• PICES Acronym: WG 54
• ICES Acronym: WGDLP
• Parent Committee: BIO
  
• Term: IGC-2025 - IGC-2028
• ICES Co-Chair: TBA
• PICES Mailing List

Plankton are the foundation of marine ecosystems, playing a crucial role in oceanic food webs, biogeochemical cycles, and carbon sequestration. Monitoring plankton communities is essential for understanding ecosystem health, climate change impacts, and fisheries dynamics. The rapid increase in high-resolution plankton imaging systems has enabled the collection of massive datasets, requiring automated tools for efficient processing and analysis. Deep learning (DL) provides a transformative solution, offering fast and accurate plankton classification, yet there is no consensus on best practices for applying DL methods, leading to inconsistencies in data interpretation, model validation, and result comparability.

This working group addresses the urgent need for the best practices in DL-based plankton image processing. The lack of consistent guidelines has resulted in challenges such as model generalizability, dataset biases, and reproducibility issues. To ensure the robustness of DL models, it is essential to develop comprehensive training libraries, establish metadata best practices, and implement transparent evaluation metrics. By fostering international collaboration, this WG will harmonize DL-based image analysis methodologies across PICES and ICES member countries.

The scientific justification for this initiative lies in the growing demand for high-quality, standardized plankton monitoring data. With climate change driving shifts in plankton populations, monitoring changes in species composition, abundance and distribution is critical for ecosystem assessments. Developing best practices for DL-based plankton classification will significantly enhance the accuracy and efficiency of monitoring programs, enabling more precise ecological modelling and predictive analyses.

Improved plankton monitoring contributes to sustainable fisheries management, early detection of harmful algal blooms (HABs), and assessments of carbon flux and climate regulation services provided by the ocean. By ensuring high-quality, reproducible data, this WG will support environmental policies, conservation strategies, and blue economy initiatives. Additionally, this initiative will facilitate capacity building by engaging early-career researchers, providing training in DL applications, and fostering cross-disciplinary collaborations between oceanographers, ecologists, and computer scientists.

By integrating deep learning into plankton research in a standardized and transparent manner, this WG will enhance the consistency of plankton image processing, enable comparison among different regions, and contribute to the broader understanding of ocean health and climate change impacts. The outcomes of this WG will provide a foundation for future advancements in AI-driven marine monitoring and ecosystem-based management.

1. Review current deep learning applications in plankton image processing.
2. Develop standardized DL model training, validation, and evaluation methodologies.
3. Establish a shared library of annotated plankton images and benchmarking datasets.
4. Foster collaboration between plankton ecologists, imaging specialists, and DL experts.
5. Organize workshops and symposiums to disseminate findings and enhance capacity building.
6. Publish a final report summarizing best practices for DL-based plankton image processing.

This working group aligns with PICES' strategic goals by fostering international collaboration (Goal 1) among experts in deep learning, plankton ecology, and imaging systems. By developing best practices for DL-based image processing, it enhances the accuracy of plankton monitoring, contributing to ecosystem assessments and resilience studies (Goal 2).

The group advances methods for analyzing marine ecosystem responses to climate change and human activities (Goal 3), supporting predictive modelling of plankton populations (Goal 4). By integrating AI with traditional oceanographic monitoring, it improves forecasting of ecological shifts, benefiting fisheries and conservation efforts.

Additionally, the working group promotes open data-sharing practices and ensures accessibility of standardized scientific information (Goal 5). It also supports early-career scientists (Goal 6) by fostering interdisciplinary training and capacity building. Through these efforts, the group strengthens global cooperation and contributes to sustainable marine ecosystem management in the North Pacific and beyond.

• WG 54 / WGDLP builds upon the work of PICES WG 48, which focused on plankton imaging systems. By expanding into deep learning methodologies, this working group will further develop standardized image processing techniques, ensuring data consistency and comparability across regions.
• ICES: Collaboration with the ICES Working Group on Zooplankton Ecology (WGZE) aims to align best practices for deep learning (DL) applications in the North Atlantic and North Pacific. Dr. James Scott, the correspondent for TOR5 (plankton imaging), has joined the proposed working group to ensure that the scope of work remains consistent.
• CPR Program (Continuous Plankton Recorder): Coordination with plankton monitoring programs to integrate automated identification methods.
• MONITOR and AP-NPCOOS: Contributions to North Pacific Ocean observation systems through advancements in imaging-based monitoring.

Year 1 (2025)

1. WG meeting (Zoom meeting in July/August after the ISB/IGC approval in May/June):
   (a) Discuss schedules, plans, and contributors for terms of reference and deliverables. (b) Discuss schedules and plans of symposium during the next PICES/ICES annual meeting.
2. PICES/ICES workshop (during PICES/ICES annual meeting: Japan or Lithuania):
   (a) Summarize developments and limitations of different plankton image processing procedures. (b) Establish subgroups for processing pipeline, library/data, and case studies. (c) Develop work plan for each subgroup.
3. Contact information:
   Make a list of experts on plankton imaging systems and plankton monitoring among PICES and ICES nations.

Year 2 (2026)

1. WG meeting (Zoom in March/April):
   (a) Revise schedules and discuss plans for terms of reference and deliverables. (b) Review available machine learning algorithms for plankton identification and enumeration. (c) Review different types of libraries
2. Special session on plankton image processing:
   (a) Expand the list of experts on plankton imaging processing. (b) Further identify data availability for comparison among different imaging systems & processing procedures. (c) Review the applications of imaging processing.
3. PICES/ICES symposium (during PICES/ICES annual meeting):
   (a) Overview machine learning for plankton identification and enumeration. (b) Overview data/management needs for plankton image processing. (c) Develop protocols and standard libraries to test and compare the performance of different algorithms.
4. Contact information:
   (a) Expand the list of contact information on experts on plankton imaging, image processing, and plankton monitoring in PICES and ICES nations.
5. Review articles:
   (a) Review different machine learning algorithms for plankton identification and enumeration (b) Review different types of libraries, model training and potential ways to compare algorithms and output
6. Compare the performance of different algorithm using standard libraries:
   (a) Expand the collection of annotated plankton images to build a shared dataset. (b) Evaluate the performance of different DL algorithms using the dataset. (c) Organize a special session at the PICES Annual Meeting to present preliminary findings.

Year 3 (2027)

1. WG meeting (just before or after PICES/ICES annual meeting):
   (a) Make a draft of PICES/ICES scientific report, including the following information on plankton image processing (b) Review of different DLs for plankton image processing, advantages and limitation of different algorithms (c) Recommendations and best practices protocols for the utilization/selection of different algorithm based on the research purposes. (d) Recommendations and best practices libraries and protocols for comparing different machine learning algorithms for plankton identification and enumeration.
2. Sessions for PICES or ICES annual meeting:
   (a) Integrate imaging systems, image processing into existing plankton monitoring programs. (b) Different platforms for monitoring plankton using imaging systems and plankton monitoring. (c) Standard protocol and library for comparing different plankton identification and enumeration procedures
3. Review articles:
   (a) Submit, revise, and publish the review articles on monitoring plankton using imaging systems with results from case studies.
4. PICES/ICES scientific report:
   (a) Submit a final scientific report to PICES/ICES. (b) Finalize and publish best practices for DL in plankton image analysis. (c) Submit a final report summarizing the WG’s findings and recommendations. (d) Promote the adoption of best practices practices through international collaboration.

Expected Deliverables

• A comprehensive review of deep learning techniques in plankton image processing.
• Protocols for best practices DL model development and evaluation.
• A shared library of annotated plankton images for benchmarking DL algorithms.
• A final report summarizing the WG’s findings, published as a PICES/ICES scientific report.

This working group is committed to adhering to the PICES Data Management Policy and ensuring the responsible collection, sharing, and dissemination of data, in alignment with the FAIR principles (Findable, Accessible, Interoperable, and Reusable).

Data Accessibility and Sharing
All best practices developed by this working group will be openly published in peer-reviewed journals and made publicly accessible to the scientific community. The image libraries compiled during the working group's activities will be hosted on **GitHub**, ensuring open access, transparency, and reproducibility. These libraries will include annotated datasets with appropriate metadata following PICES data-sharing guidelines to facilitate interoperability and broad usage.

Compliance with PICES Data Policy
The working group will ensure that all collected and processed data adhere to the PICES data policy by:

• Utilizing recognized open-access repositories and platforms for data storage and dissemination.
• Providing comprehensive metadata and documentation for reproducibility.
• Complying with data licensing and citation best practices.
• Ensuring that the datasets meet FAIR data principles.

Data Repositories
The working group encourages the use of well-established repositories for plankton image datasets, including:

• GitHub for dataset hosting and model sharing.
• Other recognized public repositories such as IEEE in compliance with international data-sharing policies.

Through these efforts, the working group aims to enhance data transparency, support international research collaboration, and facilitate the integration of deep learning methodologies into global plankton monitoring programs.