*EASTBIO* Towards greater food security from tropical-freshwater and mesopelagic-marine fish resources

Fish provide important protein for human consumption, either directly (humans eat fish) or indirectly via animal feed (humans eat other animals fed on fish). Long-term sustainable fish harvest is therefore a major contributor to human food security and agriculture (which includes the cultivation and breeding of animals). This PhD will apply AI and big-data techniques to improve acoustic identification and assessment of fish in 1) Lake Victoria (Africa’s largest lake), and 2) the mesopelagic zone (200 to 1,000 m depth) in the global ocean.

About 35 million people in the Lake Victoria basin depend on fish, fishing and fish products. Although the introduced Nile Perch is well known (generally and scientifically), this species is not so important for local, African consumption (it is exported as a luxury food for to the developed world). In a local nutritional context, the Silver Cyprinid (Rastrineobola argentea; known locally as ‘dagaa’) is much more important. Dagaa is a schooling fish. We have an ongoing Global Challenges Research Fund project on Lake Victoria. In this PhD, we will develop automated techniques to extract dagaa schools from over 10 years of acoustic survey data, and progress towards improved biomass estimates for the species. The school detection approach will be based on previous work (by Brierley) on automated detection of Antarctic krill and (by Fernandes) on random forest classification to discriminate various species of marine schooling fish. We will also use image analysis techniques to improve the signal processing element of Nile Perch detection, which although not as vital for direct food security, provides much needed local income for poverty alleviation.

The second component of this PhD will investigate methods to improve the estimation of mesopelagic fish biomass. The mesopelagic zone has been described as representing a ‘dark hole’ in our collective understanding of the ocean. However, it has been suggested that there may be 1 billion tonnes of mesopelagic fish, and these fish may offer an as-yet untapped source of protein. There is already growing commercial interest in targeting these fish, but confidence in the biomass estimate is low. We have developed a global biogeography of the mesopelagic using unsupervised learning (clustering) of echosounder data. In this PhD we will progress towards supervised learning and application of ensemble decision trees to refine the biogeography and progress towards improved biomass estimation. Accurate biomass estimates are absolutely essential for any future sustainable long-term harvest of mesopelagic resources, and for preservation of ecosystem function in the face of exploitation: there is thus an ecological imperative as well as a food-security imperative.

This PhD will provide training in collection and analysis of acoustic survey data (including fieldwork on Lake Victoria), and in application of AI/machine learning to classification of large data sets.