Stocking density and welfare of cage farmed Atlantic salmon: application of a multivariate analysis

The welfare of fish is receiving increasing attention and attempts have been made to control welfare in farmed fish through regulation of management practices, including stocking density. However, there is little published information on the influence of stocking density on welfare of fish in marine cages. This present study examined welfare in Atlantic salmon (Salmo salar) in cages on a commercial marine farm, exposed to densities ranging from 9.7 to 34 kg m−3. On three occasions over a period of 10 months, fish were sampled from each cage, weighed and measured; their fin condition assessed and blood samples taken for measurement of glucose and cortisol. A multivariate analysis was used to combine four commonly used measures of fish welfare (condition of body and fins and plasma concentrations of glucose and cortisol) into a single welfare score. As well as objectively reflecting a coherence within the data, this score was consistent with the evaluation of welfare by experienced farmers. A generalized linear model indicated that the median welfare score for each cage was significantly related to sampling period, to stocking density (mean over the previous 3 months) and to location of the cage. A model with all the data from individual fish proved to be more robust and also identified sample period, stocking density (mean over 3 months) and position of the cage as significant predictors of the welfare score. There was no significant association between the welfare score and the length of time since grading or lice treatment. Further analysis of the relationship between stocking density and the welfare score suggested that there was no trend up to an inflection point ca. 22 kg m−3, after which increasing stocking density was associated with lower welfare scores. This study suggests that, while stocking density can influence the welfare of Atlantic salmon in production cages, this is only one influence on their welfare and on its own cannot be used to accurately predict or to control welfare.