Machine Learning for Fisheries Resources

Fisheries monitoring is a time consuming manual process. Applying ML automated image recognition can reduce human burden and improve accuracy.

The Problem: Fish population distributions are shifting in response to climate change (Hare et al. 2016; Nye et al., 2009), which can negatively impact fishing communities, coastal economies, and world food supplies. Estimating fisheries stocks and species distributions requires surveying and monitoring to determine species occurrence, abundance, and migratory patterns. In general, fisheries monitoring is a time consuming manual process involving large data sets prone to human error. State and federal agencies have been interested in machine learning assisted monitoring of fisheries, but development of such systems has shown limited success. Applying machine learning to fisheries management has become a priority by state and federal agencies.

The Solution: Applying machine learning automated image recognition to fish monitoring processes can reduce human burden, improve efficiency and accuracy, improve assessments of fisheries stocks and climate impacts on fisheries, and enable coastal communities to become more resilient to climate impacts on coastal and marine resources.

World Change: Automated fisheries monitoring systems reduce time, effort, cost, and potential for human error. The result is improved efficiency and accuracy of data used by resource managers in stock assessments, resource conservationists in coastal community resource assessments, and the general public in promoting awareness and stewardship through education and citizen science. More accurate and timely assessments of fisheries resources enables resource management and industry to support sustainable resources, coastal communities, and economies. As fish stocks shift, communities and economies on the receding end of the species range will be negatively impacted, while communities and economies on the extending edge of the species range will be positively impacted.  Communities on both end of the shifting range must be prepared in order to adapt to change, maintain resilience, and continue to thrive. In addition, fish stocks do not acknowledge international boarders, which can lead to conflicts over resource allocations and sustainable fisheries management as fisheries ranges shift across international boarders. Machine learning enhanced monitoring and processing enables more timely and accurate assessments of altered fisheries distributions in response to climate change, enabling resource managers, industry, and community planners to better prepare for impacts to coastal communities, economies, and marine resources.

Jonathan A. Hare, W.E. Morrison, M.W. Nelson, M.M. Stachura, E.J. Teeters, R.B. Griffis, M.A. Alexander, J.D. Scott, L.Alade, R.J. Bell, A.S. Chute, K.L. Curti, T. Curtis, D. Kircheis, J.F. Kocik, S.M. Lucey, C.T. McCandless, L.M. Milke, D.E. Richardson, E. Robillard, H.J. Walsh, M.C. McManus, K.E. Marancik, C.A. Griswold. 2016.  A Vulnerability Assessment of Fish and Invertebrates to Climate Change on the Northeast U.S. Continental Shelf.. PLOSOne  

JA Nye, JS Link, JA Hare, WJ Overholtz. 2009. Changing spatial distribution of fish stocks in relation to climate and population size on the Northeast United States continental shelf. Marine Ecology Progress Series 393, 111-129

The machine learning technology we are developing is new technology applied to image recognition for use with fisheries management.  We are expanding on existing machine learning image recognition applications, and incorporating secondary variables and technological modifications to reduce false positive and negatives, and improve the likelihood of detecting target species. We will adapt machine learning technology from other industries such as medical and manufacturing, and apply those techniques to improve machine learning capabilities for fisheries management and climate impact to fisheries resources.

We are developing and applying machine learning for fisheries monitoring, stock assessments, and assessments of climate impacts to fisheries resources. Currently fisheries surveys using images are manually reviewed, resulting in many hours and increased likelihood for human error.  Automation through machine learning would reduce human burden, increase processing time and timeliness of data availability, reduce uncertainty and error in results, and improve fisheries assessments and management. More accurate assessments support long-term sustainability of fisheries resources, adaptation to climate impacts, and assists industry with operations planning and resource allocations in support of local communities and economies.

In 12 months we plan to finish prototype improvements and development, and apply the technology to three user groups: (1) local fish wardens; (2) state agency monitoring of commercial and recreational fisheries; (3) federal commercial fisheries monitoring programs. We will accomplish this by hiring two programmers and purchasing necessary hardware for development and testing. We will work closely with user groups to ensure needs are met, and for product testing. We will test in parallel with current monitoring and assessment methods to demonstrate the technology and verify results. We will then work to implement the technology with international partners.

To grow over the next three-five years we plan to:

1. Apply the finished product to local, state, and federal fisheries monitoring and assessment programs nationally.

2. Apply the technology to international fisheries monitoring and assessments programs in Canada, Iceland, Ireland, and Netherlands.

3. Apply the technology to coastal and offshore aquaculture operations.

4. Work with the additional end users mentioned to identify specific group needs. We will then work to modify the technology as needed to address specific needs of these additional user groups, enabling use of the technology and its benefits for coastal communities and economies globally.

We currently with local, state, and federal resource managers, coastal communities, and industry partners, and we have been working closely with them during prototype development. We will continue to work with these groups to apply the technology to their specific monitoring programs, and demonstrate our product through workshops, training, and outreach programs.

We are currently working with one state and two federal agencies, and three local municipalities to develop the product.

We expect to serve at least five local municipalities, two state agencies, and two federal agencies within 12 months. 

We expect to serve up to 12 local municipalities, three state agencies, and three federal agencies within three years, as well as industry participants. We also expect to serve national and international fisheries monitoring programs within three years.

MIT Sea Grant offers fisheries and computer scientists, programmers, electrical and mechanical engineers. We have discussed collaborations with others at MIT to support product design, and development, and have commitments by two MIT labs that specialize in machine learning.  We developed prototype machine learning software last year, with initial results of 93% success with scallops, and 86% success with adult river herring.  Results partners find encouraging. We will improve river herring results, results for flounders, and apply to juvenile river herring.  We will then develop a user interface, transfer the technology to fisheries managers, and conduct user training and outreach.

We are a non-profit organization (MIT Sea Grant).  Our goal is secure funding that will enable us to further develop the technology and distribute to user groups, allowing them to overcome technological barriers, improve resource management processes, and better predict and prepare for impacts from climate change. Distributing the technology to user groups will also support local communities, economies, and sustainable coastal and marine resources. Fisheries is a world-wide industry, providing a food supply of the world population, and support for local, regional, and national communities and economies. Need for accurate fisheries assessments is high, as is the need to assess and prepare for climate impacts on resources, coastal communities and economies. Our local, state, and federal partners and stakeholders have expressed the need for this technology and their continued interest in working us for development and implementation. Demonstrating advanced machine learning applications that improve fisheries management processes and climate impact assessments can attract additional funding from federal resource management agencies and international partners.

We are applying to Solve to acquire the funds needed to support students to improve our prototype and finish product development; to purchase needed hardware; for testing and implementation; and to enable training and distribution to user groups. Funding from Solve will enable us to partner with other MIT labs that specialize in machine learning technology, and in doing so, enable us to improve our prototype by applying advanced machine learning applications from other industries that will advance the level of technological applications and tools available for fisheries monitoring and climate impact assessments.

We have commitment from local, state, federal and industry stakeholders, with whom we have been working closely, and they are eager to have technological solutions such as machine learning applications to improve their coastal resource management and climate impact assessment needs.  Currently, funding to support students and engineers working on the project, and purchase of needed hardware and equipment, is the key barrier for our solution to succeed.