Deep learning (DL) is increasingly integrated into quantitative ecology, particularly for automating the classification of sensor data in biodiversity monitoring. In addition to substantially reducing data processing effort, DL models often achieve high classification performance. However, despite ongoing improvements, certain species or classification tasks remain challenging, and predictions are rarely error-free. Manual verification is frequently included into data processing pipelines to mitigate misclassifications, but this approach may mask rather than quantify uncertainty. Here, we propose directly incorporating classification uncertainty into ecological inference, rather than filtering it afterwards. Specifically, we treat model predictions as probabilistic outputs rather than fixed class assignments, by using Monte Carlo simulations to propagate uncertainty from the classification process into downstream ecological models. We illustrate this approach using two case studies. The first estimates stochastic population growth rates for a penguin population using detection time series derived from Radio Frequency IDentification (RFID). The second propagates uncertainty in species identification from camera trap images into occupancy estimates. For both, we compare results obtained propagating classification uncertainty with those from conventional single-class attribution at two confidence score thresholds. Our findings show that propagating uncertainty typically leads to higher, more optimistic ecological estimates compared to the single-class confidence approach. Importantly, this method expands the total uncertainty interval by explicitly introducing the confidence score produced by automatic classification as a representation of uncertainty. Quantifying this uncertainty in parameter estimation allows for more informed and reliable ecological interpretations. Monte Carlo simulations offer a flexible and accessible means to integrate classification uncertainty into diverse ecological modelling workflows. ### Competing Interest Statement The authors have declared no competing interest.

Deep learning (DL) is increasingly integrated into quantitative ecology, particularly for automating the classification of sensor data in biodiversity monitoring. In addition to substantially reducing data processing effort, DL models often achieve high classification performance. However, despite ongoing improvements, certain species or classification tasks remain challenging, and predictions are rarely error-free. Manual verification is frequently included into data processing pipelines to mitigate misclassifications, but this approach may mask rather than quantify uncertainty. Here, we propose directly incorporating classification uncertainty into ecological inference, rather than filtering it afterwards. Specifically, we treat model predictions as probabilistic outputs rather than fixed class assignments, by using Monte Carlo simulations to propagate uncertainty from the classification process into downstream ecological models. We illustrate this approach using two case studies. The first estimates stochastic population growth rates for a penguin population using detection time series derived from Radio Frequency IDentification (RFID). The second propagates uncertainty in species identification from camera trap images into occupancy estimates. For both, we compare results obtained propagating classification uncertainty with those from conventional single-class attribution at two confidence score thresholds. Our findings show that propagating uncertainty typically leads to higher, more optimistic ecological estimates compared to the single-class confidence approach. Importantly, this method expands the total uncertainty interval by explicitly introducing the confidence score produced by automatic classification as a representation of uncertainty. Quantifying this uncertainty in parameter estimation allows for more informed and reliable ecological interpretations. Monte Carlo simulations offer a flexible and accessible means to integrate classification uncertainty into diverse ecological modelling workflows.

Education and public engagement are increasingly recognized as essential levers for biodiversity conservation. Over the past decade, biodiversity-related education and engagement initiatives have expanded considerably, yet their evaluation remains fragmented and methodologically uneven. This diversity complicates assessments of effectiveness, comparability, and long-term conservation relevance. We conducted a scoping review of peer-reviewed empirical studies published between 2015 and 2025 to examine how evaluation practices are designed, implemented, and reported in biodiversity-related education and engagement initiatives. Following PRISMA-ScR guidelines, 74 articles were selected from the Web of Science Core Collection and analyzed in terms of implementation contexts, target publics, educational practices, evaluation objectives, methodologies, temporalities, reported outcomes, limitations, and levers. The review reveals a strong concentration of evaluations in formal education settings, primarily targeting children and young people. Objectives are largely structured around the knowledge–attitudes–behaviors (KAB) framework, with limited attention to competencies, empowerment, collective processes, or ecological outcomes. Most studies rely on short-term, self-reported quantitative measures, while longitudinal, qualitative, participatory, and process-oriented approaches remain underrepresented. Although experiential, place-based, creative, and participatory practices are identified as promising levers, their impacts are rarely assessed beyond immediate individual-level outcomes. Overall, the findings expose a structural tension between the diversity of biodiversity education practices and the absence of shared, integrative evaluation frameworks. Advancing evaluation in this field requires systemic, reflexive, and multi-scalar approaches linking individual learning, collective action, institutional change, and conservation outcomes. Such a shift is essential to strengthen the contribution of education to conservation and to support more robust, cumulative, and policy-relevant evidence.

Studying the demographic processes that shape how populations respond to environmental changes has long provided insights for conservation biology. Recent theoretical advances have deepened our understanding of these processes, yet their application in conservation remains unclear. We conducted a literature search to examine how six key demographic concepts — life-history trade-offs, the fast–slow continuum, temporal covariation among demographic parameters, demographic buffering and lability, individual heterogeneity and transient dynamics — have been used in conservation, and discussed their potential benefits and limitations.Their applications fall into three main categories: improving estimates of demographic parameters, population dynamics, and extinction risk; predicting the magnitude and duration of population responses to disturbances or conservation actions; and identifying the demographic processes most relevant for guiding conservation decisions. Individual heterogeneity and the fast–slow continuum were widely used, likely due to their low data and analytical requirements, allowing broad predictions of species' vulnerability and informing conservation decisions. Trade-offs explained how populations adapt to anthropogenic disturbances, invasions or conservation actions. Conversely, temporal covariation and buffering–lability were rarely applied, despite their value for improving projections and assessing populations' capacity to cope with environmental variability. Limited use reflects data and modelling needs, and, for temporal covariation, lack of direct conservation guidance. Transient dynamics, highlighting short-term responses and demographic resilience, are relevant because they match the timescale of many conservation projects.We argue that even modest monitoring efforts can capture essential demographic processes, and that their systematic integration, directly or via inference from related systems, could strengthen long-term conservation outcomes.

Population size is a key metric for management and conservation. This is especially true for large carnivore populations for which management decisions are often based on population size estimates. In France, gray wolves (Canis lupus) have been monitored for more than two decades using non-invasive genetic sampling and capture-recapture models. Population size estimates directly inform the annual number of wolves that can be killed legally. It is therefore key to use appropriate methods to obtain robust population size estimates. To track the recent numerical and geographical expansion of the population, a substantial increase in sample collection was performed during the winter 2023/24 within the entire wolf distribution range in France. A total of 1964 samples were genotyped and assigned to 576 different individuals using microsatellites genetic markers. During the winter 2023/24, spatial capture-recapture models estimated the wolf population size in France to be likely between 920 and 1125 individuals (95% credible interval). Detection probability varied spatially and was positively influenced by snow cover and accessibility. Wolf density was strongly associated with the recent presence of the species, reflecting the ongoing recolonization process from the Alps. This work illustrates the usefulness of non-invasive genetic data and spatial capture-recapture for large-scale population assessment. It also lays the ground for future improvements in monitoring to fully exploit the potential of spatial capture-recapture models.

Ecologists are increasingly expected to inform management decisions under uncertainty, yet most analytical workflows stop at statistical inference. Bayesian decision theory provides a coherent framework to bridge this gap by propagating posterior uncertainty to evaluate alternative actions through utility functions, but remains underused in ecology. Here, we present a practical workflow for implementing Bayesian decision theory using standard Bayesian tools, illustrated with two case studies: wolf management in France, where the decision is the number of wolves to remove under uncertain population dynamics, and invasive muskrat management in the Netherlands, where control effort is allocated across space. In both cases, expected utility integrates posterior uncertainty and management trade-offs. Optimal decisions emerge as compromises between competing objectives. For wolves, optimal harvest balances removal benefits and population risk. For muskrats, optimal effort increases with the importance of population reduction and is unevenly allocated across provinces. Bayesian decision theory provides a formal interface between scientists, who characterize ecological systems and uncertainty, and decision-makers, who define objectives, values and trade-offs. By making these trade-offs explicit, it enhances transparency and relevance for management. It also provides a common framework for bringing together Bayesian statistics, decision analysis and risk analysis, strengthening the link between ecological inference and action.

Understanding the spatial ecology of invasive species is essential for effective management, particularly when ecological damage and sanitary risks are involved. GPS tracking offers valuable insights but remains challenging for semi-aquatic rodents. Here, we describe an exploratory glue-on GPS tagging protocol for coypu ( Myocastor coypus ), applied without anaesthesia and adapted from techniques used on Eurasian beavers ( Castor fiber ). The method is rapid to deploy, minimizes handling time, and limits animal disturbance. We tested this approach on 15 individuals, with contrasted outcomes, allowing us to assess both feasibility and constraints. While successful deployments yielded usable movement data, tag loss and performance issues were observed, likely related to species-specific behaviour such as burrow use and frequent immersion in water. We provide a detailed description of the protocol, field performance, and discuss limitations and recommendations for future applications. We do not draw ecological inference, but share practical methodological insights to inform future GPS studies on coypu and similar species.

Europe has seen the recovery of many species of wild herbivores, which are now widespread across much of the continent. In addition, large carnivores are also recolonising many European countries. Most ungulates are managed through hunting, but natural predation can also have a significant influence in many areas. Therefore, the management of large herbivores must increasingly account for both hunting pressure and the impact of predation. Recent studies suggest that lynx predation can have a significant impact on roe deer population dynamics, both by targeting reproductive individuals and by exerting consistently high predation pressure across a wide range of prey densities. Here, we develop a two-species predator-prey matrix population model that integrates lynx and roe deer through functional and numerical responses. We test a set of management rules, applied to both prey and predators, to examine whether joint hunting of both species can prevent prey declines and stabilise the population dynamics. Our simulations show that protecting (i.e. not hunting) either species increases the predator population, which in turn leads to a decline in the prey population. Hunting only the prey worsens their fate due to the addition of hunting and predation. However, simultaneous hunting of predators and prey, adaptively regulated through simple heuristics, does help prevent prey declines. We also show that the initial densities of the predator and prey population have significant impact on the outcome of the simulations. The importance of relative predator and prey population densities highlights the need for adaptive harvesting that monitors and adjusts to current predator and prey population levels.

Failing to account for ecological processes such as dispersal and connectivity when modeling distributions can lead to biased inference about environmental drivers and reduced predictive performance. Spatial dynamic occupancy models are promising to study range dynamics while accounting for dispersal and connectivity, but they currently rely on restrictive formulations of the colonization process, and computational constraints prevent their application at large spatial scales. Here, we propose a process-based dynamic occupancy model to study the distribution of range-expanding species while accounting for connectivity and effects of the environment. We introduce a formulation based on dispersal-pressure that provides a flexible and ecologically interpretable representation of the colonization process, and develop a computational approach based on sparse distance matrices that enables its application to national and transnational scales. We conducted a simulation study that showed unbiased parameter estimation across various ecological scenarios. We also applied our model to two range-expanding carnivores offering complementary insights: the grey wolf and the Eurasian otter. Our model revealed contrasting colonization dynamic, with wolves primarily constrained by altitude and forest cover while otters where only marginally affected by the environment, suggesting that their distribution is limited by dispersal history rather than habitat preferences. By explicitly disentangling the influence of dispersal and environment on distributions, our model provides better insight into occupancy-environment relationships under non-equilibrium conditions, and help identifies what limits species distributions. In light of the increasing availability of large-scale biodiversity data, our framework offers opportunities to study range dynamics using mechanistic approaches across entire landscapes.

The persistence of large carnivores in human-dominated landscapes is particularly sensitive to the impact of humans on key demographic parameters, such as survival, which is strongly influenced by legal and illegal killing. However, estimating demographic parameters is challenging, requiring remarkable logistical and economic efforts. We developed a Bayesian integrated population model for wolves persisting in a humandominated landscape at the edge of their European range. We incorporated multiple long-term datasets (2007-2019) of population counts, fecundity, molecular individual identification, and dead recoveries of GPScollared wolves. With the IPM, we estimated population size and structure, growth rate, survival and emigration rates, the latter two being largely unknown parameters for wolves persisting in these landscapes. We estimated a growth rate of 1.04 (95%BCI: 1.03-1.05). Estimated survival rates were 0.72 (95%BCI: 0.66-0.77) for adults and 0.53 (95%BCI: 0.30-0.71) for pups, and the emigration rate was 0.13 (95%BCI: 0.06-0.21). The low emigration rates estimated may contribute to explaining the cryptic population structure documented in the Iberian wolf population. By forecasting population growth over a ten-year period (2020-2029) under different fecundity and survival rate scenarios, we identified adult survival as a key factor influencing wolf persistence in human-dominated landscapes. Conflicts around wolf depredation on livestock, which may boost retaliatory killing and public pressure to reduce wolf populations, together with other anthropogenic mortality causes and the impacts of infrastructure development, can reduce wolf survival rates to an unknown extent and, therefore, jeopardize wolf recovery.

Semi-aquatic mammals lie at the intersection of several key conservation issues such as wetland deterioration or species invasions, and monitoring their distribution in space and time is essential to inform conservation strategies. However, gathering information about their presence is challenging due to their elusive lifestyle and generally low abundance. The Eurasian otter (Lutra lutra), a near-threatened and strictly protected species in Europe, is currently recolonizing part of its historical range. Its high conservation interest, combined with a dynamic more commonly associated with range-expanding or invasive species, makes it a particularly compelling case study. Otter monitoring has traditionally relied on scat surveys, but recent environmental DNA (eDNA) and camera-trapping initiatives have emerged offering promising complementary tools. Yet, these approaches have rarely been formally compared, either to one another or across regions. Here, we compared the efficiency of spraint surveys, camera traps, and eDNA for detecting otters, and assessed how their performance varied among four catchments in southern France where the species is known to be present. All three methods provided otter detections with varying efficiency. Scat surveys were the most effective method, with an average detection probability of 0.71 and no strong variability between catchments. Although camera-traps had the lowest detection rate, they provided detections at two of the four sites where no spraint was found, highlighting the complementarity of these two approaches. Detection rates varied greatly between individual cameras rather than between catchments, underscoring sensitivity to camera placement. eDNA showed important variability between catchments, with detection probabilities differing by roughly sixfold across regions. All in all, our results highlight differences in efficiency between methods and across environmental conditions, and show the value of combining approaches for future monitoring programs.

Monitoring species in time and space is vital to identify changes in their status and to help define management measures to improve their conservation. Abundance is often used to monitor species at the population level. However, this metric is costly and requires intensive fieldwork, reaching limits in the context of elusive territorial species distributed over large areas. As a cost-effective alternative, we developed a spatial metric to monitor such species using the grey wolf in France as a case study. We built a dynamic occupancy model for the wolf population in France and calibrated some parameters to turn site occupancy probabilities into occupied versus unoccupied sites. Then, we evaluated the performance of the model under different monitoring-focus scenarios to catch, both at the national and local scales, trends in wolf demography in order to monitor the population and inform its status for the current and previous years. We calibrated the spatial metric to achieve the best trade-off between avoiding false positives and false negatives. To do so, we aimed to match known occurrences of wolf permanent presence areas (packs and non-packs) identified from na & iuml;ve detection, as well as population sizes estimated from capture-recapture analysis. The best scenario performed efficiently to predict wolf presence and absence, as well as new permanent presence area occurrences. However, it detected less the disappearance of such areas, which are usually quickly recolonized when located in the core area of the species distribution. The spatial metric results were presented through three elements: annual maps of the predicted wolf presence, annual estimates of the total area occupied by the wolves at the national scale and evaluation of the changes in occupancy at a local scale from year to year. Practical implication. This spatial metric could be efficient to monitor species trends at a large scale while being more cost-effective in data collection compared to capture-recapture analysis at such a large scale. While accounting for imperfect detection with presence-absence data, the different elements of the spatial metric also provide dynamic maps over the years, allowing for management processes at both national and local scales. Le suivi spatiotemporel des esp & egrave;ces est essentiel pour identifier des changements dans leur statut et d & eacute;finir des mesures de gestion visant & agrave; am & eacute;liorer leur conservation. L'abondance est souvent utilis & eacute;e pour le suivi des populations. Cependant, cette mesure est co & ucirc;teuse et n & eacute;cessite un travail de terrain intensif, atteignant des limites dans le contexte d'esp & egrave;ces territoriales faiblement d & eacute;tectables r & eacute;parties sur de vastes zones. Nous avons d & eacute;velopp & eacute; une m & eacute;trique spatiale, comme alternative plus efficiente, pour le suivi de ces esp & egrave;ces et pr & eacute;sentons le loup gris en France comme cas d'& eacute;tude. Nous avons construit un mod & egrave;le d'occupation dynamique pour la population de loups en France et calibr & eacute; certains param & egrave;tres pour transformer les probabilit & eacute;s d'occupation en sites occup & eacute;s ou inoccup & eacute;s. Ensuite, nous avons & eacute;valu & eacute; la performance du mod & egrave;le selon diff & eacute;rents sc & eacute;narios de suivi pour saisir, & agrave; la fois & agrave; l'& eacute;chelle nationale et locale, les tendances de la population de loups et d'informer son statut pour l'ann & eacute;e en cours et les ann & eacute;es pr & eacute;c & eacute;dentes. Nous avons calibr & eacute; la m & eacute;trique spatiale afin d'obtenir le meilleur compromis entre les faux positifs et les faux n & eacute;gatifs. Pour ce faire, nous avons cherch & eacute; la correspondance avec les zones de pr & eacute;sence permanente des loups (meutes et non meutes) identifi & eacute;es & agrave; partir d'une d & eacute;tection na & iuml;ve, ainsi qu'avec les estimations de taille de population obtenues par mod & egrave;les de capture-recapture. Le meilleur sc & eacute;nario a permis de pr & eacute;dire efficacement la pr & eacute;sence et l'absence des individus, ainsi que l'apparition de nouvelles zones de pr & eacute;sence permanente. Cependant, il a moins bien d & eacute;tect & eacute; la disparition de ces zones, qui sont g & eacute;n & eacute;ralement recolonis & eacute;es rapidement lorsqu'elles sont situ & eacute;es dans la zone c oe ur de la distribution de l'esp & egrave;ce. Les r & eacute;sultats de la m & eacute;trique spatiale ont & eacute;t & eacute; pr & eacute;sent & eacute;s & agrave; travers trois & eacute;l & eacute;ments: des cartes annuelles de la pr & eacute;sence pr & eacute;dite des loups, des estimations annuelles de la zone totale occup & eacute;e par les loups & agrave; l'& eacute;chelle nationale, et l'& eacute;valuation des changements dans l'occupation & agrave; l'& eacute;chelle locale d'une ann & eacute;e sur l'autre. Implication pratique. Cette m & eacute;trique spatiale pourrait & ecirc;tre efficace pour suivre les tendances de l'esp & egrave;ce & agrave; grande & eacute;chelle tout en & eacute;tant plus efficiente en ce qui concerne la collecte de donn & eacute;es compar & eacute; aux mod & egrave;les de capture-recapture & agrave; une telle & eacute;chelle. Tout en tenant compte de la d & eacute;tection imparfaite avec les donn & eacute;es de pr & eacute;sence-absence, les diff & eacute;rents & eacute;l & eacute;ments de la m & eacute;trique spatiale fournissent & eacute;galement des cartes dynamiques au fil des ans, permettant d'informer le processus de gestion & agrave; la fois &

Camera traps and other sensors allow continuous-time biodiversity observation, raising new questions and opportunities for modelling detection in hierarchical models such as occupancy (for species presence) and N-mixture models (for abundance). We focused on a rarely considered aspect: how the temporal treatment of detection covariates affects inference. Through simulations and a five-month case study on an research center, we examined the effects of covariate temporal resolution, discretisation scale in discrete-time (DT) models, and interpolation methods in continuous-time (CT) models. While occupancy and abundance estimates were largely unaffected by these choices, detection estimates were more sensitive to them. DT models with fine temporal discretisation closely matched CT models. Simulations showed that when detection covariates had no effect on detectability, the considered modelling choices had little impact. But when covariates did influence detection, bias and error increased if their temporal variation was not accurately retained. The case study revealed more complex patterns, highlighting the consequences of temporally simplifying both observations and detection covariates. Overall, our results suggest that when detectability is of ecological interest, exploring a range of temporal treatments of detection covariates, from fine-scale to coarser resolutions, can reveal complementary insights into scale-dependent patterns in detection. ### Competing Interest Statement The authors have declared no competing interest. Association Nationale de la Recherche et de la Technologie, https://ror.org/00ht2ab73, CIFRE N°2022/0428 Programme d'Investissement d'Avenir, 2182D0406-A European Commission, GA No. 101052342

The lethal control of large carnivores is criticized regarding its efficiency to prevent hotspots of attacks on livestock. Previous studies, mainly focused on North America, provided mixed results. We evaluated the effects of wolf lethal removals on the distribution of attack intensities in the French Alps between 2011 and 2020, using a Before After Control‐Impact approach with retrospective data. We built an original framework combining both continuous spatial and temporal scales and a 3D kernel estimation. We compared the attack intensities observed before and after the legal killings of wolves over a period of 90 days and a range of 10 km, and with control situations where no removal occurred. The analysis was corrected for the presence of livestock. A moderate decrease in attack intensity was the most common outcome after the lethal removal of a single wolf. This reduction was greatly amplified when removing two or three wolves. The scale of analysis also modulated this general pattern, with decreases being generally amplified at a small spatio-temporal range. Contextual factors (e.g., geographical or seasonal variations) could also lead to deviations from this general pattern. Overall, between 2011 and 2020, lethal control of wolves in France generally contributed to reducing livestock attacks, but mainly locally and to a minor extent. Our results highlight the importance of accounting for scale in such assessments and suggest that the evaluation of the effectiveness of lethal removals in reducing livestock predation might be more relevant in a local context. As a next step, we recommend to move forward from patterns to mechanisms by linking the effects of lethal control on wolves to their effects on attacks through analysis of fine-scaled data on wolves and livestock.