当前位置: X-MOL 学术Remote Sens. Ecol. Conserv. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Ecoacoustics: acoustic sensing for biodiversity monitoring at scale
Remote Sensing in Ecology and Conservation ( IF 3.9 ) Pub Date : 2020-08-02 , DOI: 10.1002/rse2.174
Dan Stowell 1 , Jérôme Sueur 2
Affiliation  

Nature sound recordings have been collected for over a hundred years, with an exponential increase since the 1950s (Ranft 2004). Most such recordings were taken in order to describe and decipher animal communication. However, the sounds of animals reveal more than behaviour: they also reflect the structure and functioning of the ecosystem of which the animals are a part. The practice of deploying remote acoustic sensors in natural environments has been systematized under the term ‘passive acoustic monitoring’ (PAM), a technical term mostly used in marine acoustics but then employed in terrestrial and aquatic acoustics (Gillespie et al. 2009; Marques et al. 2012). Acoustic sensing has distinct advantages which make it complementary to other sampling modalities. Like camera trapping, acoustic sensing can be used on land or under water, in all type of habitats. An acoustic sensor has the advantage that it can capture a wide spatial range (often 360° and about 100 m in terrestrial habitats), and is much less affected by occlusion than imagery. It can also record continuously or regularly over a long time period and can collect information of a full assemblage of species as it captures all the sounds in the surrounding environnment. These properties ensure a high sampling effort with a rather low technical investment (Ciira wa Maina 2016; Hill et al. 2018).

In many studies, acoustic data are analysed manually or with simple energy‐based detectors, and with the goal of targeted monitoring of a single species (Dawson and Efford 2009; Gillespie et al. 2009; Digby et al. 2013). However, ambient sound recordings such as those obtained with automatic devices contain evidence for a long list of ecological information, such as: species absence/presence, population density, population structure, community structure, landscape architecture, animal phenology, reproduction period, migration period, species interactions or ecosystem functions. Many of these only become evident through large‐scale studies, with analysis methods tailored to acoustic data. Benefiting from growth in recent decades of the scale of data capture and processing, the focus in acoustic monitoring can shift to broader ecosystem‐level questions, while using audio as a prime source of evidence. This is the main goal of ecoacoustics (Sueur and Farina 2015).

Ecoacoustic methods can cover all types of environment from deep sea to tropical forest, and are complementary to other biodiversity monitoring techniques such as camera trapping, LIDAR, satellite‐based remote sensing or environmental DNA. Research in ecoacoustic methods has grown massively over the past 15 years, developing methodology in hardware devices, signal processing, machine learning and visualization (see Sugai et al. 2019, this issue, for a review). Particularly important is the move from ecoacoustics as a fundamental to an applied science, with such methods being deployed in practical conservation and ecosystem monitoring (Ciira wa Maina 2016; Gordon et al. 2019; Sertlek et al. 2019; Znidersic et al. 2020). Within the context of the United Nations Sustainable Development Goals (UN SDGs), ecoacoustic methods have already been demonstrated to contribute useful evidence, which can complement other evidence sources. Within SDG 14 ‘Life below water’ and SDG 15 ‘Life on land’, these include monitoring threatened species (Braune et al. 2008; Hill et al. 2018), invasive species (Grant and Grant 2010), poaching (Hill et al. 2018), noise pollution (both on land and below water) (Fairbrass et al. 2018; Sertlek et al. 2019), land degradation and mountain ecosystems (Helbig‐Bonitz et al. 2015).

Much of the recent technical progress has been at the level of signal analysis and sound classification, in particular the development of acoustic indices on the one hand, and the use of deep learning tools on the other (Stowell et al. 2019; Joly et al. 2019), and from signal processing engineering work on representations and transformations of audio data (Sueur et al. 2014; Phillips et al. 2018). At the sensor level, recent progress is in low‐cost innovation (Hill et al. 2018), and the main challenge is now to have connected sensors so that data streams can be integrated continuously (Roch et al. 2017; Sethi et al. 2018) or to have integrated systems being able to run signal analyses and classification directly on board.

Large‐scale acoustic methods should now be transferred to application, and used more widely for conservation and management. It is now time to use ecoacoustics as a tool. Acoustic sensors should be included in large scale (i.e. national and international) monitoring programmes, in complementary fashion to other standard methods and in particular to design acoustic monitoring into long term programs.

As cited above, there are many documented case studies and methodological developments that support this move. We are pleased to introduce this special issue of Remote Sensing in Ecology and Conservation on ecoacoustics, demonstrating across many different ecosystems the value and maturity of ecoacoustic methods.

Methodologically, there are two broad paradigms in ecoacoustics, and they are reflected in this issue. One paradigm measures the acoustic diversity of a soundscape through the computation of acoustic indices: algorithmically straightforward and highly scalable, these indices yield evidence of biodiversity that is implicit, but holistic across many taxa. Sánchez‐Giraldo et al. (2020) and Roca and Van Opzeeland (2019) conduct large‐scale studies in very different ecological contexts – respectively in forests in the Columbian Andes, and underwater in the Southern Ocean – and quantify the reliability of such indices. Sánchez‐Giraldo et al. (2020) tackle the widely encountered issue of the effect of rain noise on index computation, while Roca and Van Opzeeland (2019) reveal acoustic significant differences between distinct Antarctic marine habitats using a set of indices. Campos‐Cerqueira et al. (2019) develop another type of index by extracting compressed data from a long‐term spectrogram representation. This study is one of the first to test for the efficiency of conservation policy, supporting the idea that ecoacoustics should now conduct applied research.

The second paradigm involves detecting or counting individual acoustic events, often limited to chosen target species. This offers a higher degree of selectivity, but can only be performed approximately when applied automatically at large scale. In a thorough review of passive acoustic monitoring techniques, Sugai et al. (2019) propose a set of good practices for designing the design of such automated surveys. This constitutes a crucial step towards the standardization of ecoacoustic data collection. Smith (2020) define a data sampling protocol suitable for very long duration (multi‐year) acoustic monitoring, which focuses on seasonally varying patterns of peak acoustic activity. They demonstrate in a field study that it can produce comparable results as manual field transects, with less than a quarter of effective survey effort. Yip et al. (2019) demonstrate that sound level measurements can improve population density estimates, serving as a proxy measure for the distance between sound event and autonomous recording unit.

Both methodological paradigms can be applied to monophonic or to multi‐channel audio recordings. In either case, there will usually be multiple animals audible on any given audio track. Lin and Tsao (2019) provide a review and roadmap of source‐separation methods including recent techniques that may help to disentangle overlapping sounds in monophonic recordings. Sumitani et al. (2020) demonstrate that interaction patterns among vocalizing individuals can be characterized with the aid of a dimension reduction algorithm coupled to a new compact microphone array, leading to automatic source localization.

The projects represented in this volume review methods in use, introduce new tools and apply with success ecoacoustic principles in very different geographic and environmental contexts. Altogether, these contributions reveal that ecoacoustic evidence can now inform at large scale national and international biodiversity policy.



中文翻译:

生态声学:用于生物多样性监测的声学传感

自从1950年代以来,自然声音记录已经收集了一百多年,并且呈指数级增长(Ranft 2004)。大部分此类录音都是为了描述和破译动物交流而拍摄的。但是,动物的声音不仅仅表现为行为:它们还反映出动物所参与的生态系统的结构和功能。在自然环境中部署远程声学传感器的做法已通过“无源声学监控”(PAM)进行了系统化,该术语主要用于海洋声学,然后又用于陆地和水生声学(Gillespie等,2009; Marques等)人2012)。声学传感具有明显的优势,使其与其他采样方式互补。像相机陷阱一样,声波感测可用于所有类型的栖息地的陆地或水下。声传感器的优势在于它可以捕获宽广的空间范围(在陆地栖息地中通常为360°和约100 m),并且与图像相比,闭塞的影响要小得多。它还可以长时间连续或定期记录声音,并且可以捕获周围环境中的所有声音,从而收集各种物种的信息。这些特性可确保以较低的技术投入进行大量采样(Ciira wa Maina 2016; Hill等人,2018)。

在许多研究中,以人工方式或使用简单的基于能量的探测器对声学数据进行分析,目标是对单个物种进行有针对性的监测(Dawson和Efford,2009; Gillespie等,2009; Digby等,2013)。)。但是,诸如通过自动设备获得的环境声音记录中包含一长串生态信息的证据,例如:物种的缺失/存在,种群密度,种群结构,群落结构,景观结构,动物物候,繁殖期,迁徙期,物种相互作用或生态系统功能。其中许多只有通过大规模研究才能变得明显,这些研究是针对声学数据量身定制的分析方法。得益于近几十年来数据捕获和处理规模的增长,声学监控的重点可以转移到更广泛的生态系统级问题,同时将音频用作主要证据。这是生态声学的主要目标(Sueur和Farina,2015年)。

生态声学方法可以覆盖从深海到热带森林的所有类型的环境,并且是其他生物多样性监测技术(例如照相机诱集,激光雷达,基于卫星的遥感或环境DNA)的补充。在过去的15年中,生态声学方法的研究得到了巨大的发展,在硬件设备,信号处理,机器学习和可视化方面开发了方法论(请参见Sugai等人,2019年,本期综述)。尤为重要的是,从生态声学作为基础科学转向应用科学,并将此类方法部署在实际的保护和生态系统监测中(Ciira wa Maina 2016 ; Gordon等2019 ; Sertlek等2019 ; Znidersic等2020)。)。在联合国可持续发展目标(UN SDGs)的背景下,生态声学方法已经被证明可以提供有用的证据,可以补充其他证据来源。在可持续发展目标14``水下生活''和可持续发展目标15``陆地生活''中,这些活动包括监测受威胁物种(Braune等人2008 ; Hill等人2018),入侵物种(Grant和Grant 2010),偷猎(Hill等人)。 (2018),噪声污染(包括陆地和水下)(Fairbrass等,2018 ; Sertlek等,2019),土地退化和山区生态系统(Helbig‐Bonitz等,2015)。

最近的许多技术进步都在信号分析和声音分类方面,特别是一方面是声学指标的发展,另一方面是深度学习工具的使用(Stowell等人2019 ; Joly等人(2019),以及信号处理工程对音频数据表示和转换的研究(Sueur等人2014 ; Phillips等人2018)。在传感器方面,低成本创新的最新进展是(Hill等人,2018年),现在的主要挑战是连接传感器,以便可以连续集成数据流(Roch等人,2017年; Sethi等人,2017年)2018年)或具有能够直接在船上运行信号分析和分类的集成系统。

现在应将大规模声学方法转移到应用中,并更广泛地用于保存和管理。现在是时候使用生态声学作为工具了。声学传感器应以与其他标准方法互补的方式包括在大规模(即国家和国际)监测计划中,尤其是将声学监测设计为长期计划。

如上所述,有许多记录在案的案例研究和方法论发展都支持这一举措。我们很高兴介绍生态学和生态学遥感专题,介绍生态声学,在许多不同的生态系统中展示了生态声学方法的价值和成熟度。

从方法上讲,生态声学有两种广泛的范式,它们反映在这个问题上。一种范例是通过计算声学指标来衡量声景的声学多样性:算法上直接且高度可扩展,这些指标提供了生物多样性的隐含证据,但在许多分类单元中都是完整的。Sánchez-Giraldo等。(2020年)和Roca和Van Opzeeland(2019年)分别在哥伦比亚安第斯山脉的森林和南大洋的水下,在非常不同的生态环境中进行了大规模研究,并对这些指标的可靠性进行了量化。Sánchez-Giraldo等。(2020年)解决了降雨噪声对指数计算的影响时遇到的广泛问题,而Roca和Van Opzeeland(2019)使用一组指标揭示了南极不同海洋生境之间的声学​​显着差异。Campos-Cerqueira等。(2019)通过从长期频谱图表示中提取压缩数据来开发另一种类型的索引。这项研究是第一个测试保护政策效率的研究,支持生态声学现在应该进行应用研究的观点。

第二种范例涉及检测或计数单个声学事件,通常仅限于所选目标物种。这提供了更高的选择性,但是只有在大规模自动应用时才能大约执行。在对无源声学监测技术的全面回顾中,Sugai等人。(2019)提出了一套设计此类自动调查设计的良好实践。这是迈向生态声学数据收集标准化的关键一步。史密斯(2020)定义适用于持续时间很长(多年)的声学监测的数据采样协议,该协议重点关注峰值声学活动的季节性变化模式。他们在野外研究中证明,它可以产生与手工野样类似的结果,而有效的调查工作不到四分之一。Yip等。(2019)证明了声级测量可以改善人口密度估计,可以作为声音事件和自主录音单元之间距离的代理测量。

两种方法范式都可以应用于单声道或多通道音频录音。无论哪种情况,在任何给定的音轨上通常都会听到多只动物的声音。Lin和Tsao(2019)提供了源分离方法的回顾和路线图,包括最近的技术,这些技术可能有助于消除单声道录音中重叠的声音。Sumitani等。(2020)证明了发声个体之间的交互模式可以借助降维算法与新的紧凑型麦克风阵列耦合来表征,从而实现自动声源定位。

本批量审查方法所代表的项目正在使用,引入了新的工具并在非常不同的地理和环境环境中成功应用了生态声学原理。总而言之,这些贡献表明,生态声学证据现在可以为大规模的国家和国际生物多样性政策提供信息。

更新日期:2020-09-22
down
wechat
bug