Biology subjects

Etienne, M.-P.

Publications and source records attributed to Etienne, M.-P..

6 recordsLinked to original sources

bioRxiv2026

Monte Carlo simulations to propagate the uncertainty of machine-learning classification into ecological models

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.

ecology↗

bioRxiv2025

When does temporal resolution matter? Including detection covariates in discrete- versus continuous-time occupancy and N-mixture models

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.

ecology↗

bioRxiv2024

Unveiling the Hidden Feast: translating molecular detection into predation rate. An example on biological control by generalist predators in agricultural fields.

Very few processes are as decisive as predation in shaping the structure and dynamics of ecological communities. For most predator species, the number of prey items killed by a predator in a day (predation rate) remains impossible to assess because direct observations are scarce or impossible to acquire. To fill this gap, we propose here a Hierarchical Bayesian Model that integrates data on the molecular detection of prey in predators (e.g. PCR results), on individuals captured in the field on the one hand, and on individuals fed in the laboratory, on the other, in a novel mechanistic framework. By explicitly combining the processes of predation and digestion, model fit provides an estimate of the slope and intercept of the digestion curve, and an estimate of the number of prey consumed by a predator in a day. In a case study targeting 25 carabid beetle species and 5 types of prey in agricultural fields (winter wheat), we use our model to estimate predation rates at species and community scales and demonstrate its advantages for studies on biocontrol and beyond. Code and Data are available on this repository : /r/HBM_PredationCode-C725/

ecology↗

bioRxiv2024

Using informative priors to account for identifiability issues in occupancy models with identification errors

Non-invasive monitoring techniques like camera traps, autonomous recording units and environmental DNA are increasingly used to collect data for understanding species distribution. These methods have prompted the development of statistical models to suit specific sampling designs and get reliable ecological inferences. Site occupancy models estimate species occurrence patterns, accounting for the possibility that the target species may be present but unobserved. Here, two key processes are crucial: detection, when a species leaves signs of its presence, and identification where these signs are accurately recognized. While both processes are prone to error in general, wrong identifications are often considered as negligible with in situ observations. When applied to passive bio-monitoring data, characterized by datasets requiring automated processing, this second source of error can no longer be ignored as misclassifications at both steps can lead to significant biases in ecological estimates. Several model extensions have been proposed to address these potential errors. We propose an extended occupancy model that accounts for the identification process in addition to detection. Similar to other recent attempts to account for false positives, our model may suffer from identifiability issues, which usually require another source of data with perfect identification to resolve them. As an alternative when such data are unavailable, we propose leveraging existing knowledge of the identification process within a Bayesian framework by incorporating this knowledge through an informative prior. Through simulations, we compare different prior choices that encode varying levels of information, ranging from cases where no prior knowledge is available, to instances with accurate metrics on the performance of the identification, and scenarios based on generally accepted assumptions. We demonstrate that, compared to using a default prior, integrating information about the identification process as a prior reduces bias in parameter estimates. Overall, our approach mitigates identifiability issues, reduces estimation bias, and minimizes data requirements. In conclusion, we provide a statistical method applicable to various monitoring designs, such as camera trap, bioacoustics, or eDNA surveys, alongside non-invasive sampling technologies, to produce ecological outcomes that inform conservation decisions.

ecology↗

bioRxiv2023

Analysing biodiversity observation data collected in continuous time: Should we use discrete or continuous-time occupancy models?

O_LIBiodiversity monitoring is undergoing a revolution, with fauna observations data being increasingly gathered continuously over extended periods, through sensors like camera traps and acoustic recorders, or via opportunistic observations. These data are often analysed with discrete-time ecological models, requiring the transformation of continuously collected data into arbitrarily chosen non-independent discrete time intervals. To overcome this issue, ecologists are increasingly turning to the existing continuous-time models in the literature. Closer to the real detection process, they are lesser known than discrete-time models, not always easily accessible, and can be more complex. Focusing on occupancy models, a type of species distribution models, we asked ourselves: Should we dedicate time and effort to learning and using these continuous-time models, or can we go on using discrete-time models? C_LIO_LIWe conducted a comparative simulation study using data generated within a continuous-time framework. We assessed the performance of five static occupancy models with varying detection processes: discrete detection/non-detection process, discrete count process, continuous-time Poisson process, and two types of modulated Poisson processes. Our goal was to assess their abilities to estimate occupancy probability with continuously collected data. We applied all models to empirical lynx data as an illustrative example. C_LIO_LIIn scenarios with easily detectable animals, we found that all models accurately estimated occupancy. All models reached their limits with highly elusive animals. Variation in discretisation intervals had minimal impact on the discrete models capacity to estimate occupancy accurately. C_LIO_LIOur study underscores that opting for continuous-time models with an increased number of parameters, aiming to get closer to the sensor detection process, may not offer substantial advantages over simpler models when the sole aim is to accurately estimate occupancy. Model choice can thus be driven by practical considerations such as data availability or implementation time. However, occupancy models can encompass goals beyond estimating occupancy probability. Continuous-time models, particularly those considering temporal variations in detection, can offer valuable insights into specific species behaviour and broader ecological inquiries. We hope that our findings offer valuable guidance for researchers and practitioners working with continuously collected data in wildlife monitoring and modelling. C_LI

ecology↗

bioRxiv2018

Identifying stationary phases in multivariate time-series for highlighting behavioural modes and home range settlements

O_LIRecent advances in bio-logging open promising perspectives in the study animal movements at numerous scales. It is now possible to record time-series of animal locations and ancillary data (e.g. activity level derived from on-board accelerometers) over extended areas and long durations with a high spatial and temporal resolution. Such time-series are often piecewise stationary, as the animal may alternate between different stationary phases (i.e. characterised by a specific mean and variance of some key parameter for limited periods). Identifying when these phases start and end is a critical first step to understand the dynamics of the underlying movement processes.\nC_LIO_LIWe introduce a new segmentation-clustering method we called segclust2d. It can segment bi-(or more generally multi-) variate time-series and possibly cluster the various segments obtained, corresponding to phases assumed to be stationary. It is easy to use, as it only requires specifying the minimum length of a segment (to prevent over-segmentation) based on biological considerations.\nC_LIO_LIAlthough this method can be applied to time-series of any nature, we focus here on two-dimensional piecewise time-series whose phases correspond at small scale to the expressions of different behavioural modes such as transit, feeding and resting, as characterised by two joint metrics such as speed and turning angles or, at larger scale, to temporary home ranges, characterised by stationary distributions of bivariate coordinates.\nC_LIO_LIUsing computer simulations, we show that segcust2d can rival and even outperform previous, more complex methods, which were specifically developed to highlight changes in movement modes or home range shifts (based on Hidden Markov or Ornstein-Uhlenbeck modelling, respectively), which, contrary to our method, require truly informative initial guesses to be efficient. Furthermore we demonstrate it on actual examples involving a zebras small scale movements and an elephants large scale movements, to illustrate the identification of various movement modes and of home range shifts, respectively.\nC_LI

ecology↗