Marine Directorate persistent elevated mortality framework in salmon farming: EIR release

Information requested

I am requesting the information below to support an independent, reproducible public-data analysis of the persistent elevated mortality framework. I am not seeking confidential producer records beyond what is necessary to understand the methodology and public-data screening outputs.

For the Marine Directorate's persistent elevated mortality framework described in the March 2026 report, please provide the following information, to the extent held:

1. Method definitions and analytical specification

• The full written methodology, protocol, technical note, or equivalent document used to define elevated, recurrent, and persistent elevated mortality.
• The analysis period, inclusion/exclusion rules, and definition of the 216 active marine salmon sites analysed.
• The mortality-rate formula used for monthly screening, including denominator treatment, fallow or near-zero biomass months, and any handling of missing or corrected SEPA data.
• The recurrence and persistence rules used to move sites from elevated to recurrent and then to model-persistent status.
• Any stocking-cycle attribution rules, including how cycle start/end dates, fallow periods, transfers, or site restocking were handled.
• Any FHI event-to-month attribution rules, including whether event start date, end date, report date, inspection date, or overlap-based assignment was used.

2. Threshold calculations and sensitivity work

• The computed 95th-centile threshold or thresholds used in the March 2026 framework.
• Any computed 90th-centile, 99th-centile, or absolute mortality thresholds considered during development, quality assurance, or sensitivity testing.
• Any tables, notebooks, spreadsheets, or internal outputs showing elevated, recurrent, or model-persistent site counts under alternative threshold definitions.

3. Site-level screening outputs

• The list of the 216 active marine salmon sites analysed, including site identifiers and site names where held.
• The list of the 28 sites identified as recurrent elevated-mortality sites by the model.
• The list of the 25 sites identified as model-persistent before FHI expert review.
• The list of the 20 sites remaining after FHI review.
• The list of the 9 sites remaining after completion of the full framework.
• For each list, please include the site identifier, site name, operator where held, year/cycle/month fields where relevant, and the rule or stage under which the site was included.

4. Cause categorisation / AI or LLM workflow

• Any prompt text, system instructions, model configuration, model name/version, temperature or decoding settings, taxonomy definitions, coding rubric, or retrieval/context materials used to categorise mortality causes.
• Any deterministic pre-processing, synonym mapping, cleaning rules, or post-processing rules applied to FHI mortality-cause text before or after AI/LLM classification.
• Any validation protocol, expert-review rubric, labelled sample design, inter-rater agreement statistics, accuracy/F1/confusion-matrix outputs, or quality-assurance results for the causecategorisation workflow.
• Any non-confidential examples of input text, assigned categories, and reviewer corrections sufficient to understand the categorisation method.

5. Code and reproducibility materials

• Any code, scripts, notebooks, SQL queries, spreadsheets, configuration files, or data dictionaries used for the public-data screening stages of the framework.
• Any version-control metadata, package/environment files, or model cards sufficient to reproduce the public-data screening and cause-categorisation stages.

6. Publication status and related outputs

• The citation, DOI, abstract, slides, poster, preprint, proceedings paper, or manuscript for Moriarty et al. 2026, described in the March 2026 report as an accepted SVEPM conference presentation / conference paper, if held.
• Any related public or internal technical reports that supersede, expand, or correct the March 2026 methodology.

Response

Questions 1, 2, 4 & 5

While our aim is to provide information whenever possible, in this instance an exception under Regulation 10(4)(d) (material in the course of completion) applies to the information requested. It may help if I explain that analysis work is still ongoing in relation to the mortality modelling and data referred to in your request with this currently subject to peer review and has subsequently not been finalised.

This exception is subject to the ‘public interest test’. Therefore, taking into account all the circumstances of this case, we have considered if the public interest in disclosing the information outweighs the public interest in applying the exception. We have found that, on balance, the public interest lies in favour of upholding the exception. We recognise that there is a public interest in disclosing information as part of an open, transparent and accountable government. However, there is a greater public interest in allowing a private space for officials to consider the draft content of documents to ensure that factual and accurate information is released into the public domain.

As we have applied Regulation 10(4)(d), under Regulation 13(d) a public authority must also outline when the information will be completed. Although the above noted peer review and mortality information has not yet been finalised, it is currently anticipated that this work will be completed by September 2026. Should you still require this information at that point, you may wish to submit a further request.

Question 3

Point 1

I have included a copy of the information you have requested in the attached Annex (B).

Points 2 to 6

While our aim is to provide information whenever possible, in this instance an exception under Regulation 10(4)(d) (material in the course of completion) applies to the information requested. It may help if I explain that analysis work is still ongoing in relation to the mortality modelling and data referred to in your request with this currently subject to peer review and has subsequently not been finalised.

This exception is subject to the ‘public interest test’. Therefore, taking into account all the circumstances of this case, we have considered if the public interest in disclosing the information outweighs the public interest in applying the exception. We have found that, on balance, the public interest lies in favour of upholding the exception. We recognise that there is a public interest in disclosing information as part of an open, transparent and accountable government. However, there is a greater public interest in allowing a private space for officials to consider the draft content of documents to ensure that factual and accurate information is released into the public domain.

As we have applied Regulation 10(4)(d), under Regulation 13(d) a public authority must also outline when the information will be completed. Although the above noted peer review and mortality information has not yet been finalised, it is currently anticipated that this work will be completed by September 2026. Should you still require this information at that point, you may wish to submit a further request.

Question 6

I have included copies of information held by the Scottish Government that would be within the scope of your request, namely a slide pack produced by Scottish Government Officials for a presentation on the topic of persistent elevated mortality presented at the 2026 SVEPM conference and a proceedings paper on the topic of salmon aquaculture mortality modelling, within the attached Annexes (C & D)

On reviewing materials within the scope of your request and provided in Annexes C & D, an exception has been applied in line with the EIRs. The exception applied can be found in the attached Annex (A).

About FOI

The Scottish Government is committed to publishing all information released in response to Freedom of Information requests. View all FOI responses at https://www.gov.scot/foi-responses.

EIR 202600515523 - Information released - ANNEX A - ANNEX D

File type

PDF

File size

5.2 MB