STATEMENT OF WORK (SOW)

Project Title: Machine Learning Assisted Scientific Electronic Monitoring of West Coast Fixed Gear Fisheries

1.0 Background/Overview

This contract supports the operations of the West Coast Groundfish Observer Program (WCGOP), managed by the National Oceanic & Atmospheric Administration (NOAA) Fisheries Northwest Fisheries Science Center

(NWFSC).

The NWFSC seeks to modernize scientific data collection across West Coast Fixed Gear Fisheries by integrating Machine Learning (ML) assisted Electronic Monitoring (EM). While traditional human observer coverage is vital, maintaining high coverage levels across the fixed gear fleet—specifically the Limited Entry Non-Trawl (LENT) and Open Access (OA) sectors—is logistically complex and cost-prohibitive. EM offers a scalable solution to augment coverage, ensuring robust discard estimates and consistent data collection across the diverse fleet.

Initial Phase: LENT Fishery Integration The primary objective of this project is to establish and refine Electronic Monitoring (EM) protocols for the LENT sector that provide a full accounting of all catch, including kept fish and all discards. While EM is currently regulated for pot gear in other fisheries, standardized protocols are lacking for hook-and-line gear, particularly regarding discards which do not come aboard the vessel.

This project will develop and implement comprehensive assessment techniques to identify, enumerate, and estimate catch length that can be used to account for all catch, whether it is brought on board or discarded at the waterline (e.g., "shaking" or "drop-offs" prior to crossing the rail). Future uses of this data include quota management and integration into stock assessments.

Machine Learning and Technical Innovation Beyond hardware, this contract aims to test the feasibility of Machine Learning (ML) and automated systems to reduce manual review costs and improve data processing efficiency. Key areas of focus include:

● ML-Assisted Video Analysis: Utilizing machine learning/computer vision for automated species identification and deriving accurate length measurements from imagery.

● Algorithmic Automation: Automatically calculating weight estimates by applying length-weight regression models to the ML-derived length data.

● Automated Event Logging: Accurately identifying and timestamping gear deployment and retrieval events, whether achieved through ML video recognition, integrated vessel sensors (if applicable), or a combination of both.

2.0 Scope of Work

The Contractor shall provide a comprehensive, ML-integrated Electronic Monitoring (EM) solution inclusive of hardware, software, and advanced data analytics to modernize scientific data collection within the West Coast Fixed Gear (WCFG) fleet.

The primary objective of this contract is to develop an image library and demonstrate that ML-assisted EM can provide high-confidence data for the Limited Entry Non-Trawl (LENT) hook-and-line fleet, specifically regarding discards which do not come aboard the vessel and biological precision that matches or exceeds traditional human observer standards.

2.1 Pilot Implementation and Dual Coverage

The project will initiate with a focus on the LENT Sablefish sector, involving 3 to 5 permitted vessels. To facilitate scientific validation, the project will use dual coverage where EM data is collected simultaneously with WCGOP human observers. This side-by-side comparison is critical for evaluating EM accuracy in detecting and quantifying catch.

2.2 ML-Driven Data Objectives

The Contractor will develop and refine automated machine learning (ML) capabilities to achieve the following:

● The Contractor will develop ML tools for automated catch quantification, including species identification, length, and enumeration of both retained and discarded catch.

● The Contractor will develop specialized waterline assessment protocols and use ML tools to identify, enumerate, and estimate the length of fish that break the surface of the water but are not brought aboard.

● The Contractor will implement automated biological processing by integrating species-specific length-weight regression models into the EM data workflow to automatically derive weight estimates from ML-derived lengths.

2.3 Scalability and Integration

The Contractor shall provide a comprehensive EM solution inclusive of hardware, software, and data analysis to support a representative sample of the West Coast Fixed Gear (WCFG) fleet.

The project will initiate with a primary focus on the LENT Sablefish sector, involving 3–5 permitted vessels.

LENT vessels average 7 trips annually, while completing 15-20 sets per trip. Vessel selection will be made with the assistance of NWFSC. This initial phase serves as a scalable model for broader implementation. The objective is to refine data collection protocols ensuring that the EM systems reliably capture catch composition and discard data commensurate with data standards established in the WCGOP Observer Training Manual.

https://repository.library.noaa.gov/view/noaa/67559

3.0 Technical Requirements & Standards

3.1 Gear & Effort Tracking

● Trip Identification: The system must automatically identify and log the beginning and end of each fishing trip.

● Effort Identification & CPUE calculation: The system must identify and timestamp all "Start Set" and "End Haul" events. The system must be capable of calculating soak duration and associate each retrieval event (Haul) to its corresponding deployment event (Set). This association data must be encoded to allow for automated reconciliation with vessel reporting records (e.g., eLogs).

● Effort Validation: The system must provide the capability to record total gear units deployed and retrieved (e.g., number of pots, hooks, or vertical lines).

● Compliance Monitoring: The system must capture imagery or data sufficient to verify the use of required seabird deterrence measures during gear setting operations for hook-and-line gear.

● System Health: The system must record system health and performance metrics which shall be provided to NWFSC.

● Geospatial Data Tracking: The system must include an integrated GPS receiver capable of recording the vessel’s position (latitude/longitude), heading, and speed. This geospatial data must be continuously logged and hard-bound to the video metadata to ensure all fishing events and catch handling are geographically anchored.

● Video Duration: The system must record all relevant fishing activity, including gear setting, retrieval, catch handling, and on-deck operations, while excluding vessel steam time, transit between sets and port from the final submission to NWFSC.

● Onboard Control Center (Vessel Interface): The system must include a dedicated onboard monitor or interface accessible to the Captain on the bridge. This interface must provide real-time status alerts, camera "live-view" functionality for clearing lenses, and an indicator to verify that GPS, cameras, and data storage are functioning correctly prior to and during fishing operations. Data stored within the EM system is encrypted and not accessible by the vessel personnel.

3.2 Catch & Discard Monitoring (The "Waterline" Requirement)

● Full Catch Coverage and Identification: The system must provide a 100% uninterrupted field of vision covering the entire catch handling process from the "waterline" (to capture "shaking" or "drop-offs") through all deck sorting and processing areas. This coverage must be sufficient to identify all catch, kept and discarded, to the lowest taxonomic level possible under all operational conditions (e.g., varying light and weather). Primary species of interest include, but are not limited to: Sablefish, Pacific Halibut, Lingcod, Dover Sole, Arrowtooth Flounder, Shortspine Thornyhead, and Redbanded Rockfish. The Contractor is responsible for determining the number and placement of cameras required to achieve this coverage.

● Data Quality for Identification: Imagery must be captured at a resolution, frame rate, and lighting configuration (including during night operations) that allows both ML and human reviewers to distinguish between similar species (e.g., various rockfish or flatfish), estimate lengths accurately, and identify gear characteristics and seabird deterrent devices (if applicable).

● Redundancy and Reliability: The system design must incorporate sufficient coverage and technical redundancy to ensure a full accounting of catch is maintained even in the event of partial visual obstructions (e.g., crew movement, sea spray) or localized hardware failures.

● Length Estimation Capability: The system must possess the technical capability to estimate the length of catch for designated species. This capability must be functional for both catch brought on deck and, to the extent scientifically feasible, for discards occurring between the surface of the water and the top rail of the vessel.

● Fleet-Wide Scalability: All assessment techniques, hardware configurations, and ML models must be scalable and adaptable to the diverse vessel configurations and operational practices of the West Coast fixed-gear fleet. The Contractor shall prioritize length estimation methods that do not require specialized, vessel-specific deck modifications.

● Review Rate: The Contractor shall conduct a 100% review of all hauls, or a percentage determined necessary by the program to meet the goals of this research, to establish a baseline for ML training and data validation for both on-deck and waterline data.

3.3 Biological Data

● Pacific Halibut Injury Assessment: Explore ways to categorize Pacific halibut releases into "Minor,"

"Moderate," "Severe," or "Dead" based on visible assessment, the method used to release the fish from the hook, or other option(s) approved by the program.

● Weight Estimation: The Contractor’s software shall automatically apply species-specific length-weight regression models to calculate total weight. A limited number of species have NWFSC established regression models, others will need to be developed.

● MMSBST Identification: The system should be capable of flagging Marine Mammals, Sea Birds, and Sea Turtles (MMSBST). Any encounter must trigger a narrative review and specific timestamps for NWFSC biological evaluation.

3.4 ML Development, Training, and Workflow Integration

● Model Development and Licensing: The Contractor shall develop or implement ML models for species identification/classification and length estimation.

○ Base Model Inclusion: Whether developed from scratch or built upon an existing "base" model (any existing model or tool, ML or otherwise), the resulting solution must be delivered in its entirety.

● Delivery and Access: The Contractor must provide NWFSC with all model files and all source code used for training and inference.

○ Completeness Requirement: This delivery must include all code necessary to independently execute, recreate, or continue training the model, including any base code, modified scripts, and environment configurations required for operation.

● Licensing: The Contractor shall grant the Government a perpetual, irrevocable, worldwide, non-exclusive, and royalty-free license to use, execute, reproduce, and practice the model, including any integrated base components, for any government purpose, including future training and sharing.

○ Hosting & Sharing: "Government Purpose" specifically includes the right for NWFSC to provide external access to the model, code, and documentation.

● Annotated Image Library: The Contractor shall deliver a complete Annotated Image Library consisting of all images and video clips used to train or validate the models under this contract.

● Annotation Standards: Annotations must follow NWFSC-defined standards or prevailing industry standards, at the discretion of NWFSC.

● Data Rights: The Government retains Unlimited Rights to the Annotated Image Library and all generated annotations.

● Performance Testing and Validation: The Contractor shall test both the species classification and length estimation models for accuracy and provide a detailed report on the methodology and results.

● EM Workflow Integration: Validated models must be integrated directly into the EM Data Processing Workflow to automate identification and length estimates prior to final data payload transmission.

3.5 Data Transmission and Security

● Data Retrieval and Delivery: To minimize reliance on the physical exchange of media (e.g., mailing hard drives), the Contractor shall prioritize remote or wireless delivery methods to extract data from the vessel to the NWFSC secure data processing environment. All wirelessly transmitted data shall route directly to, and reside solely within, a NWFSC-owned cloud environment; no data shall be stored, hosted, or cached on the contractor's infrastructure.

● Chain of Custody & Security: The Contractor must maintain a secure, verifiable, and tamper-evident chain of custody for all data from the moment of capture on the vessel until final delivery. All data must be encrypted at rest and in transit.

○ Data Hosting and Review Accessibility: The Contractor will host all collected data (including raw video, sensor data as applicable, and analyzed metadata) in a secure environment.

○ NWFSC Access: This data and cloud environment will be accessed by must provide authorized NWFSC personnel associated with the project with continuous, on-demand access to the Contractor’s video review platform.

○ Access Capability: This access must allow NWFSC to view, audit, and verify both the raw imagery and the contractor’s analyzed data (including ML annotations, GPS streams, and human-reviewer tags) throughout the duration of the project.

● System Maintenance and Data Loss Mitigation: In the event of a system failure identified during or after a trip, the Contractor shall prioritize technical support and hardware repair to minimize further data loss. For the duration of this pilot, the Contractor shall maintain a log of system failures and the "time-to-resolution" to help NWFSC evaluate system reliability for future fleet-wide implementation.

● Data Retention and Final Disposition: The Contractor shall maintain all collected data (raw video, images, and metadata) for the entire duration of the contract period of performance within the NWFSC approved system.

○ Final Delivery: Upon expiration or termination of the contract, the Contractor shall provide a complete, organized transfer of all raw video and image files to NWFSC in a format specified by the agency.

○ Data Destruction: Following confirmed receipt of the final data transfer by NWFSC, the Contractor shall permanently remove and delete all project-related video and images from their local and cloud-based systems. The Contractor shall provide a written Certificate of Data Destruction to NWFSC upon completion.

● API Integration: The Contractor shall be responsible for securely formatting and transmitting finalized data payloads to the NWFSC-provided Application Programming Interface (API) ingestion into the

WCGOP database within 30 days of trip completion. Exact data formatting specifications will be provided to the Contractor upon contract award.

● Feedback Loop: The Contractor shall operate as a self-sufficient liaison to the commercial fishing industry. This entails direct, professional engagement with vessel owners, captains, and crew to foster a culture of compliance and mutual understanding. The Contractor is expected to manage these relationships independently, providing technical support and data reconciliation without the daily intervention of NWFSC. For example, provide a direct feedback mechanism to vessel captains to ensure EM set numbers match official logbook entries.

● Logbook Access: NWFSC will provide electronic logbook access (e.g. webservice) to the Contractor for data validation and reconciling discrepancies between the vessel captain's entries and the EM system.

3.6 Vessel Monitoring Plan (VMP) Requirements

The Contractor shall develop a vessel-specific VMP for each participating vessel. To ensure the VMP supports the scientific needs of the West Coast Fixed Gear fishery, it must include:

3.6.1. Vessel Operator Responsibilities

The Vessel Operator is responsible for the continuous operation and maintenance of the Electronic Monitoring (EM) system throughout the duration of the project.

● Pre-Trip:

○ Perform a system function test to ensure all cameras, sensors (if applicable), and storage drives are operational.

○ Night Operations Prep: Verify that deck lighting configurations are functional and positioned to maintain video quality sufficient for ML-assisted identification during 24-hour operations.

○ Ensure the EM system is powered and recording prior to departing the dock.

● During Trip:

○ Maintain clear camera lenses.

○ Monitor the system display to ensure continuous recording and GPS signal strength.

○ Adhere to designated Discard Control Points (see Section 3.6, Subsection 3).

● Trip End:

○ Verify the successful recording of the final haul.

○ Follow data offload protocols and ensure all hard drives or cellular uploads are initiated per program guidelines.

3.6.2. EM Installation & Technical Specifications

● Waterline Coverage: Cameras must be installed according to the vessel-specific schematic, ensuring 100% visibility of the gear retrieval station. The field of view must capture "shaking" or "drop-offs" occurring prior to crossing the rail.

● EM Installation Summary: A record of all hardware components, sensor types (if applicable), and camera serial numbers, must be kept onboard.

● Technical Support & Troubleshooting: Vessel operators must have access to technical support and a basic troubleshooting guide for common system errors (e.g., Camera Malfunction or lack of data storage).

● System Checks: System health and malfunction tests must be completed prior to departure. Best efforts must be made to resolve malfunctions prior to departure.

3.6.3. Catch Handling & Discard Control

● Discard Control Points: All catch must be sorted and discarded within defined areas. These points must be aligned with HD camera placements to ensure images are sufficient for species identification and length estimation by both human and ML software.

● Handling Protocols: Operators must ensure that discards are handled in a manner that allows for clear visual documentation before they are returned to the sea.

3.6.4. Logbook & Fishing Effort Documentation

● Logbook Requirements: The vessel must document fishing effort using the electronic non-trawl groundfish electronic logbook.

● Fishing Effort Documentation: Every fishing event must include precise "Start Set" and "End Haul" times within the eLog to allow for accurate syncing between sensor data (if applicable) and video footage.

● Data Consistency: Ensure that catch estimates in the logbook align with the visual data captured at the Discard Control Points.

3.7 Final Report

The Contractor shall provide a comprehensive final report that includes the following:

● Project overview with high-level results and key recommendations.

● Project results and technical and operational feasibility analysis.

● Video review software documentation, including specifications for the software provided to review produced video data.

● ML Performance & Validation Report: Statistical analysis of ML accuracy (Species ID and Length) compared to the human-expert baseline.

● Annotated Image Library documentation, including metadata schemas and taxonomic validation results.

● Technical Documentation: Full documentation for the ML models and source code, including environment configurations.

● User Manual for the video review platform and API transmission protocols.

● Comprehensive cost comparison and detailed pricing, featuring a three-line breakout for:

○ Software acquisition

○ Licensing (if applicable)

○ Training on the use of the software

● Separate price estimate specifically for the training sessions on the video review software.

● Lessons learned and other supporting information.

4.0 Deliverables Schedule (2027)

Due Date Deliverable Activity Description

01/01/2027 Final Fleet Implementation Plan

Written summary of NWFSC planning meetings, final equipment/software specifications, and the proposed installation schedule for the LENT fleet.

02/15/2027 Vessel Monitoring Plans

(VMP)

Custom VMPs for each of the 3–5 LENT vessels, detailing camera angles (including waterline redundancy), discard control points, and lighting for night operations.

03/01/2027 Data Mapping & API Schema

Data dictionary and mapping documentation linking Contractor data fields to NWFSC database schemas and API requirements.

04/01/2027 System Deployment Completion of hardware installation, data transmission testing and technician-led vessel personnel training.

Bi-Monthly Status & Feedback Reports Written summary of operational challenges, maintenance performed, and feedback from vessel operators to minimize impacts on fishing operations.

Trip + 7 Days Initial Trip Success Reports Detailed reports outlining successes/problems for the first two trips of each vessel to allow for corrective action before subsequent hauls.

08/01/2027 ML Feasibility & Scientific Audit

Technical report assessing ML accuracy in species identification, waterline length estimation, and Halibut injury categorization.

08/01/2027 Image Library Annotated ML Image Library: A comprehensive dataset of annotated images, structured and formatted for immediate use in ML model training and validation.

09/01/2027 Final Pilot Synthesis Final report comparing EM vs. Observer data, along with a cost analysis, and a roadmap for OA expansion.

5.0 Performance Requirements Summary

Task Performance Standard Acceptable Quality Level

(AQL)

VMP Adherence 100% uninterrupted view from the waterline to the rail and all sorting areas as defined in the vessel-specific VMP (Section 3.6).

100% Compliance

Geospatial Integrity GPS data is hard-bound to video metadata with coordinates, heading, and speed logged at intervals defined in 3.1.

99% of all logged video data.

System Health & UI The Onboard Control Center provides real-time functional status and "live-view" access to the Captain (Section 3.1).

100% functional at trip departure.

Waterline Discard Accounting

The system must capture imagery of sufficient quality to successfully identify, enumerate, and estimate the length of all catch, including intentional discards and drop-offs, at the waterline under all operational conditions (Section 3.2).

95% of all haul events where catch is present on the gear.

Deck Imagery Quality

Cameras provide clear, unobstructed views of all sorting and processing areas.

Resolution must be sufficient for species ID and gear characteristic collection (Section 3.2).

95% of all trip footage.

Data Completeness & Effort Association

Successful capture of all fishing events (Sets/Hauls) with documented association between Set and Haul for CPUE calculation (Section 3.1).

95% of all trips

Technical Accuracy

(ML)

ML species identification and length estimation matches the human-expert review baseline (Section 3.4).

90% Correlation

API Transmission Timeliness

Finalized data payloads successfully formatted and transmitted to the NWFSC API (Section 3.5).

Within 30 days of trip completion for 95% of trips

6.0 Period of Performance

January 1, 2027 to December 31, 2027-Base Period

January 1, 2028 to December 31, 2028-Option Period I

7.0 Place of Performance

The contract for installation will be performed at the location of vessels. The contractor shall arrange for a webinar or video telecom conference (VTC) for software discussion and demonstration with NWFSC personnel.

8.0 Government Furnished Information (GFI)

The Government will provide: (1) WCGOP Observer Manual, (2) API Documentation for database ingestion, and

(3) A list of permitted LENT vessels.

9.0 IT Data and Security Requirements

The Contractor shall coordinate with the vessel operator/owner to maintain the security and integrity of the EM system and data. To ensure the security of confidential data, NWFSC and the Contractor shall develop a data protocol governing the secure, wireless transmission of data upon trip completion and the verification of system readiness prior to subsequent trips. Participation by vessels is voluntary; however, any vessel electing to participate shall be required to execute a Data Sharing Agreement prior to the start of the pilot.

9.1 Information Technology (IT) Security

• The Contractor shall comply with all Federal, Department of Commerce, NOAA, and NWFSC Security

Policies.

• The Contractor shall ensure that all system and testing documentation is accurate and current.

9.2 Special Requirements

Due to the sensitivity of data associated with the business of fishing, a confidentiality agreement must be signed by the Contractor upon award of the contract to ensure complete confidentiality of the details of the work. All outreach and information specific to this statement of work will be approved by NWFSC prior to release by the Contractor.

10.0 Other Relevant Information

Government Ownership of Data, Hardware, and Intellectual Property All raw video, images, sensor data (if applicable), and metadata collected under this contract are the sole property of the U.S. Government. This ownership extends to the Annotated Image Library and all generated annotations (Section 3.4). Furthermore, all vessel hardware purchased by the Contractor for the execution of this contract, and all custom source code and ML model files developed under this contract (Section 3.4), shall be delivered to and remain the property of the Government. Data may not be released by any party without written consent from the NWFSC POC. Personnel must comply with all Federal, DOC, NOAA, and NWFSC policies relating to IT and data security.

3.7 Final Report

10.0 Other Relevant Information