Abstract Onboard observer programs are key to collect data for fishery dynamics analysis and species bycatch and discard estimation. Major sources of bias of onboard sampling for catch data collection concern the vessel selection method, representativeness of catch sampling data and changes in fishing behavior in the presence of observers. This study was motivated by the worldwide issue of low number of vessels included in the sampling frame due to several types of refusals, and the representativeness of the samples for obtaining accurate catch estimates. Analyses were performed using trips from a bottom otter trawl (target fleet) and from the sampling frame, with and without observers onboard. Multivariate analysis of trips was conducted using logbook data, including landings, trip duration, fishing hours and spatial information, but not discards. Two groups of fishing trips were identified within the target fleet during the study period (2012−2015) with distinct fishing regimes, total landings and main landed species. Vessels from the sampling frame were representative of the target fleet and no observer effects occurred in the sampled trips. These findings support the assertion that the sampling frame used is an indicator fleet and analogous to a reference fleet. Cluster-based discard approaches improved the discard estimates precision, when compared to the fleet-based, in particular the cluster stratified approach. Our study suggests that the implementation of a threshold on the species prevalence in discards may potentially introduce bias in the estimates. Discard estimation without applying a threshold, with fleet- or cluster-based approach, should be ultimately dictated by end-users need, based on the trade-off between bias and variance.

中文翻译：

---

调查船上采样行程的代表性和基于聚类的丢弃物估计

摘要 船上观察员程序是收集渔业动态分析和物种兼捕和丢弃估计数据的关键。捕获数据收集船上抽样偏差的主要来源涉及船只选择方法、捕获抽样数据的代表性以及观察员在场时捕鱼行为的变化。这项研究的动机是由于几种类型的拒绝而导致抽样框架中包含的船只数量少的世界性问题，以及获得准确产量估计的样本的代表性。分析是使用底水獭拖网（目标船队）和抽样框架的行程进行的，船上有和没有观察员。使用日志数据对航行进行多变量分析，包括上岸量、航行持续时间、捕鱼时间和空间信息，但不包括丢弃物。在研究期间（2012-2015 年），在目标船队内确定了两组捕鱼活动，它们具有不同的捕鱼制度、总上岸量和主要上岸物种。抽样框架中的船舶代表了目标船队，并且在抽样航行中没有发生观察者效应。这些发现支持这样的断言，即所使用的抽样框架是一个指标车队并且类似于参考车队。与基于车队的丢弃方法相比，基于集群的丢弃方法提高了丢弃估计的精度，尤其是集群分层方法。我们的研究表明，对丢弃物中的物种流行率实施阈值可能会导致估计出现偏差。在不应用阈值的情况下放弃估计，使用基于车队或集群的方法，最终应由最终用户需求决定，