Assessment of Error Rates in Acoustic Monitoring with the R package monitoR
Assessment of Error Rates in Acoustic Monitoring with the R package monitoR
复制标题
DOI：
10.1080/09524622.2015.1133320
复制
发表时间：
2016-05-03
影响因子：
1.8
通讯作者：
Donovan, Therese
中科院分区：
文献类型：
--
作者：
Katz, Jonathan;Hafner, Sasha D.;Donovan, Therese
检测人口规模对气候变化和土地利用变化的反应可能需要对许多地点进行多年的监测，这一过程适合自动化系统。我们开发并测试了 monitoR，这是一个用于长期、多类群声学监测项目的 R 包。我们用两种东北鸣禽物种测试了 monitoR：黑喉绿莺 (Setophaga virens) 和烤箱鸟 (Seiurus aurocapilla)。我们将在美国佛蒙特州和纽约州 10 个地点记录的 52 个 10 分钟调查中的 monitoR 检测结果与人类识别的歌曲子集进行比较，这些歌曲属于单一歌曲类型并具有视觉可识别的频谱图（例如，信噪比至少 10 dB：黑喉绿莺总共 439 首歌曲中的 166 首，烤箱鸟总共 990 首歌曲中的 502 首）。 monitoR 的自动检测过程使用“分数截止”，这是将未知事件视为检测所需的最低匹配，并导致真阳性、真阴性、假阳性或假阴性检测。在选定的分数截止点上，monitoR 在使用二进制点匹配的 52 项调查中分别正确识别了黑喉绿莺和烤箱鸟的存在，正确识别率分别为 64% 和 72%，在使用频谱图互相关的 52 项调查中分别正确识别了 73% 和 72%。在个别歌曲中，72% 的黑喉绿莺歌曲和 62% 的烤箱鸟歌曲是通过二进制点匹配识别的。频谱图互相关识别出 83% 的黑喉绿莺歌曲和 66% 的烤箱鸟歌曲。歌曲事件检测的误报率< 1%。
Detecting population-scale reactions to climate change and landuse change may require monitoring many sites for many years, a process that is suited for an automated system. We developed and tested monitoR, an R package for long-term, multi-taxa acoustic monitoring programs. We tested monitoR with two northeastern songbird species: black-throated green warbler (Setophaga virens) and ovenbird (Seiurus aurocapilla). We compared detection results from monitoR in 52 10-minute surveys recorded at 10 sites in Vermont and New York, USA to a subset of songs identified by a human that were of a single song type and had visually identifiable spectrograms (e.g. a signal: noise ratio of at least 10 dB: 166 out of 439 total songs for black-throated green warbler, 502 out of 990 total songs for ovenbird). monitoR's automated detection process uses a 'score cutoff', which is the minimum match needed for an unknown event to be considered a detection and results in a true positive, true negative, false positive or false negative detection. At the chosen score cutoffs, monitoR correctly identified presence for black-throated green warbler and ovenbird in 64% and 72% of the 52 surveys using binary point matching, respectively, and 73% and 72% of the 52 surveys using spectrogram cross-correlation, respectively. Of individual songs, 72% of black-throated green warbler songs and 62% of ovenbird songs were identified by binary point matching. Spectrogram cross-correlation identified 83% of black-throated green warbler songs and 66% of ovenbird songs. False positive rates were < 1% for song event detection.