Health monitoring of Atlantic salmon (Salmo salar L.) using weekly mortality data and dynamic linear models

Introduction: We have applied a dynamic linear model (DLM) to weekly Atlantic salmon mortality data from sites across Scotland. Previous literature has established that DLMs can be a useful tool for monitoring salmon mortality when applied to monthly data. This study aims to evaluate the effectiveness of DLMs in
identifying significant changes in mortality trends in Scottish Atlantic salmon farms based on weekly mortality counts, providing timely warnings for possible intervention.

Methodology: The DLM was implemented to model and forecast the weekly log-transformed mortality rates for Atlantic salmon farming sites. The log transformation was used to ensure that the data followed a normal distribution, as is necessary when applying DLMs. The model parameters were dynamically updated to capture changing patterns in the data over time. To generate alarms, 95% confidence intervals (CIs) for the log-transformed mortality estimates were calculated. Alarms were triggered when observed log mortality rates exceeded the upper bound of the 95% CI, indicating a significant deviation from expected mortality levels.

Results: We compared models trained on our weekly mortality counts with a model trained on the monthly mortality proportion data used in a previous project. Using weekly log-transformed mortality rates allowed for more precise and timely identification of mortality events compared to the broader trends observed with monthly data. This finer temporal resolution revealed patterns and deviations missed in the coarser monthly assessments. The weekly model also reduced uncertainty, meaning that it captured some spikes in mortality that the monthly model missed.

Conclusions: Using weekly mortality data and dynamic linear models with log-transformed rates significantly improves the detection and response to abnormal mortality patterns in Atlantic salmon farming in Scotland. Compared to monthly mortality proportions, weekly data’s higher resolution captures short-term changes, enabling earlier intervention and potentially reducing fish losses. This study highlights the value of more frequent data collection and advanced modelling in aquaculture management. Future efforts will integrate additional environmental variables to enhance the accuracy of weekly mortality monitoring systems.