NCSE2025 Programme
Location: School of Mathematics and Statistics
Workshop and Talks: Theatre D, top floor towards the back of the building
Lunch, Breaks and Registration: Room 1A, take the back stairs to the bottom floor, through the double door and second room on the left
Link to University shop for Registration
Wednesday 17th December
Chairs: Amber Cowans and Fergus Chadwick
University of Edinburgh
Firstly, we present a framework for abundance estimation using AI-satellite surveys in large-scale heterogeneous landscapes. Current aerial surveys are time consuming and costly, whereas satellites which easily image huge areas are attractive for consideration and may represent a cost/effort saving. By design, detections from a point-in-time satellite image differ from detections in aircraft moving through the landscape, and these changes must be accounted for in the abundance estimation framework. Several species have been shown to be satellite-detectable, including Polar Bears, Elephants and Wildebeest, however these focus on counts where wildlife are predominantly in open terrain, such that counts can be treated as a final estimate. We examine the framework required for an AI-satellite abundance survey of elephants in northern Botswana, a large (120,000km\(^2\)) and highly heterogeneous region. We cover all stages from survey design and AI detection, to modelling corrections for heterogeneous visibility vegetation obstruction.
Secondly, we investigate whether Capture-Recapture (CR) abundance methods can be used to obtain confidence bound counts for object detection in aerial imagery. CR methods typically enable a confidence bound estimation of abundance, accounting for inevitable false negatives (missed animals). We take the CR principles of multiple observation occasions, and apply this to object detection of generic objects in a single image. Using different object detection algorithms as individual observers, we draw links with ensemble modelling and investigate whether an extended CR methodology can be applied to generate confidence bounds for counts with object detectors.
Bielefeld University
Markov-switching models are powerful tools that allow capturing complex patterns from time series data driven by latent states. Recent work has highlighted the benefits of estimating components of these models nonparametrically, enhancing their flexibility and reducing biases, which in turn can improve state decoding, forecasting, and overall inference. Formulating such models using penalised splines is straightforward, but practically feasible methods for a data-driven smoothness selection in these models are still lacking. Traditional techniques, such as cross-validation and information criteria-based selection suffer from major drawbacks, most importantly, their reliance on computationally expensive grid search methods, hampering practical usability for Markov-switching models. An alternative approach treats spline coefficients as random effects and employs marginal likelihood maximisation via the TMB R package, avoiding grid search but introducing a computationally demanding nested optimisation problem and potential numerical instability.
As an alternative, we propose using the so-called extended Fellner-Schall method for smoothness selection, which leverages the relatively simple structure of penalised splines treated as random effects. This method provides an efficient and general mechanism for smoothness selection, avoiding the need for nested optimisation and higher-order derivatives, improving numerical stability, and significantly reducing computational costs. Our approach enables the practical estimation of flexible Markov-switching models, even in complex settings.
Cardiff University
Raising awareness of statistics anxiety reduces students’ feelings of this. However, some students are too anxious to attend statistics anxiety workshops. To teach all students about this, one could run a dedicated workshop within a module. However, students who view statistics positively may see little benefit in a whole session dedicated to this. Instead, in this multi-institutional study, lecturers “drip-fed” content on statistics anxiety throughout an existing course, reaching those who needed it while not exasperating those who didn’t. The effect of these “Stat-tastic thoughts” (named to be jovial, if a little awkward!) was compared to other features that the lecturers used, based on what existing literature suggested could reduce statistics anxiety, such as the use of humour or the lecturer demonstrating a positive attitude. Staff approachability, lecturer attitude towards statistics, and the use of humour all scored high in positively affecting students’ feelings towards statistics, and examples featuring lecturers’ pets was an outright winner. Features such as references to statistics history or famous statistics quotes had less positive effect. Our results are useful for other statistics educators deciding what aspects of their teaching to focus time and energy on, or what general approaches to take, to mitigate statistics anxiety.
Thursday 18th December
Chairs: Mia Goldman and Chris Sutherland
University of St Andrews
CEFAS
Often there are multiple models to describe a specific system and ultimately any decision can be sensitive to the model used. Choosing a single model from a suite of models is potentially throwing away information, meaning that the decision is not as well informed as it could be. Further, ignoring alternative models does not rigorously quantify the “true” uncertainty. An alternative to choosing a single model is to combine them using an ensemble model.
In this talk I will discuss common ways of combining models, demonstrating them on a number of examples including a decision problem from fisheries management.
Keywords: ensemble modelling, uncertainty quantification, Fisheries management, risk
University of St Andrews and Universidade NOVA de Lisboa
Wildfire events are increasing throughout the globe. Climate change, land use pressures, and human activity are contributing to these events occurring more often and with greater severity. These events pose a severe risk to life, cause ecological and environmental damage, and affect forestry-based economic activity. Therefore paramount to develop models that can help predict under what conditions such destructive events occur. This study explored the spatial distribution of wildfire ignitions in the United Kingdom using a point process modelling framework, with particular attention to spatial autocorrelation and latent spatial structure. Log-Gaussian Cox processes were applied to model the spatial distribution of wildfire ignitions using the inlabru package in R, which builds on the Integrated Nested Laplace approximation framework. This approach allowed for the modelling of unobserved spatial variation through a latent Gaussian random field and enabled efficient inference even in complex spatial models. The analysis focused on comparing spatial patterns of ignition with variation both spatially and temporally. The modelling framework supported the representation of spatial heterogeneity and offered insight into structural patterns in the ignition data. This approach provides a foundation for further refinement and interpretation as model development progresses. By applying a consistent spatial framework to wildfire data from the UK, this study contributes to the wider effort to understand fire occurrence through statistically rigorous and spatially aware methods, capable of adapting to evolving fire-prone landscapes.
University of St Andrews
Ecologists regularly ask questions about cause and effect in natural systems and are often called upon by funding and governmental agencies to help find causal links in service of sustainable development policy. High complexity and ethical concerns largely prevent the use of randomised, controlled experiments, however, so we must find innovative ways to derive causal information from observational data. Causal graphs are tools that visually represent relationships of cause and effect between variables. These graphs help users to encode and communicate their causal assumptions and then select covariates for models to estimate causal effects that are unbiased by confounding, given those assumptions are met. In this talk, we will introduce causal graphs and motivate their use in ecological modelling via an interactive example. We will address barriers to and concerns regarding causal inference and outline current efforts in the field. No prior knowledge of causal inference is necessary or expected.
University of St Andrews
Chairs: Louis Backstrom and Regina Bispo
CEFAS
Spurdog in the Northeast Atlantic have been assessed as being in poor condition for the past several decades. Management measures including zero catch have led to an improvement in stock status in recent years. Despite this there is a need to avoid spurdog bycatch in mixed and multispecies fisheries like those in the Celtic Sea to protect the improved stock status and allow for further recovery. This is further complicated because spurdog are an aggregating species, with most catches being low or zero but some including hundreds or even thousands of spurdog. We have used generalised additive models (GAM) to predict spurdog bycatch on a fine temporal and spatial scale using data from a self-reporting scheme by fishers. Environmental variables for which fine-scale spatial and temporal predictions exist were used as predictors in the model. Our model is well calibrated predicting low, medium and high bycatch events at rates consistent with the data. Though high bycatch events are rare and hard to predict we demonstrate via a simulation study how high spurdog catches could be significantly reduced based on a simple model of fisher response to predictions from the GAM.
University of Kent
Line transect surveys, fixed paths along which observers record detections of animals or signs of their presence, are a core tool in ecological monitoring. In occupancy studies, line transect surveys are used to determine species’ presence/absence whilst accounting for imperfect detection and varying availability for detection along the transect. However, which research questions can be answered using such surveys depends on how detections are recorded, ranging from binary presence/absence in transect segments to continuous-time detections. These differing data collection protocols give rise to structurally distinct statistical models, with varying identifiability properties and inferential performance. Using algebraic analysis and data cloning, we show that widely used models are often either non-identifiable or not practically identifiable. We develop a novel, data-based diagnostic to detect this issue in applied settings. An extensive simulation study illustrates the practical implications of data-collection approaches, model choice, and corresponding inference. We make recommendations regarding choice of detection protocol and line transect survey design. Our results offer methodological innovation and actionable guidance for ecologists and statisticians.
Lancaster University
University of Kent
Accurate estimation of population size is essential for policymaking, resource allocation and research. However, traditional censuses are sometimes infeasible, often expensive and have poor uptake, which has led to an increasing number of countries adopting a register-based approach to either replace or complement censuses. We have proposed a novel approach to estimate the true population size from register data, building on capture-recapture (CR) models, with a hidden Markov model (HMM) formulation, providing an efficient model-fitting framework. The model accounts for temporary emigration and incorporates an arbitrary number of, possibly interacting, observation registers. While the application offers new insights into population dynamics and individual trajectories, there are possible concerns about computational time as we work at an individual-level. As an alternative, we propose working at a population-level with a model based in MSE but extended to allow knowledge exchange over time, creating a longitudinal approach. Instead of using log-linear models for contingency tables, we consider latent class models (LCMs) which by design assume individuals belong to two (or more) latent classes, each of which behave differently in the observation process. We illustrate both approaches using Swedish population registers where issues of overcoverage - individuals registered as living in the country but no longer living there - provide motivation for this project and highlight the need for accurate estimation.
University of Kent
Citizen science data is an abundant repertoire of low-quality observations which, when aggregated, can be a valuable source of data for inference. In particular, species distribution modelling presents unique challenges without these data, since systematic surveys are expensive and sometimes prohibitively impractical.
Bayesian neural network models, which can ingest big data with minimal preprocessing and provide uncertainty quantification on inferred parameter values and estimates (of population, richness, distribution, etc.), are especially suitable to this application due to both the largeness, and inherent uncertainty arising from the inconsistent quality of, the data.
We continue the work of Saad et al. in their 2024 paper introducing the Bayesian Neural Field, a tool for modelling spatiotemporal field values with intrinsic uncertainty quantification. Our work aims to adapt this tool to species distribution modelling, by implementing a hierarchical structure that separately reflects the underlying ecological and observation processes. One key advantage of this structure is that by minimally reconfiguring the final layers of the modelling hierarchy, the same set of model parameters can be used interchangeably with both count and occupancy data, for either predictive inference or for training.
Since the ground truth of real-world citizen science data is inaccessible, we have implemented a synthetic ecological data generating process to rigorously evaluate the model and perform hyperparameter search. We are currently working to improve model performance on occupancy data which is spatiotemporally sparse, but we have already demonstrated effective interpolation of large-scale patterns in count data in regions with densely available observations.
Chairs: Rebecca Supple and Len Thomas
University of St Andrews
Understanding how human recreation influences wildlife behaviour and habitat selection is crucial for effective conservation, as sensitive species may alter their behaviour to minimise human contact. We conducted a landscape-scale experiment in Cairngorms Connect, Scotland’s largest habitat restoration project, to investigate how woodland passerines respond to varying levels of recreation. Using a network of autonomous recording units placed on and off main trails, we employed the AI species-labelling tool BirdNET to process over 13,000 hours of acoustic data. We developed hierarchical models grounded in the theory of hierarchical habitat selection to examine factors affecting (1) species occupancy (2nd-order selection), (2) species availability to be detected in recordings (3rd-order selection), and (3) call rates (4th-order selection). Fine-scale habitat features were quantified using LiDAR, while human recreation intensity was derived from the global fitness tracking platform Strava. Our findings reveal species-specific responses to recreation that may reshape community composition and behaviour in shared landscapes. Sensitive species adjusted their activity patterns to avoid disturbance, whereas others showed little change. This study demonstrates how emerging technologies can generate powerful ecological insights, enhancing our understanding of human–wildlife interactions in regenerating landscapes and supporting evidence-based strategies to balance conservation and public access in protected areas.
University of St Andrews
Identifying which species have gone extinct (and when) is a core component of biodiversity conservation. In spite of this, accurately inferring extinction remains a significant challenge, as evidenced by the relative frequency with which “extinct” species are rediscovered, often after extensive periods without a confirmed sighting. A number of quantitative approaches to inferring extinction have been developed by conservation biologists and statisticians over the past several decades, many of which model a species’ sighting record and use statistical tests to determine whether extinction has occurred or not. This talk provides an overview of the nearly 50 published sighting records-based models of extinction dynamics, highlights some key statistical challenges associated with making accurate models of extinction, and explores the performance of these models when applied to a real-world dataset of sightings of the recently extinct Slender-billed Curlew.
CEFAS
Patagonian toothfish (Dissostichus eleginoides) are a long-lived, slow-growing deep-sea species endemic to the Southern Hemisphere, and the focus of longline fisheries managed to ensure sustainable exploitation. Managing the fisheries relies on accurate stock assessments that include pertinent biological and exploitation processes. Length and age at first maturity are key biological parameters used in the stock assessments for Patagonian toothfish in South Georgia (Subarea 48.3). The conventional approach to estimating length and age at first maturity relies on macroscopic gonadal staging, but this may be confounded by the difficulty in distinguishing immature and mature resting (i.e., skipped spawning) individuals. Here, we extend existing length-at-age breakpoint models to incorporate sex and individual temperature experience, while accounting for increasing variation in length with age. We fit and compare a set of candidate models to assess the empirical evidence for temperature experience effect on sex-specific length at first maturity estimates using data from 3,683 Patagonian toothfish sampled by both commercial longline fisheries and scientific surveys around South Georgia between 2010 and 2023. Our analysis supports an expected non-linear effect of age on length, with a higher age at first maturity for females compared to males, and demonstrates that temperature experience explains significant variation unaccounted for by age and sex alone, showing an inverse relationship between the temperature experienced by Patagonian toothfish and their length-at-age. These findings have direct implications for improving the biological realism and predictive accuracy of stock assessment models, particularly under climate change.
University of St Andrews
Citizen science datasets are evolving with new technologies, requiring updated analytical methods to address emerging biases. eBird, the world’s largest citizen science project, now operates alongside Merlin, an acoustic identification app from the Cornell Lab of Ornithology. This raises questions about how automated identification tools shape citizen science data. We examine this by modelling how Merlin use is correlated with species reporting rates in eBird. Reporting rates increased with Merlin use for species more often detected by sound, particularly among less experienced observers. Conversely, rates declined for species usually identified visually, perhaps because attention to a phone reduces visual awareness. These heterogeneous effects across users and species highlight the need to account for changing reporting behaviour when drawing ecological inferences, especially trend estimation. As AI tools become more common, explicitly incorporating their effects will be essential for maintaining robust conclusions from citizen science datasets.
CEFAS
Typically, tactical management, such as setting annual quotas, requires making short-term predictions, whereas strategic management, involves setting long-term goals and objectives for managing fish populations. In the short-term fish stocks, most of the information about short-term dynamics is about learning about the current state of an individual stock, i.e. a single-species approach. However, on a longer-term, stocks start to interact with one another through predator-prey interactions, mixed fisheries and environmental changes. Therefore, a multispecies approach is required to make long-term predictions to inform strategic management. A potential issue is that the single-species and multispecies approaches are not compatible with one another, possibly giving conflicting predictions. In this presentation we use the complementary strengths of single-species and multispecies approaches to develop and implement a single framework that make seamless predictions across timescales predictions across all time scales.
Chairs: Oliver Hartley and Danielle Harris
CEFAS
Litter on the sea and the beaches is often bad news for animals living there. I give a summary of current beach and seafloor litter monitoring issues from a UK and international perspective. Along the way, I delve into the murky worlds of EU diplomacy (how not to do it) and academic journal politics.
Bielefeld University
Hidden Markov models (HMMs) are widely used tools for analysing animal behaviour based on movement, acceleration and other sensor data. In particular, they allow inference on how the animal decision-making process interacts with internal and external drivers, by relating the probabilities of switching between distinct behavioural states to covariates. A key challenge in covariate-driven HMMs is the models’ interpretation, especially with more than two states, as then several relationships between state-switching probabilities and covariates need to be jointly interpreted. The model-implied probabilities of occupying the different states, as a function of a covariate of interest, constitute a much simpler and hence useful summary statistic.
A pragmatic approximation of the state occupancy distribution, namely the hypothetical stationary distribution of the model’s underlying Markov chain for fixed covariate values, has in fact routinely been reported in HMM-based analyses of ecological data. However, for stochastically varying covariate processes with relatively little persistence, we show that this approximation can be severely biased, hence potentially invalidating ecological inference based on the approximate version of this important summary statistic of interest.
We develop two alternative approaches to obtain the state occupancy distribution as a function of a covariate: one based on an additional model fitted to the covariate process and another using regression analysis of model-implied state probabilities. The methods are demonstrated through simulations and a case study on Galápagos tortoises. They enable unbiased inference on the relationship between animal behaviour and general types of covariates, helping to uncover the drivers of behavioural decisions.
University of Exeter
Identifying the components of movement paths is essential for studying key processes such as space use, habitat selection, and connectivity. Although animal movement operates across scales, most segmentation approaches do not differentiate how fine-scale behaviours depend on broader movement phases.
We introduce a hidden Markov model (HMM) formulation that jointly analyses fine-scale activity modes (inactive, moving) and broader movement phases (resident, non-resident). Our goal is to provide a flexible and accessible method for segmenting movement trajectories into behaviourally meaningful states within longer-term movement patterns.
Our framework uses variables directly derived from GPS tracks—step lengths and turning angles to identify activity modes, and residence time to characterize broader phases. An asymmetric coupled HMM (ACHMM) structure allows activity modes to depend on movement phases, but not vice versa. We demonstrate the model with two illustrative examples and apply it to telemetry data from Cantabrian brown bears, a species with diverse movement strategies. Here, we modelled state transition probabilities using individual identity and time of day (via splines) to examine interindividual variation in diel activity.
The model effectively segmented trajectories into interpretable states—residence areas, stepping stones, dispersals, excursions—and provided a biologically informed approach to identifying home ranges and core areas. It also showed that the characteristics of activity modes vary with movement phase and revealed interindividual differences and phase-dependent shifts in bear diel activity patterns.
University of St Andrews
Changes in breeding phenology influences population dynamics, yet the processes shaping birth timing remain unknown for many species. British grey seals (Halichoerus grypus) are asynchronous breeders: pupping occurs in a clockwise coastal cline from southwest to southeast UK. Standard pup production models assume global parameters across colonies, ignoring spatiotemporal variation and limiting inference on phenology. To address this, we develop several hierarchical Bayesian models that incorporate colony-specific variation. Using multi-year aerial survey data, we estimate spatiotemporal trends in pup production. Accounting for colony-level spatiotemporal variation improves inference for sparsely monitored colonies and provides uncertainty estimates around birth timing. This framework offers a novel approach to capture phenological structure and signals of environmental change, shifts in colony size, or recruitment through time.
Friday 19th December
Chair: Gabby Arrieta
BioSS
High pathogenicity avian influenza virus (HPAIV) is a rapidly evolving virus causing significant economic and environmental harm. Wild birds are a key viral reservoir and an important source of viral incursions into animal populations, including poultry. However, we lack a thorough understanding of which species drive incursions and whether this changes over time. The ability to rapidly identify potential drivers of incursions may be key to focussing biosecurity efforts and targeting limited surveillance resources.
Under the assumption that species responsible for viral incursions are likely to be found at higher densities where outbreaks are more common, several statistical methodologies are available to explore such associations. However, each have their merits and pitfalls, and must be balanced against stakeholder constraints.
In this talk I discuss three methodologies: spatial random forest, spatial survival analysis, and spatial generalised additive models. I utilise species abundance distributions sourced from the citizen science initiative eBird, alongside farm localities and poultry densities from the Great British Poultry Registers, to explore potential associations between the abundances of 152 avian species and outbreaks of highly pathogenic avian influenza (HPAI) in poultry premises across Great Britain between October 2021 and January 2023. Through post-hoc analyses I also demonstrate how associations were explored across species group aggregations, and how associations changed between outbreak periods.
University of St Andrews
Environmental DNA (eDNA) is a promising approach for assessing the distribution and abundance of many marine species. However, for relatively rare targets, it is unclear whether large-scale eDNA surveys would be feasible, since detection rates tend to be very low. To better understand the feasibility of eDNA sampling for cetaceans, we conducted a spatially explicit simulation that accounts for the distribution and movement of individual animals, production, dispersal, and loss of eDNA, and investigates different possible sampling approaches over a large spatial scale. We used fin whales (Balaenoptera physalus) off the US West Coast as a case study and simulated whale movement and eDNA production over a 5-day period. Then, we used an oceanographic model to track eDNA particle movement over time and applied different possible decay functions. Finally, we simulated eDNA sample collection and investigated the expected precision in estimates of fin whale abundance under various eDNA sampling scenarios. This work illustrates the scale of sampling that would be required to assess cetacean population size using eDNA. We anticipate that this work will guide future survey design and expect the methods and results presented here to be applicable to other relatively rare marine taxa.
University of Edinburgh
For statistical convenience, we often assume that individual heterogeneity within ecological models is modelled via a normal distribution. However, there is typically little validation of this assumption; or investigation of the impact of such model misspecification. We present results from some exploratory analyses that investigate the impact of such misspecification in the context of capture-recapture models. We also discuss how we may more (in)formally assess the normality assumption and how we can fit more general error structures.
This is joint work with Lisa McFetridge (Queens University, Belfast) and Rachel McCrea (Lancaster University)
BioSS
Time-Varying Autoregressive (TVAR) processes are powerful functions which assess how the autoregression of a time-series changes over time. Usually fitted as a non-parametric smoothed function of time, TVAR processes provide insight into the stability of an autoregressive process and assess if there has been a change in serial correlation of the time-series. For example, periods of increased or decreased autocorrelation usually reflect periods of growth or decline in the time-series. Along with TVAR processes, Generalised Additive Model for Location, Shape, and Scale (GAMLSS) provide a flexible modelling procedure which relaxes model assumptions and allows parameters to be modelled directly as smoothed non-parametric functions. Incorporating TVAR processes with the GAMLSS framework provides a complementary model which models the TVAR process of discrete data while also modelling the overdispersion parameter of a Negative Binomial distribution as a function of time.
To better understand changes in aphid abundance across Scotland over time (1967-2023), we model aphid counts from several Scottish suction trap sites (Ayr, Dundee, Edinburgh) using a TVAR(1) process. We incorporate a TVAR(1) process into a GAMLSS framework and use a Bayesian approach to estimate the overdispersion parameter and the TVAR(1) processes as a function of time. Where smoothed functions are penalised using localised shrinkage via Horseshoe priors. Our results show that increases in aphid abundance over time results in increases in autocorrelation. This suggests as aphid abundance increases, aphids are more likely to survive overwintering, highlighting the appropriate application of TVAR(1) processes to model aphid abundance.
Chairs: Rachel Drake and Luca Borger
University of Sheffield
Animal movement data are often characterised in terms of the step lengths and turning angles between observations. While sometimes a convenient summary, this does not directly correspond to a model of the actual movement process. An alternative is to model a path as a sequence of straight-line segments, with turns between them at times that do not necessarily match the observation times. This kind of continuous-time step-and-turn or velocity-jump model poses challenging problems for statistical inference, especially when the timescale of observations is close to that of the turns. I will describe a statistical approach to reconstructing such trajectories and simultaneously estimating the parameters of the underlying movement process, and illustrate it with some real applications. I will also discuss the connection between such models and so-called bouncy particle particle samplers in continuous-time Markov chain Monte Carlo algorithms. This connection gives a coherent way of developing step-and-turn models that incorporate resource selection.
University of St Andrews
The rapid global expansion of wind energy has raised increasing concerns about its environmental impacts, particularly on bird and bat populations. Post-construction fatality monitoring is a key tool for assessing collision-related mortality at wind energy facilities. However, accurate mortality estimation critically depends on accounting for spatial bias in carcass detection. Carcasses rarely fall in a uniform radial pattern around turbines. Because field surveys typically cover limited radii around turbines, failing to adjust for non-uniform spatial distribution and partial coverage can lead to substantial bias in mortality estimates. Existing approaches often rely on parametric assumptions about carcass fall distributions, which may not hold across different contexts. In this study, we present a flexible modelling approach that combines a bulk non-parametric density estimator, based on penalized splines, with a parametric tail to model carcass distances from turbines. In this talk, I will summarize the main results of a simulation study that varies mortality rates, spatial distributions, detection probabilities, and truncation distances to examine how well the method recovers the true underlying distribution and under which conditions it provides the most reliable estimates. Our results highlight the importance of flexible models in addressing partial coverage in wind farm mortality estimation, offering improved tools for environmental impact assessments and wildlife conservation planning.
BioSS
Advances in animal tracking technologies are transforming our ability to study movement, survival, and space use in wild populations. For seabirds such as the Black legged kittiwake (Rissa tridactyla), these questions are both ecologically important and central to consenting for offshore wind energy, where cumulative impacts must be understood to meet renewable targets.
Key gaps remain around juvenile survival, colony connectivity (movement between breeding colonies), year round links between wind farms and Special Protection Areas (SPAs), and overall windfarm use. GPS and satellite devices provide detailed spatio temporal data, but their short lifespan restricts deployment to adults during breeding. Addressing year round use requires approaches that capture ecological processes across space and time over longer periods.
One solution is the Motus automated radio telemetry network, where lightweight ultra-high frequency (UHF) tags on leg rings, combined with receiver stations in the colonies and within offshore windfarms, generate intra and inter annual data on movement and survival. These data can be incorporated into capture–mark–recapture (CMR) frameworks to estimate survival and connectivity, either between colonies or between colonies and windfarms.
Receiver networks provide dynamic detections across time and space, depending on coverage. Mobile receivers expand spatial reach but add analytical complexity, as detections reflect both bird movement and shifting receiver positions. This talk will explore CMR approaches that are suitable for analysis detections of tagged individuals in continuous time (such as continuous time CMR) and/or in continuous space (such as search encounter spatial capture–recapture models), illustrating key points through simulation scenarios.
University of St Andrews
Chair: Luca Borger