Community science includes the participation of people who are not professional scientists in the collection, analysis or interpretation of scientific data. Two broad approaches were distinguished in the review. In guided projects, scientists lead data collection through an established protocol and work directly with volunteers or organizations, sometimes through platforms such as Zooniverse. In open projects, individuals work largely independently and share records through social media or apps such as iNaturalist. These approaches can expand the geographic coverage and volume of ecological data available to researchers, particularly when observations are contributed over many locations and across extended periods.

Orthopterans are prominent members of the insect group that represents a large share of Earth’s biodiversity, a group for which reports of severe insect declines have raised concern. They occupy rural and suburban backyards, urban parks and vacant lots, where the calls of many males form an acoustic feature of summer. Their short life cycles and, in some species, specialization in habitat, food and egg-laying make them useful indicators in climate-change research. However, the review notes that there is still too little information about extant insect biodiversity to understand the causes and consequences of population declines, strengthening the potential value of continuous community-science monitoring.

Many male orthopterans call to attract mates, which makes it relatively straightforward to quantify species richness, abundance and emergence times from acoustic surveys. Scientists can identify species from their acoustic profiles, while community scientists can record sounds from trails and roadsides without collecting specimens. Acoustic monitoring can be especially useful in fragile habitats or for threatened species because it may reduce the need for repeated physical disturbance. A single recording can also contain calls from several taxa, potentially allowing existing recordings to be used more extensively, increasing the rate of new data collection, reducing costs and encouraging collaboration.

Smartphones provide many of the tools needed for such work, including audio and video recording, global positioning and internet access. Tools such as TADARIDA, a toolbox for animal detection in acoustic recordings that integrates discriminant analysis, as well as artificial-intelligence methods, can make large collections of acoustic and photographic data more useful. The review describes nationwide community-science acoustic monitoring in some locations and taxa, including FrogID and bird-monitoring programs. These examples show how the same basic technology can support repeated observations, but the reviewed orthopteran literature itself contained only four studies that directly used community-collected acoustic data.

The scale of open participation was illustrated by iNaturalist records available on July 14, 2022. The platform contained 1,128,486 orthopteran records accompanied by photographs and observation localities for 5,732 species. Such records can help researchers identify occurrences, document range expansion and track other distributional patterns. Local participation may also provide valuable knowledge about where singing insects occur, while residents can contribute observations throughout the year, including periods when distant research teams are unlikely to be conducting fieldwork. Community-science initiatives have also been used to monitor conservation efforts, rediscover species thought to be extinct, identify new species and document invasive species or range expansion.

The review characterized 43% of the 14 studies as guided and 50% as open, leaving the type unclear in the remaining 7%. Most major orthopteran groups were represented, including grasshoppers, crickets, katydids and wetas, but taxonomic coverage within each group was limited largely to new, rare or invasive species. Guided studies generally involved groups of fewer than 15 people, whereas open studies typically did not report the number of nonprofessional participants. This omission prevents consistent evaluation of project reach and the amount of volunteer labor contributed, making participant numbers a relevant measure when comparing the scale of different initiatives.

Online analysis provides another route for community scientists to contribute to orthopteran research. The ongoing Cricket Wing project on Zooniverse asks participants to classify microscope images of live and dead cricket sperm cells and immune cells, or hemocytes. The underlying research examines how traffic noise may affect immune and reproductive traits, areas for which relatively little physiological evidence about orthopterans is available. At the time of the source’s July 14, 2022 access, the project had generated 12,304 sperm-cell images and 1,917 hemocyte images. Each image was classified ten times before the final count was recorded, a redundancy measure intended to control classification error and bias.

By July 14, 2022, Cricket Wing had registered 700 participants since its May 10, 2022 launch and recorded 38,497 classifications, comprising 37,356 sperm counts and 1,141 hemocyte counts. Guides, tutorials and an open discussion forum help participants troubleshoot tasks and learn about the broader science. The project is also being used in high-school outreach, while its developers plan to extend it to questions involving rapid adaptation through song analysis and machine learning. These are future directions rather than reported experimental results; the current project’s demonstrated role is the large-scale community analysis of images for the research into cricket immune and reproductive traits.

Effective programs require clear, concise protocols that specify how volunteers access study organisms, where target species can be found, how sounds should be recorded and how observations should be documented. Instructions should cover recording distance and duration, locality, date, time, habitat and general notes, as well as procedures for backing up or sharing files. Workshops, field notes, video demonstrations, online tutorials and example recordings can provide training, while an “unsure” option may reduce forced identifications. Training duration, practice and the mode of instruction can affect data quality, and replication through independent observers or repeated classifications can help reveal error and observer variation.

Projects must also account for spatial and temporal sampling bias, as well as the difficulty of correcting uneven coverage without introducing further problems. The authors therefore recommend matching analytical techniques to the question, planning sampling in advance and determining the appropriate balance between community-generated and professionally collected data. Despite the potential of community scientists to track species abundance, richness, emergence time and distributional change through sound, organized long-term orthopteran programs remain scarce. With training, replication and sampling plans designed to reduce avoidable harm, community science can turn the calls of crickets, grasshoppers and other orthopterans into sustained ecological evidence while engaging the public in the research process.

Subject of Research: Community science for orthopteran ecological monitoring and image-based analysis

Article Title: The role of community science in orthopteran research

Article References: Byerly, A. R., Firneno Jr., T. J., Beard, R., & Larson, E. L. (2023). The role of community science in orthopteran research. Journal of Orthoptera Research, 32(2), 109-114. https://doi.org/10.3897/jor.32.90444

Image Credits: AI Generated

DOI: 10.3897/jor.32.90444

Keywords: community science, orthopterans, acoustic monitoring, biodiversity, grasshoppers, crickets, katydids, smartphone sensors, citizen science, conservation, role, community

The story begins with a puzzle. Oecanthus pellucens, a pale, delicate tree cricket first described in 1763, had been recorded on Madeira Island since 1938. But in 1993, Russian entomologist Andrei Gorochov described a new species, Oecanthus dulcisonans, from specimens collected in Tenerife and the Arabian Peninsula. The two species are morphologically so similar that field records made before 1993 cannot be reliably assigned to either one. Because Madeira’s historical records of O. pellucens predate the description of O. dulcisonans, and because online photographic records rarely capture the one structure that separates the two species, genuine doubt had arisen about which cricket actually inhabited the island.

The key to the mystery lies in a small but decisive anatomical feature: the metasternal plate, a structure on the underside of the thorax. In O. dulcisonans, this plate is longer and thinner than in O. pellucens. Male genitalia, specifically the epiphallus, offer a second distinguishing character, while body proportions provide supporting evidence. O. dulcisonans males measure roughly 14 to 17 millimeters in body length with hind wings protruding 3 to 4 millimeters beyond the forewings, whereas O. pellucens males are smaller, around 10 to 13 millimeters, with protruding wings of only about 1 millimeter. Armed with these criteria, the research team examined both freshly collected specimens and older material held in the University of Madeira Insect Collection.

The fieldwork itself relied on one of the most charming methods in entomology: eavesdropping on cricket song. During night surveys in June 2023 on Madeira, the researchers walked paths through vegetation between 22:00 and midnight, tracing the source of each stridulation. On Porto Santo in October 2022, they drove slowly along roads at 15 kilometers per hour between 20:00 and 23:00, stopping whenever a calling song reached them. Once located, insects were captured by hand or with a net, and each collection point was logged with GPS precision of three meters. This acoustic approach is particularly effective for tree crickets, whose songs are species-specific and can be analyzed quantitatively.

The results were striking. The team collected 21 specimens of O. dulcisonans on Madeira Island, at Ponta de São Lourenço on the island’s eastern tip and at Reis Magos on the south coast, plus a single male on Porto Santo at Campo de Baixo—the first record of any Oecanthus species on that island. Four confirmed specimens of O. pellucens were also documented on Madeira, from Boca da Corrida at 1,202 meters elevation, Garachico, and Ponta do Pargo, along with a nymph found on the endemic plant Rumex maderensis in the Funchal Ecological Park. In total, the archipelago’s Orthoptera list now stands at 29 species, with 27 on Madeira and 13 on Porto Santo.

Song analysis added an independent line of evidence. O. dulcisonans produces continuous long echemes without gaps, a signature trait confirmed by recordings made on Porto Santo at 23°C using a smartphone with 48 kHz sampling frequency and 16-bit resolution. The researchers generated oscillograms with the seewave package in R and compared their recordings with reference data for O. pellucens from the Orthoptera Species Files Online, which show regular gaps between echemes at an echeme repetition rate of approximately one per second. Because the two species’ songs differ in this continuous versus non-continuous pattern, acoustic monitoring offers a practical tool for mapping their distributions without needing to capture every individual.

Morphometric data from the new specimens revealed intriguing variation. The Porto Santo male of O. dulcisonans measured 17.34 millimeters in body length, slightly exceeding the 14.00 to 17.00 millimeter range reported elsewhere, while its hind femur at 7.03 millimeters fell below the 8.70 to 10.00 millimeter range documented in previous studies. Madeira males averaged 14.44 millimeters in body length with hind femora of 7.98 millimeters, also deviating from literature values. The mean oviscape length in Madeira females, 5.48 millimeters, exceeded the value reported by earlier authors. These discrepancies hint that the archipelago’s populations may be morphologically distinct, and the researchers propose genetic comparisons with continental European specimens to explore this possibility.

Ecologically, the two species showed a consistent altitudinal segregation that mirrors patterns observed in the Canary Islands and Mallorca. O. dulcisonans was prevalent at lower altitudes, stridulating on stems within patches of tall herbaceous vegetation dominated by the grass Hyparrhenia hirta, often between buildings on Porto Santo and near lighthouses and coastal roads on Madeira. O. pellucens, by contrast, was encountered at higher elevations, consistent with its reported association with tall trees and shrubs. This habitat association matters for conservation planning, since Oecanthus species are closely tied to specific vegetation types and microhabitats that can be protected or restored through informed land management.

The biogeographical significance of the discovery is considerable. With both species now confirmed on Madeira, the archipelago—together with the Canary Islands—marks the westernmost limit of the Palearctic ranges of O. dulcisonans and O. pellucens, meaning their distributional extension within the region is larger than previously thought. Yet these crickets appear surprisingly rare across Macaronesia. Since its first record in 1938, only 17 specimens of O. pellucens have been collected on Madeira, and reports from the Azores have not been repeated since 1999. Whether this apparent rarity reflects genuinely small populations, limited survey effort, or habitat constraints remains an open question that further research in the Macaronesian islands must address.

Conservation concerns loom over these findings. Only around 20 percent of Madeira’s native Laurel Forest survives today, and the island’s south coast has been largely converted to urban and agricultural land. Porto Santo, the oldest island in the archipelago at 14 million years, retains only scattered remnants of native vegetation on cliffs and mountaintops, while erosion and nutrient runoff have degraded its soils since the fifteenth century. In the past decade, intensive tourist construction has further diminished the coastal herbaceous habitats these crickets depend on. The researchers stress that confirming the species’ presence on Madeira’s north coast and conducting comprehensive surveys across both islands are urgent priorities, and they see the charismatic singing crickets as an opportunity to engage the public and students in biodiversity education on these increasingly fragile islands.

Subject of Research: New records of Oecanthus tree crickets in the Madeira archipelago

Article Title: New records of Oecanthus species ( Orthoptera , Oecanthidae ) in the Madeira archipelago

Article References: Silva, H. M., Avivar-Lozano, L., Gomes, G., Rhee, H., Dellinger, T., & Aguín-Pombo, D. (2024). New records of Oecanthus species (Orthoptera, Oecanthidae) in the Madeira archipelago. Journal of Orthoptera Research, 33(1), 95-102. https://doi.org/10.3897/jor.33.113947

Image Credits: AI Generated

DOI: 10.3897/jor.33.113947

Keywords: Oecanthus dulcisonans, Oecanthus pellucens, tree crickets, Orthoptera, Madeira, Porto Santo, Macaronesia, biogeography, acoustic monitoring, biodiversity, conservation, records

The research team, led by Vernon S. Peters, deployed six autonomous recording units across the burn area from fall 2018 through spring 2020. Rather than relying on labor-intensive visual surveys in rugged terrain, the researchers used KALEIDOSCOPE PRO acoustic analysis software with an advanced classifier to detect and quantify nutcracker calls across thousands of hours of audio. This passive acoustic approach allowed continuous monitoring across seasons and habitat types that would be nearly impossible to survey on foot at comparable intensity, and it captured the birds’ vocal activity as a proxy for habitat use.

The study was designed around a natural experiment in pine reproductive timing. Both whitebark and limber pine are masting species, meaning they produce cones in intermittently massive crops rather than steady annual amounts. Fall 2018 was a mast year, when cone production surged and nutcrackers descended to harvest and cache seeds. Spring 2019 followed, when those cached seeds would be retrieved, either to be eaten or, crucially, forgotten in ways that allow seedlings to establish. The researchers contrasted habitat use between this mast year and the subsequent non-mast year, between whitebark and limber pine habitat, and between burned and unburned stands, testing whether distance from remaining seed sources shaped visitation patterns.

Model selection using Akaike’s information criterion revealed that nutcracker habitat use depended on masting status, habitat type, burned habitat, distance from seed sources, and interactions among these factors in both fall and spring foraging periods. In the fall of the mast year, when both pine species produced high numbers of cones in close proximity, nutcrackers used whitebark pine habitat more heavily, particularly unburned stands. This pattern suggests concentrated caching behavior, as birds harvest seeds and bury them in sites they may later fail to recover, which is precisely the mechanism these pines depend upon for regeneration.

The spring results told a different and equally important story. Nutcracker habitat use was greater in the non-mast year, particularly in burned whitebark pine habitat. This shift makes ecological sense: after a mast year, birds return to retrieve the seeds they cached, and burned areas may offer favorable conditions for cache recovery. For the pines, lower habitat use the following spring carries a hidden benefit. When caches are retrieved, seeds are eaten rather than dispersed, so reduced nutcracker presence in spring means fewer caches are raided and more seeds survive to germinate. The 2018 mast year thus attracted seed dispersers when it mattered most, and the lower habitat use the following spring protected those caches from seed predation.

One of the most striking findings concerns the relative attractiveness of the two pine species. Whitebark pine attracted more nutcracker habitat use with fewer trees and fewer cones per tree than limber pine. This suggests that whitebark pine, despite its endangered status and often sparse distribution, exerts a disproportionately strong pull on its disperser, possibly through cone characteristics, stand structure, or landscape position. For limber pine, which co-occurs with whitebark at this northern range limit of five-needle pine overlap, the results imply that it may rely on the same nutcracker traffic drawn by whitebark, making the conservation of whitebark stands doubly important for the recovery of both species.

The prescribed burn itself emerged as a long-term asset rather than a temporary disturbance. The 2009 fire created attractive habitat for Clark’s Nutcracker nine to eleven years after the burn, demonstrating that opportunities for seed dispersal persist well beyond the immediate post-fire window. This finding challenges assumptions that nutcracker activity fades quickly after fire and instead shows that burned landscapes continue to function as dispersal corridors and caching grounds during the period when five-needle pine management is actively planned. For recovery strategies that use prescribed fire or manage natural burns, the results indicate that the window for natural regeneration on burns extends across at least a decade.

The study also highlights a critical intersection with disease management. White pine blister rust, caused by the fungal pathogen Cronartribicola, is a major threat to whitebark pine across its range, and current recovery strategies center on planting seedlings bred for rust resistance. The researchers emphasize that continued use of rust-resistant trees by dispersers is essential for natural regeneration to be effective. As existing plantings of rust-resistant seedlings reach cone production age in the future, attracting nutcrackers to those trees will leverage the investment in resistant stock, allowing the birds to disperse seeds from trees that can survive the disease and establish the next generation.

By combining passive acoustics with rigorous model selection, the study provides a template for monitoring disperser behavior across large, remote landscapes without the cost and bias of visual surveys. The approach captures seasonal dynamics that ground-based counts often miss, and it can be scaled to other systems where vocal birds serve as keystone dispersers. As climate change alters fire regimes and masting patterns across western North America, understanding when and where nutcrackers cache seeds becomes increasingly urgent for the survival of whitebark and limber pine at their northern range limit.

The findings carry direct implications for Alberta’s recovery plans for both endangered pines. Burns intended for five-needle pine management should be evaluated not just for immediate planting opportunities but for their capacity to attract nutcrackers during mast years, when cone crops in proximity to burned habitat create the conditions for caching. The study’s evidence that both pine species benefited from attracting dispersers in the 2018 mast year, and from reduced cache predation the following spring, suggests that synchronizing management actions with masting cycles could amplify natural regeneration. In a landscape where these pines face rust, fire exclusion, and climate stress, the humble nutcracker call recorded by a small device on a decade-old burn may prove to be one of the most valuable signals conservationists can listen for.

Subject of Research: Acoustic monitoring of Clark's Nutcracker habitat use during post-fire recovery of endangered whitebark and limber pines

Article Title: Acoustic analysis shows Clark’s Nutcracker seasonal and spatial habitat use during post-fire recovery of two endangered pine species

Article References: Peters, V. S., Vriend, J. L., Proppe, D. S., Buist, E., Greidanus, K., & Peters, C. E. (2026). Acoustic analysis shows Clark’s Nutcracker seasonal and spatial habitat use during post-fire recovery of two endangered pine species. Nature Conservation, 63, 99–126. https://doi.org/10.3897/natureconservation.63.171832

Image Credits: AI Generated

DOI: 10.3897/natureconservation.63.171832

Keywords: Clark's Nutcracker, whitebark pine, limber pine, seed dispersal, post-fire recovery, acoustic monitoring, masting, Alberta, endangered species, forest ecology, prescribed burn, seed caching

The challenge the researchers address is known in machine learning as open-set recognition, and it is fundamentally harder than the closed-set classification that dominates the literature. In closed-set scenarios, every test sample is guaranteed to belong to one of the training categories, so the model only needs to draw boundaries between known classes. In open-set scenarios, the model must simultaneously perform two tasks: classify samples from known classes correctly and identify samples from classes it has never seen as unknown. The scarcity of labeled sound samples makes both tasks harder, because learning discriminative representations and reliable class prototypes requires enough examples to characterize what each class sounds like and, just as importantly, where its boundaries lie.

Environmental sound classification has advanced rapidly with deep learning, with convolutional neural networks, attention-based residual networks, and transformer architectures all achieving strong results on benchmark datasets. But most of these studies assume closed-set conditions. The open-set problem in acoustic scene classification was formally identified nearly a decade ago, and subsequent work has explored rejection-based multi-class classification, CNN-based open-set audio recognition, and generative approaches. Prototype-based methods have emerged as a particularly promising family for open-set recognition because they represent each class by a point, or prototype, in feature space, and classify samples by their proximity to these points. Samples that fall far from every prototype can be flagged as unknown. The convolutional prototype network, originally developed for open-set recognition in computer vision, forms the foundation of the new approach.

The method proposed by Jiang and Li contains three complementary components, each targeting a specific weakness of prototype networks under limited data conditions. The first is a time-frequency attention mechanism that emphasizes perceptually salient components of the spectrogram, the two-dimensional representation of sound in which time runs along one axis and frequency along the other. Not all parts of a spectrogram are equally informative for distinguishing one sound from another. A dog bark may be characterized by transient broadband bursts, while a siren may be defined by narrow tonal lines that sweep across frequency. Attention mechanisms allow the network to learn which time-frequency regions carry the most discriminative information and to weight them more heavily during feature extraction, enhancing the quality of the learned representations.

The second component is a Mahalanobis distance-based complementary prototype loss, which exploits information from misclassified samples to refine prototype representations. In conventional prototype learning, prototypes are updated based on correctly classified samples, and misclassified samples are treated as noise. But misclassified samples carry valuable information: they reveal where the decision boundaries between classes are ambiguous and where prototypes are poorly positioned. The complementary loss draws on the complementary information contained in these difficult samples, using Mahalanobis distance, a metric that accounts for the covariance structure of the feature distribution rather than treating all directions in feature space equally, to reshape prototypes in ways that reduce confusion between classes. This is especially important when data is limited, because with few examples per class, every sample, including the ones the model gets wrong, matters for characterizing the class distribution.

The third component is a learnable Mahalanobis metric for adaptive similarity modeling between feature embeddings and class prototypes. In standard prototype networks, similarity is measured using Euclidean distance, which assumes that all feature dimensions contribute equally and independently to the distance calculation. The Mahalanobis metric relaxes this assumption by learning a full distance function that adapts to the geometry of the feature space. Features that are correlated, redundant, or noisy can be downweighted, while discriminative dimensions can be emphasized. Because the metric is learnable, the network can discover the most appropriate way to compare embeddings with prototypes for the specific sound classification task at hand, rather than relying on a fixed geometric assumption that may not match the structure of acoustic features.

Experiments on environmental sound datasets demonstrate that the proposed method consistently improves both known-class classification accuracy and unknown-class rejection, with the largest gains appearing in low-data open-set scenarios. This is a significant result because it addresses the two failure modes that plague prototype networks when data is scarce: prototypes that are poorly estimated from too few examples, and feature representations that are not discriminative enough to separate known classes from unknown ones. By combining attention-enhanced feature learning with prototype refinement from misclassified samples and an adaptive distance metric, the method builds a more robust geometric structure in feature space, one in which known classes form tight, well-separated clusters and unknown samples fall clearly outside all clusters.

The implications extend beyond the laboratory. Environmental sound recognition is increasingly deployed on embedded devices for applications including wildlife monitoring, urban noise assessment, industrial fault detection, and smart home systems. These deployments often face exactly the conditions the new method targets: limited labeled data, because annotating sound recordings is time-consuming and expensive, and open-set conditions, because the acoustic environment is never fully predictable. A monitoring system for a nature reserve may be trained on a handful of species calls but must reject the sounds of wind, rain, passing vehicles, and human activity that it was never trained on. An industrial acoustic fault detector may know the signatures of a few machine states but must flag novel anomalies rather than misclassifying them as normal operation.

The work also connects to a broader research program on prototype learning and few-shot recognition. Prototypical networks were originally developed for few-shot learning, where models must generalize to new classes from only a handful of examples, and they have since been applied to open-set recognition, noisy label learning, and contrastive learning. The new study builds on the convolutional prototype network for open-set recognition, which was developed in the computer vision community, and adapts it to the acoustic domain with modifications that address the specific challenges of sound data. The authors’ related work on cognitive prototype learning for few-shot open-set sound recognition and on attention-fused prototypes for few-shot open-set audio classification suggests a sustained research effort in this direction, with the current study representing a step toward more general and data-efficient acoustic recognition systems.

As deep learning continues to push the frontiers of what machines can hear and understand, the ability to say “this sound is unlike anything I know” may prove as important as the ability to name the sounds it does know. The improved convolutional prototype network proposed by Jiang and Li offers a concrete technical path toward that goal, demonstrating that with careful attention to feature learning, prototype estimation, and distance modeling, machines can learn to classify known sounds and reject unknown ones from limited data. For the growing community of researchers building acoustic monitoring systems for the real world, where the soundtrack is never fully known in advance, this work provides both a practical tool and a conceptual framework for building systems that know the limits of their own knowledge.

Subject of Research: Machine learning model for open-set environmental sound classification with limited data

Article Title: An improved convolutional prototype network for open-set sound classification with limited data

Article References: Jiang, Y., & Li, X. (2026). An improved convolutional prototype network for open-set sound classification with limited data. Cluster Computing, 29(15), Article 869. https://doi.org/10.1007/s10586-026-06724-3

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06724-3

Keywords: open-set recognition, environmental sound classification, prototype network, deep learning, few-shot learning, Mahalanobis distance, attention mechanism, acoustic monitoring, improved, convolutional, prototype, network

The echolocation grouping approach deserves particular attention. Bats do not respond to disturbance as a single homogeneous class; species differ in call frequency, call structure, foraging strategy, and tolerance of anthropogenic noise. By assigning species to groups such as Pipistrelloid and Nyctaloid, the study acknowledges that a noise level harmless to one group may suppress the activity of another. The observation that the larger ConVecDro hexacopter deterred both Pipistrelloid and Nyctaloid bats, while the smaller DJI Mavic 2 Pro affected only Nyctaloid groups, suggests that heavier aircraft with more powerful rotors produce noise profiles that overlap more broadly with the hearing and call ranges of multiple species. This has implications for UAS selection in ecological monitoring: a lighter, quieter platform may be preferable not only for flight stability reasons but because it minimizes the behavioral footprint of the survey itself.

The threshold value identified, 4.5 dB at 23 kHz, is notable because it represents the point below which no echolocation group showed any detectable impact relative to a control condition. In practical terms, this means that once the multicopter’s noise at 23 kHz falls below this level at the recording location, the survey can proceed with reasonable confidence that bat activity in the sampled volume is not being suppressed by the aircraft itself. The frequency of 23 kHz is significant because it falls within the range used by many bat species for echolocation, particularly those in the Nyctaloid group, which tend to call at lower frequencies than some other groups. The fact that this single threshold sufficed across all groups tested suggests that the most noise-sensitive species in the study set the limit for everyone else, a conservative approach that errs on the side of protecting the most vulnerable species.

The directional properties of the recording device add another layer of complexity to the placement question. Ultrasonic microphones are not omnidirectional in practice; their sensitivity varies with the angle of incoming sound, and orienting the microphone toward or away from a noise source changes how much of that noise enters the recording. The study tested both orientations, and the placement distances recommended assume the microphone faces the multicopter, which is the orientation that maximizes noise pickup. This is a conservative choice: if the microphone were oriented away from the aircraft, the effective noise level would be lower, and the required placement distance could potentially be shorter. However, survey designers rarely have full control over microphone orientation relative to where bats actually fly, so the conservative recommendation provides a margin of safety across varied field conditions.

The broader context of wind energy development frames why this methodological refinement is timely. Bat mortality at wind turbines is estimated at roughly 12 bats per turbine per year in North America and 14 in Europe, and these figures are expected to rise as turbine numbers grow and rotors sweep larger volumes of airspace. Current mitigation relies heavily on curtailment, shutting down turbines during periods of expected bat activity, which reduces energy production and is based on incomplete information about when and where bats actually fly within the rotor’s risk area. Acoustic monitoring from the ground or from the nacelle after construction covers only a fraction of the risk area, and acoustic shadowing from the tower and nacelle further limits coverage. A mobile platform capable of sampling the entire risk volume, both before construction for planning purposes and during operation for refining curtailment strategies, would address a significant gap in current practice.

The study’s approach of using a sound attenuation model to predict placement distances is methodologically elegant because it separates the physical problem from the behavioral one. The physical component, how multicopter noise diminishes with distance at each frequency, can be measured once for a given aircraft and then applied repeatedly. The behavioral component, at what noise level each echolocation group stops responding, requires field observation but yields thresholds that can be generalized across similar conditions. By combining these two components, the researchers created a framework that can be updated as new multicopters are tested or as new behavioral data become available, without needing to repeat the full experimental protocol for every aircraft and every species combination.

It is worth noting that the study found no evidence of attraction to multicopter noise across any of the conditions tested. All observed effects were deterrent, which aligns with previous research showing that bats generally avoid anthropogenic sound sources rather than investigating them. This consistency matters for survey interpretation: if multicopter noise were attracting bats, recorded activity near the aircraft would be inflated, leading to overestimates of bat presence in the risk area. Deterrent effects, by contrast, risk underestimating activity, which could lead planners to conclude that an area is less important to bats than it actually is. The placement distances recommended by the study are therefore designed to ensure that recorded activity reflects genuine bat presence rather than a behavioral artifact of the survey method itself.

The distinction between the two multicopters tested also highlights how aircraft design choices propagate into ecological outcomes. The ConVecDro hexacopter, being larger and heavier, requires more rotor power to maintain flight, producing higher sound pressure levels across the frequency spectrum. The DJI Mavic 2 Pro, a compact quadcoopter, generates less noise overall and its noise profile overlaps less with the call frequencies of Pipistrelloid species. The result is that the Mavic 2 Pro requires only 20 meters of separation from the recording device, while the ConVecDro requires 36 meters. For survey teams operating in constrained environments, where hovering at 36 meters may be impractical due to airspace regulations, obstacles, or wind conditions, the choice of aircraft can determine whether a compliant survey is feasible at all.

Looking ahead, the framework established by this study could be extended to other UAS platforms and other taxa sensitive to noise disturbance. The principle that a measured noise profile combined with a sound attenuation model can predict safe operating distances is not specific to bats; it could apply to birds, marine mammals, or any wildlife whose behavior is altered by anthropogenic sound. As UAS become increasingly central to ecological monitoring, studies like this one provide the evidence base needed to ensure that the tools used to observe wildlife do not themselves distort the phenomena being observed.

Subject of Research: Science

Article Title: Determining recorder distances from multicopters to avoid bat disturbance

Article References: Huamán Roswag, M., Fietz, J., Roswag, A., Kunz, V. D., & Taefi, T. T. (2026). Determining recorder distances from multicopters to avoid bat disturbance. Frontiers in Ethology, 5, Article 1868324. https://doi.org/10.3389/fetho.2026.1868324

Image Credits: AI Generated

DOI: 10.3389/fetho.2026.1868324

Keywords: drones, bats, echolocation, multicopter, noise, wind energy, acoustic monitoring, sound pressure level, wildlife conservation, UAS, Determining, recorder

The platform under study is no small fixture. Stretching 101 meters long and 47.5 meters wide, with an aquaculture draft of 15 meters and roughly 60,000 cubic meters of farming water, it represents the new generation of offshore infrastructure designed to relieve pressure on saturated nearshore waters. Powered by a hybrid system of photovoltaic panels and diesel generators, these truss platforms are becoming the mainstream equipment for deep-sea aquaculture across coastal provinces like Guangdong, Shandong, and Hainan. Yet until now, almost nothing was known about how their operations reshape the underwater soundscape that marine organisms depend on for foraging, communication, predator avoidance, and navigation.

The research team deployed hydrophones at three strategic locations: inside the central net pen where fish are actually farmed, near the propulsion hatch on the platform deck, and at a distant reference station 1.5 kilometers away. A sound level meter simultaneously recorded airborne noise on the deck corridor, while a self-contained current meter and high-precision tide gauge tracked the hydrodynamic conditions. Everything ran continuously for 46 hours under unusually calm sea conditions, with mean wind speeds of just 4.3 meters per second and sea state below Force 2, providing a clean baseline for isolating the platform’s acoustic signature from natural background noise.

What emerged from the data was striking. During generator operation, the underwater near-field exhibited a discrete harmonic skeleton centered on a fundamental frequency of 24.41 Hz, with power spectral density reaching 122.29 dB re 1μPa²/Hz, approximately 16 dB higher than the idle baseline. Sharp narrowband peaks appeared at harmonic multiples including 48.8, 73.2, 100, and 122 Hz, forming a fingerprint unmistakably tied to the platform’s propulsion system. Above 500 Hz, the interference rapidly faded, and beyond 2 kHz the spectral curves before and after startup completely overlapped with the natural environmental background, which remained within the 72 to 86 dB range.

Perhaps the most compelling evidence came from the moment of generator startup and shutdown. The researchers observed pixel-level temporal synchronization between the deck’s airborne sound field and the underwater near-field radiation. When the 25 Hz spectral line abruptly vanished from the air spectrum, the bright high-energy band in the underwater low-frequency range collapsed in a vertical cliff-like manner. Simultaneously, the baseline at the 1.5-kilometer distant station experienced a spontaneous drop to the quietest state of the entire monitoring period. This triple alignment across three independent measurement domains pointed to a single conclusion about how the acoustic energy reaches the water.

Because air and seawater have an acoustic impedance mismatch of roughly 3,600 to 1, atmospheric noise incident at the sea surface experiences significant reflection losses, making direct airborne transmission into the water negligible. Instead, the structural elastic waves excited by the propulsion system travel as solid-borne vibrations through the rigid truss pile foundation, conducted along the steel columns to their underwater outer walls, where they radiate into the water column. This finding has immediate engineering implications: noise control for such platforms should prioritize structural isolation and vibration damping at equipment foundations rather than relying solely on sound-absorbing barriers within deck machinery rooms.

The distant acoustic environment told its own story. Within the observation window, the low-frequency soundscape at the 1.5-kilometer station appeared modulated by semidiurnal tidal dynamics, with noise peaks lagging behind tidal peak levels by approximately 6 hours. During neap tide periods, when current velocity dropped below 0.3 meters per second, the acoustic background fell below -40 dB re 1μPa²/Hz, forming regular low-energy troughs. Conversely, during stronger tidal-current transitions, the low-frequency energy rose in phased association with the surging flow. The researchers interpret this as flow noise excitation driven by turbulent boundary layer pressure fluctuations and vortex shedding across the seabed topography, processes that decay substantially when currents weaken below the empirical threshold.

Day-night differences added another layer of complexity. The team measured sound-speed profiles in the shallow-water channel and found a distinct minimum at approximately 4 meters depth during the day, creating a natural acoustic waveguide that confines sound energy. At night, surface cooling and evaporative heat loss flipped the profile into a positive gradient with upward refraction, forcing sound rays to strike the sea surface repeatedly. Under the rougher nocturnal sea surface, this enhanced incoherent scattering plausibly explains the observed decay of the distant-station characteristic sound after dark, allowing the ambient soundscape to return to its natural quiet trough. The total sound-speed change across the water column at night reached approximately 1.96 meters per second, significantly greater than daytime fluctuations.

The ecological stakes are considerable. Fish rely on acoustic signals for essential life functions, and anthropogenic underwater noise can disrupt these behaviors through auditory masking, physiological stress, and behavioral avoidance. Repeated exposure to vessel noise has been shown to damage sensory hair cells in fish inner ears and trigger irreversible declines in auditory sensitivity, while noise exposure alters daily swimming patterns and escape responses. Because modern aquaculture management increasingly depends on acoustic technologies, including sonar-based biomass estimation, acoustic virtual fencing, and passive monitoring of fish behavior, understanding the platform’s own noise characteristics becomes a core prerequisite for optimizing equipment deployment and improving detection accuracy.

The researchers are careful to frame their conclusions appropriately. These findings derive from a single 46-hour campaign under calm conditions and should be regarded as process-level evidence rather than climatological statistics, applicable only to the observation window and not extrapolatable to other seasons or sea states without further measurements. Limitations include single-range sampling, the absence of direct in-situ structural vibration measurements, and the lack of sediment acoustic parameters. Future work will employ multi-node, multi-range arrays and full-waveform inversion to achieve precise reconstruction of the three-dimensional acoustic radiation, propagation, and attenuation boundaries of large fixed offshore facilities. For now, the study provides foundational data for acoustic environment monitoring, ecological impact assessment, and the optimized deployment of aquaculture acoustic equipment on similar deep-sea platforms worldwide.

Subject of Research: Underwater noise characteristics and tidal modulation of the acoustic environment around a large truss-type deep-sea aquaculture platform

Article Title: A study on the evolutionary patterns of the acoustic environment of large-scale truss-type deep-sea aquaculture platforms

Article References: Wu, D., Qu, K., Yang, B., Deng, X., Chen, X., Ling, W., & Qi, Y. (2026). A study on the evolutionary patterns of the acoustic environment of large-scale truss-type deep-sea aquaculture platforms. Frontiers in Marine Science, 13, Article 1962198. https://doi.org/10.3389/fmars.2026.1962198

Image Credits: AI Generated

DOI: 10.3389/fmars.2026.1962198

Keywords: deep-sea aquaculture, underwater noise, truss platform, passive acoustics, tidal dynamics, solid-body transmission, underwater soundscape, acoustic monitoring, marine ecology, Frontiers in Marine Science, evolutionary, patterns

The stakes are far from academic. Industrial anomaly detection is a cornerstone of modern manufacturing, where catching the faint acoustic signature of a worn bearing or a misaligned belt before it fails can save enormous maintenance costs and prevent catastrophic downtime. Acoustic and vibration monitoring is especially attractive because it is non-intrusive and sensitive to subtle mechanical degradation. Yet classical pipelines built on support vector machines hit a wall when multiple anomaly types coexist and their feature distributions intermingle in Euclidean space. The result is an unacceptably high false positive rate, which in practice means operators learn to ignore the alarms, defeating the entire purpose of the monitoring system.

The researchers built their study around two physical test rigs with deliberately contrasting difficulty. The first, an Open Belt Drive (OBD) system, consists of rubber and metal chain belt drives in which wooden chopsticks are inserted through pre-drilled holes, producing sudden, loud breaking sounds that define the anomaly class. The second, a Mini 4WD (M4W) course, sends a miniature four-wheel drive vehicle around a three-lane track where it first strikes a one-millimeter wooden stick, generating a small impact sound, and then passes over Magic Tape, producing a faint scratching noise. While chopstick breaks are obvious even to human ears, the stick and tape sounds of the M4W rig are extremely difficult to discern, and the two anomaly types occur sequentially within each roughly five-second lap.

From five-minute recordings of each system, the team extracted 30 normal samples per dataset and converted the audio into compact feature vectors using autoregressive (AR) modeling. Each 10-second segment was resampled from 48 kHz to 1100 Hz, cut into non-overlapping 0.4-second windows, and characterized by AR coefficients estimated with the Yule-Walker method, with median aggregation across windows providing robustness against transient noise spikes. The AR model order was varied from 2 to 10, and information-theoretic criteria, the Akaike Information Criterion and Bayesian Information Criterion, independently identified orders 9 to 10 as optimal, validating the experimental range. Because real industrial equipment rarely fails often enough to supply labeled anomaly data, the team adopted an unsupervised One-Class SVM that learns a decision boundary from normal operation data alone, flagging significant deviations as anomalies.

Against this pipeline, three kernels competed: the classical RBF baseline, and two quantum kernel architectures. Quantum Kernel 1 (QK1) uses linear entanglement, with nearest-neighbor CNOT gates connecting adjacent qubits in a chain, creating pairwise correlations between consecutive features at a circuit depth that scales linearly with qubit count. Quantum Kernel 2 (QK2) employs all-to-all entanglement, connecting every qubit pair to capture global, higher-order dependencies at the cost of quadratic circuit depth. Both use a data re-uploading strategy in which AR coefficients are repeatedly encoded into rotation gates across two layers, producing quantum amplitudes that depend non-linearly on the input through interference. Importantly, the quantum kernels were implemented via classical simulation, using adjusted RBF gamma parameters that approximate the correlation structures of the two entanglement topologies, a choice the authors candidly flag as a limitation that leaves confirmation on actual quantum hardware to future work.

On the OBD dataset, where anomalies are cleanly separable, all three methods eventually achieved perfect classification, but the quantum kernels got there faster. QK1 reached a perfect F1 score of 1.0 with just four AR features, QK2 matched it at seven, while the RBF kernel needed eight. Wilcoxon signed-rank tests across the full feature range confirmed these differences were statistically significant, with p-values of 0.0420 for QK1 versus RBF and 0.0203 for QK2 versus RBF. In practical terms, the quantum kernels offered a computational efficiency advantage, roughly halving the feature extraction overhead, but no fundamental capability gap, since classical methods converged to the same asymptotic solution.

The M4W dataset told an entirely different story. There, QK2 achieved an F1 score of 0.893 with a false positive rate of just 0.033, while the RBF kernel collapsed to an F1 of 0.428 with a staggering false positive rate of 0.600, meaning 60 percent of normal operation was flagged as anomalous. The statistical significance was emphatic, with a p-value of 0.0023. QK1 landed in between at F1 = 0.661 but became unstable at higher feature orders. Most tellingly, the RBF kernel’s F1 trajectory remained flat, never exceeding 0.47 despite a fivefold increase in feature count, a signature of structural failure rather than insufficient data. ROC trajectory analysis made the mechanism visible: RBF wandered horizontally in ROC space, oscillating between false positive rates of 0.6 and 1.0 without ever approaching the ideal operating point, while QK2 systematically drove its false positive rate down from 0.93 to 0.03 as features were added.

To explain why, the researchers turned to t-SNE visualization of the AR feature space, backed by rigorous statistical testing. For OBD, Kruskal-Wallis tests confirmed complete class separation in both projected dimensions, with all pairwise Mann-Whitney comparisons passing Bonferroni-corrected significance thresholds. For M4W, the picture was one of fundamental inseparability: the second dimension showed no significant structure at all (H = 0.4, p = 0.942), and Magic Tape anomalies proved statistically indistinguishable from every other class. A striking paradox emerged in the decision score distributions: QK2 achieved its superior performance despite a centroid separation between normal and anomalous classes roughly three times smaller than RBF’s. The quantum kernel’s advantage lies not in pushing class means apart but in reshaping the distributions to minimize their overlap, exploiting non-linear correlations through high-dimensional Hilbert space projections that Euclidean geometry simply cannot represent.

The study also connects its unsupervised findings to earlier supervised work by the same group, including apple defect detection in shipping inspection that achieved F1 scores of 0.8 to 0.9 with only 24 training samples per class, far outperforming classical RBF. Across both paradigms, a consistent pattern emerges: quantum kernel benefits appear in small-data regimes where classical feature distributions overlap, while showing diminishing returns beyond moderate feature dimensionality, a phenomenon attributed to exponential concentration that may impose fundamental scalability limits. The authors propose a practical screening framework: run a preliminary t-SNE analysis with classical features, and if Kruskal-Wallis tests reveal poor separation or multiple inseparable class pairs, quantum kernels may be worth the investment. With false alarms triggering unnecessary inspections and production stoppages, a drop from a 60 percent to a 3 percent false positive rate is not a marginal improvement but the difference between a monitoring system operators trust and one they switch off. As quantum hardware matures, this work suggests that messy, overlapping, real-world sensor data, not pristine synthetic benchmarks, may be where quantum machine learning first earns its keep.

Subject of Research: Quantum kernel methods for unsupervised industrial acoustic anomaly detection with overlapping data distributions

Article Title: Potential of multi-anomalies detection using quantum machine learning

Article References: Tomono, T., & Tsujimura, K. (2026). Potential of multi-anomalies detection using quantum machine learning. Quantum Machine Intelligence, 8(2), Article 106. https://doi.org/10.1007/s42484-026-00448-8

Image Credits: AI Generated

DOI: 10.1007/s42484-026-00448-8

Keywords: quantum machine learning, quantum kernels, anomaly detection, One-Class SVM, acoustic monitoring, false positive rate, t-SNE, feature space, industrial condition monitoring, RBF kernel, entanglement, Hilbert space

The study, published in the journal Discover Conservation, is a systematic review of more than three decades of bat bioacoustics research. Researchers Luna Pino-Aedo and Andrés Muñoz-Sáez of the Universidad de Chile searched the ISI Web of Science database for studies published between 1990 and 2022, the period since passive acoustic monitoring of terrestrial ecosystems began in earnest. Their initial queries returned 531 candidate articles, and after screening each one at full text, 207 studies met the criteria: they analyzed bat echolocation calls and specified exactly how those calls were analyzed. From this body of work the authors catalogued 34 distinct echolocation analysis methods, spanning commercial software packages and a growing arsenal of supervised machine learning techniques.

The review reveals a field that has expanded explosively. Most of the included studies appeared after 2010, a trend the authors attribute to the falling cost and rising availability of automated ultrasonic recorders, with a notable peak in publications in 2021 that may reflect pandemic-era restrictions pushing researchers toward data analysis rather than fieldwork. Geographically, the science is heavily skewed: the United States accounted for 36.28 percent of the articles, followed by China with 12.56 percent. That concentration matters, because reference libraries of bat calls and the classifiers trained on them are richest for Nearctic and Palearctic species, leaving tropical hotspots of bat diversity comparatively underserved.

Echolocation itself is a remarkable biological technology. Bats generate ultrasonic pulses in the larynx and emit them through the mouth or nose; the returning echoes build an acoustic image of the surroundings that the animals use to navigate, forage and communicate. Because each species tends to produce calls with characteristic frequencies, durations and shapes, researchers realized decades ago that recordings could be used to identify species without ever catching them. Bat detectors from manufacturers such as Pettersson, Titley Scientific, Wildlife Acoustics and Avisoft dominated the reviewed studies, capturing calls at high sampling frequencies for later analysis through spectrograms and oscillograms.

But turning a recording into a species name is where things get complicated. Among the 118 studies whose primary goal was species identification, only 29 relied on fully automated identification; 55 used manual identification by a specialist, and 34 used a mixed approach in which experts verified or complemented machine output. The most frequently used software packages were Kaleidoscope, Avisoft, Analook, BatSound and SonoBat, often made by the same companies that build the detectors. While all of these tools visualize calls and filter noise, only some, notably Kaleidoscope and SonoBat, were actually used for automated species classification; the rest served mainly for manual inspection and parameter measurement.

Alongside commercial software, the review documented a parallel rise of supervised learning methods implemented in platforms such as Python and RStudio. Discriminant function analysis, a statistical technique that models predefined classes from numerical acoustic features, was among the most common, followed by artificial neural networks, random forests and support vector machines. Each brings distinct trade-offs. In a benchmark study by Armitage and Ober, artificial neural networks achieved the highest training accuracy at 86.1 percent of calls correctly identified but demanded the most computational resources and training time. Random forests delivered the best overall combination of accuracy, sensitivity and specificity at 85.5 percent while remaining fast to train. Discriminant function analysis was quickest but least accurate on average at 75.9 percent. In Uruguay, Botto Nuñez and colleagues found random forests outperformed both support vector machines and neural networks, correctly identifying 92.9 percent of calls with lower variance than either alternative.

Head-to-head comparisons under controlled conditions expose the gap between machine learning and commercial packages most starkly. Tabak and colleagues built a deep learning model based on convolutional neural networks that achieved an average accuracy of 91 percent, while Kaleidoscope Pro and BCID managed only 65 percent and 61 percent respectively on the same task. Earlier discriminant function and neural network approaches also outperformed the modern software evaluated by Rydell and colleagues, whose study found that the best automated program still fell short of the top human expert, though the software did reliably classify calls at the genus level and distinguished species with highly distinctive calls. Brabant and colleagues, testing four programs on nine European species, found BatIdent achieved the highest proportion of correct species-level identifications but also produced more misclassifications, while Kaleidoscope left the largest share of calls unidentified. Sensitivity settings materially changed results: Kaleidoscope’s accuracy rose under more sensitive configurations, whereas SonoChiro stayed relatively stable across settings.

Perhaps most disquieting is what happens when there is no ground truth at all. Studies by Lemen and by Nocera that assessed agreement between software packages without a reference identification found concordance rates of only around 40 percent, varying by species and sampling period. Agreement between programs, the authors conclude, is not a reliable proxy for accuracy. Even software version matters in unpredictable ways: Goodwin and Gillam found the best performance from Kaleidoscope Pro version 4.3.0 and SonoBat version 3.2.2, meaning newer releases were not necessarily better classifiers. Accuracy reporting itself remains rare, with only 28 of the 207 studies quantifying how often their methods classified calls correctly.

The review also catalogues the tangled web of factors that influence identification beyond the algorithms themselves. Call type matters, since search, approach, feeding and social calls differ within a single species, and many species produce calls so similar that even experts struggle. Recording context matters, whether calls come from free-flying animals or hand-released individuals. Weather matters, with temperature shifting peak frequencies. So do geography, age, morphology and even the roost a population occupies, all of which inject intraspecific variation that classifiers trained on narrow libraries cannot handle. Habitat structure and detector placement also shape outcomes: models trained on calls recorded at structurally complex forest edges outperformed those trained in open spaces, and recording angle significantly affected correct classification.

The authors’ prescription is a hybrid future. Automated classifiers should be treated as accelerators that triage enormous audio volumes and flag candidate events for expert review, not as stand-alone oracles, and they should be deployed unsupervised only for taxa with highly distinctive calls. Capture-based methods such as mist-netting remain indispensable, both for species with highly directional sound beams that evade standard detectors and for building verified reference datasets; combined acoustic-and-capture designs consistently detect more species than either method alone. The researchers call for open, regionally representative call libraries with detailed metadata on habitat and recording context, versioned benchmark datasets spanning detectors and ecoregions, and transparent reporting of complete analysis pipelines. With bat populations declining worldwide amid accelerating biodiversity loss, the stakes of getting this right could hardly be higher: misidentified calls can translate directly into misinformed conservation decisions for one of the planet’s most important, and most imperiled, groups of mammals.

Subject of Research: Systematic review of bat echolocation call analysis methods for species identification and conservation monitoring

Article Title: A systematic review of bat echolocation analysis methods and their implications for monitoring and conservation

Article References: Pino-Aedo, L., & Muñoz-Sáez, A. (2026). A systematic review of bat echolocation analysis methods and their implications for monitoring and conservation. Discover Conservation, 3(1), Article 17. https://doi.org/10.1007/s44353-026-00087-x

Image Credits: AI Generated

DOI: 10.1007/s44353-026-00087-x

Keywords: bats, echolocation, bioacoustics, machine learning, acoustic monitoring, species identification, conservation, neural networks, random forests, Kaleidoscope, passive acoustic monitoring, biodiversity

The problem with current detection methods is timing. The standard techniques — alcohol wash, sticky board counting, and manual inspection — are invasive, labor-intensive, and deliver results one to three days after sampling. By then, mite populations may have grown past the point where low-impact treatments remain effective. When infestations are caught late, beekeepers often resort to synthetic miticides such as amitraz or coumaphos, which accumulate in beeswax, carry documented resistance risks in multiple countries, and can harm colony health. When mite loads are identified while still low, beekeepers can instead turn to organic acids, which at correct dosage leave no honey residue and have no documented resistance. The economic threshold for action is 2 to 3 mites per 100 adult bees; an alert issued at 1 to 2 mites per 100 bees gives a beekeeper roughly four to six days before that threshold is crossed.

Previous automated approaches have each tackled one piece of the problem. A vision-based system by Bjerge and colleagues achieved 91 percent classification accuracy using a convolutional neural network on hive entrance video, but required a custom GPU workstation and a dedicated optical tunnel, pushing total cost above $500. A microscopy-based method by Babic and colleagues reached 94 percent accuracy but requires physical bee samples and offline laboratory processing. Acoustic research showed that varroa-stressed colonies produce measurably different sound signatures, particularly in the 200–500 Hz band corresponding to disrupted wing-beat and fanning behavior, and environmental studies demonstrated that carbon dioxide concentration and temperature at the hive entrance correlate with mite-induced stress. The new system is, to the author’s knowledge, among the first to fuse all three sensing modalities in a single, affordable, edge-deployed device.

The hardware is deliberately modest. A Raspberry Pi 5 runs all inference locally, paired with a 12-megapixel Arducam mounted horizontally at the hive entrance, an I2S MEMS microphone, and an SCD41 sensor that measures true carbon dioxide using non-dispersive infrared technology rather than cheaper proxy estimates. A custom PCB routes the sensor buses, and a battery-backed power module keeps the unit running through outages. The complete bill of materials came to $189.68, and a proposed mass-production redesign built around a Raspberry Pi Zero 2 W would cut that to an estimated $49.50. The device needs no internet connection, storing records on a microSD card and serving a dashboard to any browser on the local Wi-Fi network.

The software pipeline runs a four-stage inference cycle every 30 seconds. Stage A is a YOLOv8n bee detector that draws bounding boxes around every bee in each video frame, achieving a mean average precision of 0.976 at 512-by-512 resolution. Stage B is a second YOLOv8n model that inspects each bee crop for visible mites — a crop-first architecture that proved essential, because at full-frame resolution a mite occupies roughly 0.3 percent of the image, making direct detection unreliable. Cropping first lifted the F1 score from 0.774 to 0.805. Stage C is a CNN-LSTM that classifies 30-second audio clips as healthy or stressed, using 43 acoustic features including MFCCs, spectral centroid, and chroma features; an earlier 157-feature version overfit at just 74 percent accuracy before the streamlined feature set raised test accuracy to 93.6 percent. Stage D is a Random Forest that classifies colony risk from temperature, humidity, and carbon dioxide readings.

The fusion layer is where the system earns its keep. Each modality’s probability score is combined through confidence-weighted voting, with weights proportional to each sensor’s validation precision: 0.28 for vision, 0.34 for audio, and 0.38 for environmental data. Because the three signals arise from biologically independent mechanisms — visible mites on bee bodies, altered colony acoustics, and metabolic drift in hive air — spurious readings in one modality are unlikely to coincide with spurious readings in the others. The result is dramatic: the false alarm rate falls from 27.0 percent for vision alone to 0.5 percent for the fused system, a reduction confirmed by McNemar’s test at p = 0.0176. Fused precision reached 0.995 with an F1 score of 0.957, and removing any single modality degraded performance, confirming that each contributes independently.

Field validation spanned nine colonies in three infestation categories, three hives each, with mite loads confirmed by alcohol wash sampling. The three healthy hives ran for 30 days apiece and produced median fused scores between 0.0196 and 0.0217, with zero threshold-positive windows and zero alert episodes across 12,920 total observation cycles. Three mildly infested hives carrying 1 to 5 mites per 100 bees also ran 30 days each; their median scores shifted upward to between 0.0759 and 0.0889 without ever crossing the 0.50 alert threshold, showing the system responds to sub-threshold mite pressure without crying wolf. Three heavily infested colonies, each carrying 8 to 10 or more mites per 100 bees, were monitored for one day and produced medians above 0.51, with 34 to 38 alert episodes per hive and threshold exceedances in roughly three-quarters of all inference cycles.

The 25-fold increase in median fused score from healthy to heavily infested colonies, with the mild category falling neatly between, indicates the system tracks infestation severity rather than behaving as a binary switch. A separate 50-hive, 14-day simulation estimated that temporal smoothing detected more than 80 percent of developing infestations before their peak, with a median lead time of 4.6 days — enough to preserve the beekeeper’s full range of treatment options. The system also survived robustness testing under low light, high noise, extreme temperature, and partial sensor failure, averaging 0.875 accuracy across all stress conditions. The full pipeline sustained 28 to 29 frames per second on the Raspberry Pi 5 with power draw under 5 watts and no thermal throttling.

The study is candid about its limitations. All nine hives sat at a single apiary over a single season, so generalization across regions, bee subspecies, and climates awaits multi-site validation. The environmental classifier was trained on a synthetic dataset built from published biological parameter ranges, a workaround made necessary by a striking methodological discovery: the only publicly available labeled real-sensor hive dataset contained a leaky column — cumulative bee weight — that carried 45.7 percent feature importance but cannot be measured at inference time. Any model trained on it without removing that column reports a suspicious 100 percent accuracy and would fail in the field. No controlled infestation experiments were conducted, for ethical reasons; all infestation levels reflect naturally occurring conditions.

What distinguishes this work is less any single algorithm than the demonstration that deployment-grade reliability is achievable at a price individual beekeepers can afford. Most of the world’s beekeepers manage fewer than 25 colonies, and no commercially available system previously offered continuous, non-destructive varroa monitoring below $200 per hive. The complete inference pipeline, trained models, and processed field datasets have been released publicly, and the modular architecture — camera, microphone, environmental sensor, late fusion — extends naturally beyond varroa to livestock respiratory disease monitoring, crop pest surveillance, and coral reef health. For a parasite that has spread to every inhabited continent since jumping from its original Asian honey bee host in the 1970s, and against which the western honey bee has evolved no effective defenses, an affordable early-warning system may be the most practical lifeline apiculture has seen.

Subject of Research: Non-invasive multimodal machine learning detection of Varroa destructor infestations in honey bee colonies

Article Title: A multimodal machine learning system for non-invasive detection of varroa destructor infestations in honey bee colonies

Article References: Gaba, P. (2026). A multimodal machine learning system for non-invasive detection of varroa destructor infestations in honey bee colonies. Smart Agricultural Technology, 15, Article 102354. https://doi.org/10.1016/j.atech.2026.102354

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102354

Keywords: varroa destructor, honey bees, machine learning, sensor fusion, precision beekeeping, edge computing, colony collapse, acoustic monitoring, computer vision, Raspberry Pi, apiculture, early detection

Estrus detection is one of the most economically consequential tasks in modern dairy management. A cow that is not inseminated at the right moment misses a breeding cycle, extending the interval between calvings and reducing lifetime milk yield. Yet identifying the behavioral signs of heat in a barn holding hundreds of animals is notoriously difficult. Progesterone testing, often considered the gold standard, is costly and cannot provide continuous monitoring. Human observation, even when conducted at regular intervals, struggles against the sheer scale of a large herd and the irregular timing of behavioral onset.

Acoustic monitoring offers a tempting alternative because cows vocalize differently during estrus, and microphones can capture those changes without any invasive sampling. The challenge has always been practical: a continuous day of barn audio is dominated by machinery noise, fans, bird calls, and collisions of equipment, with only brief bursts of high-pitched cow calls embedded in the acoustic clutter. Storing and processing all of that raw audio for an entire herd quickly becomes an enormous data-management problem, and conventional lossy compression risks distorting the very timing and spectral cues that matter.

The research team, led by Jun Wang of Henan University of Science and Technology, built a custom dual-channel acoustic tag around an STM32 microcontroller. One unidirectional microphone, pressed against the cow’s neck, picks up vibrations from the animal’s own vocalizations, while a second omnidirectional microphone records airborne calls and ambient barn sound. An accelerometer tracks movement simultaneously. The waterproof device weighs 115 grams, runs for more than twelve hours per charge, and stores each audio channel on a separate memory card. Over two spring recording periods, the researchers equipped 116 Holstein cows with the tags and had experienced veterinarians score estrus behavior from ceiling-mounted surveillance video at three-hour intervals.

From 27 cows with complete 24-hour records spanning an entire estrus episode, the team catalogued vocalizations and split them by pitch. Low-pitched calls showed only a moderate association with estrus stage, with a Spearman rank correlation of 0.38, but high-pitched calls tracked the reproductive cycle far more closely, with a correlation of 0.74. During estrus, high-pitched vocalizations occurred in 99.3 percent of hourly intervals, and their mean duration more than doubled compared with the post-estrus period. Notably, first-time (primiparous) mothers were considerably more vocal than older cows, averaging about 161 seconds of high-pitched calling per hour of active vocalization versus 88 seconds for multiparous animals.

The heart of the method lies in what the authors call an enhanced SAX strategy—a reworking of the classic Symbolic Aggregate Approximation technique used in time-series mining. Each two-minute segment of audio, a window length chosen after testing alternatives and balancing correct recognition accuracy against both error rates and runtime, is transformed into a waveform image. The algorithm first normalizes the energy of every sample, then applies a gain compression whose strength varies sample by sample according to that energy. It then squeezes the waveform horizontally by dividing it into segments and keeping one representative value per segment. This dual vertical-and-horizontal compression produces 265 candidate images per window, and a selection rule based on structural similarity (SSIM) minus a penalty for average gain retains just one image per window.

The numbers behind this filtering are striking. Processing 46,080 minutes of raw audio generated more than six million temporary candidate images, from which the algorithm kept exactly 23,040 representatives. Crucially, this aggressive compression preserved the biologically relevant signal: 99.09 percent of the measured high-pitched vocalization duration survived, and a Jensen-Shannon divergence measure confirmed that the distributional change was far below the threshold that would indicate meaningful information loss. Cow calls were kept; machinery noise and birdsong were discarded.

Classification then proceeds on the retained images. A convolutional neural network with 12.8 million trainable parameters serves double duty, both assigning its own labels and providing a 512-dimensional feature embedding that feeds four additional classifiers: random forest, k-nearest neighbors, support vector machine, and XGBoost. A hard-voting scheme combines the five opinions into a single verdict for each image, and 30 consecutive image labels are aggregated into one cow-hour decision, with an hour classified as estrus when at least ten of the thirty images were flagged as high-pitched vocalization. On a pooled test set of 790 images, the ensemble correctly classified 782, achieving 98.99 percent accuracy with a 95 percent confidence interval of 98.01 to 99.49 percent, modestly outperforming every individual learner.

Timing matters as much as detection. In a same-farm comparison across 32 cows, the system’s predicted onset of estrus came before the veterinarian-reconstructed onset in 22 cases, with a mean prediction lead of roughly 89 minutes, and lagged behind in 10 cases by an average of about 28 minutes. Earlier alerts give farm staff a wider window to schedule insemination at the optimal moment, though the authors caution that the veterinary onset itself was reconstructed from three-hour observation intervals rather than minute-resolved measurements.

The researchers are careful to frame these results as an internal, single-farm evaluation. All animals were Holsteins managed under the same conditions, the image-level data split was not documented as cow-disjoint, and no external farm, breed, or climate was tested. The authors also note that dense or overlapping calls produced the highest error rates, that no measured edge-hardware benchmarks exist for the CNN, and that formal ablation studies are still needed to isolate the contribution of the gain term. Nevertheless, the framework demonstrates that a thoughtfully compressed acoustic representation can carry nearly all of the estrus-relevant information while making the data burden of continuous herd monitoring manageable, and it points toward acoustic estrus surveillance as a viable complement to accelerometers and hormonal testing in precision livestock farming.

Subject of Research: An acoustic estrus detection method for large-herd dairy cows combining enhanced SAX audio compression with heterogeneous ensemble learning

Article Title: Towards precise estrus identification of large-herd dairy cows: An acoustic detection method using enhanced SAX strategy and heterogeneous ensemble learning

Article References: Wang, J., Yue, Y., Wang, H., Aboelmaaty, A. M., Si, P., Zhao, K., Zhao, Y., & Shakweer, W. M. E.-S. (2026). Towards precise estrus identification of large-herd dairy cows: An acoustic detection method using enhanced SAX strategy and heterogeneous ensemble learning. Smart Agricultural Technology, 15, Article 102524. https://doi.org/10.1016/j.atech.2026.102524

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102524

Keywords: dairy cows, estrus detection, acoustic monitoring, SAX compression, ensemble learning, precision livestock farming, bioacoustics, machine learning, Holstein cattle, reproductive management, wearable sensors, convolutional neural network