AI Oceanography — The fourth industrial revolution comes in the form of oceanographic science
Artificial Intelligence — In 2020 the term is practically inescapable. From self-driving cars to AI-enabled smartphones; The explosion in the functioning and trading of AI games, expert game-playing computers, machine learning techniques for an endless variety of tasks.
Led by companies such as Amazon, Google, Apple and Samsung, companies around the world are revolutionizing their businesses, with recent developments in deep learning (large neural network) algorithms and a steady decline in computing power. Scientists in many disciplines are at the forefront of finding new ways to exploit powerful model-detection and statistical reasoning enabled by modern AI.
In Life Sciences, AI provides insights into the human genome, predicts cancer development, and accelerates drug discovery at an unprecedented place.
In the earth sciences, AI improves atmospheric and environmental time series by representing missing data and integrating conflicting observations, correcting bias, and building better models of attendance than previously possible.
The long-standing and uninterrupted geological research question of when and when the next major earthquake will occur, for the first time, maybe soluble using neural networks. The promise of machine learning solutions for problems that are impossible or too challenging to use traditional methods is appealing.
Marine science has its unique challenges and uncertainties — collecting data on vast spatiotemporal scales, tracking and isolating the effects of different water masses in highly dynamic systems, or accessing remote and often dangerous areas.
Thanks to pioneering projects such as high-resolution, long-term in situ observational datasets, Argo now boasts about 4000 widely dispersed autonomous platforms, which cover the upper 2000 m of the ocean. Biogeochemical arc floats are also increasingly online, allowing measurements of chlorophyll fluorescence, acoustic backscatter, dissolved oxygen, and other crucial marine properties.
For large datasets, such as the Global Argo Observational Output, deep learning techniques can be used to address turbulent processes, sub-surface flows, air-sea currents, and energy transport; Interpolating dimensions in areas with little coverage; improving model parameterization; Automatic quality control and more. Data-gathering is also greatly assisted by new autonomous marine vehicles, enabling intelligent auto-piloting and adaptive modeling using ocean gliders and other AUVs in current AI research.
The primary driver of these recent developments is significant progress in deep learning architecture models. Among these, flexible convolutional neural networks (CNNs) are popular for identifying features in multidimensional data such as images.
CNN’s are networks of spatially-interconnected neurons — simple algorithms that take multiple inputs and have one output, the simplest function of the inputs, and the bias value. By building multiple layers of neurons and optimizing the output functions of neurons, behavior mimics the structure of the human visual system. For processing, such as acoustic spectrograms and subsea imagery, traditionally performed by researchers with the help of undergraduate research assistants, CNNs can provide exceptional improvements in both speed and accuracy.
Where it takes time to manually select small features in mostly uninteresting seafloor images, CNNs can segment images intelligently and output only the features of interest for later detection. Also, if sufficient labeled data is available for optimization of the CNN, the presence of an animal to be classified taxonomically, a network can be established to distinguish between taxa and perform detection.
An exciting application of deep learning for biological oceanography and for the Ph.D. project I follow, using CNNs to process Microscope images. Phytoplankton is a critical indicator of water quality, climate change, and primary production and is detrimental to marine and human life in harmful algal bloom events.
Despite their global importance, it is difficult to model and classify plankton at the standards required for an accurate representation of their diversity and distribution. My project is heading towards solving this problem by developing an open-source, low-cost device for micro plankton’s high-speed, high-performance digital microscopy.
I aim to allow the development of robust CNNs that can automatically identify and classify interesting species by allowing the wider oceanographic community to collect and publish their data on millions of phytoplankton images in mixed, untreated seawater samples. My project is part of the Nexus (Next Generation Unmanned and Autonomous Systems Science) Center for Doctoral Training at the National Oceanography Center Southampton, one of many programs designed to support intense collaboration among those computer scientists, engineers, and oceanographers.
Other exciting NEXUS projects include the use of CNNs for seafloor image classification, the prediction of sea migration patterns, the design of new ocean sensors and platforms, and the development of autonomous navigation and control for AUVs.
There is a significant global drive towards collecting and making available high-quality image databases for deep learning optimization. The excellent Seafloor Explorer project developed by the Woods Hole Oceanographic Institution has influenced civilian scientists and worked to create an enormous training dataset for automated seafloor analysis.
Other marine projects include creating a large online database of civilian scientist-labeled zooplankton images and publishing over 4 million pre-labeled phytoplankton microscopy images. As more data becomes available for its optimization, AI will greatly improve oceanographic understanding, increase the scope of research, and automate expensive and laborious manual tasks. Shortly, a global maritime surveillance system that measures and uses AI to describe physical, geological and biological marine characteristics could also move sensing platforms for optimal coverage/resolution.
This new GOOS will rapidly communicate its results to oceanographers via the Internet, which will integrate with models to predict spot patterns and future trends in the enormous datasets it produces. This new era of oceanography is possible based on the individual components currently in development.