Abstract Passive acoustic monitoring (PAM) is an effective remote sensing approach for sampling acoustically active animal species and is particularly useful for elusive, visually cryptic species inhabiting remote or inaccessible habitats. Key advantages of PAM are large spatial coverage and continuous, long‐term monitoring. In most cases, a signal detection algorithm is utilized to locate sounds of interest within long sequences of audio data. It is important to understand the demographic/contextual usage of call types when choosing a particular signal to use for detection. Sampling biases may result if sampling is restricted to subsets of the population, for example, when detectable vocalizations are produced only by a certain demographic class. Using the African forest elephant repertoire as a case study, we test for differences in call type usage among different age‐sex classes. We identified disproportionate usage by age‐sex class of four call types—roars, trumpets, rumbles, and combination calls. This differential usage of signals by demographic class has implications for the use of particular call types in PAM for this species. Our results highlight that forest elephant PAM studies that have used rumbles as target signals may have under‐sampled adult males. The addition of other call types to PAM frameworks may be useful to leverage additional population demographic information from these surveys. Our research exemplifies how an examination of a species' acoustic behavior can be used to better contextualize the data and results from PAM and to strengthen the resulting inference.

The impacts of human activities and climate change on animal populations often take considerable time before they are reflected in typical measures of population health such as population size, demography, and landscape use. Earlier detection of such impacts could enhance the effectiveness of conservation strategies, particularly for species with slow population growth. Passive acoustic monitoring is increasingly used to estimate occupancy and population size, but this tool can also monitor subtle shifts in behavior that might be early indicators of changing impacts. Here we use data from an acoustic grid, monitoring 1250 km2 of forest in northern Republic of Congo, to study how forest elephants (Loxodonta cyclotis) assess risk associated with human impacts across a landscape that includes a national park as well as active and inactive logging concessions. By quantifying emerging patterns of behavior at the population level, arising from individual-based decisions, we gain an understanding of how elephants perceive their landscape along an axis of human disturbance. Forest elephants in relatively undisturbed forests are active nearly equally day and night. However, they become more nocturnal when exposed to a perceived risk such as poaching. We assessed elephant perception of risk by monitoring changes in the likelihood of nocturnal vocal activity relative to differing levels of human activity. We show that logging is perceived to be a risk on moderate time and small spatial scales, but with little effect on elephant density. However, risk avoidance persisted in areas with relatively easy access to poachers and in more open habitats where poaching has historically been concentrated. Increased nocturnal activity is a common response in many animals to human intrusion on the landscape. Provided a species is acoustically active, passive acoustic monitoring can measure changes in human impact at early stages of such change, informing management priorities.

African forest elephants (Loxodonta cyclotis) are a critically endangered and visually cryptic species that inhabits Central African rainforests. Using a 1250 sq-km grid of 50 acoustic sensors in Republic of Congo, we investigated the landscape-scale behavioural response of forest elephants to poaching events. We detected eight automatic weapon fire events over 1.5 years of recording (2017-2019). We examined detections of elephant vocalisations across grid sites within 10 km of the gunfire events in the 48 h surrounding the gunfire, and in paired 48-h control periods free of gunfire. Fewer vocalisations were detected before gunfire events than during control periods, suggesting elephants show behavioural responses to poacher presence before any shots are fired. Immediately following the gunfire events, a significant increase in elephant calls was detected, which may indicate increased communication to facilitate group cohesion and movement. Elevated call levels dropped to lower-than-baseline rates after several hours, suggestive of a sustained response. These patterns indicate forest elephants respond to both poacher presence and to gunfire events themselves. As these patterns may reflect behaviours that mediate population processes, conservation strategies should account for the potential of poaching to impact populations via indirect effects on nontarget elephants in the area.

Artificial intelligence is increasingly used in ecological contexts to monitor animal and insect populations. Species of interest are those in danger of extinction, and those that play pivotal roles in agriculture. Noticing population declines or geographical shifts early enough for intervention can prevent local famine and disruption to the global food chain. Traditionally, data are collected in the field using human labor or sensors. Applicable classification models then analyze the data on central servers. The most expensive, and sometimes dangerous part of the remote sensing solution is the human labor of visiting the sensors, retrieving data, and changing batteries. Constantly sending all readings by radio is expensive in power. Instead, having AI in the sensors process readings, and only transmitting results could lead to an indefinitely autonomous, renewably powered solution. We implemented an elephant vocalization detector on a small processor board, and demonstrate that such a device can be operated at low enough power levels with considerable freedom of choice among AI technologies. We achieved a mean of 1.6W, in the best case staying within 75% of memory limits. Measurements covered three inference models, two batch sizes, and two floating point word width settings.

Social relationships are shaped by ecological conditions, giving rise to diverse societies even among related species. Among elephants, females exhibit close, often familial bonds with some degree of philopatry and fission-fusion dynamics. Forest elephant sociality is the least studied among elephant species, and their ecology differs from that of savannah and Asian elephants. We use over 15 years of data from an individually identified population using the Dzanga Bai forest clearing in Central African Republic to characterize social patterns of adult female elephants in two contexts, when entering the forest clearing and within the clearing. Social properties were nonrandom and highly stable over the study period in both contexts. Composition of aggregations within the clearing indicated considerable mixing among females, while association in groups entering the bai together were limited in composition, likely representing core social units. The co-entry data set exhibited distinct communities that were not apparent in the co-occurrence data set. In contrast to savannah elephants, network centrality and age were negatively correlated, suggesting a muted or absent matriarch social hub role in forest elephants at a broader level than family units. Temporal and community analyses indicated that bai co-entry and cooccurrence contexts may represent distinct processes, with the former social and the latter driven by shared resource use. Our results point to social patterns potentially driven by the need for families to fission into small units because of resource competition and highlight the importance of aggregation context in interpreting association indices. (C) 2020 Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour.

Acoustic monitoring is an extremely useful tool for conservationists to track and identify elephants deep within densely wooded environments. However, traditional acoustic audio event detection approaches achieve limited success for two reasons. First, elephant calls are rare in continuous audio data and vary significantly in their duration, frequency, and harmonics. Second, other rare forest noises can easily be mistaken for elephant calls. To tackle class imbalance, we use majority class undersampling and introduce a data augmentation technique that generates additional realistic positive elephant call training examples. Moreover, to handle rare, challenging background sounds, we introduce a novel curriculum driven training strategy that identifies hard to classify background, and uses such sounds to incrementally improve the model. We show that the curriculum approach significantly improves upon our baseline model. Additionally, we introduce the limit case of the curriculum strategy, captured by a new two-stage modeling framework. Our results show that combining these approaches of data augmentation and curriculum-driven learning leads to significant model improvement, achieving a test F1 score of 0.74 compared to 0.45 for the baseline model.

Natural habitats are being impacted by human pressures at an alarming rate. Monitoring these ecosystem-level changes often re-quires 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 em-bed 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 identi-fication of anomalous sounds in playback experiments, providing a potential route for real-time automated detection of irregular envi-ronmental 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 col-laborative autonomous ecosystem monitoring efforts.

SummaryPassive acoustic monitoring is rapidly gaining recognition as a practical, affordable and robust tool for measuring gun hunting levels within protected areas, and consequently for its potential to evaluate anti-poaching patrols’ effectiveness based on outcome (i.e., change in hunting pressure) rather than effort (e.g., kilometres patrolled) or output (e.g., arrests). However, there has been no report to date of a protected area successfully using an acoustic grid to explore baseline levels of gun hunting activity, adapting its patrols in response to the evidence extracted from the acoustic data and then evaluating the effectiveness of the new patrol strategy. We report here such a case in Cameroon’s Korup National Park, where anti-poaching patrol effort was markedly increased in the 2015–2016 Christmas/New Year holiday season to curb the annual peak in gunshots recorded by a 12-sensor acoustic grid in the same period during the previous 2 years. Despite a three- to five-fold increase in patrol days, distance and area covered, the desired outcome – lower gun hunting activity – was not achieved under the new patrol scheme. The findings emphasize the need for adaptive wildlife law enforcement and how passive acoustic monitoring can help attain this goal, and they warn about the risks of using effort-based metrics of anti-poaching strategies as a surrogate for desired outcomes. We propose ways of increasing protected areas’ capacity to adopt acoustic grids as a law enforcement monitoring tool.

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.

Understanding why related species combine calls in different ways could provide insight into the selection pressures on the evolution of combinatorial communication. African savannah elephants ( Loxodonta africana ), African forest elephants ( Loxodonta cyclotis ), and Asian elephants ( Elephas maximus ) all combine broadband calls (roars, barks, and cries) and low-frequency calls (rumbles) into single utterances known as “combination calls.” We investigated whether the structure of such calls differs among species and whether any differences are better explained by phylogenetic relationships or by socioecological factors. Here, we demonstrate for the first time that the species differ significantly in the frequency with which they produce different call combinations using data from multiple study sites. Elephas maximus and L. africana mostly produced roar–rumble combinations, whereas L. cyclotis produced a more even distribution of roar–rumble, rumble–roar, and rumble–roar–rumble combinations. There were also significant differences in favored structure among populations of the same species. Moreover, certain call orders were disproportionately likely to be given in particular behavioral contexts. In L. africana , rumble–roar–rumble combinations were significantly more likely than expected by chance to be produced by individuals separated from the group. In E. maximus , there was a nonsignificant trend for rumble–roar–rumbles to be given more often in response to a disturbance. Site-specific socioecological conditions appear more influential for call combination structure than phylogenetic history.

In this work, we consider applying machine learning to the analysis and compression of audio signals in the context of monitoring elephants in sub-Saharan Africa. Earth's biodiversity is increasingly under threat by sources of anthropogenic change (e.g. resource extraction, land use change, and climate change) and surveying animal populations is critical for developing conservation strategies. However, manually monitoring tropical forests or deep oceans is intractable. For species that communicate acoustically, researchers have argued for placing audio recorders in the habitats as a cost-effective and non-invasive method, a strategy known as passive acoustic monitoring (PAM). In collaboration with conservation efforts, we construct a large labeled dataset of passive acoustic recordings of the African Forest Elephant via crowdsourcing, compromising thousands of hours of recordings in the wild. Using state-of-the-art techniques in artificial intelligence we improve upon previously proposed methods for passive acoustic monitoring for classification and segmentation. In real-time detection of elephant calls, network bandwidth quickly becomes a bottleneck and efficient ways to compress the data are needed. Most audio compression schemes are aimed at human listeners and are unsuitable for low-frequency elephant calls. To remedy this, we provide a novel end-to-end differentiable method for compression of audio signals that can be adapted to acoustic monitoring of any species and dramatically improves over naive coding strategies.

BACKGROUND:Videographic material of animals can contain inapparent signals, such as color changes or motion that hold information about physiological functions, such as heart and respiration rate, pulse wave velocity, and vocalization. Eulerian video magnification allows the enhancement of such signals to enable their detection. The purpose of this study is to demonstrate how signals relevant to experimental physiology can be extracted from non-contact videographic material of animals.RESULTS:We applied Eulerian video magnification to detect physiological signals in a range of experimental models and in captive and free ranging wildlife. Neotenic Mexican axolotls were studied to demonstrate the extraction of heart rate signal of non-embryonic animals from dedicated videographic material. Heart rate could be acquired both in single and multiple animal setups of leucistic and normally colored animals under different physiological conditions (resting, exercised, or anesthetized) using a wide range of video qualities. Pulse wave velocity could also be measured in the low blood pressure system of the axolotl as well as in the high-pressure system of the human being. Heart rate extraction was also possible from videos of conscious, unconstrained zebrafish and from non-dedicated videographic material of sand lizard and giraffe. This technique also allowed for heart rate detection in embryonic chickens in ovo through the eggshell and in embryonic mice in utero and could be used as a gating signal to acquire two-phase volumetric micro-CT data of the beating embryonic chicken heart. Additionally, Eulerian video magnification was used to demonstrate how vocalization-induced vibrations can be detected in infrasound-producing Asian elephants.CONCLUSIONS:Eulerian video magnification provides a technique to extract inapparent temporal signals from videographic material of animals. This can be applied in experimental and comparative physiology where contact-based recordings (e.g., heart rate) cannot be acquired.

Quantitative assessments of the structure of vocalizations are a fundamental prerequisite to understand a species’ vocal communication system and, more broadly, the selective pressures shaping vocal repertoires. For example, to reduce ambiguity in signal interpretation in the absence of auxiliary visual cues, species in densely vegetated habitats should exhibit more discrete vocal signals than species in open habitats. To test this “ambiguity reduction hypothesis”, we conducted the first quantitative assessment of the rumble vocalizations of the forest elephant. Based on 686 forest elephant rumbles recorded with autonomous acoustic recording units at four sites across Central Africa, we used model-based cluster analyses paired with subsequent evaluation of cluster-discreteness and discriminant function analyses to quantify the structure of rumbles based on 23 source- and filter-related acoustic parameters. Model-based cluster analyses suggest that rumbles can be classified into five to eight types. Similar to previous findings in savannah elephants and contrary to the ambiguity reduction hypothesis, average silhouette coefficients below 0.34 indicated that these rumble types were highly intergraded. However, discriminant function analyses predicted rumble types with at least 75% accuracy whereby the location of the minimum fundamental frequency, middle slope and peak frequency contributed most to separation between types. In line with an increasing number of studies highlighting that a distinction between discrete and graded repertoires may have little biological significance, we propose that ambiguity reduction may take place through the evolution of perceptual and cognitive mechanisms, rather than acting on vocal production.

Forest elephants Loxodonta cyclotis aggregate in large numbers in forest clearings. Whether they maintain contact as they move through the forest and are able to coordinate these aggregations, similar to the fission-fusion sociality of the well-studied savanna elephants Loxodonta africana, is currently unknown. Since sound attenuates faster in closed as compared to open habitats, the low-frequency rumble vocalizations of forest elephants may exhibit smaller detection ranges than measured for those of savanna elephants, which may restrict the ability of forest elephants to coordinate interactions between separated family units. Here, we modeled the attenuation of forest elephant rumbles using amplitude measurements of rumbles recorded in a rainforest in Gabon and estimated the distances at which elephants might be able to detect them under observed ambient sound conditions. Our results suggest an attenuation rate less than predictions of spherical spreading loss, suggesting that reflection of the sound waves within the forest results in constructive interference. Nevertheless, we found that forest elephant rumbles of average dominant frequency (31.07 Hz) under average ambient sound levels would not be detectable farther than 0.8 km. Moreover, for 50% of analyzed rumbles, the harmonic structure was completely attenuated at only 100 m. However, we estimated detection distances of up to 3.2 km for rumbles of average dominant frequency when ambient sound was at its lowest. Our findings suggest that long-distance communication to coordinate interactions among separated family units may be limited in forest elephants, with potentially important consequences for their social organization. Significance statement The challenges associated with the extent of, and variation in, detection distances of long-distance vocalizations used by animals to mediate interactions between separated group members has rarely been investigated. While it has been suggested that forest elephants exhibit a fission-fusion sociality similar to savanna elephants, our results indicate shorter detection distances for forest elephant rumbles, suggesting a limited ability to mediate interactions between separated family units. However, under optimal ambient sound conditions, detection distances increased considerably. The long detection distances estimated for savanna elephants may reflect the optimal conditions under which the playback experiments were conducted. On average, savanna elephants may be much more limited in communication distance. Further studies on the constraints and opportunities that the different environments impose on these species' communication capability may be critical to understanding potential differences in the social complexity they express.

African forest elephants face severe threats from illegal killing for ivory and bushmeat and habitat conversion. Due to their cryptic nature and inaccessible range, little information on the biology of this species has been collected despite its iconic status. Compiling individual based monitoring data collected over 20 years from the Dzanga Bai population in Central African Republic, we summarize sex and age specific survivorship and female age specific fecundity for a cohort of 1625 individually identified elephants. Annual mortality (average = 3.5%) and natality (average = 5.3%) were lower and markedly less variable relative to rates reported for savanna elephant populations. New individuals consistently entered the study system, leading to a 2.5% average annual increase in the registered population. Calf sex ratios among known birth did not differ from parity. A weak seasonal signal in births was detected suggesting increased conceptions during the wet season. Inter-calf intervals and age of primiparity were longer relative to savanna elephant populations. Within the population, females between the ages of 25-39 demonstrated the shortest inter-calf intervals and highest fecundity, and previous calf sex had no influence on the interval. Calf survivorship was high (97%) the first two years after birth and did not differ by sex. Male and female survival began to differ by the age of 13 years, and males demonstrated significantly lower survival relative to females by the age of 20. It is suspected these differences are driven by human selection for ivory. Forest elephants were found to have one of the longest generation times recorded for any species at 31 years. These data provide fundamental understanding of forest elephant demography, providing baseline data for projecting population status and trends.

El Nino Southern Oscillation events (ENSO) and the subsequent opposite weather patterns in the following months and years (La Nina) have major climatic impacts, especially on oceanic habitats, affecting breeding success of both land and sea birds. We assessed corticosterone concentrations from blood samples during standardized protocols of capture, handling and restraint to simulate acute stress from 12 species of Galapagos Island birds during the ENSO year of 1998 and a La Nina year of 1999. Plasma levels of corticosterone were measured in samples collected at capture (to represent non-stressed baseline) and subsequently up to 1 h post-capture to give maximum corticosterone following acute stress, and total amount of corticosterone that the individual was exposed to during the test period (integrated corticosterone). Seabird species that feed largely offshore conformed to the brood value hypothesis whereas inshore feeding species showed less significant changes. Land birds mostly revealed no differences in the adrenocortical responses to acute stress from year to year with the exception of two small species (<18 g) that had an increase in baseline and stress responses in the ENSO year - contrary to predictions. We suggest that a number of additional variables, including body size and breeding stage may have to be considered as explanations for why patterns in some species deviated from our predictions. Nevertheless, comparative studies like ours are important for improving our understanding of the hormonal and reproductive responses of vertebrates to large scale weather patterns and global climate change in general. (C) 2018 Elsevier Inc. All rights reserved.

Summary African forest elephants Loxodonta cyclotis are experiencing persistent declines driven by illegal killing and range loss. Despite the importance for policy debates regarding elephant trade managed through the Convention on International Trade in Endangered Species (CITES), little is known about forest elephant demography and, consequently, the impacts of offtake and subsequent population recovery potential. Using 23 years of individually based demographic data from Dzanga, Central African Republic, we found low reproductive potential resulting from annual birth rates averaging 4·3% (SD: 1·4%), a median inter‐birth interval of 68 months and a median primiparous age of 23 years. Average mortality was 3·1% per year (SD: 1·0%) during the study, with approximately 1·4% of that attributed to human killing. This population of forest elephants demonstrated concerningly slow growth rates, with a doubling time of nearly 60 years under current conditions (41 years excluding human impacts), amounting to three times that reported for savanna elephants. As such, forest elephants appear to be significantly more sensitive to human‐induced mortality than their congeneric species. Such slow intrinsic growth challenges current perceptions of historic and contemporary ivory trade impacts on forest elephants, highlighting the urgent need to stem poaching and institute long‐term protective measures. Policy implications. Debates regarding the sustainability of the ivory trade for the species appear to have overestimated growth rates of forest elephants. The information presented here indicates that sustainable offtake models for forest elephants need reassessment.

Hunting for wild meat is pervasive in Afrotropical forests and often the primary threat to large-bodied mammals even within protected areas, the effectiveness of which (at reducing hunting) is correlated with law enforcement efforts, especially anti-poaching patrols. However, although tropical protected areas are critically underfunded and law enforcement accounts for a substantial proportion of their budgets, few have mechanisms in place to evaluate and adaptively adjust anti-poaching strategies based on robust field evidence.

Summary The accelerating loss of biodiversity worldwide demands effective tools for monitoring animal populations and informing conservation action. In habitats where direct observation is difficult (rain forests, oceans), or for cryptic species (shy, nocturnal), passive acoustic monitoring ( PAM ) provides cost‐effective, unbiased data collection. PAM has broad applicability in terrestrial environments, particularly tropical rain forests. Using examples from studies of forest elephants in Central African rain forest, we show how PAM can be used to investigate cryptic behaviour, mechanisms of communication, estimate population size, quantify threats, and assess the efficacy of conservation strategies. We discuss the methodologies, requirements, and challenges of obtaining these data using acoustics. Where applicable, we compare these methods to more traditional approaches. While PAM methods and associated analysis are maturing rapidly, mechanisms are needed for processing the dense raw data efficiently with standard computer hardware, speeding development of detection algorithms, and harnessing communication networks to move data from the field to research facilities. Passive acoustic monitoring is a viable and cost‐effective tool for conservation and should be incorporated in monitoring schemes much more broadly. The capability to quickly assess changes in behaviour, population size, and landscape use, simultaneously over large geographical areas, makes this approach attractive for detecting human‐induced impacts and for assessing the success of conservation strategies.