Impacts of existing and planned roads on terrestrial mammal habitat in New Guinea

New Guinea is one of the last regions in the world with vast pristine areas and is home to many endemic species. However, extensive road development plans threaten the island's biodiversity. We quantified habitat fragmentation due to existing and planned roads for 139 terrestrial mammal species in New Guinea. For each species, we calculated the equivalent connected area (ECA) of habitat, a metric that takes into account the area and connectivity of habitat patches in 3 situations: no roads (baseline situation), existing roads (current), and existing and planned roads combined (future). We assessed the effect of roads as the proportion of the ECA remaining in the current and future situations relative to the baseline. To examine whether there were patterns in these relative ECA values, we fitted beta‐regression models relating these values to 4 species characteristics: taxonomic order, body mass, diet, and International Union for the Conservation of Nature Red List status. On average across species, current ECA was 89% (SD 12) of baseline ECA. Shawmayer's coccymys (Coccymys shawmayeri) had the lowest amount of current ECA relative to the baseline (53%). In the future situation, the average remaining ECA was 71% (SD 20) of baseline ECA. Future remaining ECA was below 50% of the baseline for 28 species. The montane soft‐furred paramelomys (Paramelomys mollis) had the lowest future ECA relative to the baseline (36%). In general, currently nonthreatened carnivorous species with a large body mass had the greatest reductions of ECA in the future situation. In conclusion, future road development plans imply extensive additional habitat fragmentation for a large number of terrestrial mammal species in New Guinea. It is therefore important to limit the impact of planned roads, for example, by reconsidering the location of planned roads that intersect habitat of the most threatened species, or by the implementation of mitigation measures such as underpasses.


INTRODUCTION
Habitat destruction threatens many species worldwide (IPBES, 2019;Jaureguiberry et al., 2022;Maxwell et al., 2016).It has been estimated that 75% of Earth's land surface is altered by humans (IPBES, 2019), which has led to an average loss of 18% of natural habitat for mammals, birds, and amphibians (Beyer & Manica, 2020).New Guinea, the second-largest island in the world, is one of the few remaining largely pristine regions and is home to a large number of endemic species (Shearman & Bryan, 2015).The island hosts one of the world's largest remaining tropical forests and large peatland areas (Page et al., 2011;Shearman & Bryan, 2015).According to the global map of International Union for the Conservation of Nature (IUCN) habitat types from 2015 (Jung et al., 2020), only 4.1% of New Guinea has been converted to anthropogenic land cover, such as cropland, pasture, and urban areas, which is much less than in, for example, Borneo (17%) and Sumatra (34.6%).Nevertheless, New Guinea is expected to experience increased human impacts in the near future (Alamgir et al., 2019;Gaveau et al., 2021;Sloan et al., 2019).
One of the main threats to New Guinea's biodiversity is the large planned expansion of its road network (Alamgir et al., 2019;Sloan et al., 2019).Roads affect species directly and indirectly.Direct effects of roads include mortality due to collisions with vehicles (Grilo et al., 2020(Grilo et al., , 2021)), disturbance of habi-tat close to roads (Benitez-Lopez et al., 2010;de Jonge et al., 2022), and movement impediment (Chen & Koprowski, 2019).Movement impediment may affect migration patterns and gene flows within and among populations (Holderegger & Di Giulio, 2010), thus, increasing habitat fragmentation effects.Indirect effects of roads on wildlife include increased rates of habitat conversion and hunting facilitated by the increased accessibility provided by roads (Clements et al., 2014;Laurance et al., 2009;Pinto et al., 2020).Given these potentially pervasive effects, it is important to quantify and understand the impacts of roads on biodiversity (de Jonge et al., 2022).
Despite the imminent threat of the expansion of the road network to New Guinea's unique biodiversity, however, knowledge of the magnitude of this threat is limited.Alamgir et al. (2019) showed that future road developments in Papua New Guinea (eastern part of New Guinea) will decrease the connectivity of natural areas in several protected areas and areas of high priority for biodiversity conservation and that these developments might lead to large additional losses of forests and peatlands.However, fragmentation effects of roads in New Guinea have not been quantified at the species level.We sought to fill this gap by quantifying species-specific habitat fragmentation due to existing and planned roads for terrestrial mammals in New Guinea.Such a species-specific approach is particularly useful for identifying species subject to the largest fragmentation threats.

General approach
We quantified species-specific habitat fragmentation due to existing and planned roads by comparing the equivalent connected area (ECA) of the habitat for each species in 3 different situations: baseline (no roads), current (existing roads), and future (existing and planned roads).
We assessed the effect of roads as the proportion of ECA in current and future situations relative to the baseline.The ECA is a network-based index that takes into account the area of habitat patches and their connectivity.It indicates the hypothetical size of a single patch that would provide the same probability of connectivity as the actual configuration of habitat patches in the landscape (Saura et al., 2011).The probability of connectivity is defined as the probability that two animals randomly placed in the landscape will occur in habitat areas that are reachable from each other (Saura & Pascual-Hortal, 2007).
We prefer the ECA to alternative fragmentation metrics for 3 reasons.First, the ECA is a composite metric of 3 components that are relevant in the context of habitat fragmentation: number of patches, connectivity of patches, and patch area.A decrease in any of these components leads to a decrease in the ECA (Saura & Pascual-Hortal, 2007).Second, connectivity is quantified based on a continuous measure of the probability of dispersal between patches; hence, ECA does not need thresholds to determine whether two separate patches are connected (as in, e.g., Santini et al., 2019) or assumptions on whether or not individuals can cross roads (as in, e.g., Ceia-Hasse et al., 2017).Third, the ECA is an area metric, which makes it intuitive and allows it to be compared with a benchmark value.

Species selection and trait data
We used geographic range maps from the IUCN (2021) to select terrestrial mammal species with at least 90% of their geographic range in New Guinea (n = 171).We excluded bats (n = 27) because we used allometric models in our analyses to estimate dispersal probabilities, which would not be valid for bats (Santini et al., 2013).We excluded species with no habitat in their geographic range according to the combination of land cover and habitat preference data (n = 5) (details in "").This resulted in 139 mammal species in our analyses.
We obtained species' body mass and diet (i.e., herbivore, carnivore, or omnivore) from the COMBINE data (Soria et al., 2021), which we used to derive median dispersal distances.For Pseudohydromys berniceae, P. patriciae, P. sandrae, we obtained body mass data from Phylacine (Faurby et al., 2018(Faurby et al., , 2020) ) because there was no body mass data in COMBINE.For Shawmayer's coccymys (Coccymys shawmayeri), we estimated the body mass as the mean from congeneric species and assumed its diet is the same as the most common diet in congeneric species because we could not find body mass and diet data for this species (Supporting Information Appendix S1).

Road data
We obtained data on the current road network from the Global Roads Inventory Project (GRIP) database (Meijer et al., 2018).We selected roads classified as open and available year round.We complemented the GRIP data by digitizing existing and planned roads from the maps published by Alamgir et al. (2019) and Sloan et al. (2019).We digitized their maps by overlaying them with the World Imagery (WGS84) layer in ArgGIS Pro 2.9.We only digitized roads that were not included in the GRIP data.Some of the digitized planned roads retrieved from Alamgir et al. (2019) and Sloan et al. (2019) were visible on the World Imagery (WGS84) layer in ArgGIS Pro 2.9, indicating these roads have already been constructed.When this was the case, we classified these digitized roads as existing roads.Otherwise, they were classified as planned roads (Figure 1).
The GRIP database distinguishes among 3 road types in New Guinea: primary, secondary, and tertiary roads.Further, GRIP characterizes the surface of each road as paved or unpaved.For the digitized existing roads, we determined the road surface based on visual inspection of the World Imagery (WGS84) layer.For the digitized planned roads, we assumed a road was paved if it was connected to at least 1 paved road.Otherwise, we assumed the road was unpaved.Similarly, we assigned a road type (primary, secondary, tertiary) to each digitized road (existing and planned) based on the most developed type of road it was connected to (primary being the most developed).
For a subsample of 200 roads for each road type, including paved and unpaved roads, we estimated road width by overlaying the road data with the World Imagery (WGS84) layer in ArgGIS Pro 2.9 and applying the measure distance function (Supporting Information Appendix S2).We then calculated the mean road width for each road type, based on this subsample of roads, and used this mean road width in our analyses (primary roads, 8.1 m; secondary roads, 5.7 m; tertiary roads, 6.8 m) (details in Supporting Information Appendix S2).

ECA calculation
We calculated ECA based on the following equation (Saura et al., 2011): where a i and a j are the areas of habitat patches i and j, p ij * is the maximum probability of dispersal between patches i and j among all possible paths between these patches (i.e., direct path between patches i and j or via any other patches), and n is the total number of patches.When two patches are completely isolated from each other, then p ij * = 0.When i = j, then p ij * = 1 because a patch can by definition be reached from itself.The ECA value decreases when either 1 or more patches decrease in size or when isolation of a patch increases.The ECA value lies between the size of the largest patch and the total area of habitat (Saura et al., 2011).

Patch identification
To identify habitat patches (a in Eq. 1), we first identified locations of habitat within the geographical range of each species based on the species' elevation limits and habitat preferences as listed in species' assessments (Crooks et al., 2017;Gallego-Zamorano et al., 2020;IUCN, 2021;Santini et al., 2019).For the habitat filtering, we used a global map of IUCN habitat types from 2015 at an approximately 100-m resolution (Jung et al., 2020).For the elevation filtering, we used the MERIT DEM elevation raster (Yamazaki et al., 2017) at an approximately 90m resolution, which we resampled to the same resolution as the IUCN habitat map by applying spatial averaging.To reduce the computation time of calculating the maximum probability of dispersal between habitat patches (p in Eq. 1), we aggregated the resulting refined species range rasters to a resolution of approximately 300 m, based on the assumption that a grid cell contains habitat when any of the underlying grid cells contains habitat within the elevation limits of a species.Finally, we defined patches as clusters of neighboring grid cells with habitat and converted these to polygons.If roads intersected these polygons, the resulting smaller polygons constituted separate patches.For each polygon, we calculated its area (a in Eq. 1) based on the original approximately 100-m resolution habitat type map by summing the areas of the parts of the grid cells with habitat that were intersected by the polygons (details in Supporting Information Appendix S2), such that the effect of raster aggregation on the ECA value was limited.

Probability of dispersal
To estimate the probability of dispersal between 2 habitat patches, we calculated the shortest distance between these 2 patches.For each species, we first estimated the median dispersal distance based on allometric relationships between median dispersal distance and body mass, fitted separately for carnivores and noncarnivores (Supporting Information Appendix S2).When there were no roads, we calculated the probability of direct dispersal between 2 patches with the following equation (Saura et al., 2018): where p ij is the probability of direct dispersal between patches i and j, d ij is the shortest distance between patches i and j, and d med is the median dispersal distance.This equation leads to a 50% probability of dispersal when the shortest distance is equal to the median dispersal distance.
When the shortest path between 2 patches was intersected by a road, we calculated the probability of dispersal by multiplying the probability of dispersal (without considering roads) by the probability of crossing that road (p rc ): To calculate p rc , we used the model from Chen and Koprowski (2019), which is based on 90 estimated road crossing probabilities of 37 mammal species and has an explained variance (R 2 ) of 53%: where BM is body mass of a species (in grams), RW is road width (in meters), and RS is a binary variable indicating whether the road is paved (1) or unpaved (0).This model was parameterized based on species with a body mass range of 5.5 g-411 kg, thus, it encompassed the range of body mass values of the species in our dataset (10.3 g-10 kg).When applying this model, we used the estimated mean road width for each road type and the road-specific information on road surface.The probability of dispersal between 2 patches via intermediate patches was calculated by multiplying the probabilities of dispersal of all dispersal steps.The dispersal path with the largest probability of dispersal among all possible dispersal paths between patches i and j, (i.e., p ij *) was then used in the calculation of the ECA (Eq.1).We limited the computation time of our calculations by calculating the shortest distance between patches only for pairs of patches that were within 10 times the median dispersal distance.We believe this is reasonable because the probability of dispersal was <1% for patches that were farther away than 10 times the median dispersal distance.To determine which patches were within 10 median dispersal distances, we added an area around each patch of 5 times the median dispersal distance and determined which of these areas overlap.

Comparison between ECA and minimum area requirements
To put the results in context, we compared the species-specific ECA values with an estimate of the minimum area requirement (MAR) of the respective species.Because the MAR defines the amount of space (habitat) that is required for the long-term persistence of a population (Pe'er et al., 2014), the ratio of ECA to MAR provides a proxy of a species' extinction risk: lower values (i.e., ECA closer to or below MAR) indicate a higher risk.We defined the MAR as the area required to support a population with an extinction probability of <1% within 40 generations.To estimate MAR values, we divided the minimum viable population size (MVP) of each species (i.e., size of a population having a probability of extinction <1% within 40 generations [Reed et al., 2003]) by its population density.We used a generic MVP of 6,020 individuals, which is the median MVP value among 52 empirically derived estimates for terrestrial mammals (Reed et al., 2003).We used this MVP value because it has been argued that this value is the most conservative MVP estimate and the most relevant value when the aim is to investigate securing longterm survival (Williams et al., 2022).We obtained population density data from Santini et al. (2022).For species with missing density data (n = 25), we used the mean values from congeneric or confamilial species.

Relationships between habitat fragmentation and species characteristics
We examined whether there were patterns in the remaining ECA (i.e., proportion of ECA in the current or future situation relative to the baseline situation) related to the species' taxonomy or other species characteristics.To that end, we fitted beta-regression models relating the remaining ECA to 4 species characteristics: taxonomic order, body mass (log 10 transformed), diet (herbivore, omnivore, carnivore), and IUCN Red List status (LC, least concern; NT, near threatened; VU, vulnerable; EN, endangered; CR, critically endangered; DD, data deficient) (IUCN, 2021).Because the response variable in beta regression models was bounded between zero and one, we transformed the remaining ECA with a Smithson-Verkuilen transformation to avoid zeros and ones (Smithson & Verkuilen, 2006): where N is the sample size (i.e., number of species [139]).We fitted models with all possible additive combinations of the 4 explanatory variables and then performed model selection based on Akaike's information criterion corrected for small sample size (AIC c ) to identify the most parsimonious model (Burnham & Anderson, 2002).We ran this model selection for both remaining ECA in the current situation and remaining ECA in the future situation.In a similar way, we identified which variables best explained the variation in ECA/MAR between species by fitting linear models between log 10 -transformed ECA/MAR and all possible additive combinations of the same 4 explanatory variables.
For the most parsimonious models, we calculated Pagel's λ based on the model residuals and a sample of 100 phylogenetic trees from the credible set of phylogenetic trees from Upham et al. (2019).This revealed there was no phylogenetic signal in the residuals of any of the most parsimonious models (Supporting Information Appendix S3).In addition, the variance inflation factors of all retained explanatory variables were <2, indicating no collinearity problems (Zuur et al., 2010) (Supporting Information Appendix S3).
All data used in the analyses above, including road data, species-specific trait data, and range data, are available on DANS (DOI: https://doi.org/10.17026/dans-zzj-gb2u).

RESULTS
The current road network in New Guinea yielded ECA values that were on average 89% of the baseline ECA values (i.e., no roads) (Figure 2, Supporting Information Appendices S1, S3).The interspecies variability in the remaining ECA was relatively large (SD 12), and the lowest value was found for Shawmayer's coccymys (53%).The planned extensions of the road network decreased the remaining ECA to 71% (SD 20) of the baseline values (Figure 2, Supporting Information Appendices S1, S3).For 28 species, the future remaining ECA dropped below 50% of the baseline values, and the lowest proportions of remaining ECA values were found for the montane soft-furred paramelomys (Paramelomys mollis) (36%), black-tailed melomys (Melomys rufescens) (36%), and long-nosed dasyure (Murexia naso) (37%) (Figure 3).Our results suggested that the interspecies variability in the remaining ECA in the current situation was related to the species' IUCN Red List category: species with a lower threat status had a lower remaining ECA (Figure 4a, Supporting Information Appendix S3).The remaining ECA in the future situation had a similar relationship with IUCN Red List category and lower values for large species and for carnivores compared with omnivores and herbivores (Figure 4b, Supporting Information Appendix S3).
For the majority of the species, the ECA values were more than 2 orders of magnitude larger than the MAR in the current and future situation (Figure 5, Supporting Information Appendices S1, S3).Seventeen species had ECA values less than the MAR estimate in the baseline situation.This number did not change when we considered existing and planned roads, although the ratio of ECA to MAR decreased with the addition of roads (Figure 5).For example, for the Woolley's three-striped dasyure (Myoictis leucura), ECA/MAR decreased from 0.53 to 0.49.For some species, such as Wallace's threestriped dasyure (Myoictis wallacei) and the western long-beaked echidna (Zaglossus bruijnii), relatively large decreases in ECA due to future road construction (remaining ECA < 80%) coincided with ECA/MAR decreasing to less than 10.In all 3 situations (baseline, current, and future), the ECA/MAR was related to the species' IUCN Red List category and body mass, such that ECA/MAR was lower for more threatened species and larger species (Figure 6).

Impacts of fragmentation by roads
Our study complements a previous study reporting that key habitats of endangered species in New Guinea will be intersected by planned roads (Alamgir et al., 2019).In that study, however, fragmentation impacts per species were not quantified.We found that the current road network in New Guinea has resulted in considerable habitat fragmentation for some species and could drop ECA values to as low as 53% of the baseline ECA values (situation without roads).Fragmentation will be amplified by the construction of new roads in the near future, resulting in 28 species for which the remaining ECA will drop below 50% of the baseline values.This result complements the findings of earlier studies that show severe threats of imminent road construction to mammals in general (e.g., Carter et al., 2020;Quintana et al., 2022) and New Guinea's biodiversity in particular (Alamgir et al., 2019;Rochmyaningsih, 2021).
The large fragmentation effects of roads we found for some species are cause for concern because many of the mammal species in New Guinea are endemic and threatened by other stressors as well (Shearman & Bryan, 2015;White et al., 2021), which we did not take into account.For example, the critically endangered western long-beaked echidna and the endangered giant bandicoot (Peroryctes broadbenti) are also highly susceptible to hunting (IUCN, 2021).Road construction can also reinforce the impacts of hunting and habitat loss, notably by giving hunters and loggers access to previously undisturbed areas (Alamgir et al., 2019;Clements et al., 2014;Laurance et al., 2009;Pinto et al., 2020).Indeed, previous studies have demonstrated increased hunting and deforestation after road construction in New Guinea (Gaveau et al., 2021;Pattiselanno & Krockenberger, 2021).Plans to upgrade a large number of existing roads in New Guinea (Government of Papua New Guinea, 2018), which we did not consider in the future situation, could further increase habitat fragmentation.Moreover, we only quantified the effect of roads on habitat connectivity, whereas roads can have several other detrimental effects on species, such as mortality due to collisions with vehicles and disturbance of the habitat close to the roads (Benitez-Lopez et al., 2010;de Jonge et al., 2022;Grilo et al., 2021;Laurance et al., 2009).For example, a global study on roadkill shows that roadkill rates of some mammals are higher than 10 individuals per kilometer of road per year, posing a significant threat to the persistence of species (Grilo et al., 2021).The increased habitat fragmentation due to future road development, in combination with other direct and indirect effects of roads, could thus bring mammal species in New Guinea closer to extinction.
In general, we found larger impacts of roads on the habitat of currently nonthreatened species (i.e., LC or NT) than on the habitat of threatened species (i.e., VU, EN, or CR).This reflects the typically larger habitat area of nonthreatened species compared with threatened species, increasing the probability of the habitat area being intersected by roads.Nevertheless, we also found relatively large fragmentation impacts for some highly threatened species, such as the critically endangered western long-beaked echidna and the endangered giant bandicoot.After accounting for the effects of IUCN Red List status and including the impact of planned road constructions, we also found a larger impact of roads on the habitat of large species with a carnivorous diet compared with small species with an omnivorous or herbivorous diet, indicating that these species will be especially affected by future road construction.
We found smaller ratios of ECA to MAR (i.e., higher extinction risks) for species with a higher threat status, indicating that ECA/MAR is a useful proxy of extinction vulnerability.For the majority of the mammal species we examined, the ECA values remained well above their MAR even when all planned roads were accounted for.Because of the large extent of natural areas in New Guinea, increased fragmentation by roads has not yet decrease the ECA to below the MAR.Nevertheless, this does not mean that planned road constructions are without consequences for mammalian fauna in New Guinea, in view of the additional threats posed by other pressures, such as habitat loss and hunting (Shearman & Bryan, 2015;White et al., 2021).We also found that ECA/MAR is generally lower for species with a larger body mass, which might also be more affected by hunting than smaller species (Benitez-Lopez et al., 2019;Ripple et al., 2016).Indeed, 22% of the species in our analysis are currently threatened with extinction according to the IUCN, and for 83% of these species hunting is one of the main threats (IUCN, 2021).There were also 17 species for which the ECA was smaller than the MAR in the baseline situation; these are all currently classified as threatened or data deficient.Hence, considerable further fragmentation of the ranges of these species due to the construction of new roads, such as for the Woolley's three-striped dasyure, is concerning.

Methodological reflections
The ECA is a comprehensive metric of connectivity, integrating patch areas, number of patches, and connectivity of the patches (Saura & Pascual-Hortal, 2007;Saura et al., 2011).Until now, this metric has mostly been used to estimate the connectivity of protected areas (Santini et al., 2016) and habitat (Kuipers et al., 2021) for groups of species, assuming similar dispersal distances in each group.We showed how this metric can also be applied at the species level, accounting for species-specific dispersal capacity.We also showed how to incorporate species-specific and road-type-specific road crossing probabilities in the ECA calculation.As far as we are aware, this is the first multispecies study that quantifies habitat fragmentation effects of roads based on such species-specific and road-type-specific road crossing probabilities.Because our approach is based on easily retrievable road and species characteristics, it can be readily applied to other geographic regions and taxonomic groups.For example, our approach could be used to explore and evaluate alternative road development plans in regions where future road developments seem inevitable as a consequence of economic and human population growth and the resulting increased demand for resources (Dulac, 2013;Laurance et al., 2014;Meijer et al., 2018).
Our approach and results come with various sources of uncertainty.For example, residual heterogeneity in the allometric relationships for median dispersal distance leads to uncertainty in the probabilities of dispersal between habitat patches, and there could be errors in the IUCN data on species' ranges and habitat preferences.However, we expect these errors to affect the 3 different situations (i.e., baseline, current, and future) equally; hence, they are likely less relevant when comparing the different situations.In addition, a previous study showed that it is appropriate to use IUCN habitat preference data in macroecological studies (Broekman et al., 2022).We further acknowledge that aggregating the species' habitat maps from an approximately 100 to 300-m resolution may lead to an overestimation of connectivity because the aggregation may merge distinct habitat patches into one.Unfortunately, the large computational requirements of the calculations precluded the use of 100-m resolution rasters.
Our estimates of road crossing probabilities (Eq.4) are also uncertain because they relied on the mean width of roads of the same type rather than road-specific estimates of road width,  which could not be obtained.Further, our estimates of road crossing probabilities were based on extrapolation of observations from other mammal species and regions, where conditions are likely different.Nevertheless, we preferred our estimates of road crossing probability over assuming that roads cannot be crossed at all, as applied in earlier studies (e.g., Borda-de-Agua et al., 2011;Ceia-Hasse et al., 2017;Pinto et al., 2018).To gain more insight into road crossing probability and its underlying drivers, we recommend acquiring more empirical data on road crossings.This might be facilitated by the increasing availability of GPS tracking data.For example, the Movebank database (http://www.movebank.org)contains tracking data of 11,700 animals, representing 189 species.Combined, these tracking data and road data (Meijer et al., 2018) offer interesting opportunities to estimate road crossing probability across multiple species and regions.Better predictions of road crossing probability might also be obtained by including traffic volume as a predictor because larger traffic volumes might decrease road crossing probability through road kills, noise, lights, and pollutants (Bennett, 2017;Jaeger et al., 2005).There was insufficient data to include traffic volume as a predictor in the road crossing model (Chen & Koprowski, 2019), and traffic volume data are also lacking for New Guinea (Slattery et al., 2018), precluding the inclusion of a traffic volume effect in our study.

Conservation implications
To limit the potential negative effects of current and planned roads on New Guinea's biodiversity, it is important to prevent or mitigate the habitat fragmentation effects of roads.This could be achieved by, for example, redirecting roads or building crossing structures such as underpasses (Rytwinski et al., 2016;Smith et al., 2015).Our results indicated that such conservation actions could be particularly effective when aimed at large species because the remaining ECA and the ratio of ECA to MAR decreased as body mass increased.Moreover, our speciesspecific approach helps to identify specific species for which these measures are most urgently needed (i.e., species with relatively large declines in remaining ECA combined with relatively low ECA/MAR).For these species, an overlay of their occurrence range with road maps helps reveal where measures need to be implemented.For example, to limit habitat fragmentation by roads for the western long-beaked echidna, road redirection or road mitigation measures are needed for the planned roads in the northwestern part of New Guinea (Figure 7a).
To limit impacts on the Wallace's three-striped dasyure, measures need to focus on the planned roads in the south-central part of the island (Figure 7b).By revealing patterns in road fragmentation effects related to species characteristics and by identifying and prioritizing species most threatened by roads, the results of our study may contribute to designing more effective conservation strategies to protect New Guinea's biodiversity.

FIGURE 1
FIGURE 1Existing (blue) and planned (orange) roads in New Guinea.

FIGURE 2
FIGURE 2 Number of species in relation to the proportions of equivalent connected area (ECA) when only existing roads are considered (current situation) and when existing and planned roads are considered (future situation) relative to the baseline ECA (no roads).

FIGURE 3
FIGURE 3Proportions of the equivalent connected area (ECA) when only existing roads are considered (current situation) and when existing and planned roads are considered (future situation) relative to the baseline ECA (no roads) for the 5 terrestrial mammal species in New Guinea with the lowest proportion of remaining ECA in the future situation.

FIGURE 4 FIGURE 5 FIGURE 6
FIGURE 4 Proportions of equivalent connected area (ECA) (a) when only existing roads are considered (current situation) and (b) when existing and planned roads are considered (future situation) relative to the baseline ECA (no roads) in relation to species body mass, diet, and International Union for the Conservation of Nature Red List status (lines, predicted values based on the most parsimonious model (Supporting Information Appendix S3).LC, least concern; NT, near threatened; VU, vulnerable; EN, endangered; CR, critically endangered; DD, data deficient.