Towards net zero: development of AI enabled biological observing

海洋环境的可持续管理是一个全球关注的问题，联合国宣布2021-2030年为海洋科学促进可持续发展十年或许是最好的突显。由于在观察和监测深海和开阔海洋环境中的海洋生物方面存在挑战，对非沿海海洋环境的生态认识一直落后于对物理、化学和地质的认识。现代海底调查和监测平台，包括自主水下航行器、遥控潜航器和自主登陆器，能够收集一系列在空间和时间上明确的多传感器数据，包括海量视频和/或图像数据集，提供高或大的空间和时间分辨率数据集。虽然这些平台的使用及其同时进行视觉和环境观测的能力已经改变了我们对海洋生态系统，特别是海山和热液喷口等坚硬底物系统的理解，但这些自主和机器人系统的全部潜力尚未实现。实现这一潜力的最大挑战之一在于克服由于需要人工(人工)解释图像和视频以提取定量生物数据而造成的瓶颈。最近人工智能和计算机视觉的发展为克服这一瓶颈提供了一种潜在的机制，提供了一种比人工注释更快、更一致、更具成本效益和可共享的替代方案。我们已经确定，深度学习(人工智能的一个分支)可以用于在合适的条件下可靠、快速地计算特定物种的数量。这一能力需要扩展到更广泛的分类群和开发的管道，这些分类群和管道可以在现场应用，将我们推向人工智能生物观察的未来。认识到这一未来对于减少海洋生物研究的碳足迹并帮助我们实现气候变化目标非常重要。在这个项目中，我们将研究最好的方法，将自主和机器人系统收集的大量数据转化为生态知识，然后输入到模型中，使我们能够预测生物多样性如何分布以及随着时间的推移可能发生变化。这将极大地改善我们对海洋生态的看法，并更好地为养护和管理措施提供信息。

Sustainable management of the marine environment is a global concern, perhaps best highlighted by the UN declaration of 2021-2030 as the Decade of Ocean Science for Sustainable Development. Ecological understanding of the non-coastal marine environment has lagged behind the physical, chemical and geological understanding due to challenges in observing and monitoring marine life at depth and in open ocean environments. Modern seafloor survey and monitoring platforms, including Autonomous Underwater Vehicles (AUVs), Remotely Operated Vehicles (ROVs), and Autonomous Landers, are able to collect an array of spatially and temporally explicit, multi-sensor data, including vast video and / or image datasets, offering either high or large spatially and temporally resolved datasets. While use of these platforms, and their ability to make concurrent visual and environmental observation have already transformed our understanding of marine ecosystems, particularly hard substrate systems like seamounts and hydrothermal vents, the full potential of these autonomous and robotic systems has not yet been realised. One of the greatest challenges to realising that potential lies in overcoming the bottleneck created by the need for manual (human) interpretation of images and video in order to extract quantitative biological data. Recently developments in artificial intelligence and computer vision have offered a potential mechanism to overcome that bottleneck, offering a faster, more consistent, cost effective and shareable alternative to manual annotation. We have established that deep learning (a branch of artificial intelligence) can be used to reliably and quickly count specific species in the right conditions. This capacity needs to be expanded to a wider selection of taxa and pipelines developed that can be applied in-situ, moving us toward a future of AI enabled biological observing. Realising this future is important to reducing the carbon footprint of marine biological research, and helping us achieve our climate change targets. In this project we will investigate the best methods to translate the large volume of data collected by autonomous and robotic systems, into ecological knowledge to then feed into models enabling us to make predictions on how biodiversity is distributed and may change over time. This will drastically improve our perception of the oceans ecology and better inform conservation and management measures.