Wildlife roadkill and COVID‐19: A biologically significant, but heterogeneous, reduction

The COVID-19 pandemic was a major driver of global change. It has also made possible an unprecedented large-scale field experiment to test the effects human mobility on wildlife ecology and behaviour (Rutz et al., 2020). Sanitary measures tackling spread SARS-CoV-2 curtailed movement world-wide (Bates 2020), reducing disturbance for animals, set rapid responses from were recorded soon after implementation lockdowns. Apart behavioural responses, lockdowns massively reduced vehicular traffic thus risk wildlife-vehicle collisions, one most important anthropogenic sources mortality many vertebrates (Hill 2019; Schwartz Some studies have quantified decrease in overall number collisions during Northern Hemisphere spring (or Southern autumn) 2020. For example, Bíl al. (2021) found that across 10 European countries Israel decreased with large mammals (mainly ungulates) by 19% compared 2015–2019, Shilling arrived at similar conclusions four states United States. Conversely, some other focused single susceptible species smaller spatial scales. Manenti (2020) reported significant reduction roadkill amphibians reptiles two sites northern Italy, Driessen 48% marsupial around Hobart (Australia) Łopucki indicated 50% white-breasted hedgehog Erinaceus roumanicus Chelm Poland. Considering availability species-specific data larger (country-wide) scale, only 58% mountain lions Felis concolor California 2020 lockdown. This emerging research can be summarized into groups effect road mortality: (a) changes total volume roadkill, is pooling together multiple and/or collected different sources, such as police records, hunting statistics or observations volunteers; (b) small-scale studies, focusing temporal All these refer lockdowns, albeit subsequent waves forced second lockdown autumn There gap species, scales, encompassing both periods Each gaps problematic, considering conservation management need robust applicable evidence. while pooled rapidly showed collision rates (Bíl 2021; 2021), estimates are fundamental management. Many terrestrial managed adaptively through quota systems culling, aiming maintain population densities ‘optimal’ values minimize their undesirable ecological socio-economic impacts, ensuring viability populations (Fryxell 2014). Therefore, quotas should periodically modified based animal populations: case substantial either increased prevent fluctuations abundance would jeopardize goals. particularly true wild ungulates, which provide several ecosystem services humans (Apollonio 2017; Csányi 2014), but provoke negative human–wildlife interactions due conflicts activities interests, often involved vehicles (Carpio 2021). Species-specific multispecies assembly because adaptive account like competition (Serrouya 2011) predation (Wikenros Moreover, quantifying scales entire year, 2020, mechanisms source-sink dynamics compensatory mortality, elements well. In this study, we quantification followed Slovenia. Our analysis explored 11-year dataset, whole country, seven weekly standardized scheme study performed throughout Slovenia, central country covering surface area 20,273 km2 harbours complex assemblage including high ungulates carnivores (Linnell 2020; Ražen Skrbinšek Stergar 2009). Due absence large, urbanized areas, rich network protected areas good environmental connectivity, live side-by-side country. results interactions, source medium-sized (Kaczensky 2003; Krofel 2015; Potočnik 2019) (Pokorny, 2006). order improve monitoring management, 2006, Hunters Association Slovenia (HAS) established online information system collecting archiving (Levanič, 2018). increase reliability completeness about traffic-related HAS Directorate Roads Republic developed mobile application registering situ (for hunters only), reporting non-hunters regional emergency call centres. Every carcass centres citizens must relayed ground where occurred; appointed hunter then removes records attributive all road-killed game mammals. Afterwards, trained officers (one per each ground; n = 415) upload sex, estimated age carnivores), date location national database. incentive report lies fact harvest plans ‘elimination quotas’ (i.e. bags, plus individuals dead, regardless cause) managers fined €4,200 €125,000 when not filled. highly motivated register helps them fulfil objectives avoid fines. over-reporting prevented collect, prepare hand over relevant proof dead (e.g. left hemimandibles ungulates). recognized public service, therefore restricted ensured constant detection effort period (including allowed reliable comparison between previous years. However, further under-reporting people (see included 415 non-professional grounds clubs (average size: 4500 ha), excluded 12 special purposes (approx. 10% Slovene surface) institutions (the Forest Service Triglav National Park). We following reasons: they employees whose activity lockdowns; very (25,000–60,000 ha) remote percentage undetected cases could higher; (c) low density roads, frequency rather low. As there small fraction studied 2019: 4.3% roe deer Capreolus capreolus, 2.6% red fox Vulpes vulpes, 4.7% badger Meles meles 0.7% boar Sus scrofa respectively), assume dataset used reliably reflects country-wide roadkill. considered 2010 2020: (n 53,259), 9,889), 5,170), brown hare Lepus europaeus 5,050), stone marten Martes foina 4,267), 1,188) Cervus elaphus 1,088). These selected had highest annual peaks Europe coincided (Canova & Balestrieri, Langbein 2011; Pokorny, 2006; Raymond Valerio capable living environments even urban (Bužan 2020) among killed Hothorn represent valuable exploring association exceptional volumes rates. our already been it did require any ethical approval. aggregated basis, decomposed time series means Bayesian generalized additive models with: linear component, accounting variation years, capturing long-term characterizing populations, seasonal modelled B-spline, cyclical rate crossings Laliberté St-Laurent, Steiner periodic variations volume, weather conditions affecting safety drivers ice, fog) autoregressive deal correlation consecutive weeks. model binomial distribution response. Model selection, knots B-spline structures, sum log-converted Conditional Predictive Ordinal (CPO) values, obtained leave-one-out cross validation (Zuur Ieno, 2019). value simplest structure selected. To appreciate associated traffic, time-series 2010–2019 data. Then periods, 16 March 30 April (spring lockdown) 20 October 31 December (autumnal lockdown), posterior predictive models. represented expected various weeks its uncertainty, given quantify impact summed differences observed values. transparency process, adopted approaches highlight importance uncertainty estimation. First, calculated those within lay outside 95% credibility interval (hereinafter, CI). absolutely anomalous, showing higher lower than plausible predicted model. and: median distribution, closest bound CI distant CI. three scenarios, numbers frequentist confidence Kruschke Liddell, 2018), equally plausible: likely model, forecasts become less plausible, reaching minimum bounds change whereas far likely. Finally, noted anomalies going beyond probably potential Discussion section), year better occurred restrictions, downloaded Google Community Mobility Reports (https://www.google.com/covid19/mobility/). daily movements users who turned Location History setting, allowing record position, default option smartphone devices using apps Maps. places traditionally reached car: retail recreation sites, parks green transit stations workplaces. compare series, relative index representing centred. Statistical analyses carried out statistical software R (R Core Team, 2021) INLA (Gómez-Rubio, A complete overview selection diagnostic checks available Supplementary Information (Appendix S1). retained second-order autocorrelation deer, without badger, first-order species. detect strong patterns residuals, against fitted covariates S1), matched (Figure 1). With exception fox, markedly 2019, component marginal 2). Data pinpointed corresponding heterogeneous 3; Table spring, significantly expected, being falling Wild pattern autumn. randomly scattered distribution. began, systematically expected. Once ended, again plummeted enforced Figure S2). beginning autumnal deer. systematic discrepancy foxes hares marten, mixture more pronounced 156 individuals, if counted CI, 321 (Table On hand, prolonged began lasted until October, badgers exceeded expectations. show road-related below expectations summer. Such summer deflated S2 column ‘Entire 2021’ first extent related affected seasons. advance qualitatively predictions generated much shorter (Abraham Mumma, (Driessen, help understand complexity mammal assemblages assessing magnitude interplay biology consequences conservation. sanitary intensity varying biological characteristics breeding phenology, heterogeneity consequences. data, measure communities misleading. Not will respond similarly and, reflect abundant ones case, certainly dominated masking lower. perspective evaluating interventions caused traffic. Following evidence wildlife, suggested acting solution Rendall consideration activity, meaningfulness effectiveness measures. key point accounted plans, patterns. clear: 156–321 animals. 4,500 5,900 die (outside lockdowns), 11.0%–14.2% registered (Oslis, Thus, correspond 2.6%–7.3% yearly might seem consider bags. clustered few hotspots, mostly eastern, low-altitude part S3). proportion [15%–20% district level (>100,000 ha large) >30% (in average: ha)]. At therefore, dynamics. Even unlikely affect trajectories, subjected mechanisms, unclear what happen regularly, longer periods. While scenario sounds extreme, pandemics surrounded (Baker restrictions needed future. situation experience road/traffic-related local-scale reductions planning culling quotas, impacts optimize outcomes science-based 2017). findings indicate phenology moderated mortality. experienced slight December, (Johann pressures humans, outdoor hunting. mating (Yamaguchi 2006), partially Badgers continued summer, became anomalously When night-time curfew imposed badgers, adults, offspring survival, size pool unfertilized females. autumn, factors overlapped date, published evaluated however, initial short-term overharvesting occurs. Wildlife tailoring response do so, urgently needed. detected (LeClair simple before-and-after designs 2021); rather, decomposition, changes, instead. raise applied questions. If others, trophic networks lockdowns? And, significant, ecosystems affected? Other perturbation experiments, removal invasive alien (Bergstrom 2009), mass events, perturbations take unforeseen directions processes (Fey 2019), time-lag (Piovia-Scott Ungulates positively role 2017) vegetation temperate (Ramirez communities. Understanding whether local webs ecosystems, mediated future research. conclusion, Temporally limited, marked widespread variable decreases vulnerable vehicles. strongly influenced species' characteristics. policies abundance, scope Standardized protocols needed, well sharing initiatives create collaborative datasets, combining structured sampling schemes citizen science allow us map potentially long-lasting ecosystems. authors thank developing implementing enabling access comprehensive They associate editor, Dr Annabel Smith, reviewers comments suggestions improved version manuscript, Charlotte Taft linguistic corrections. funded by: Slovenian Research Agency (project V4–1825 V1–2031, programme P4–0107 P1–0386, Open Access publication fee); Ministry Agriculture, Forestry Food (both above-mentioned projects); (several projects aimed Slovenia); (d) STEPCHANGE Union's Horizon Innovation Program under grant agreement No. 101006386, oriented accelerate collaboration scientists wildlife/biodiversity no conflict interest declare. B.P., J.C. E.B. conceived ideas designed methodology; B.P. provided data; analysed J.C., led writing manuscript. contributed critically drafts gave final approval publication. via Science Framework Digital Repository https://osf.io/ak3vt/ (Pokorny S1 Please note: publisher responsible content functionality supporting supplied authors. Any queries (other missing content) directed author article.