What commercial fisheries need from electronic monitoring
In brief
- Commercial value comes from reducing evidence-production cost and latency, not from model accuracy alone.
- The same event record must serve vessel operations, reviewers, managers, buyers and assurance workflows.
- Standards mapping should be explicit and versioned, without implying certification by the software.
A commercial electronic-monitoring platform serves several actors at once: vessel crew, fleet operators, reviewers, fisheries managers, buyers and assurance/certification teams. Their interfaces differ, but the underlying system should share one evidence graph.
The core record
For each catch or interaction event I propose to retain:
- vessel/trip/operation context;
- timestamp and location policy;
- source camera/sensor;
- original media reference + integrity metadata;
- species/taxon, count/weight and disposition where available;
- model/version/confidence;
- reviewer decision and correction history;
- eLogbook/observer/landing links;
- export/standard mapping and data-retention class.
This allows one event to support operational review, catch reporting, traceability and later audit without copying inconsistent data into multiple products.
Requirements by user
Standards as schemas, not marketing claims
MSC’s Evidence Requirements Framework focuses on the quality and handling of information used in assessment. That naturally motivates provenance, completeness, independent verification and auditability features.
Traceability standards such as GDST similarly encourage consistent exchange of key data elements and critical tracking events. A product can map fields and exports to these schemas; it should not imply that schema compatibility automatically establishes certification or regulatory compliance.
Where AI creates economic value
The primary operational value of AI-assisted electronic monitoring is review compression without loss of evidence quality:
raw video → activity detection → candidate events → species/quantity proposal → risk/uncertainty prioritisation → human verification.
NOAA notes that manual video review, transmission and storage are major EM programme costs and identifies computer vision/ML as a route to more automated extraction. The commercial experiment should therefore measure the whole workflow.
A useful A/B evaluation compares assisted and unassisted review on the same blinded sample:
- reviewer minutes per validated event;
- event recall including random-audit footage;
- correction rate;
- unresolved-event rate;
- report turnaround;
- compute/storage/network cost.
Impact
The product claim I propose to want is not “we automate species detection.” It is:
we reduce the cost and latency of producing a reviewable fisheries record, while maintaining or improving measured evidence quality.
That is a commercial outcome and a scientific claim that can both be tested.