IntroductionThe natural world is experiencing immense change, driven by increasing human populations, climate change, and the degradation in quality of key resources such as natural habitats and water (Maja and Ayano, 2021). These rapid and widespread environmental changes have created conditions that far exceed the adaptive capacities of many species, leading to significant declines in wildlife populations and the degradation of biodiversity globally (Isbell et al. 2023). This loss of biodiversity not only threatens the survival of individual species but also undermines the integrity of ecosystems that provide essential services to both people and nature (Isbell et al. 2023). Systems that monitor and manage wildlife in areas of land-use change are crucial for disentangling the complex effects that change has on wildlife populations and their habitats (McRae et al. 2008; Murphy et al. 2023a). Without these systems in place, we will fail to understand how rapid change impacts key ecological and economic pillars, such as agricultural productivity, zoonotic disease emergence, and the resilience of natural resources (McMahon et al. 2018). In many countries, preventative systems are in place to identify likely impacts of land-use change and mitigate against them, in the form of policy instruments such as Environmental Impact Assessment (Glasson, and Therivel, 2013). Unfortunately, these processes typically fail to identify multivariate consequences of land-use change that occur over time and space (Murphy et al. 2022).Wildlife responses to human-mediated land-use changes, such as construction, forestry, and agriculture, can occur over varying temporal and spatial scales (Allen et al. 2019; Murphy et al. 2022). These changes may drive species into smaller refugia near human populations, increasing densities and exacerbating conflicts, often mistakenly attributed to wildlife abundance rather than the initial land-use change (Pozo et al. 2017; Fowler et al. 2019; Murphy et al. 2023a). When planned, land-use changes like road construction or forest removal offer opportunities to predict wildlife responses and validate hypotheses with empirical data, informing proactive mitigation strategies (Gaughran et al. 2021; Murphy et al. 2022). One important ecological process impacted by landscape disturbance is disease dynamics (Brearley et al. 2013). Landscape changes, such as habitat fragmentation and alteration, can influence wildlife spatial ecology, leading to increased interactions between wildlife, domestic animals, and humans, thereby facilitating the spread of diseases (Daszak et al. 2001). One notable example is bovine tuberculosis (bTB), a zoonotic disease with a global distribution and complex epidemiology due to the presence of multiple wildlife hosts, for instance, deer (Cervus & Odocoileus spp.) in North America (O’Brien et al. 2011), brush-tailed possums (Trichosurus vulpecula) in New Zealand (Tweddle and Livingstone 1994), buffalo (Syncerus caffer ) in South Africa (Davey 2023), and European badgers (Meles meles ) in Ireland and the UK, where they are the primary wildlife hosts implicated in the maintenance and transmission of bTB to local cattle herds (Allen et al. 2018; Chang et al. 2024). Recent studies have shown disturbances adjacent to agricultural landscapes have been linked to increased risk of bTB breakdowns in cattle herds (Barroso et al. 2022).However, our understanding of the mechanisms behind fluctuations in bTB risk remains incomplete due to conflicting evidence from different studies. It is hypothesized that disturbances alter disease risk by modifying wildlife behaviour and social structures, potentially forcing species like badgers to change their movement patterns and increasing inter/intra species interactions (Murphy et al. 2022). For instance, Barroso et al. (2022) observed an increased risk of bTB in cattle herds near roadworks, suggesting that such disturbances may contribute to disease spread. Conversely, Gaughran et al. (2021) found no significant changes in wildlife territoriality due to road construction and suggest that fully mitigated major road upgrades are unlikely to cause a perturbation effect substantial enough to increase TB in local cattle. This discrepancy underscores the complexity of disease dynamics in disturbed landscapes and emphasises the need for more integrated research that combines ecological, epidemiological, and human factors. The example of how bTB continues to proliferate despite surveillance and control of the disease highlights the necessity for deeper investigation into how macroecological human-mediated landscape disturbances may impact disease and the need to develop effective strategies for mitigating disease risk and spread.Previous research in Ireland has explored the link between landscape disturbances, such as forest clearfells, and bTB risk in cattle herds. Byrne et al. (2022) found that the percentage of herds testing positive for bTB increased after clearfelling, potentially due to disturbance of badger populations, a key wildlife host (Mullen et al. 2019). While unable to infer the mechanism, their study highlighted the need to integrate epidemiology with wildlife ecology. Murphy et al. (2022) extended this work by incorporating ecological covariates, showing that the relationship between clearfells and bTB risk was dynamic, depending on the extent and timing of clearfell activities relative to cattle farms and surrounding habitats. Murphy et al. (2022) hypothesised that badgers may initially leave disturbed areas but return as vegetation regenerates, altering disease transmission in time and space.Finally, Khouri et al. (2023) examined the role of wildlife densities, including badgers and deer, in influencing bTB risk after clearfell. Their study found that active badger sett density consistently predicted increased bTB risk, particularly when clearfells occurred within 2 km of farms, 24-36 months prior. Khouri et al. (2023) call for simulation approaches using agent-based models (ABMs) to simulate individual wildlife behaviours in response to habitat disturbances like clearfells. ABMs can reflect adaptive behaviours and environmental dynamics, offering insights into wildlife dispersal and interactions (Murphy et al. 2020). Additionally, GPS tracking data for wildlife hosts could provide valuable insights on how habitat disturbances affect wildlife movement and contacts with cattle, thereby informing more effective bTB management strategies (Conteddu et al. 2024).In this study, we developed a highly realistic agent-based model (ABM) to simulate wildlife responses to clearfell forestry at 100 sites across Ireland, focusing on the potential link between ecological disturbance and bTB risk. Clearfell forestry, where all marketable trees are harvested at the end of a rotation (typically 30–50 years for conifers), potentially displaces wildlife, thus altering their movement patterns and contact rates with cattle. We hypothesised that such disturbances modulate wildlife movement, increasing the risk of new encounters between wildlife hosts and cattle. Our ABM was parametrised using historical data from 202 individually tracked badgers across two study regions in Ireland, which had experienced both Test and Vaccinate or Remove (TVR) operations and a large road improvement scheme. We also integrated high-resolution GIS data, including spatial information on farms, clearfell sites, habitats, forestry, and badger setts.Methods To simulate the behavioural response of European badgers to clearfell forestry in Ireland and assess the relative impact on bovine tuberculosis, we gathered a mixture of tracking data and field survey data (e.g. badger sett locations and activity) for badgers, and remote sensing spatial data for habitats, farms and forestry operations. We used these data to build a realistic agent-based model. The data, subsequent analysis and an overview of the agent-based model is described below however a comprehensive description of the agent-based model is described using the ODD protocol proposed by Grimm et al (2010) in Supplementary Material 1. All data management and statistical analysis was conducted in R version 4.4.1 (R Core Team, 2023) and all simulation modelling was done in Netlogo version 6.4.0 (Wilensky, 1999)Wildlife DataWe used data from GPS-collared badgers in two regions in Ireland and Northern Ireland, respectively. We used data from a badger tracking study conducted in Co. Wicklow conducted by Trinity College Dublin and from another badger tracking study in Co. Down conducted by the Veterinary Epidemiology Unit in the Department of Agriculture, Environment and Rural Affairs (DAERA). The Tellus Light GPS collar (Followit Wildlife, Lindsberg, Sweden) was used to capture data at one-hour fixes with an even spread of data accumulated across both regions (53% Co. Wicklow; 47% Co. Down). Data from both studies were compiled into a single database to be examined and screened for errors (i.e we removed incorrect fixes and incomplete tracks). Further details about each specific tracking study can be found at MacWhite et al. 2013 and Gaughran et al. 2021 for the Co. Wicklow study and O’Hagan et al. 2021 for the Co. Down study, respectively.Our final database consisted of 202 unique badgers tracked for the period 2010-2017 (Co. Wicklow study period 2010-2016; Co. Down study period 2014-2017), covering all times of the year (68% Apr-Oct; 32% Nov-Mar), age classes (88% Adults; 12% Juveniles), sexes (45% Females; 55% Males) and across a representative heterogenous landscape consisting of mixed farmland (grassland, arable, farm yards), mixed forestry, urban areas, among other habitats. Full details of the data analysis of the badger telemetry database are described below.We obtained badger sett survey data from the Department of Agriculture, Food and the Marine (DAFM) Wildlife Unit. The database contained 55,672 locations of surveyed setts across Ireland from the period 1996-2023. These data were filtered to exclude any sett type that was not a main sett (e.g. annex, secondary, outlier) so that only active main setts were included in the final database (see Thornton, 1988). Sett data are collected relative to cattle herd breakdowns, and therefore there are known areas without survey effort (Byrne et al. 2014). Therefore, the raw sett data was compared against the predictions from a sett relative abundance model to generate an accurate sett density for each site as laid out in Supplementary Material 1.Habitats Data We used CORINE 2018 land-use polygons data for both Ireland and Northern Ireland (UK) to classify habitat types using a consistent methodology across the study area. This dataset, available from the European Union’s Copernicus Land Monitoring Service (DOI: 10.2909/71c95a07-e296-44fc-b22b-415f42acfdf0) provides detailed classifications of land cover types at a 100-meter spatial resolution. We downloaded each national dataset and then merged the data from both Ireland and Northern Ireland into an integrated all-island database.Farm Data Spatial and administrative data for all farms in Ireland were also provided by DAFM via the Land Parcel Information System (LPIS). LPIS serves as a comprehensive database, consisting of over 135,000 farms, that registers and maps agricultural land parcels across the country, encompassing information on land use, boundaries, ownership and information on herd numbers & enterprise type. These data were provided by DAFM and are subject to stringent general data protection regulation (GDPR) requirements, whereby the minimum amount of data required for the purpose of the study is used and all personal data (e.g. herd numbers) are anonymised in the analysis.Forestry Data We collected all relevant spatial and administrative forestry data from Coillte, Ireland’s semi-state forest custodian. Coillte manage 57% (440,000 ha) of Ireland’s forestry, which is predominantly conifer high forest in clearfell rotations. Clearfell forestry is a rotational forestry system where all marketable trees are harvested within a forest at the end of a set rotation. Rotations generally between take place when trees are at an ages of 30 and 50 years in conifer forests, once trees are felled, they are replanted and the site typically regenerates over a period of 12-36 months. The species mixtures in our clearfell sites are typically comprised on non-native conifer species such as Sitka spruce Picea sitchensis , Japanese larch Larix kaempferi , lodgepole pine Pinus contorta , and Douglas fir Pseudotsuga menziesii . We build a forestry database that included information on private forestry, public forestry, clearfell sites and forest boundaries. This database included the date of each forest clearfell event, the administrative boundary of the forest and the ID of each forest subplot felled within the property for the period 2015-2017. Bovine Tuberculosis Testing Data The relevant bTB testing and breakdown history data for all farms in Ireland was obtained from the Department of Agriculture, Food and the Marine (DAFM) in collaboration with the UCD Centre for Veterinary Epidemiological Risk Analysis (CVERA). Every cattle herd in Ireland has at least one herd test per year. All cattle present on the farm on the day of the test are tested, with the exception of calves aged under 6 weeks, which were born on the farm. The test used is the single intradermal comparative tuberculin test (SICTT). The SICTT test is the main diagnostic test used in Ireland, though it is imperfect. The sensitivity and specificity of the test is known to vary with estimates of 52.9%–60.8% and 99.2%–99.8% respectively from field trials in Ireland (Clegg et al., 2011). The animals showing a positive reaction to the test are known as reactors, and farms with animals that test positive are known as breakdown herds.Wildlife GPS data analysis We used the amt package in R (Signer et al. 2019) to turn the database of badger GPS locations into tracks for each animal to extract movement data that would be used to parameterise badger movement in the ABM. Each badger had a unique ID in the database to identify the animal and each collar had an ID to track repeated deployments. To ensure that our parameters were extracted accurately, and since badgers do not leave the sett during the day, we transformed each nightly track night into a burst and gave each badger and nightly track a burst ID which would be used for the analysis. For each badger-night, we calculated the time (in minutes) between each step, the distance (in metres) between each step, and then used that data to calculate speed of movement. We calculated the relative turning angle to assess directionality of movement and the net squared displacement to assess the percentage of the population that were long range dispersers (Singh et al. 2016). To assess if any badgers dispersed over long ranges we used net squared displacement for the study period rather than for each night (Borger and Fryxell, 2012). Finally, we created unique databases for each class of animal (male, female, adult, juvenile) in order to assign the correct movement parameters to the appropriate class of badger.For each unique badger we also calculated their home range throughout the study period using the adehabitatHR package in R as a proxy for territory (Calenge and Fortmann-Roe, 2023). Space use for each individual badger was estimated through Kernel density estimation (KDE) models, and the 95% contour was used to indicate each badger’s home range, indicating where the badger is most likely to spend its time (ref here). The area of each polygon, measured in hectares (ha), was computed to quantify the area of the home range for each individual badger. Thest_intersection function from the sf package (Pebesma and Bivand, 2023) was used to intersect the CORINE habitats data with each badger’s home range to assess the presence, diversity and area of habitats within each badger’s home range. Using these data we conducted basic resource selection function (RSF) analysis (Manly et al. 2007). We computed the availability of each habitat type within badger home ranges and calculated the proportion of GPS fixes within each habitat type during overall survey period, and the summer, and winter periods specifically to assess seasonal changes in habitat selection. These proportions were standardised by availability to compute selection ratios, indicating relative habitat preferences of badgers in each survey area. These selection ratios were used in the ABM to inform badger habitat selection.Spatial data analysis (sites generation) Each study site represented a clearfell area in Ireland, identified by selecting 100 clearfell sites from a database of 1,939 clearfell events in the Coillte Harvest data. For each site, the centroid of the clearfell area was calculated and buffered by 5 km, resulting in a total site area of approximately 78.54 km². To incorporate relevant spatial information, we used the st_intersection function to create spatial databases for each site, integrating key data layers such as farm boundaries from the LPIS system, habitat information from CORINE, main badger setts from DAFM wildlife surveys, and clearfell boundaries from Coillte Harvest. These spatial databases were then transformed into shapefiles, which were included in the agent-based model (ABM) for simulating wildlife movement and bTB risk dynamics.Simulation (Agent Based Model) We built a spatially and temporally explicit agent-based model, also referred to as individual-based model, called an ABM hereafter, in the Netlogo environment to assess the behavioural response of badgers to clearfell forestry activities taking place in an agro-forestry mosaic. The purpose of our ABM was to simulate badger movement after a clearfell and calculate the number of alien farm encounters (i.e. entering farms that fall outside of their own territory) from badgers on 100 real, heterogeneous landscapes and to record encounter rates by badgers and by farm (using the herd number of every unique farm) to understand the ecological role of badgers in elevated disease risk for farms located close to clearfells hypothesised in previous studies (see Murphy et al. 2022, Byrne et al. 2022, Khouri et al. 2023).We expected the role of badger density, and habitat selection for more favourable environments (hypothesised by Murphy et al. 2022) to cause higher encounter rates after a clearfell. To consider our model sufficient for its purpose, all parameters were either defined from real world GPS or spatial data and any parameter that was derived from the literature was subjected to sensitivity analysis to ensure that behaviours replicated what expert stakeholders expected to see in the field. The ABM incorporated key badger behaviours such as seasonality of activity, home ranges, resource selection, dispersal, mating and mortality, with the aim of modelling encounter rates and locations that may be representative of disease transmission on a heterogeneous landscape. State variables for the spatial patches and for the badger are given in the ODD protocol.We ran 25 models (optimal number of models was decided by sensitivity analysis – whereby the number of encounters over 1,5,10,25,50,100 models were assessed and 25 models was when the estimate stabilised and was hence chosen) for each site (100 sites x 25 models, 2500 simulations in total) each lasting one calendar year after clearfell (based on top performing models in Murphy et al. 2022). Each model output included site level statistics and farm level statistics, to assess both the macro-ecological variables and the relative impact at a farm level. Model development is described comprehensively using the standardised reporting ODD protocol proposed by Grimm et al (2010) in Supplementary Material 1. See Fig. 1 for an example of the modelling framework and see Fig. 2 for locations of all 100 sites and the size in hectares of the clearfell included. The simulation generated a range of variables that allowed us to assess the impact of clearfell forestry on badger movement, population dynamics, and dispersal patterns. Each simulation run was replicated across multiple sites, with the replicate number recorded to ensure comparisons within sites. Sett and badger densities were recorded for each model run, both of which were calculated per km2at the end of the simulation year. These values provided a snapshot of the spatial distribution of badgers and their setts within each study area. In addition to these final densities, the simulation tracked changes in both badger and sett densities over the study period. This allowed for an assessment of how populations fluctuated from the start to the end of the simulation year, offering insights into population stability or growth in response to forestry disturbances. Two critical dispersal metrics were also captured: the total number of badgers dispersing to new territories and the subset of individuals that engaged in long-distance dispersal, defined as those moving beyond neighbouring territories to join or establish new clans. We also measured the direct disturbance to the badger population, measured by the number of individuals disturbed by clearfell activities. The county in which the site was located was also saved for each site. These variables collectively provided a comprehensive view of how clearfell forestry affects badger populations, ranging from local population density changes to broader landscape-level dispersal patterns within the subsequent modelling exercise, see below.
