Publications
18 results found
Ross SRP-J, O'Connell DP, Deichmann JL, et al., 2023, Passive acoustic monitoring provides a fresh perspective on fundamental ecological questions, FUNCTIONAL ECOLOGY, ISSN: 0269-8463
Norman DL, Bischoff PH, Wearn OR, et al., 2022, Can CNN-based species classification generalise across variation in habitat within a camera trap survey?, METHODS IN ECOLOGY AND EVOLUTION, ISSN: 2041-210X
Cretois B, Rosten CM, Sethi SS, 2022, Voice activity detection in eco-acoustic data enables privacy protection and is a proxy for human disturbance, METHODS IN ECOLOGY AND EVOLUTION, Vol: 13, Pages: 2865-2874, ISSN: 2041-210X
Sethi SS, Kovac M, Wiesemueller F, et al., 2022, Biodegradable sensors are ready to transform autonomous ecological monitoring, NATURE ECOLOGY & EVOLUTION, Vol: 6, Pages: 1245-1247, ISSN: 2397-334X
- Author Web Link
- Cite
- Citations: 2
Sethi SS, Ewers RM, Jones NS, et al., 2021, Soundscapes predict species occurrence in tropical forests, OIKOS, Vol: 2022, Pages: 1-9, ISSN: 0030-1299
Accurate occurrence data is necessary for the conservation of keystone or endangered species, but acquiring it is usually slow, laborious and costly. Automated acoustic monitoring offers a scalable alternative to manual surveys but identifying species vocalisations requires large manually annotated training datasets, and is not always possible (e.g. for lesser studied or silent species). A new approach is needed that rapidly predicts species occurrence using smaller and more coarsely labelled audio datasets. We investigated whether local soundscapes could be used to infer the presence of 32 avifaunal and seven herpetofaunal species in 20 min recordings across a tropical forest degradation gradient in Sabah, Malaysia. Using acoustic features derived from a convolutional neural network (CNN), we characterised species indicative soundscapes by training our models on a temporally coarse labelled point-count dataset. Soundscapes successfully predicted the occurrence of 34 out of the 39 species across the two taxonomic groups, with area under the curve (AUC) metrics from 0.53 up to 0.87. The highest accuracies were achieved for species with strong temporal occurrence patterns. Soundscapes were a better predictor of species occurrence than above-ground carbon density – a metric often used to quantify habitat quality across forest degradation gradients. Our results demonstrate that soundscapes can be used to efficiently predict the occurrence of a wide variety of species and provide a new direction for data driven large-scale assessments of habitat suitability.
Heath R, Orme DS, Sethi CSL, et al., 2021, How index selection, compression, and recording schedule impact the description of ecological soundscapes, Evolutionary Ecology, Vol: 11, Pages: 13206-13217, ISSN: 0269-7653
Acoustic indices derived from environmental soundscape recordings are being used to monitor ecosystem health and vocal animal biodiversity. Soundscape data can quickly become very expensive and difficult to manage, so data compression or temporal down-sampling are sometimes employed to reduce data storage and transmission costs. These parameters vary widely between experiments, with the consequences of this variation remaining mostly unknown.We analyse field recordings from North-Eastern Borneo across a gradient of historical land use. We quantify the impact of experimental parameters (MP3 compression, recording length and temporal subsetting) on soundscape descriptors (Analytical Indices and a convolutional neural net derived AudioSet Fingerprint). Both descriptor types were tested for their robustness to parameter alteration and their usability in a soundscape classification task.We find that compression and recording length both drive considerable variation in calculated index values. However, we find that the effects of this variation and temporal subsetting on the performance of classification models is minor: performance is much more strongly determined by acoustic index choice, with Audioset fingerprinting offering substantially greater (12%–16%) levels of classifier accuracy, precision and recall.We advise using the AudioSet Fingerprint in soundscape analysis, finding superior and consistent performance even on small pools of data. If data storage is a bottleneck to a study, we recommend Variable Bit Rate encoded compression (quality = 0) to reduce file size to 23% file size without affecting most Analytical Index values. The AudioSet Fingerprint can be compressed further to a Constant Bit Rate encoding of 64 kb/s (8% file size) without any detectable effect. These recommendations allow the efficient use of restricted data storage whilst permitting comparability of results between different studies.
Sethi S, Ewers R, Jones N, et al., 2020, SAFE Acoustics: an open-source, real-time eco-acoustic monitoring network in the tropical rainforests of Borneo, Methods in Ecology and Evolution, Vol: 11, Pages: 1182-1185, ISSN: 2041-210X
1. Automated monitoring approaches offer an avenue to unlocking large‐scale insight into how ecosystems respond to human pressures. However, since data collection and data analyses are often treated independently, there are currently no open‐source examples of end‐to‐end, real‐time ecological monitoring networks. 2. Here, we present the complete implementation of an autonomous acoustic monitoring network deployed in the tropical rainforests of Borneo. Real‐time audio is uploaded remotely from the field, indexed by a central database, and delivered via an API to a public‐facing website.3. We provide the open‐source code and design of our monitoring devices, the central web2py database, and the ReactJS website. Furthermore, we demonstrate an extension of this infrastructure to deliver real‐time analyses of the eco‐acoustic data. 4. By detailing a fully functional, open source, and extensively tested design, our work will accelerate the rate at which fully autonomous monitoring networks mature from technological curiosities, and towards genuinely impactful tools in ecology.
Sethi SS, Ewers RM, Jones NS, et al., 2020, Soundscapes predict species occurrence in tropical forests, Publisher: Cold Spring Harbor Laboratory
Accurate occurrence data is necessary for the conservation of keystone or endangered species, but acquiring it is usually slow, laborious, and costly. Automated acoustic monitoring offers a scalable alternative to manual surveys, but identifying species vocalisations requires large manually annotated training datasets, and is not always possible (e.g., for silent species). A new, intermediate approach is needed that rapidly predicts species occurrence without requiring extensive labelled data.We investigated whether local soundscapes could be used to infer the presence of 32 avifaunal and seven herpetofaunal species across a tropical forest degradation gradient in Sabah, Malaysia. We developed a machine-learning based approach to characterise species indicative soundscapes, training our models on a coarsely labelled manual point-count dataset.Soundscapes successfully predicted the occurrence of 34 out of the 39 species across the two taxonomic groups, with area under the curve (AUC) metrics of up to 0.87 (Bold-striped Tit-babbler Macronus bornensis). The highest accuracies were achieved for common species with strong temporal occurrence patterns.Soundscapes were a better predictor of species occurrence than above-ground biomass – a metric often used to quantify habitat quality across forest degradation gradients.Synthesis and applications: Our results demonstrate that soundscapes can be used to efficiently predict the occurrence of a wide variety of species. This provides a new direction for audio data to deliver large-scale, accurate assessments of habitat suitability using cheap and easily obtained field datasets.
Sethi S, Jones NS, Fulcher B, et al., 2020, Characterising soundscapes across diverse ecosystems using a universal acoustic feature set, Proceedings of the National Academy of Sciences of USA, Vol: 117, Pages: 17049-17055, ISSN: 0027-8424
Natural habitats are being impacted by human pressures at an alarming rate. Monitoring these ecosystem-level changes often requires labor-intensive surveys that are unable to detect rapid or unanticipated environmental changes. Here we have developed a generalizable, data-driven solution to this challenge using eco-acoustic data. We exploited a convolutional neural network to embed soundscapes from a variety of ecosystems into a common acoustic space. In both supervised and unsupervised modes, this allowed us to accurately quantify variation in habitat quality across space and in biodiversity through time. On the scale of seconds, we learned a typical soundscape model that allowed automatic identification of anomalous sounds in playback experiments, providing a potential route for real-time automated detection of irregular environmental behavior including illegal logging and hunting. Our highly generalizable approach, and the common set of features, will enable scientists to unlock previously hidden insights from acoustic data and offers promise as a backbone technology for global collaborative autonomous ecosystem monitoring efforts.
Fulcher B, Lubba C, Sethi S, et al., 2020, A self-organizing, living library of time-series data, Scientific Data, Vol: 7, ISSN: 2052-4463
Time-series data are measured across the sciences, from astronomy to biomedicine, but meaningful cross-disciplinary interactions are limited by the challenge of identifying fruitful connections. Here we introduce the web platform, CompEngine, a self-organizing, living library of time-series data, that lowers the barrier to forming meaningful interdisciplinary connections between time series. Using a canonical feature-based representation, CompEngine places all time series in a common feature space, regardless of their origin, allowing users to upload their data and immediately explore diverse data with similar properties, and be alerted when similar data is uploaded in future. In contrast to conventional databases which are organized by assigned metadata, CompEngine incentivizes data sharing by automatically connecting experimental and theoretical scientists across disciplines based on the empirical structure of the data they measure. CompEngine’s growing library of interdisciplinary time-series data also enables the comprehensive characterization of time-series analysis algorithms across diverse types of empirical data.
Signorelli A, Sethi S, Orme D, 2020, aaronSig/rainforest-rhythms: Version 1.0.0 Snapshot Release
This release is a snapshot of the SAFE Acoustics website for the publication of the paper describing the system.
Sethi S, Jones N, Fulcher B, et al., 2019, Combining machine learning and a universal acoustic feature-set yields efficient automated monitoring of ecosystems, Publisher: bioRxiv
Natural habitats are being impacted by human pressures at an alarming rate. Monitoring these ecosystem-level changes often requires labour-intensive surveys that are unable to detect rapid or unanticipated environmental changes. Here we developed a generalisable, data-driven solution to this challenge using eco-acoustic data. We exploited a convolutional neural network to embed ecosystem soundscapes from a wide variety of biomes into a common acoustic space. In both supervised and unsupervised modes, this allowed us to accurately quantify variation in habitat quality across space and in biodiversity through time. On the scale of seconds, we learned a typical soundscape model that allowed automatic identification of anomalous sounds in playback experiments, paving the way for real-time detection of irregular environmental behaviour including illegal activity. Our highly generalisable approach, and the common set of features, will enable scientists to unlock previously hidden insights from eco-acoustic data and offers promise as a backbone technology for global collaborative autonomous ecosystem monitoring efforts.
Lubba CH, Sethi SS, Knaute P, et al., 2019, catch22: CAnonical time-series CHaracteristics, Data Mining and Knowledge Discovery, Vol: 33, Pages: 1821-1852, ISSN: 1384-5810
Capturing the dynamical properties of time series concisely as interpretable feature vectors can enable efficient clustering and classification for time-series applications across science and industry. Selecting an appropriate feature-based representation of time series for a given application can be achieved through systematic comparison across a comprehensive time-series feature library, such as those in the hctsa toolbox. However, this approach is computationally expensive and involves evaluating many similar features, limiting the widespread adoption of feature-based representations of time series for real-world applications. In this work, we introduce a method to infer small sets of time-series features that (i) exhibit strong classification performance across a given collection of time-series problems, and (ii) are minimally redundant. Applying our method to a set of 93 time-series classification datasets (containing over 147,000 time series) and using a filtered version of the hctsa feature library (4791 features), we introduce a set of 22 CAnonical Time-series CHaracteristics, catch22, tailored to the dynamics typically encountered in time-series data-mining tasks. This dimensionality reduction, from 4791 to 22, is associated with an approximately 1000-fold reduction in computation time and near linear scaling with time-series length, despite an average reduction in classification accuracy of just 7%. catch22 captures a diverse and interpretable signature of time series in terms of their properties, including linear and non-linear autocorrelation, successive differences, value distributions and outliers, and fluctuation scaling properties. We provide an efficient implementation of catch22, accessible from many programming environments, that facilitates feature-based time-series analysis for scientific, industrial, financial and medical applications using a common language of interpretable time-series properties.
Lubba CH, Sethi SS, Knaute P, et al., 2019, <i>catch22</i>: CAnonical Time-series CHaracteristics <i>selected through</i> highly comparative time-series analysis, Publisher: Cold Spring Harbor Laboratory
<jats:title>Abstract</jats:title><jats:p>Capturing the dynamical properties of time series concisely as interpretable feature vectors can enable efficient clustering and classification for time-series applications across science and industry. Selecting an appropriate feature-based representation of time series for a given application can be achieved through systematic comparison across a comprehensive time-series feature library, such as those in the <jats:italic>hctsa</jats:italic> toolbox. However, this approach is computationally expensive and involves evaluating many similar features, limiting the widespread adoption of feature-based representations of time series for real-world applications. In this work, we introduce a method to infer small sets of time-series features that (i) exhibit strong classification performance across a given collection of time-series problems, and (ii) are minimally redundant. Applying our method to a set of 93 time-series classification datasets (containing over 147 000 time series, including biomedical datasets) and using a filtered version of the <jats:italic>hctsa</jats:italic> feature library (4791 features), we introduce a generically useful set of 22 CAnonical Time-series CHaracteristics, <jats:italic>catch22</jats:italic>. This dimensionality reduction, from 4791 to 22, is associated with an approximately 1000-fold reduction in computation time and near linear scaling with time-series length, despite an average reduction in classification accuracy of just 7%. <jats:italic>catch22</jats:italic> captures a diverse and interpretable signature of time series in terms of their properties, including linear and non-linear autocorrelation, successive differences, value distributions and outliers, and fluctuation scaling properties. We provide an efficient implementation of <jats:italic>catch22</jats:italic>, accessible from many programming environments, tha
Sethi S, Ewers R, Jones N, et al., 2018, Robust, real-time and autonomous monitoring of ecosystems with an open, low-cost, networked device, Methods in Ecology and Evolution, Vol: 9, Pages: 2383-2387, ISSN: 2041-210X
1. Automated methods of monitoring ecosystems provide a cost-effective way to track changes in natural system's dynamics across temporal and spatial scales. However, methods of recording and storing data captured from the field still require significant manual effort. 2. Here we introduce an open source, inexpensive, fully autonomous ecosystem monitoring unit for capturing and remotely transmitting continuous data streams from field sites over long time-periods. We provide a modular software framework for deploying various sensors, together with implementations to demonstrate proof of concept for continuous audio monitoring and time-lapse photography. 3. We show how our system can outperform comparable technologies for fractions of the cost, provided a local mobile network link is available. The system is robust to unreliable network signals and has been shown to function in extreme environmental conditions, such as in the tropical rainforests of Sabah, Borneo. 4. We provide full details on how to assemble the hardware, and the open-source software. Paired with appropriate automated analysis techniques, this system could provide spatially dense, near real-time, continuous insights into ecosystem and biodiversity dynamics at a low cost.
Sethi SS, Zerbi V, Wenderoth N, et al., 2017, Structural connectome topology relates to regional BOLD signal dynamics in the mouse brain, Chaos, Vol: 27, ISSN: 1054-1500
Brain dynamics are thought to unfold on a network determined by the pattern of axonalconnections linking pairs of neuronal elements; the so-called connectome. Prior work has indicatedthat structural brain connectivity constrains pairwise correlations of brain dynamics(“functional connectivity”), but it is not known whether inter-regional axonal connectivity isrelated to the intrinsic dynamics of individual brain areas. Here we investigate this relationshipusing a weighted, directed mesoscale mouse connectome from the Allen Mouse BrainConnectivity Atlas and resting state functional MRI (rs-fMRI) time-series data measured in 184brain regions in eighteen anesthetized mice. For each brain region, we measured degree,betweenness, and clustering coefficient from weighted and unweighted, and directed and undirectedversions of the connectome. We then characterized the univariate rs-fMRI dynamics ineach brain region by computing 6930 time-series properties using the time-series analysis toolbox,hctsa. After correcting for regional volume variations, strong and robust correlationsbetween structural connectivity properties and rs-fMRI dynamics were found only when edgeweights were accounted for, and were associated with variations in the autocorrelation propertiesof the rs-fMRI signal. The strongest relationships were found for weighted in-degree, which waspositively correlated to the autocorrelation of fMRI time series at time lag s ¼ 34 s (partialSpearman correlation q ¼ 0:58), as well as a range of related measures such as relative high frequencypower (f > 0.4 Hz: q ¼ 0:43). Our results indicate that the topology of inter-regionalaxonal connections of the mouse brain is closely related to intrinsic, spontaneous dynamics suchthat regions with a greater aggregate strength of incoming projections display longer timescalesof activity fluctuations.
Sethi SS, Ewers RM, Jones NS, et al., Robust, real-time and autonomous monitoring of ecosystems with an open, low-cost, networked device
<jats:title>Abstract</jats:title><jats:p><jats:list list-type="order"><jats:list-item><jats:p>Automated methods of monitoring ecosystems provide a cost-effective way to track changes in natural system’s dynamics across temporal and spatial scales. However, methods of recording and storing data captured from the field still require significant manual effort.</jats:p></jats:list-item><jats:list-item><jats:p>Here we introduce an open source, inexpensive, fully autonomous ecosystem monitoring unit for capturing and remotely transmitting continuous data streams from field sites over long time-periods. We provide a modular software framework for deploying various sensors, together with implementations to demonstrate proof of concept for continuous audio monitoring and time-lapse photography.</jats:p></jats:list-item><jats:list-item><jats:p>We show how our system can outperform comparable technologies for fractions of the cost, provided a local mobile network link is available. The system is robust to unreliable network signals and has been shown to function in extreme environmental conditions, such as in the tropical rainforests of Sabah, Borneo.</jats:p></jats:list-item><jats:list-item><jats:p>We provide full details on how to assemble the hardware, and the open-source software. Paired with appropriate automated analysis techniques, this system could provide spatially dense, near real-time, continuous insights into ecosystem and biodiversity dynamics at a low cost.</jats:p></jats:list-item></jats:list></jats:p>
Sethi SS, Ewers RM, Jones NS, et al., SAFE Acoustics: an open-source, real-time eco-acoustic monitoring network in the tropical rainforests of Borneo, Publisher: Cold Spring Harbor Laboratory
<jats:title>Abstract</jats:title><jats:p><jats:list list-type="order"><jats:list-item><jats:p>Automated monitoring approaches offer an avenue to deep, large-scale insight into how ecosystems respond to human pressures. Since sensor technology and data analyses are often treated independantly, there are no open-source examples of end-to-end, real-time ecological monitoring networks.</jats:p></jats:list-item><jats:list-item><jats:p>Here, we present the complete implementation of an autonomous acoustic monitoring network deployed in the tropical rainforests of Borneo. Real-time audio is uploaded remotely from the field, indexed by a central database, and delivered via an API to a public-facing website.</jats:p></jats:list-item><jats:list-item><jats:p>We provide the open-source code and design of our monitoring devices, the central web2py database and the ReactJS website. Furthermore, we demonstrate an extension of this infrastructure to deliver real-time analyses of the eco-acoustic data.</jats:p></jats:list-item><jats:list-item><jats:p>By detailing a fully functional, open-source, and extensively tested design, our work will accelerate the rate at which fully autonomous monitoring networks mature from technological curiosities, and towards genuinely impactful tools in ecology.</jats:p></jats:list-item></jats:list></jats:p>
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.