Abstract Understanding plant behaviour requires the integration of multiple phenotypic and physiological signals measured over time under controlled conditions. However, different plant signals are typically studied using separate experimental setups, limiting temporal alignment and integrative analyses. We present Mind(the)Plant, a modular experimental facility designed for the synchronized, long‐term acquisition of multimodal plant data, including three‐dimensional shoot kinematics, above‐ and below‐ground volatile organic compounds (VOCs) and root imaging. Its modular architecture is designed to accommodate additional acquisition modules, such as electrophysiological signalling, as future extensions. The platform integrates a controlled growth environment with stereovision imaging, high‐resolution time‐of‐flight mass spectrometry and custom rhizocameras. These components are connected through a unified network infrastructure that ensures synchronized acquisition and centralized data handling. We validate the performance of each acquisition module through multi‐week recordings, demonstrating high‐temporal stability, reliable stereovision synchronization, effective isolation of VOCs signals and robust operation of below‐ground imaging. We further illustrate the analytical potential of the platform using a one‐day continuous multimodal acquisition combining shoot kinematics, above‐ground VOC emissions, rhizocameras observations and environmental data. Mind(the)Plant provides a novel methodological framework for studying plant behaviour, signalling and phenotypic plasticity in ecological and evolutionary research. By enabling coordinated measurements of multiple plant response modalities, the platform supports investigations of dynamic plant‐environment and plant–plant interactions from a behavioural perspective.
Abstract The hidden Markov model (HMM) is a central framework for identifying behavioural state changes from animal movement data. In this context, movement is typically represented as a sequence of observed metrics, such as step length and turning angle distributions, generated by an unobserved behavioural state process. Each hidden state corresponds to a distinct behavioural mode (e.g. foraging, resting), and transitions between states are governed by probabilistic rules. Typically, these models have been applied to data collected at relatively low frequency, for example one location every few minutes or hours, so that turning angle and step length distributions are, to some extent, an artefact of the data‐gathering frequency. However, with the growing availability of high‐frequency biologging data (often greater than 1 Hz), it is now possible to determine the precise places where an animal has turned, enabling behaviourally informed step length and turning angle distributions to be fed into HMMs. In this study, we introduce a fast and accurate method for identifying animal behavioural states from high‐frequency data (e.g. ≥1 Hz) within the HMM framework. We first use an existing algorithm to segment the animal's high‐frequency movement path into steps, where each ‘step’ is defined as the straight‐line trajectory between successive real turning points. We then develop a new HMM model for identifying behavioural states that accounts for steps that are of differing durations. We then extend our technique to allow state‐switching to be driven by environmental effects. To evaluate the accuracy of our method, compared with methods that use lower‐frequency data, we apply it to both simulated trajectories and high‐frequency data from free‐ranging goats (Capra aegagrus hircus). Our technique greatly improves the accuracy of inference compared with using step lengths and turning angles derived from fix‐to‐fix steps in lower‐frequency data. Indeed, we demonstrate that the latter can lead to quite notable inaccuracies, and be very sensitive to sampling frequency, in situations where our method achieves >95% accuracy in state identification.
Abstract The ability to accurately measure heart rates is vital for studying the ecological and evolutionary pressures on animals because cardiovascular regulation is associated with key physiological responses to environmental change. Heart rates, especially in ectotherms, are influenced by multiple factors over both the short and long term, including environmental temperatures, handling frequency, metabolic rates and hemodynamic demands. Traditional methods for measuring heart rate in small vertebrates often involve invasive techniques or unreliable non‐invasive tools. In this study, we present a novel, non‐invasive methodology for stabilizing and measuring heart rates in a small vertebrate ectotherm, the lizard Sceloporus consobrinus. We measured heart rates at both 28 and 32°C using a Doppler ultrasound system fixed within a custom‐built restraint chamber and a commonly employed egg heart monitor (The Buddy®). Compared to the conventional egg monitor device, the Doppler system measured heart rates with greater accuracy and precision while providing greater flexibility in experimental design and organismal size. Our design provides an improved non‐invasive monitoring technique for physiological research and a standardized approach for studying cardiovascular responses in small vertebrates.
Abstract Understanding trophic positions is essential for analysing complex food webs. While calculating these positions can be straightforward when feeding relationships and their proportions are known, determining diet coefficients traditionally requires substantial time and effort. Modern analytical techniques, such as stable isotope analysis, provide an efficient alternative for estimating trophic positions. Stable isotope analysis provides trophic position estimates, but this information alone does not reveal prey–predator relationships. To help reconstruct food webs, a method is needed to convert trophic positions into diet coefficients. Our method employs a superposition of potential prey pairs, with weighting coefficients given by probabilities computed using Bayesian statistics. We present a mathematical inverse analysis approach to estimate diet coefficients using a minimal dataset consisting only of trophic positions. This approach provides a robust and parsimonious framework for reconstructing potential feeding networks from minimal data, offering a quantitative bridge between, but not limited to, stable isotope measurements and ecological network structure.
Abstract Formal undergraduate research experiences (UREs) have provided students with authentic research experiences in ecology for decades, but accessibility to these resources is inherently limited, as evidenced during the COVID‐19 pandemic. Moreover, UREs often do not reflect the diversity of locations of current ecological publications. Open online educational resources (OERs) can address the need for more accessible resources, but they usually rely on the curation of static content and can be functionally restricted by a lack of pedagogical instructions. The OCELOTS network (Online Content for Experiential Learning of Tropical Systems) addresses these gaps by shifting the OER model from static curation to a dynamic, scalable ‘ incubator matrix ’ . Rather than serving as a passive repository for disconnected teaching files, OCELOTS acts as a structured workshop and collaborative peer review network where active researchers translate their own peer‐reviewed primary literature into modular, narrative‐driven OER case studies, hereafter referred to as ‘modules’. A combination of factors contributes to OCELOTS' novelty. (a) Researchers who published the study are part of the module‐creating team. (b) Modules are place‐based and globally networked and designed for any type of instructional setting. (c) Peer review is conducted through a living collaborative model rather than a static editorial review. (d) Embedded real‐time data tools use applications that incorporate actual field datasets collected by the module authors. (e) Instructors publish implementation plans and teaching notes for using modules in undergraduate courses. (f) The design is explicitly decolonial in its multilingual accessibility. We summarize the various ways in which 29 OCELOTS modules, representing the work of 68 authors in various tropical field sites in 12 countries, have been created, implemented, adapted, and disseminated. We reflect on how this network has enabled these activities that portray a range of scientific content, processes, and methods in an immersive way. Moving beyond the limitations of standard static text portals, the OCELOTS network presents a highly transferable methodological framework. It demonstrates how global research coordination networks can actively create, review, and implement high‐quality interactive open materials that democratize complex biological data across diverse institutional types around the world.
Abstract Using machine learning models to classify bioacoustic signals of animal species is increasingly important for conservation monitoring because manual expert labelling is time‐consuming and tedious. Monitoring endangered species is particularly challenging because these species are rare, making it difficult to collect the large, representative datasets needed to train conventional machine learning models. Zero‐shot learning (ZSL) offers a promising solution by enabling models to recognise new species that are absent from the training data. We developed a multi‐stage ZSL approach to detect and classify frog calls in natural soundscapes recorded by an automated acoustic monitoring device. Our multi‐label ZSL method uses a denoising model for embedding generation. It is designed for complex real‐world conditions, including substantial background noise, long periods without frog vocal activity, and overlapping multi‐species choruses. These settings are rarely represented in the curated benchmark datasets used in many previous ZSL studies. To evaluate our approach, we curated and annotated a new dataset of Australian frog calls and conducted a three‐part ablation study. The ZSL model was trained on AnuraSet, a large corpus of Neotropical anuran calls, and therefore represents a zero‐shot transfer of anuran call classification across continents. Using our evaluation set, the model achieves an F1 score of 65.54% for frogs overall and 64.14% for the endangered green and golden bell frog (Litoria aurea), outperforming a distribution‐matched random baseline of 25.9% by more than double. Our findings show that ZSL can support practical acoustic monitoring of rare or previously unrecorded species under real‐world conditions. We present a proof‐of‐concept system that operates in noisy, multi‐label soundscapes and demonstrates cross‐continental transfer of anuran call classification from the Neotropics to eastern Australia. These results highlight the feasibility of applying ZSL to biodiversity monitoring and identify key design considerations for future implementations.
Abstract The ecological dynamic regime (EDR) framework was recently proposed as an alternative to equilibrium‐based approaches for assessing ecological resilience in empirical systems, explicitly incorporating dynamic regimes as a reference for assessing the system's deviation during disturbances. Yet the lack of predictive capacity of the EDR framework limits its applications, especially when long‐term data are unavailable or the disturbed system is not well represented by frequently observed dynamics. Here, we extend the EDR framework by introducing an algorithm (PETRA‐EDR: Predicted Ecological TRAjectories in Ecological Dynamic Regimes) and a metric (MPD: Mean Predicted Deviation) to forecast ecological dynamics and estimate prediction accuracy. Our method employs multivariate analyses and can be applied to any ecological system characterized by a set of state variables (e.g. species abundances, functional traits). We conducted a simulation study to evaluate the performance of our method and illustrated its application using empirical data from Canadian boreal forests. Our results demonstrate the method's ability to forecast ecological dynamics and the effectiveness of distance‐weighting functions in improving predictions while mitigating the effects of insufficient sampling, observation noise and hidden variables. Finally, we discuss the assumptions of our method and its applications for assessing ecological resilience to pulse disturbances and detecting regime shifts from a multidimensional, dynamic perspective.
Abstract Wildlife populations are undergoing unprecedented changes to their spatial distributions as a result of human activities. Innovative approaches are needed to document these changes, especially in cases where sampling is sparse. Here, we contribute to the tools available in spatial ecology by developing novel metrics based on the coefficient of variation (CV) applied to location data to help characterize population spatial structure. These threshold‐free methods distinguish between population elongation and fragmentation using simple summary statistics. Simulations confirm the analytical results and demonstrate robustness across diverse scenarios, with reasonable power to detect trends even with small sample sizes. An R package (shapeshiftr) implementation makes these tools more widely accessible. By characterizing the spatial distribution of individuals, these tools allow space use to be readily compared across time and species and provide early warning signals of population fragmentation.
Abstract Systematic reviews and meta‐analyses are key evidence synthesis methods for informing future research, interventions and policy. As the validity of their conclusions depends on the primary studies they synthesise, assessing the internal validity of the included studies is essential. In some fields, such as medicine, this is the norm and is commonly done using Risk of Bias assessment tools. Risk of Bias (RoB) assessment is however rare in ecology and evolutionary biology (EEB) even though several RoB tools have been developed in some ecological subfields and related fields. To identify potential reasons for a limited uptake, we conducted a survey of ecologists and evolutionary biologists with evidence synthesis experience and reviewed 275 journals that publish EEB research for guidelines on performing RoB or related assessments. Only 28 of 232 (12%) survey respondents had correct interpretation of the RoB concept, while 46 (40%) of 116 that had heard of RoB have confused RoB with publication bias. Just 10 (4%) had conducted a RoB assessment, most of whom found it challenging. Out of the 209 EEB journals that explicitly solicit evidence synthesis (N = 58) or reviews (N = 151), only five (2%) directly mentioned standards for conducting evidence synthesis, which include RoB or related (e.g. critical appraisal) assessments. An additional 45 (22%) journals indirectly linked to RoB or a related assessment via referring to the guidelines for reporting evidence synthesis (e.g. PRISMA), despite such reporting guidelines not providing information on how to conduct RoB assessments. To increase its uptake in EEB we recommend making RoB assessment: (1) known and recognised as an essential component of a reliable evidence synthesis by including it in training materials, and journals' and funders' guidelines and policies; (2) easy to perform by bringing the synthesis community together to determine the need for developing new or adjusting existing RoB tools; and (3) possible by further improving reporting standards for primary studies so that RoB assessment can be done on these studies. For those unfamiliar with the RoB assessment, we provide five key RoB questions that existing tools often cover. These questions can be considered to understand the basic composition of the evidence included in evidence synthesis.
Abstract Passive acoustic monitoring (PAM) is widely used to collect large‐scale wildlife data encompassing many species, but efficiently identifying all species—particularly rare and potentially endangered ones—without exhaustive manual examination of the entire dataset remains a major challenge. We formally introduce the machine learning task of novel class identification (NCI), which aims to identify as many classes (e.g. species) as possible while examining as few samples as necessary, and we propose a comprehensive evaluation framework for both overall and class‐presence‐specific performance. NCI uses human‐in‐the‐loop querying strategies based on the active learning paradigm, while its objective of efficiently identifying classes aligns with open‐world learning assumptions and is closely related to the goals of novel class discovery. Drawing on insights from the literature, we identify four promising query method families: embedding‐based methods, self‐supervised learning, similarity learning and audio‐language models. To explore the potential of these directions, initial experiments evaluate 44 query method configurations from the identified query method families across five multi‐label PAM datasets. The results demonstrate that the proposed evaluation framework effectively captures both overall and class‐presence‐specific performance, enabling robust comparison of different methods. These experiments further indicate that audio‐language models achieve the highest performance, providing a foundation for future research. Our findings highlight the potential of deep learning methods for rapid species identification, supporting faster and more informed conservation decisions from large PAM datasets.
Abstract Ecoacoustics mainly aims at monitoring soundscapes by means of non‐invasive protocols. Despite the widespread adoption of machine and deep learning techniques, existing ecoacoustic models predominantly rely on supervised learning and, consequently, face two primary limitations: (1) the necessity of annotated data; and (2) the restriction to fixed pre‐defined classes. In this work, we leverage recent advances in contrastive language‐audio pretraining (CLAP) and envision its first application to large‐scale soundscape analysis. As trained on extensive datasets of paired audio and global text descriptions using contrastive learning, CLAP allows computing similarity scores between audio and text prompts without the constraints of pre‐defined categorical lists. This flexibility enables a comprehensive investigation of various elements within the recordings from coarse‐grained (e.g. mammals, weather, humans, vehicles) to fine‐grained (e.g. dog, rain, speech, airplane) descriptions. Here, we first conducted a preliminary experiment on a calibration dataset, featuring audio events likely to occur in soundscapes, which is shared with the community and constituted from an online, free sound library. Then, we developed a methodology to define reproducible, bounded, independent and interpretable Contrastive Ecoacoustic Indices (CEI), which can characterize the prevalence of four primary sound categories in soundscapes—biophony, geophony, anthropophony and technophony. We finally computed these new CEI on 9‐month field recordings (189,137 1‐min excerpts) monitoring both tropical (Ecuador) and temperate (France) soundscapes, portraying an anthropic gradient from protected forests to urban city centres. This experiment reveals clear soundscape patterns associated with human population density, suggesting that the CEI could be used in other ecological contexts.
Abstract Macroevolutionary adaptation of a continuous trait to different discrete states across species can be modelled using a phylogenetic state‐dependent Ornstein–Uhlenbeck (OU) process. Existing inference methods face two challenges. First, although efficient likelihood algorithms for continuous trait evolution models exist, none have been specifically described for a state‐dependent OU model. Second, the commonly adopted sequential inference approach, where the state‐dependent OU model parameters are inferred conditionally on a discrete character history, treats the character history as known and does not allow the continuous trait to inform character history estimates. We present two mathematically equivalent approaches to compute the likelihood under a state‐dependent OU process: a variance–covariance approach and a pruning approach. We demonstrate the computational efficiency of our pruning algorithm for likelihood calculation of a state‐dependent OU model implemented in RevBayes. Coupled with a data augmentation approach to sample character histories of a discrete character, our model can jointly infer continuous and discrete trait evolution using Markov chain Monte Carlo. We validate our derivation and implementation through simulation tests. Using a case study of tooth crown evolution in ruminants, we compare joint and sequential inference approaches. We show that the root state estimates and some OU parameter estimates are qualitatively different between joint and sequential approaches. This indicates that the continuous trait can be informative for estimating the character history, which in turn impacts the OU parameter estimates. Our state‐dependent OU process and pruning algorithm enable studies using large phylogenies. Together with data augmentation, our joint inference approach provides a more robust and flexible method of macroevolutionary adaptation of continuous traits to underlying discrete states.
Abstract Team‐based synthesis is a collaborative, data‐intensive, often interdisciplinary approach to conducting science that capitalizes on existing data to answer novel questions. In ecology, synthesis has advanced understanding about the functioning and structure of ecological and environmental systems. Interest in ecological synthesis has greatly expanded over the past decades due to the increasing volume of accessible data, new collaboration tools, targeting funding calls, and computational workflows that facilitate working with large, complex data. The team‐based nature of synthesis fosters human connections and mentorship opportunities that make the work meaningful and enjoyable, but also challenges including harmonizing disparate datasets, coordinating scientists with a range of expertise and backgrounds and managing multi‐author projects that evolve over multi‐year time spans among others. Here, we have compiled a list of ten simple rules to help team members engage and excel throughout the entire process as they embark on ecological synthesis journeys.
Abstract Simulation modelling is becoming increasingly widespread to evaluate population biological responses to global changes such as habitat loss or habitat fragmentation. To produce reliable results, this approach requires great care in setting up the model. In the case of spatially explicit modelling, this involves all spatial parameters, but particularly habitat patches, the spatial unit at which demographic processes and dispersal occur. We developed a method to delineate habitat patches that are biologically relevant for simulating demographic processes of the species of interest. Based on the Voronoï tessellation principle, this fully automatized method splits contiguous tracts of habitat such that the resulting patches meet the required surface (e.g. average size of one or several home ranges of the study species). We investigated the consequences of patch definition on modelling outputs (population density, habitat occupancy, dispersal success, straightness and distances), comparing predictions when modelling was based on patches as they are currently usually defined (contiguous sets of habitat cells) with our species‐adapted definition. To do so, we used the spatially explicit individual‐based modelling platform, RangeShifter, applied to five study areas in the south of France, taking the European red squirrel as an example. Our results showed that our framework was effective in dividing large contiguous tracts of habitat into smaller, species‐adapted patches. They also showed that simulation outputs strongly depended on habitat patch delimitation, with effects being mostly consistent among the different landscapes we tested. We also show that mean dispersal distances and population densities required species‐adapted patch definition to be as close as possible to empirical data. Finally, in order to be able to use predicted data for conservation or reintroduction purposes, species‐adapted parameters need to be adjusted after having chosen a species‐adapted definition of habitat patches.
Abstract Many ecosystems potentially exhibit alternative stable states, where distinct states can coexist under identical environmental conditions. While simulation models have generated key hypotheses in alternative stable states theory, they often rely on scale‐free parameters disconnected from real ecosystems. Controlled experiments remain the gold standard for identifying alternative stable states, but are time‐intensive due to transient dynamics. Observational data have become increasingly important due to their rich and flexible insights into community structure and ecosystem processes, as well as their rapidly expanding availability. However, current methodological frameworks using field survey data—particularly snapshot surveys—have not yet incorporated recent advances and best practices, which limits their practical utility. To address this gap, we summarized recent progress and propose an updated, practicable framework that links alternative stable states theory directly to empirical data. Specifically, we focused on (i) sampling strategy, (ii) selecting state variables, (iii) environmental factors, (iv) statistical evidence of alternative stable states, (v) the procedure of determining drivers and (vi) inferring regulating mechanisms. We aim to provide an easy‐to‐use template for studying alternative stable states with snapshot field survey data, which can be tailored to individual study systems.
Abstract Many migratory animals use spatial variation in the Earth's magnetic field for orientation and navigation. This raises the possibility of exploiting these same cues to geolocate and track migratory animals throughout their annual journeys. Magnetic‐based geolocation could be particularly valuable for small aerial and aquatic species, for which the most precise pressure‐based geolocation cannot be applied. We developed a novel geolocation method, implemented in the R package GeoMagR (https://geopressure.org/GeoMagR/) that uses three‐axis magnetic field measurements from lightweight multi‐sensor tags. The workflow consists of: (i) tilt compensation and magnetic calibration; (ii) extraction of time series of magnetic intensity and inclination; and (iii) generation of spatial likelihood maps using the World Magnetic Model (WMM). We evaluated the method using field data to assess achievable spatial precision and potential complementarity with other geolocation techniques. Magnetic geolocation achieved a typical spatial accuracy of ~150 km in latitude but provided little constraint in longitude. When combined with light‐based geolocation, which offers high longitudinal accuracy but poor latitudinal resolution, the integration markedly improved overall position estimates. Magnetic geolocation is a viable complementary tool for tracking small avian species, particularly in contexts where pressure‐based methods cannot be used. By independently resolving latitude, it enhances spatial inference when integrated with other geolocation data sources. This processing of 3D magnetic and acceleration data from geolocators also extends their use beyond positioning, enabling applications in studies of behaviour and navigation.
Abstract Automatic re‐identification of animals has significant potential to address pressing ecological and conservation challenges through improved population monitoring, individual health assessment and detailed behavioural analyses. Although numerous computer‐vision‐based solutions have been proposed and many achieve high accuracy, most remain unsuitable for real‐time analysis and deployment on low‐power edge devices (e.g. drones, camera traps). Here, we address both aspects and introduce an open‐source tool for Real‐time Animal Pattern re‐Identification on edge Devices (RAPID). RAPID processes over 40–60 cropped bounding box images per second on a standard PC or laptop and more than 10 images on an inexpensive off‐the‐shelf edge device. The algorithm operates efficiently in data‐ and compute‐limited environments, relying solely on CPU, leaving GPU resources available for other tasks, all while maintaining or even surpassing state‐of‐the‐art accuracy. Furthermore, each prediction is accompanied by a data‐driven confidence score, facilitating reliable downstream use. Our approach leverages SIFT (scale‐invariant feature transform) descriptors, which continue to demonstrate competitive robustness and accuracy against recent traditional and deep‐learning‐based methods. To overcome SIFT's main limitation—its high‐dimensional feature vectors and the associated computational cost—we integrate recent advances in vector similarity search beyond constructing a database of feature vectors rather than database images, thereby accelerating the query processing. The resulting pipeline is carefully designed to be intentionally minimalistic yet highly effective, retaining only the key components essential for accurate and fast re‐identification. We evaluate RAPID on six datasets: four publicly available animal re‐identification benchmark datasets and two new identity‐labelled datasets we release alongside this paper, namely ZebraStereoID, which contains multiview video footage of zebras, and JaguarID, a small camera trap dataset consisting of day‐ and night‐time videos. Our evaluation demonstrates strong generalisability for species, camera systems and environmental conditions. Additionally, we introduce a RAPID‐based tool, FalseTagFinder, for cleaning benchmark dataset labels, providing corrected labels for the StripeSpotter dataset as an example. Data, code and video abstract available at https://keeper.mpdl.mpg.de/d/228be8688e6146bd8221/?p=%2F&mode=list and https://www.youtube.com/watch?v=O6NWzLEivr8.
Abstract Recent advances in airborne environmental DNA (eDNA) passive sampling methodologies and their successful application in monitoring terrestrial plant and animal biodiversity have sparked growing research interests in extending this technique to fungal community surveys. Yet concerns have been raised that inherent disparities between fungi and macroorganisms may create divergent temporal representation of airborne eDNA, with fungal eDNA prone to distinctive accumulation dynamics. While fungi differ substantially from macroorganisms in the production, dispersal and environmental persistence of their propagules, fungal spores exhibit striking similarities to pollen from wind‐pollinated plants in multiple key characteristics. Such parallelisms enable the transfer of empirical and methodological insights from airborne eDNA‐based plant community monitoring to fungal research. Previous temporal analyses of airborne plant eDNA have demonstrated that even with continuous pollen accumulation on passive samplers, airborne eDNA can capture near‐real‐time variations in plant pollination activities. This pattern implies that newly deposited biological materials can obscure legacy eDNA signals, thereby supporting high‐resolution tracking of species phenological dynamics. Notably, the correlation between airborne fungal eDNA and actual community composition can be modulated by fungal functional traits, atmospheric processes and diverse environmental conditions. Future investigations targeting airborne fungal eDNA across varied temporal sampling windows will deepen our mechanistic understanding of the relationships between aerial eDNA signals and their corresponding biological sources.
Abstract Passive acoustic monitoring (PAM) is an important tool for wildlife monitoring. Deep learning, particularly convolutional neural networks (CNNs), has become the standard approach for developing bioacoustic classifiers. However, real‐time classification remains challenging due to the high computational complexity of these models and the need for resource‐intensive preprocessing steps. One way to reduce model size and computational complexity is to decrease the input dimensions, specifically by reducing the mel‐spectrogram size—the visual representation of sound. We introduce Evolutionary Spectrogram Optimisation (ESO), an evolutionary algorithm that automatically identifies and extracts narrow frequency bands from mel‐spectrograms that are informative enough to allow CNN‐based species detection. ESO is designed to optimise the selection and size of each of these bands, while maximising classification score, and minimising the number of CNN parameters. It eliminates preprocessing steps such as low‐pass filtering and downsampling during mel‐spectrogram creation, optimising for real‐time processing. We evaluated ESO on three distinct PAM case studies targeting the detection of Hainan gibbon, Thyolo Alethe and Pin‐tailed Whydah vocalisations across diverse natural soundscapes. Compared to a baseline CNN trained on full mel‐spectrograms with preprocessing, ESO reduces mel‐spectrogram size by up to 57% while simultaneously decreasing CNN parameters by up to 72% and improving the F1‐score by up to 6%. Our findings revealed that ESO increases inference speed by 45%, reduces peak memory usage by 47% and lowers energy consumption by 56% for the Pin‐tailed Whydah dataset. To facilitate its use, we provide ESO as a Python package and a user‐friendly graphical user interface. A comprehensive user guide is available on the project's GitHub page https://github.com/ufuk‐cakir/ESO. This study serves as stepping stones towards facilitating the deployment of onboard models for PAM and advancing research in this direction. By reducing model complexity while maintaining high classification accuracy, ESO is a promising solution for real‐time PAM applications. It enhances computational efficiency by reducing model inference time, memory usage and power consumption, making the method well suited for onboard deployment. While the evolutionary algorithm requires computational resources to identify the optimal solution, its reusability and intuitive design make it a valuable tool for further research and practical applications.
Abstract Quantifying biological motion is fundamentally tied to quantifying the biological structures that produce that motion. Yet, this dependence makes it essential to decouple motion from static morphology to enable general, comparable analyses across individuals and conditions. In this work, we present a methodological pipeline to study biological motion data obtained from experimental motion capture of human breathing kinematics before and after maximal exercise (gold‐standard optoelectronic plethysmography markers). We used four‐dimensional geometric morphometrics (4DGM; 3D shape through time) to characterize breathing patterns while standardizing torso shape and expiratory and inspiratory times. Results showed not only differences in breathing pattern before and after maximal exercise, but also differences in motion pattern associated with BMI and torso shape. Specifically, we found higher thoracoabdominal asynchrony in low‐BMI subjects compared with high‐BMI subjects. We also found a strong covariation between torso shape and the 3D trajectory of motion, demonstrating the power of the method to detect a shape‐to‐function signal even with a small sample size. This method is proposed as a flexible tool to separate biological motion from its underlying structure, which is particularly useful for studying complex systems. In the case of breathing pattern, it is proposed to investigate possible applications in clinical settings to test whether it detects differences between healthy and pathological kinematics, in sport sciences to attempt to link respiratory function to performance, and in evolutionary studies as a possible tool for inferring respiratory function in extinct morphologies.
