Monitoring monthly mortality of maricultured Atlantic salmon (Salmo salar L.) in Scotland II. A hierarchical dynamic linear model

The sustainability of the salmon farming industry is being challenged by increased mortality rates. Scotland’s open-source salmon production data provides the possibility of developing an industry-wide mortality monitoring model, valuable
for identifying and addressing unexpected increases in mortality without needing
data sharing agreements across different companies. This study aimed to utilize
these data to develop a hierarchical dynamic linear model (DLM) for monitoring
monthly mortality of maricultured Atlantic salmon in Scotland. We evaluated
whether considering the hierarchical structure present in the data (country,
region, and site) would improve mortality predictions when compared to the
production cycle level DLMs developed in a previous study. Our findings
demonstrated that the hierarchical DLM outperformed the production cycle
level DLMs, confirming the value of this more complex modelling approach.
Nevertheless, the hierarchical model, like the production cycle level DLMs,
exhibited some uncertainty in the mortality predictions. When mortality is
higher than expected, site level warnings are generated, which can encourage
producers and inspectors to further investigate the cause. Between 2015 and
2020, approximately 25% of the production cycles and 50% of the sites
encountered at least one warning, with most warnings happening in the
summer and autumn months. Additionally, the hierarchical model enabled
monitoring mortality at multiple levels. This information is useful for various
stakeholders as part of a monitoring system, offering insights into mortality trends
at national, regional, and sites levels that may benefit from strategic resource
management. Recommendations for model improvements include utilizing
shorter data aggregation periods, such as weekly, which are not currently
available as open-source data.