Comparing supervised classification methods on a multispecies fishery of juvenile fish: galaxiid whitebait classification
Degree GrantorUniversity of Canterbury
Degree NameMaster of Science
Post-larvae (whitebait) of the genus Galaxias constitute an iconic fishery in New Zealand. The fishery is essentially nationwide but is largely concentrated on the West Coast of the South Island. It is known that there are wide geographic differences in morphometrics of the whitebait catch, but this is complicated by the fact that five Galaxias species constitute the catch. The five species are īnanga (Galaxias maculatus), kōaro (Galaxias brevipinnis), banded kōkopu (Galaxias fasciatus), giant kōkopu (Galaxias argentus), and shortjaw kōkopu (Galaxias postvectis). One of these five species, banded kōkopu, is relatively easy to identify because of its size and markings, and one species, shortjaw kōkopu, is especially rare and considered endangered. Around 88% of the whitebait catch is one species, called īnanga. Whitebait are caught in enormous abundances as they return from development in the oceanic environment to freshwater streams and rivers. Because these fish are small, post-larval juveniles it is difficult to tell the species apart to characterise the exploitation level of each species (Figure 1). Microscopic examination and, in the case of shortjaw kōkopu, genetic analyses can give identifications with high accuracy, but these techniques are expensive and time consuming. This work aims to use statistical classification to reliably distinguish species with morphological measures. It is a novel classification method for whitebait. These methods use data from extensive sampling of whitebait from 15 regions around New Zealand collected in another study (Yungnickel 2017). It is a data-rich source of morphological measurements by geographic region and is based on around 17500 observation and measurements. This thesis is laid out as follows: a background of the whitebait fishery, life histories and current knowledge; a description of each method with example data to demonstrate the differences in methods; a description of the data; results from the methods; followed by discussion and recommendations for future work.