Road Permeability Index as a tool for mitigation planning of road impacts on wildlife in Colombia: a case study using mammals

Mitigation planning for road projects in Colombia has been largely based on actions aimed at reducing wildlife roadkills. Nonetheless, the efficiency of these actions is compromised because of the absence of robust empirical studies supporting their implementation. In this work, I used the Road Permeability Index (RPI) in conjunction with expert knowledge information to estimate the strength of the barrier effect imposed by an under-construction road (Yuma road, Santander department, Colombia) on nine functional groups of medium and large-sized mammals. The influence of 12 landscape variables on the permeability of each functional group was assessed at 30 locations along the road. The RPI was calculated for each functional group


INTRODUCTION
Roads are a significant concern in conservation management as they represent physical barriers to animal movement that cannot be circumvented but rather must be crossed (Beyer et al. 2016).The overall barrier effect imposed by roads varies as a function of its permeability, that is, the degree to which a road reduces the capacity of animals to navigate and move across landscapes (Nathan et al. 2008, Beyer et al. 2016, Assis et al. 2019).Road permeability is not homogeneous through landscapes and varies among species according to intrinsic (e.g., body size, age, behaviour, and navigation ability) and extrinsic factors affecting animal movement capabilities such as landscape configuration, resource availability, vehicular traffic, among others (Nathan et al. 2008, Reding et al. 2013, Gardiner et al. 2018, Assis et al. 2019).Therefore, certain road sections are easier to traverse for some animals, often presenting high rates of wildlife mortality due to vehicle collisions (i.e., roadkills); while others are unpassable, limiting ecological and genetic flows through landscapes (Jaeger et al. 2005, Ramp et al. 2005, Assis et al. 2019).Thus, a critical question when devising mitigation measures is whether they are sufficiently effective to increase road permeability while reducing roadkill rates (e.g., van der Ree et al. 2007).
Understanding how roads influence animal movement is essential for developing more cohesive road planning strategies for wildlife conservation (Beyer et al. 2016, Doherty andDriscoll 2017).For example, in fragmented landscapes of the Middle Magdalena Valley-an intermontane basin in north-central Colombia-mammal roadkill hotspots match areas with low regional structural connectivity, where the current flow is influenced by the presence of highly human-modified habitats (Meza-Joya et al. 2019).This suggests that local wildlife presents altered movement patterns to respond to changes in resource availability and landscape configuration (e.g., Doherty and Driscoll 2017).In this case, patch-dependent species are probably forced to travel through lower quality habitats when searching for resources, while habitat generalists likely used the anthropogenic matrix as habitat or movement path (Meza-Joya et al. 2019).
Furthermore, habitat patches alongside roads may also be perceived by animals as movement corridors or navigational features of the landscape, increasing the risk of direct mortality due to collisions with vehicles (Gardiner et al. 2018).
Palabras clave: Conocimiento de experto, efecto barrera, evaluaciones de impacto ambiental, medidas de mitigación, movimiento de fauna.Consequently, most roads in the country lack mitigation measures (e.g., crossings structures such as ecopassages and warning signs), and when required by the national environmental legislation, they are installed intuitively in the absence of robust empirical studies supporting their location and effectiveness (Payan et al. 2013, Meza-Joya et al. 2019).Unfortunately, the most implemented mitigation strategy on regional roads around the study area (i.e., warning signs) is ineffective (Meza-Joya et al. 2019).As implementing mitigation measures is expensive, studies based on reliable 'before mitigation' empirical data (e.g., movement, behavioural, population, and roadkill counts) are needed to inform the best actions to manage the potential effects of roads on wildlife (Teixeira et al. 2016).
Gathering reliable 'before mitigation' data, however, can be extremely challenging and time-consuming (Grilo et al. 2018, Assis et al. 2019), especially when planning decisions must be taken rapidly despite the shortage of empirical data (Assis et al. 2019).In such cases, alternative approaches integrating the available knowledge are required to assist the decision-making process (Grilo et al. 2012, Assis et al. 2019).In this paper, I used the Road Permeability Index (RPI; Assis et al. 2019) to estimate the strength of the barrier effect imposed by an under-construction road on medium and large-sized mammals in the Magdalena Valley of Colombia.This novel method estimates road permeability for fauna by combining biotic and abiotic environmental data and expert knowledge in situations where 'before mitigation' empirical data are not available (Assis et al. 2019).This work shows that the RPI represent a promissory tool that can be used for both a fast and reliable identification of priority species (or faunal groups) for road mitigation strategies and to inform deci- sion-makers about the best mitigation strategy and locations for implementing management actions.

Study area
This study was conducted in the municipality of Barrancabermeja, located in the Middle Magdalena Valley, Santander department, Colombia (Fig. 1).This area encompasses extensive lowland alluvial plains with swampy ecosystems interspersed with non-flooded areas (Garzón and Gutiérrez 2013).The climate is warm and humid, with a bimodal rainfall pattern, mean annual precipitation of 2917 mm, mean annual temperature of 27.9 °C, and relative humidity of 80 % (IDEAM c2016).The vegetation corresponds to Tropical Moist Forest (Holdridge 1987).Native vegetation has been historically transformed, today remaining less than 15 % of its original coverage (Etter et al. 2006, Garcia-Ulloa et al. 2012, Garzón and Gutiérrez 2013).Roads are ubiquitous landscape features in the study area, with three two-lane paved roads (NR66, NR45, and DR01)with an average pavement width of ten m-connecting the city of Barrancabermeja with other regions of the country (Fig. 1).In addition, a four-lane highway road-Gran Vía Yuma (hereafter Yuma road)-has been under construction since 2009 following the alignment of an existing tertiary unpaved road that was primarily used for local traffic and transport of agricultural products to the city of Barrancabermeja (Supplementary Material, Fig. S1).This 30-kmlong road corridor seeks to reduce cargo and passenger transportation times, boosting the region's competitiveness, productivity, and economic inputs (ANI c2020).

Data collection
Medium and large-sized native mammals were chosen as study system because in the area: (i) there is information about species richness for this assemblage (Meza-Joya et al. 2020), and (ii) these animals are highly prone to vehicle collisions and roadkills (Meza-Joya et al. 2015, 2019).The studied mammal assemblage includes species with different mobility abilities and movement behaviours; therefore, for the analyses I used nine functional groups (Fig. 2) defined by Meza-Joya et al. (2020).These groups were delimited based on species-specific traits (trophic category, life-history characteristics, social structure, body size, and environmental sensitivity; for details see Meza-Joya et al. 2020) by applying Gower distances (Gower 1966) and the Calinski criterion (Calinski and Harabasz 1974).I also selected twelve environmental variables representing landscape features (Table 1) believed to influence animal displacements (Brady et al. 2011, Assis et al. 2019, Silva et Figure 2. Functional diversity dendrogram grouping medium and large-sized mammals in the study area.The nine main functional groups (FG) used for the analyses are shown in decreasing order.These groups are roughly defined as terrestrial mesocarnivores (FG1), terrestrial apex carnivores (FG2), mid-sized terrestrial omnivores/semiarboreal insectivores (FG3), mid-sized semiarboreal omnivores (FG4), medium and large-sized terrestrial omnivores (FG5), mid-sized arboreal frugivores (FG6), large-sized semiacuatic herbivores (F7), large-sized arboreal frugivores/folivores (F8), medium and large-sized terrestrial herbivores/frugivores (FG9).Relative Road Mortality Index (RMI) was taken from Meza-Joya et al.To apply the RPI, I divided the Yuma road into 30 locations separated by 1-km, where the influence of each predictor variable on road permeability was assessed.To capture the potential influence of habitat amount in the surrounding landscape on species navigability (Plante et al. 2019), I evaluated each variable within a 200 m-radius (sampling window; Supplementary Material, Fig. S2).This scale was chosen because it corresponds to the smallest radius at which roadkill aggregations occur on other roads in the study area (Meza-Joya et al. 2019).Larger scales were avoided to prevent overlap between adjacent locations and control for non-independence in permeability data.I performed a field assessment of the potential influence of each landscape variable on road permeability for each functional group at each location.Then, I contacted via email eight expert mastozoologist from different institutions (non-governmental organisations and academic institutions) to validate field observations.Five experts responded to the call and three agreed to participate in the knowledge elicitation exercise (see acknowledgements).The elicitation process was conducted through individual online interviews based on a questionnaire designed to assess road permeability and supporting information, including updated satellite imagery (CNES/Airbus 2020) from Google Earth Pro (https://www.google.com/earth)for each sampling window (Supplementary Material, Fig. S2).Analyses were made using a 'consensus variable influence matrix' resulting from aggregating knowledge statements of experts from the elicitation process.

Data analyses
Experts' opinions about the influence of each variable on road permeability for each functional group at each location (sampling window) were coded as follows: −1 when it was negative, 1 when it was positive, 0 when it was indifferent, and NA when it was absent or not available in that location (Assis et al. 2019).I quantified among-expert variability (i.e., uncertainty) as the number of experts disagreeing about the influence of a given variable on habitat permeability divided by the number of locations where it was assessed.Lower values (near to 0) indicate low uncertainty, while higher values (close to 1) indicate high uncertainty.I calculated the RPI for each functional group and the whole studied mammal assemblage (average value) us-ing the consensus matrix resulting from the expert knowledge elicitation process.Lower RPI values (near to −1) indicate locations with reduced road and landscape permeability, while higher values (close to 1) indicate highly permeable locations where more roadkills are expected (Assis et al. 2019).The relative influence of each variable was calculated for each functional group and the overall mammal assemblage (average magnitude).The resulting RPI values were mapped to detect differences among functional groups and identify frequently crossed locations (i.e., potential roadkill hotspots).Estimates were made using the R code provided by Assis et al. (2019).I checked for Pearson's correlations between functional groups using a scatterplot matrix of RPI values.The relationship between species' roadkill rate (Road Mortality Index, RMI) and abundance (Relative Abundance Index, RAI) was assessed with non-parametric Spearman correlation (raw data) and linear regression (log-transformed data) using data from Meza-Joya et al. (2019, 2020), respectively.All statistical analyses were performed using R environment (R Core Team c2020).

RESULTS
Expert knowledge validation of field data was highly consistent (uncertainty ≤ 0.2), except for complex and hardto-assess variables such as 'housing' and 'exotic predators' (Supplementary Material, Fig. S3).As expert knowledge elicitation was conducted remotely, a higher weight was given to field observation for these variables but in agreement with the expert panel.Overall, forest cover had the strongest positive influence on road permeability, followed by the presence of herbaceous vegetation and silviculture (Fig. 3a).Conversely, road permeability is negatively influenced by exotic predators (dogs), housing, and industrial infrastructure.As expected, the influence of each predictor variable on road permeability differed among functional groups (Supplementary Material, Fig. S4).Regarding the relative importance of each functional group (FG) to overall permeability, our analysis showed that groups including terrestrial, semiarboreal, and habitat and diet generalist species (e.g., FG3-5), presented the highest contribution to overall road permeability, while groups including arboreal and habitat specialist species (e.g., FG6-9) presented the lowest contribution (Fig. 3b).
Mean RPI values considering all functional groups were positive in most sampled sites (24 windows), while the remaining six presented a negative direction, although the magnitudes varied (Fig. 4).Negative values were related to high-density housing and industrial infrastructure located around the periphery of the city of Barrancabermeja.In contrast, high positive values were largely related to forest cover and wetlands.A gradient of possibly auto-correlated RPI values was observed along the southern stretch of the road, where it disrupts a large forested area.The RPI identified six highly permeable locations for animal movement, indicating more frequently crossed locations where more roadkills are expected (Fig. 4).Overall, RPI values were highly variable across functional groups (Supplementary Material, Fig. S5) but in most cases highly correlated among them (Supplementary Material, Fig. S6).Functional groups including arboreal species showed lower Pearson correlation values (r < 0.6) when compared with those including terrestrial species.Correlations were higher than 0.8 between all other functional groups.There was a significant positive relationship between RAI and RMI values (Spearman's rho = 0.76; P = 0.01; linear regression adjusted R 2 = 0.87, P < 0.001).

DISCUSSION
Roads are usually considered homogeneous linear structures in environmental risk assessments, which may cause their effects on landscape connectivity and wildlife to be greatly underestimated; therefore, challenging the effectiveness of environmental impact studies to make qualified decisions aimed at avoiding, mitigating, monitoring, and compensating for any potential adverse impact on biodiversity (Viloria-Villegas et al. 2018).This is exemplified in the environmental impact study for the licencing and construction of the Yuma road, which proposed the construction of wildlife crossing structures to avoid 'increasing vehicular traffic on faunal populations' (see Resolution 0829 of 2016, ANLA c2019) but lacks robust conceptual and methodological approaches to guarantee their effectiveness in reducing wildlife mortality and restoring or enhancing landscape connectivity.
This study highlights the suitability of expert-based approaches such as the RPI for providing early and rapid management guidelines while other empirical data such as roadkill counts and population trends are collected (Assis et al. 2019).One of the most interesting applications of this index is the reliable and fast identification of target species or faunal groups for mitigation strategies.For the medium and large-sized mammal assemblage analysed here, functional groups including mainly terrestrial, semiarboreal, and habitat and diet generalist species (i.e., FG3-5) presented higher contribution values to overall road permeability, indicating that they represent priority targets to be considered when designing mitigation actions (Fig. 3b).These results are not surprising as these functional groups include abundant species with high roadkill rates on the roads of the study area (Meza-Joya et al. 2019, 2020; Fig. 2) such as Didelphis marsupialis Linnaeus, 1758, Tamandua mexicana (Saussure, 1860), Cerdocyon thous (Linnaeus, 1766), and Procyon cancrivorus (Cuvier, 1798).
Table 1.Description of landscape variables used to calculate the Road Permeability Index (RPI) for Yuma road.Although the variable 'exotic predators' is not a landscape feature, it was included because the presence of dogs affects fauna movement patterns (Vanak and Gompper 2010).

Forest cover
Native Tropical Moist Forest, in some cases highly disturbed, small patches (<1 ha) The RPI also provide a consistent approach for identifying management strategies depending on the proposed mitigation goals (e.g., to increase habitat connectivity and road permeability and/or prevent roadkills).Clearly, mitigation should not solely focus on reducing mortality counts, as it underestimates or neglects other critical processes affecting wildlife population dynamics, such as habitat availability and connectivity.Based on RPI results (Fig 4, Supplementary Material, Fig. S4), an integrated multi-scale management strategy appears to be the best choice to diminish the impact caused by the Yuma road on biodiversity.From a local scale, mitigation should be focused on installing crossing structures, overpasses and underpasses-the latter accompanied by sufficiently long guide fencing (Plante et al. 2019)-in locations with high RPI values to minimise roadkills and improve landscape connectivity (Fig. 4).From a regional scale, mitigation should be focused on increasing habitat connectivity by protecting, restoring, and connecting existing forest patches as they were identified as critical for road permeability (Fig. 3a).Furthermore, environmental education campaigns to improve driver attitudes to benefit wildlife conservation are desirable as complementary actions.
Although the results of this work support the utility of the RPI for road mitigation planning and management, some crucial points should be considered before its application.
The RPI requires a rigorous knowledge elicitation process to generate reliable outputs applicable for decision-making; thus, surveying an expert panel for each taxon, group, or species of interest (this index can be applied to single species) is highly recommended (Assis et al. 2019).Expert knowledge uncertainty must be addressed to improve reproducibility and transparency (Drescher and Edwards 2019).Empirical knowledge of species abundances and roadkill rates is also desirable for RPI validation.Although the RPI could be calculated as a continuous variable along the road, it is recommendable to use an adequate scale to avoid or at least diminish spatial correlation while including as many locations as possible to accurately estimate the relative contributions of variables to overall road permeability (Assis et al. 2019).
Some complex variables might be problematic as they are hard to assess objectively, leading experts to infer their effects less consistently.This is the case of the variable 'housing', whose effects vary depending on factors like people's attitudes toward wildlife and human-derived food subsidies to wild animals (Newsome et al. 2015).Similarly, the effects of 'exotic predators' depend on dog densities and how they interact with wildlife being either predators, prey or competitors (Vanak and Gompper 2010).Seasonality may also influence how some variables affect animal movements.For instance, the presence of wetlands positively affects the movement of most species during the dry season but may constitute an impassable barrier when flooded.Access to resources and reproductive patterns may vary seasonally, influencing animal movement and increasing mortality risk on roads of the study area (Meza-Joya et al. 2019).Despite these facts, most variables considered here were easily spatialised and assessed, favouring the utility of this index for road planning strategies for wildlife conservation.Furthermore, integrating expert knowledge with empirical data is essential to refine the inferences about the relative contributions of variables on permeability and gain wider acceptance of study results.
There is an urgent need to strengthen the technical quality of environmental studies and the evaluation system used by national environmental agencies to effectively address the impacts of roads on biodiversity (Viloria-Villegas et al. 2018).Environmental impact studies should include clearly stated questions and objectives, adequate sampling-at appropriate spatial scales-and rigorous analyses that serve as the basis for making qualified decisions (Teixeira et al. 2016).Methodological transparency and robustness are critical for risk assessments to accurately foresee outcomes and allow for risk inferences validation and repeatability.Road projects and concessions should also be required to systematically collect open-access roadkill data for further environmental studies and road ecology research (Schwartz et al. 2020).Although roadkill is the most evident effect of roads on biodiversity, many others (e.g., habitat loss and fragmentation, rapid spread of alien species and diseases) need to be considered by road managers and policy and decision-makers.Indeed, the absence of roadkills might indicate population-level effects (e.g., population depletion and subdivision) rather than low impacts or successful mitigation (van der Ree et al. 2007, Ascensão et al. 2019).Thus, knowledge about the status of wildlife populations (e.g., sizes and vital rates) inhabiting roadscapes or adjacent habitats may result in integral assessments of the impacts of roads on biodiversity (van der Ree et al. 2007).Nevertheless, mitigation plans should be the object of a continuous monitoring process to ensure that mitigation measures are properly implement-ed, operated, and maintained (Riffat andKhan 2006, Toro et al. 2010).

SUPPLEMENTARY MATERIAL
Figures S1-S6 are presented as supplementary material.
The impact of mortality due to vehicle collisions on wildlife has received little attention in Colombia and its effects remain vastly underestimated.Most studies have provided both systematic and incidental counts of roadkilled animals on national roads (e.g., de La Ossa-V and Galván-Guevara 2015, Meza-Joya et al. 2015, 2018, Monroy et al. 2015), whilst just a few have attempted to test spatio-temporal patterns of wildlife-vehicle collisions and estimate roadkill rates (e.g., Payan et al. 2013, Ramos and Meza-Joya 2018, Meza-Joya et al. 2019, Rincón-Aranguri et al. 2019).Adding to the rarity of road ecology research, environmental licensing in Colombia was recently regulated (Law 99 of 1993, Decree 1220 of 2005, Toro et al. 2010).

Figure 1 .
Figure 1.Study area in the Middle Magdalena Valley, Santander, Colombia.
al. 2020), regardless of the reasons for movement or the potential response of species or individuals towards fragmentation or matrix effects.

Figure 3 .
Figure 3. Influence of landscape attributes on road permeability and importance of each functional group to overall permeability.a. Relative influence of landscape variables on overall road permeability determined throughout expert opinion elicitation, varying between −1 and 1. Signs (−) for negative and (+) for positive influence.b.Relative contribution of each functional group to overall Road Permeability Index (RPI).Bars represent mean values across locations, and whiskers represent standard errors.

Figure 4 .
Figure 4. Mean Road Permeability Index (RPI) for each location analysed.Red and blue represent negative and positive RPI values, respectively.Circle size represents the magnitude of RPI.