Predicting wildlife road-crossing probability from roadkill data using occupancy-detection models

Using the dead to infer about the living: Amphibian roadkill spatiotemporal dynamics suggest local populations' reduction

Roads represent one of the main sources of wildlife mortality, population decline, and isolation, especially for low-vagility animal groups. It is still not clearly understood how wildlife populations respond to these negative effects over space and time. Most studies on wildlife road mortality do not consider the spatial and temporal components simultaneously, or the imperfect roadkill detection, both of which could lead to inaccurate assumptions and unreliable mitigation actions. In this study, we applied a multi-season occupancy model to a 14-year amphibian mortality dataset collected along 120 km of roads, combined with freely available landscape and remote sensing metrics, to identify the spatiotemporal patterns of amphibian roadkill in a Mediterranean landscape in Southern Portugal. Our models showed an explicit general decrease in amphibian roadkill. The Iberian painted frog (Discoglossus galganoi) experienced roadkill declines over time of ∼70 %, while the spiny common toad (Bufo spinosus) and the fire salamander (Salamandra salamandra) had a loss of nearly 50 %, and the Southern marbled newt (Triturus pygmaeus) had 40 %. Despite the decreasing trend in roadkill, spatial patterns seem to be rather stable from year to year. Multi-season occupancy models, when combined with relevant landscape and remote sensing predictors, as well as long-term monitoring data, can describe dynamic changes in roadkill over space and time. These patterns are valuable tools for understanding roadkill patterns and drivers in Mediterranean landscapes, enabling the differentiation of road sections with varying roadkill over time. Ultimately, this information may contribute to the development of effective conservation measures.

Can highway tunnel constructıon change the habitat selection of roe deer (Capreolus capreolus Linnaeus, 1758)?

One of the main things wildlife does for survival is movement. Wild animals need movement to meet their needs, such as reproduction, breeding, foraging, and dispersal. Although wildlife species use roads for various purposes, they also use them when moving from one habitat to another. In recent years, especially when it comes to habitat fragmentation brought about by urbanization, wild animals frequently use highways. Highways have a wide range of effects on factors such as biodiversity, wildlife, and ecology. Roads can cause habitat loss, habitat fragmentation, and habitat degradation; alter the composition of vegetation; act as barriers to the flow of genes and movement; increase human access to pristine areas; and even increase the risk of extinction for many threatened species. Species belonging to the family Cervidae also include the species most affected by road networks. Roe deer (Capreolus capreolus Linnaeus, 1758) is the smallest of the 3 Cervid species living in Turkey. Roe deer are often injured or die in road accidents, and they are one of the most important species affected by the adverse effects of roads in Turkey. For this reason, it was investigated whether the road tunnel construction affected the distribution of roe deer in the region. In the study, the general distribution of roe deer in the Ilgaz Mountain, and the factors affecting their possible distribution were determined by ecological niche modeling. Data were taken between before (2012-2015) and after the highway tunnel built (2020-2022) in Ilgaz Mountain, which connects the Western Black Sea and Central Anatolia and is located in the middle of Kastamonu and Çankırı provinces. As a result of the modeling, it was found that before the construction of the tunnel, the most influential factor in the distribution of the deer was road density. After the tunnel construction, roads ceased to be the main factor affecting the distribution of the species. This study showed that roe deer are disturbed by the density of vehicles on the road passing through the middle of their habitat. With the decrease in the number of vehicles, they are more willing to cross the road and tend to use the areas close to the road as they are less disturbed.

The distribution of vertebrate roadkill varies by season, surrounding environment, and animal class

Due to rapid human expansion in the last century, wildlife roadkill is becoming a concerning threat to biodiversity and human safety. The frequency of roadkill events depends on factors related to specific traits of the road—tortuosity or the presence of fences, among others—and the animal ecology—such as activity patterns, reproductive season, or thermoregulation. These, in turn, are related to environmental factors, with seasonal variations. Here, we assessed roadkill mortality of terrestrial vertebrates over the year. To do this, we sampled 10 road sections (of 3 km, by walk) in the south of Spain for a full year, registering the carcasses of run-over vertebrates. Then, we analysed the spatiotemporal patterns of roadkill events for the four vertebrates’ classes and the effects of road traits (presence of fence, tortuosity, distance to water point) and environmental variables (mean temperature and precipitation). Mammals suffered the highest mortality by roadkill (45.72%). The frequency of collisions was independent of tortuosity, presence of fences, and precipitation, while mean temperature significantly increased the probability of collision of mammals, birds, and reptiles. There was a seasonal effect in the number of collisions, which spatial pattern depended on the class of vertebrates. All this leads us to conclude that, to reduce the impact caused by roadkill mortality on wildlife, we need specific measures to be taken timely in each critical place and for each vertebrate group.

Spatially prioritizing mitigation for amphibian roadkills based on fatality estimation and landscape conversion

Roads cause biodiversity loss and the effects of wildlife-vehicle collisions may ripple from individuals and populations to ecosystem functioning. Amphibians are threatened worldwide and, despite being particularly prone to roadkill impacts, they are often neglected in assessments. Here, we develop a sampling and analytical framework for spatially prioritizing mitigation actions for anuran amphibian roadkills based on fatality estimation and landscape conversion. The framework is composed of the six following steps: (1) pre-selection of segments to survey using the wetland coverage in the surroundings and the presence of roadkills of aquatic reptiles as a proxy for wet areas; (2) spatiotemporally replicated counts with a dependent double-observer protocol, that is, each segment is sampled multiple times by two pairs of people on foot; (3) extraction of covariates hypothesized to affect spatial and temporal variation in roadkill rates and persistence; (4) hierarchical open-population N-mixture modelling to estimate population dynamics parameters, which accounts for imperfect detection and spatiotemporal heterogeneity in removal, detection, and roadkill rates, and explicitly estimates carcass entries per time interval. (5) Assessment of land cover transition to infer landscape stability; and (6) prioritization of segments based on higher fatality rates and lower landscape conversion rates. We estimated a mean of 136 (95%CrI = 130–142) anurans roadkill per km per day in the 50 sample sites selected. The initial number of carcasses had a positive relationship with the percentage occupied by wetlands and a negative association with the percentage occupied by urban areas. The number of entrant carcass per interval was higher in the presence of rainfall and had a positive association with the wetlands cover. Carcass persistence probability was higher at night and lower in sites with high traffic volume. Ten segments (~1% of road extension) were prioritized using the median as threshold for fatality estimates and landscape conversion. It is urgent to appropriately evaluate the number of amphibians roadkilled aiming to plan and implement mitigation measures specifically designed for these small animals. Our approach accounts for feasibility (focused on sites with greater relevance), robustness (considering imperfect detection), and steadiness (less prone to loss of effectiveness due to landscape dynamics).

A spatiotemporal risk prediction of wildlife-vehicle collisions using machine learning for dynamic warnings

INTRODUCTION

The technology in the automotive industry is becoming increasingly safer in the age of automated driving, but the number of accidents is still high, especially in wildlife-vehicle collisions (WVCs). To better avoid these accidents, patterns involved in these accidents must be detected.

METHOD

This paper presents a spatiotemporal risk prediction of WVCs, including various road and environmental characteristics. A process of data preparation using GIS automated by Python scripts was developed to enable a spatiotemporal link of diverse features for the subsequent predictive data analysis. Different machine learning (ML) approaches were applied- random forest (RF), feedforward neural networks (FNN), and support vector machine classifier (SVM) - including automated ML to predict the risk of WVCs. Therefore, a dataset of approximately 731,000 accidents reported to the police in Bavaria over a period of 10 years (2010-2019) was used. In addition, non-accidents were randomly generated in Python over time and space for the supervised ML processes. As the actual risk probability for WVCs and non-WVCs is not entirely known, the impact of different training ratios between accidents and non-accidents was tested on the risk prediction quality (RPQ) (25%, 50%, 75%, 90% WVCs) of the double-weighted sensitivity and single-weighted specificity rate.

RESULTS

The test yielded high mean values of RPQ as an indicator for a suitable WVC prediction. Both RF (86.6%) and FNN (86.7%) were identified as suitable choices for WVC risk prediction in terms of RPQ. The SVM yielded a lower prediction quality, even though acceptable results could be achieved within a shorter runtime.

PRACTICAL APPLICATIONS

Spatial transferability was verified since the algorithm was trained on the dataset of Bavaria (excluding Upper Bavaria) and successfully tested in Upper Bavaria. WVC forecasts were also proven through training with datasets from 2010-2017 and in prediction for 2018 and 2019.

Estimating Road Mortality Hotspots While Accounting for Imperfect Detection: A Case Study with Amphibians and Reptiles

Wildlife road mortality tends to aggregate spatially at locations commonly referred to as road mortality hotspots. Predictive models can be used to identify locations appropriate for mitigation measures that reduce road mortality. However, the influence of imperfect detection (e.g., false absences) during road mortality surveys can lead to inaccurate or imprecise spatial patterns of road mortality hotspots and suboptimal implementation of mitigation measures. In this research, we used amphibians and reptiles as a case study to address imperfect detection issues when estimating the probability of road mortality hotspots using occupancy detection modeling. In addition, we determined the survey effort needed to achieve a high probability of detecting large roadkill events. We also assessed whether vehicle travel reductions associated with the COVID-19 pandemic travel restrictions led to reductions in road mortality. We conducted surveys at 48 sites throughout Rhode Island, USA, from 2019–2021. In total, we observed 657 carcasses representing 19 of Rhode Island’s 37 native species. Of the 19 native species, eight species of frogs, four species of salamanders, four species of snakes, and three species of turtles were observed. We documented a reduction in roadkill density and the proportion of dead versus live amphibians and reptiles in pandemic years (2020 and 2021), but we were unable to link reductions in roadkill density to reductions in traffic volume. Our model results indicated that large roadkill events were more likely to occur on roads near wetlands and with low traffic volume and were more likely to be detected as daily precipitation increased. We determined that there was a low probability of detecting large roadkill events, suggesting that imperfect detection influences detection of large roadkill events, and many were likely missed during our surveys. Therefore, we recommend using occupancy modeling to account for the influence of imperfect detection when estimating road mortality hotspots. This approach will more effectively guide the implementation of mitigation measures.

Roadkills as a Method to Monitor Raccoon Dog Populations

The raccoon dog (Nyctereutes procyonoides) is one of the most frequently killed species on Lithuanian roads. As an invasive species, up-to-date knowledge of population size, trends and spatial distribution is critically important both for species assessment and for the planning of control measures. In Lithuania, however, raccoon dog surveys have not been carried out since 1997. We investigated, therefore, whether roadkill counts on predefined routes could be used as a proxy for a survey. Our dataset includes survey numbers for the period 1956-1997, hunting bag sizes for 1965-2020 (including the spatial distribution of the hunting bag in 2018-2020) and roadkill data relating to 1551 individuals between 2002-2020. At the most local scale, that of the hunting areas of hunting clubs, correlations between the numbers of hunted and roadkilled individuals were negative and insignificant or absent. At the country scale, however, we found significant correlation both between the numbers surveyed and hunted in 1965-1997 (r = 0.88), and between those hunted and the number of roadkills in 2002-2020 (r = 0.56-0.69). Therefore, we consider that roadkill counts on predefined and stable routes may be used as a proxy for a survey at the country scale. Practical implementation of the method is proposed.

Sustainable Landscape Planning to Mitigate Wildlife–Vehicle Collisions

Road development, traffic intensification, and collisions with wildlife represent a danger both for road safety and species conservation. For planners, deciding which mitigation methods to apply is often problematic. Through a kernel density estimate, we analyzed 715 crossing locations and wildlife–vehicle collisions (WVCs) involving brown bears, lynx, wolf, red deer, roe deer, and wild boar in the Southeastern Carpathian Mountains. We identified 25 WVC hotspots, of which eight require urgent mitigation of existing infrastructure. Moreover, many of these hotspots are in Natura 2000 sites, along road sections where vegetation is in close proximity, animal movement is the highest, and driver visibility is low. Our study is the first in Romania to recommend practical solutions to remediate WVC hotspots and benefit sustainable landscape management.

Roadkill-Data-Based Identification and Ranking of Mammal Habitats

Wildlife–vehicle collisions, as well as environmental factors that affect collisions and mitigation measures, are usually modelled and analysed in the vicinity of or within roads, while habitat attractiveness to wildlife along with risk to drivers remain mostly underestimated. The main goal of this study was the identification, characterisation, and ranking of mammalian habitats in Lithuania in relation to 2002–2017 roadkill data. We identified habitat patches as areas (varying from 1 to 1488 square kilometres) isolated by neighbouring roads characterised by at least one wildlife–vehicle collision hotspot. We ranked all identified habitats on the basis of land cover, the presence of an ecological corridor, a mammalian pathway, and roadkill hotspot data. A ranking scenario describing both habitat attractiveness to wildlife and the risk to drivers was defined and applied. Ranks for each habitat were calculated using multiple criteria spatial decision support techniques. Multiple regression analyses were used to identify the relationship between habitat ranks, species richness, and land cover classes. Strong relationships were identified and are discussed between the habitat patch ranks in five (out of 28) land cover classes and in eight (out of 28) species (97% of all mammal road kills). We conclude that, along with conventional roadkill hotspot identification, roadkill-based habitat identification and characterisation as well as species richness analysis should be used in road safety infrastructure planning.

Impact of Road Fencing on Ungulate–Vehicle Collisions and Hotspot Patterns

The number of road traffic accidents decreased in Lithuania from 2002 to 2017, while the ungulate–vehicle collision (UVC) number increased and accounted for approximately 69% of all wildlife–vehicle collisions (WVC) in the country. Understanding the relationship between UVCs, traffic intensity, and implemented mitigation measures is important for the assessment of UVC mitigation measure efficiency. We assessed the effect of annual average daily traffic (AADT) and wildlife fencing on UVCs using regression analysis of changes in annual UVCs and UVC hotspots on different categories of roads. At the highest rates, annual UVC numbers and UVC hotspots increased on lower category (national and regional) roads, forming a denser network. Lower rates of UVC increase occurred on higher category (main) roads, forming sparser road networks and characterized by the highest AADT. Before 2011, both UVC occurrence and fenced road sections were most common on higher-category roads. However, as of 2011, the majority of UVCs occurred on lower-category roads where AADT and fencing had no impact on UVCs. We conclude that wildlife fencing on roads characterized by higher speed and traffic intensity may decrease UVC numbers and at the same time shifting UVC occurrence towards roads characterized by lower speed and traffic intensity. Wildlife fencing re-allocates wildlife movement pathways toward roads with insufficient or no mitigation measures.

Developing a correction factor to apply to animal–vehicle collision data for improved road mitigation measures

Abstract ContextRoad mitigation to reduce animal–vehicle collisions (AVCs) is usually based on analysis of road survey animal carcass data. This is used to identify road sections with high AVC clusters. Large mammals that are struck and die away from a road are not recorded nor considered in these analyses, reducing our understanding of the number of AVCs and the cost–benefit of road mitigation measures. AimsOur aim was to develop a method to calculate a correction factor for large mammal carcass data reported through road survey. This will improve our understanding of the magnitude and cost of AVCs. MethodCitizen scientists reported animal carcasses on walking surveys along transects parallel to the highway and reported observations using a smartphone application at three sites over a 5-year period. These data were compared with traditional road survey data. Key resultWe found that many large mammals involved in AVCs die away from the road and are, therefore, not reported in traditional road surveys. A correction factor of 2.8 for our region can be applied to road survey data to account for injury bias error in road survey carcass data. ConclusionsFor large mammals, AVCs based on road survey carcass data are underestimates. To improve information about AVCs where little is known, we recommend conducting similar research to identify a correction factor to conventionally collected road survey carcass data. ImplicationsIdentifying road mitigation sites by transportation agencies tends to focus on road sections with above-threshold AVC numbers and where cost–benefit analyses deem mitigation necessary. A correction factor improves AVC estimate accuracy, improving the identification of sites appropriate for mitigation, and, ultimately, benefitting people and wildlife by reducing risks of AVCs.

Reducing whale-ship collisions by better estimating damages to ships

Collisions between ships and whales raise environmental, safety, and economic concerns. The management of whale-ship collisions, however, lacks a holistic approach, unlike the management of other types of wildlife-vehicle collisions, which have been more standardized for several years now. In particular, safety and economic factors are routinely omitted in the assessment of proposed mitigation solutions to ship strikes, possibly leading to under-compliance and a lack of acceptance from the stakeholders. In this study, we estimate the probability of ship damage due to a whale-ship collision. While the probability of damage is low, the costs could be important, suggesting that property damages are significant enough to be taken into consideration when assessing solutions. Lessons learned from other types of wildlife-vehicle collisions suggest that the whale-ship collision should be managed as wildlife-aircraft collisions. For several years, the International Civil Aviation Organization (ICAO) manages collisions between aircrafts and wildlife at the international level. We advocate that its United Nations counterpart, namely the International Maritime Organization (IMO), get more involved in the whale-ship collision management. Further research is needed to more precisely quantify the costs incurred to ships from damages caused by whale-ship collisions.

Predicting hedgehog mortality risks on British roads using habitat suitability modelling

Road vehicle collisions are likely to be an important contributory factor in the decline of the European hedgehog (Erinaceus europaeus) in Britain. Here, a collaborative roadkill dataset collected from multiple projects across Britain was used to assess when, where and why hedgehog roadkill are more likely to occur. Seasonal trends were assessed using a Generalized Additive Model. There were few casualties in winter-the hibernation season for hedgehogs-with a gradual increase from February that reached a peak in July before declining thereafter. A sequential multi-level Habitat Suitability Modelling (HSM) framework was then used to identify areas showing a high probability of hedgehog roadkill occurrence throughout the entire British road network (∼400,000 km) based on multi-scale environmental determinants. The HSM predicted that grassland and urban habitat coverage were important in predicting the probability of roadkill at a national scale. Probabilities peaked at approximately 50% urban cover at a one km scale and increased linearly with grassland cover (improved and rough grassland). Areas predicted to experience high probabilities of hedgehog roadkill occurrence were therefore in urban and suburban environments, that is, where a mix of urban and grassland habitats occur. These areas covered 9% of the total British road network. In combination with information on the frequency with which particular locations have hedgehog road casualties, the framework can help to identify priority areas for mitigation measures.

Predicting deer–vehicle collision risk across Victoria, Australia

The risk of deer–vehicle collisions (DVCs) is increasing in south-east Australia as populations of introduced deer expand rapidly. There are no investigations of the spatial and temporal patterns of DVC or predictions of where such collisions are most likely to occur. Here, we use an analytical framework to model deer distribution and vehicle movements in order to predict DVC risk across the State of Victoria. We modelled the occurrence of deer using existing occurrence records and geographic climatic variables. We estimated patterns of vehicular movements from records of average annual daily traffic and speeds. Given the low number of DVCs reported in Victoria, we used a generalised linear regression model fitted to DVCs in California, USA. The fitted model coefficients suggested high collision risk on road segments with high predicted deer occurrence, moderate traffic volume and high traffic speed. We used the California deer model to predict collision risk on Victorian roads and validated the predictions with two independent datasets of DVC records from Victoria. The California deer model performed well when comparing predictions of collision risk to the independent DVC datasets and generated plausible DVC risk predictions across the State of Victoria.

Where will it cross next? Optimal management of road collision risk for otters in Italy

Collisions with vehicles represent the main conflict between infrastructures and wildlife, causing damages to both humans and animals. As to the latter, road mortality is a growing phenomenon and the largest single cause of death for many vertebrates. When focusing on endangered species, the Eurasian otter (Lutra lutra) is among the most vulnerable to road-kills, which represent the predominant cause of deaths recorded in Europe. We propose a large scale spatially-explicit assessment of road-kill risk for the Eurasian otter in Italy as a tool to identify road stretches at high collision risk, thus optimizing the location of mitigation measures. The modelling approach was produced for South Central Italy, hosting the only remnant viable population of otters in Italy. We used a maximum entropy approach including 56 road collision events recorded between 2004 and 2016 through a citizen science initiative, along with seven environmental predictors measured on 1 km grid cells. Four predictors were selected to describe roads characteristics, i.e. density of highways, and of state, regional and local roads. The remaining three variables referred to the quality of otter habitat in the surrounding of the collision sites, i.e. elevation, density of freshwater bodies, and a measure of landscape heterogeneity calculated on land-cover categories. The model achieved a good predictive accuracy (AUC > 0.8; Boyce index > 0.8). The collision probability was mostly affected by elevation, density of state roads, and density of freshwater bodies. Specifically, collision risk was higher in areas at low elevation and medium density of state roads located near rivers and wetlands. In addition, model predictions evidenced that implementing mitigation measures along 10% of road network in the study area could have potentially hampered ca. 50% of otter casualties recorded during the study period.

Predicting spatiotemporal patterns of road mortality for medium-large mammals

We modelled the spatiotemporal patterns of road mortality for seven medium-large mammals, using a roadkill dataset from Mato Grosso do Sul, Brazil (800 km of roads surveyed every two weeks, for two years). We related roadkill presence-absence along the road sections (1000 m) and across the survey dates with a collection of environmental variables, including land cover, forest cover, distance to rivers, temperature, precipitation and vegetation productivity. We further included two variables aiming to reflect the intrinsic spatial and temporal roadkill risk. Environmental variables were obtained through remote sensing and weather stations, allowing the estimate of the roadkill risk for the entire surveyed roads and survey periods. Overall, the models could explain a small fraction of the spatiotemporal patterns of roadkills (<0.23), probably due to species being habitat generalists, but still had reasonable discrimination power (AUC averaging 0.70 ± 0.07). The intrinsic spatial and temporal roadkill risk were the most important variables, followed by land cover, climate and NDVI. We show that identifying spatiotemporal roadkill patterns may provide valuable information to define specific management actions focused on road sections and time periods, in complement to permanent road mitigation measures. Our approach thus offers a new insight into the understanding of road effects and how to plan and strategize monitoring and mitigation.

Accounting for Connectivity Uncertainties in Predicting Roadkills: a Comparative Approach between Path Selection Functions and Habitat Suitability Models

Functional connectivity modeling is increasingly used to predict the best spatial location for over- or underpasses, to mitigate road barrier effects and wildlife roadkills. This tool requires estimation of resistance surfaces, ideally modeled with movement data, which are costly to obtain. An alternative is to use occurrence data within species distribution models to infer movement resistance, although this remains a controversial issue. This study aimed both to compare the performance of resistance surfaces derived from path versus occurrence data in identifying road-crossing locations of a forest carnivore and assess the influence of movement type (daily vs. dispersal) on this performance. Resistance surfaces were built for genet (Genetta genetta) in southern Portugal using path selection functions with telemetry data, and species distribution models with occurrence data. An independent roadkill dataset was used to evaluate the performance of each connectivity model in predicting roadkill locations. The results show that resistance surfaces derived from occurrence data are as suitable in predicting roadkills as path data for daily movements. When dispersal was simulated, the performance of both resistance surfaces was equally good at predicting roadkills. Moreover, contrary to our expectations, we found no significant differences in locations of roadkill predictions between models based on daily movements and models based on dispersal. Our results suggest that species distribution models are a cost-effective tool to build functional connectivity models for road mitigation plans when movement data are not available.