Stratification of catch-at-length data using tree based regression: an example using Antarctic toothfish
(Dissostichus mawsoni) in the Ross Sea

This paper presents a new approach to the stratification of catch-at-length data of Antarctic toothfish (D. mawsoni) in the Ross Sea.
Tree based regression techniques were used to stratify the sampled catch based on the median length of Antarctic toothfish for each set using the observer length frequency data. The median lengths were weighted within the regression by the inverse of the variance, rather than giving equal weights to all tows. Two variables (depth and SSRU) were used by the tree regression model to determine the strata.
The resulting stratification effectively split the fishery into 4 regions, consisting of shallow inshore regions where predominantly smaller fish were found, to deeper offshore regions where only larger fish were found. The paper presents the new estimates of Antarctic toothfish catch-at-length and catch-at-age from the Ross Sea up to the end of the 2003–04 fishing season.