Data set: All detections.

The F-POD, an echolocation-click logging device, is commonly used for passive acoustic monitoring of cetaceans. This paper presents the first assessment of the error-rate of fully automated analysis by this system, a description of the F-POD hardware, and a description of the KERNO-F v1.0 classifier which identifies click trains. Since 2020, twenty F-POD loggers have been used in the BlackCeTrends project by research teams from Bulgaria, Georgia, Romania, Türkiye, and Ukraine with the aim of investigating trends of relative abundance in populations of cetaceans of the Black Sea. Acoustic data from this project analysed here comprises 9 billion raw data clicks in total, of which 297 million were classified by KERNO-F as Narrow Band High Frequency (NBHF) clicks (harbour porpoise clicks) and 91 million as dolphin clicks. Such data volumes require a reliable automated system of analysis, which we describe. A total of 16,805 Detection Positive Minutes (DPM) were individually inspected and assessed by a visual check of click train characteristics in each DPM. To assess the overall error rate in each species group we investigated 2,000 DPM classified as having NBHF clicks and 2,000 DPM classified as having dolphin clicks. The fraction of NBHF DPM containing misclassified NBHF trains was less than 0.1% and for dolphins the corresponding error-rate was 0.97%. For both species groups (harbour porpoises and dolphins), these error-rates are acceptable for further study of cetaceans in the Black Sea using the automated classification without further editing of the data. The main sources of errors were 0.17% of boat sonar DPMs misclassified as harbour porpoises, and 0.14% of harbour porpoise DPMs misclassified as dolphins. The potential to estimate the rate at which these sources generate errors makes possible a new predictive approach to overall error estimation.F-POD是一种回声定位点击记录装置（echolocation-click logging device），通常用于鲸类的被动声学监测。本文首次评估了该系统全自动分析的错误率，详细阐述了F-POD的硬件结构，并介绍了用于识别点击序列（click trains）的KERNO-F v1.0分类器。自2020年以来，保加利亚、格鲁吉亚、罗马尼亚、土耳其、乌克兰的研究团队在BlackCeTrends项目中布设了20台F-POD记录器，旨在调查黑海鲸类种群相对丰度的变化趋势。本研究分析的该项目声学数据总计包含90亿条原始点击数据，其中2.97亿条被KERNO-F分类为窄带高频（Narrow Band High Frequency, NBHF）点击（即港湾鼠海豚点击），另有9100万条被分类为海豚点击。如此庞大的数据量需要可靠的自动化分析系统，本文即对此类系统进行了说明。研究人员对总计16805个检测阳性分钟（Detection Positive Minutes, DPM）进行了逐一核查，通过目视检视每个DPM内的点击序列特征完成评估。为评估每个物种类别的整体错误率，我们分别选取了2000个被分类为包含NBHF点击的DPM，以及2000个被分类为包含海豚点击的DPM进行分析。存在误分类NBHF序列的NBHF类DPM占比低于0.1%，而海豚类别的对应错误率为0.97%。对于港湾鼠海豚和海豚这两个物种类别而言，该错误率均满足黑海鲸类研究中采用自动化分类的需求，无需对数据进行额外编辑即可用于后续研究。主要的错误来源包括：0.17%的船载声呐DPM被误归类为港湾鼠海豚DPM，以及0.14%的港湾鼠海豚DPM被误归类为海豚DPM。对这些错误来源的错误产生速率进行估算，可为整体错误率估算提供一种全新的预测路径。