Preliminary analysis of seabird by-catch using CCAMLR observer data

We used data collected by CCAMLR International Scientific Observers in 1997 and 1998 to examine potential relationships between seabird incidental mortality rates on longline vessels fishing for Dissostichus eleginoides and various factors, including the nature and use of mitigating measures as well as environmental variables such as time of day, time of year.
Out of 3283 longline sets analysed only 311 caught birds (9.4%). Data conformed most closely to a Delta distribution (many zero values and log-normal distribution of non-zero values) and were analysed using two Generalised Linear Models, a binomial model for presence/absence of seabird catches and a Gamma model for the magnitude of non-zero catches.
Sparsity of data precluded analysis of seabirds at a taxon level more detailed than albatrosses and petrels combined. Other analytical difficulties, particularly in using generalised linear models, related to the large number of potentially important factors, the lack of overlap between factors and the fact that fishing has purposely avoided making catches of seabirds. There are, for instance, only three records in the entire dataset where none of the mitigation measures have been used.
The only factors consistently significant were time of year (very few birds caught after April) and use of streamer lines but the effects of most other factors cannot be fully analysed with the present data. Even vessels using streamer lines and setting at night were found to catch albatrosses occasionally. The “residual” mortality associated with using all prescribed mitigation measures provides a more useful indication of the success of mitigation measures than general bird bycatch rates.
Given the difficulties of analysing this dataset, especially the problem of very low numbers of hauls not using mitigation measures and hauls catching birds, experimental approaches to identifying effective mitigation measures may be preferable to post-hoc analysis of observer data.