, the understanding of fine scale species distribution drivers and how that knowledge can be applied to conservation is of particular importance.

Deep-sea ecological surveys are logistically complex, time consuming and costly. Thus, increasing the sampling effort cannot be the solution and researchers have to make the best possible use of their time at sea to gather enough data. While, the most commonly used sampling systems have largely evolved and improved since the early days of deep-sea exploration, they still haven’t been able to tackle that specific challenge.

Autonomous Underwater Vehicles (AUV) offer numerous opportunities to collect large amount of data with a wide range of instrument to survey deep-sea benthic ecosystems and their direct environments. However, the technology is new and has yet to be widely applied to ecological datasets. To be more than theoretical promises, AUV-based methodologies have to be implemented in field studies with objectives to only develop new tools but explore the results and include then within the much wider frame of ecological research, where their results need to be comparable to other methodologies, available structures, gears, budgets and skills.

Furthermore, experience with AUVs has shown that the sheer volume of data collected is beyond the analysis capacity of most ecology research teams who currently rely on manual processing. This is particularly true for image data, which is an essential component of benthic ecological surveys. By consequence, means to automate the analysis of that data as well as the collection also need to be investigated to complete these objectives.

This thesis will attempt to tackle the challenge in a case study of deep benthic epifauna diversity and fine scale distribution. To that end, it will make use of the large amount of data collected by an AUV, particularly, the very high number of images that give near-full coverage of station ground. This naturally leads to investigating the automation of the image analysis process with artificial intelligence in order to annotate that big data set. The near full coverage of a study area with different types of data will enable a very detailed investigation of species diversity and fine scale distribution in an attempt to answer some fundamental questions in ecology that, so far, have eluded scientists. Eventually, all the different types of data will be combined into a habitat map to allow study of environmental drivers of species distribution and present a synthesis of all the results.