Time-series clustering of cage-level sea lice data.

Abstract

Sea lice Lepeophtheirus salmonis (Krøyer) are a major ectoparasite affecting farmed Atlantic salmon in most major salmon producing regions. Substantial resources are applied to sea lice control and the development of new technologies towards this end. Identifying and understanding how sea lice population patterns vary among cages on a salmon farm can be an important step in the design and analysis of any sea lice control strategy. Norway's intense monitoring efforts have provided salmon farmers and researchers with a wealth of sea lice infestation data. A frequently registered parameter is the number of adult female sea lice per cage. These time-series data can be analysed descriptively, the similarity between time-series quantified, so that groups and patterns can be identified among cages, using clustering algorithms capable of handling such dynamic data. We apply such algorithms to investigate the pattern of female sea lice counts among cages for three Atlantic salmon farms in Norway. A series of strategies involving a combination of distance measures and prototypes were explored and cluster evaluation was performed using cluster validity indices. Repeated agreement on cluster membership for different combinations of distance and centroids was taken to be a strong indicator of clustering while the stability of these results reinforced this likelihood. Though drivers behind clustering are not thoroughly investigated here, it appeared that fish weight at time of stocking and other management practices were strongly related to cluster membership. In addition to these internally driven factors it is also possible that external sources of infestation may drive patterns of sea lice infestation in groups of cages; for example, those most proximal to an external source. This exploratory method proved useful as a pattern discovery tool for cages in salmon farms.