Multi-year Use of the Standard Gauge Railway Underpasses by Mammals in Tsavo, Kenya: Influence of Underpass Dimensions, Infrastructure and Ecology


 Rail and road infrastructure is essential for economic growth and development but can cause a gradual loss in biodiversity and degradation of ecosystem function and services. We assessed influence of underpass dimensions, fencing, proximity to water and roads, Normalized Difference Vegetation Index (NDVI), presence of other species and livestock on underpass use by large and medium-sized mammals. Results revealed hyenas and leopards used the underpasses more than expected whereas giraffes and antelopes used the underpasses less than expected. Generalized linear mixed effects models (GLMMs) revealed that underpass height influenced their use by wildlife, with several species preferring to use taller underpasses. Electric fencing increased underpass use by funneling species towards underpasses, except for elephants and black-backed jackal for which it reduced underpass passage. GLMMs also revealed that the use of underpasses by livestock reduced the probability of their use by nearly 50% of wildlife species. Carnivore species were more likely to cross underpasses used by their prey. Buffalo, livestock, and hyenas used underpasses with higher NDVI and near water sources while baboons, dik-diks and antelope avoided underpasses with high NDVI. The findings suggest a need for diverse, and comprehensive approach for mitigating the negative impacts of rail on African wildlife.


Introduction
Railway and road infrastructure is essential for economic growth and development 1 , but it is also widely regarded as a catalyst for biodiversity loss in natural ecosystems [2][3][4][5][6][7] . Generally, railway tra c, like road tra c, may negatively impact wildlife and wildlife habitats through increased wildlife mortality from wildlife-train collisions, loss of population connectivity, habitat fragmentation, pollution (e.g., noise, chemical and light), and habitat loss 6,8−12 . Although railways may have similar impacts as roads, little attention has been paid towards mitigation of the effects of railways on wildlife. Perhaps, this is because the impacts of railways are perceived to be negligible due to much lower tra c ow than roads, characterized by long tra c-free intervals 13 . However, with increasing railway tra c volumes and the expansion of high-speed trains, wildlife mortality from railways will likely increase 8,14 , and this will demand implementation and evaluation of mitigation measures to reduce wildlife mortality. Moreover, an increase in infrastructure is predicted to occur mostly in the tropics where there are high levels of biodiversity and fragile ecosystems 15,16 .
Wildlife corridors (e.g., underpasses, overpasses, culverts) along highways traversing wildlands are a valuable mitigation tool for enhancing the permeability of transportation infrastructure for wildlife while preventing wildlife mortalities from vehicle collisions [17][18][19][20] . Several research studies, mostly in North America and Europe, have examined the effectiveness of wildlife underpasses along highways, but limited data exists for railways. Moreover, there is a dearth of information on wildlife use of underpasses associated with roads or railways by wildlife in the African continent where these designs and technologies are increasingly being adopted but see examples: 21,22 . The North American and European studies reveal that the effectiveness of underpasses is dependent on design factors such as size dimensions (i.e., height, length, and width) and location [23][24][25][26] . In addition, ecological factors such as the presence of vegetation cover, forage, species involved, species interactions (e.g., predator-prey) and human activities are also important 25,27−29 . The optimal characteristics for wildlife underpasses along highways are known to be species-speci c 30,31 , suggesting that it may be di cult to create universally optimal designs in areas with diverse wildlife species. Moreover, this is further compounded by species and individuals differing in their propensity to use underpasses 32,33 . An understanding of how railway underpass design in uences use by African savannah wildlife is lacking. There is an urgent need to understand how modern railway infrastructure is differentially impacting the connectivity and conservation of various African species. This is accomplished speci cally by identifying species that are negatively impacted by railway development and may need alternative interventions, especially for species for which wildlife corridors traversing railways appear to have little to no positive in uence.
The use of fencing to funnel species towards underpasses, so as to minimize collisions with automobiles on highways, has been demonstrated to be effective in Europe 34,35 . While fencing is an effective method for funneling species, it might limit the migration of species with routine migratory routes if these are fenced, enhancing genetic isolation [36][37][38] . This calls for an urgent need to understand how different African savannah wildlife species use wildlife corridors and the in uence of fencing on their effectiveness. Moreover, because railways are frequently co-aligned with roads, this forms infrastructure corridors 10 , and the impact of such parallel road and railways on the effectiveness of wildlife passages and fencing along highways is less known.
Here we examine for the rst time in an African savannah ecosystem, the in uence of a Standard Gauge Railway's (SGR hereafter) underpass design (type, height, and width), proximity to roads, fencing, livestock, and associated ecological factors on the likelihood of crossing by large-and medium-sized mammals in the Tsavo Conservation Area (TCA), Kenya. Speci cally, we address six key questions: 1) Are some species more likely than others to use underpasses? 2) Is fencing effective in funneling wildlife and livestock species towards underpasses? 3) Do infrastructural factors such as, type of underpass (bridge or culvert), proximity of underpass to a paved road, and the dimensions (i.e., width and length) of the underpass, enhance or inhibit their use by different wildlife species? 4) Do ecological variables such as the proximity of underpass to perennial water sources, NDVI (green biomass index) around the underpass in uence their use by different wildlife species? 5) Does the presence of livestock or wildlife predators along underpasses reduce the probability of their use by other noncarnivore wildlife species? 6) Does the presence of prey species in uence the probability of carnivores using underpasses? Answers to these questions are key to addressing the knowledge gap that exists regarding impacts of rail on African savannah wildlife.

Transport infrastructure
The study area includes a section of the Standard Gauge Railway (SGR) which bisects Tsavo East and Tsavo West National Parks. The SGR runs from the coastal city of Mombasa to the interior cities and towns in Kenya, including Nairobi and Naivasha. There is an old Meter Gauge Railway (MGR) that runs parallel to the (SGR) from Mombasa to the interior cities. Additionally, major roads transverse the protected area including the Mombasa-Nairobi highway, which parallels the SGR and MGR, the Voi-Taveta highway, and the Manyani-Malindi highway (Figure 1b). Of the transportation infrastructure in the TCA, the recently built SGR (construction started March 2015 and operational June 2017) is the primary interest of this study. The SGR, unlike the MGR and adjacent highways, was designed to facilitate wildlife movement, speci cally continual migration, and dispersal of wildlife within the TCA landscape. This was achieved initially by mapping traditional paths used by elephants, a agship species in the TCA, then designing and constructing bridges along six crossing points, namely the Maungu, Bachuma, Ndara, Kenani, Manyani and Kanga wildlife underpass corridors. Also included as a wildlife corridor under the SGR is where the SGR bridges the Tsavo River. These bridges have varied lengths and heights, with some ranging up to 2km in length (Table S1). Further, several culverts were constructed for drainage purposes and to facilitate wildlife crossings. As part of the structural design, the SGR is constructed on raised ground adjacent to bridges and culverts, creating steep embankments on either side of the railway track; there are no wildlife overpasses along the SGR. On either side of the embarkment is an electric fence, erected January 2018, to funnel wildlife to the various underpasses to reduce the risk of trains colliding with wildlife and to minimize the risk of injury to wildlife due to falling from the embankments. Generally, tracks were divided into hooved and pawed impressions. Pawed animal tracks were identi ed using the size of the track and presence or absence of claws. For similarly sized animals, the shape of the paws and proportion of the interdigital pad to the paws and other distinguishing characteristics were used to differentiate species 40 . Unclassi ed paw tracks were pooled as carnivores. Hoofed tracks were classi ed to into species or appropriate taxon based on the presence and absence of cloven hooves, size of the hooves, number of toes, and the shape of the hoofs. Hoofed tracks were recorded as antelope if they could not be identi ed to species or genus level. If tracks in the underpass were not clearly identi able, they were followed to where the substrate could allow identi cation. To avoid double counts on subsequent days, footprints were erased with a feather duster so as to prevent recount 41 .Scats were only recorded if they were fresh 40 . To avoid double counts scats were marked with white chalk.

Underpass utilization
For this study, underpasses were mapped using center GPS locations ( Figure S1) grouped into those located in the section of the railway crossing between the parks and those located in a section of the SGR crossing between the park and community ranches. Wildlife crossings were also classi ed into three categories: bridges, culverts, and embankments ( Figure 2. A, B, & C). Bridges were de ned as raised sections of the railway supported by piers and abutments and spanning more than 6 meters in length and more than 6.5 meters high. A culvert was a tunnel structure built to allow water and wildlife to pass and are usually embedded in the soil and are less than 6 meters in length with varying height. Embankments were compacted earth material that raised the grade line of a highway or railway. We monitored wildlife crossings in 14 bridges, 58 culverts and 69 embankments in two sections of the SGR namely, park-ranch interface (VB), and park-park interface (VM) ( Figure S1).

Species or taxon abundance
Data on the abundance of medium-sized to large-sized mammals (species with a mass greater than ca. 2 kg, See: 42 ) near TCA roads were collected from monthly road counts taken from July 2008 to July 2015. Vehicle road counts were conducted by driving along road transects at a xed speed of 20km-hr while recording all wildlife species sighted on either side of the road up to 250m (measured with a Bushnell Scout DX 1000 Laser Range nder). The road count was carried out along three, 10 km long transects consisting of the unpaved road network within Tsavo East National Park. Road counts started at 6.00 am and ended when the transect distance was achieved. All mammals, from dik-dik to elephant were recorded during the count, locations were mapped, and behavior was recorded. These data were used to derive the expected frequency of mammal use of the underpasses from the proportion of their proximate abundance estimates along roads.

Ecological factors
Ecological variables associated with underpasses (i.e., Normalized Difference Vegetation Index (NDVI), and proximity to water sources) were extracted from remotely sensed data and from drainage maps of the Tsavo ecosystem, respectively. NDVI products derived from SPOT VGT were downloaded from ESA (European Space Agency) (https://earth.esa.int/web/guest/data-access/browse-data-products). The NDVI products were downloaded from January 2016 to December 2019 as to obtain one averaged product per month for the study period.
The proximity of SGR underpasses to water sources were obtained from drainage (natural streams and rivers) of the Tsavo East and West digitized from 1:250000 toposheets including Voi SA-37-14, Kili SA-37-15, Garsen SA-37-11, and Kibwezi SA-37-10. Additional data on locations of water tanks, boreholes, dams, troughs, and pans for the Tsavo East and Tsavo West National parks were obtained using Garmin GPS (GPSMAP® 64). Straight-line distances between underpasses and the nearest water sources were measured in ArcGIS Toolkit.
2.5. Underpass type, size, and infrastructure Data on infrastructure variables including underpass dimensions (width, length, and height), proximity of underpass to roads, and presence of a functional electric fence along the embankments adjacent the underpasses were obtained. The length and widths measurements for both culverts and bridges were provided by the China Road and Bridge Company (the company in charge of building the SGR) and this information is labelled on some of the underpasses. The major roads were digitized from 1:250000 toposheets including Voi SA-37-14, Kili SA-37-15, Garsen SA-37-11, and Kibwezi SA-37-10. We calculated the distance of each unique underpass point to the nearest highway road. Distances were recorded to the nearest kilometer.

Statistical analyses
To answer the question of whether some species are more likely than others to use the underpass, we employed a chi-square analysis using observed data on the frequency of underpass use by the top 20 most sighted species (Table S2). We calculated the expected frequency of crossing using wildlife abundance data along roads using R software for statistical computing 43 .
To determine whether fencing, underpass type and dimensions, NDVI, proximity to rivers and other water sources, and roads affected crossing by different species or taxa we modelled covariate effects with a generalized linear mixed effect model (GLMM) framework using a logit link function and binomial error structure. Underpass ID was employed as random effect. As independent variables, we used species or taxonomic groups which had 20 or more sightings.
These species include, savannah elephant, African buffalo, plain's zebra, yellow baboon, Kirk's dik-dik, lion, leopard, spotted hyena, African civet, impala, waterbuck, and lesser kudu. We also grouped species into the following categories: antelope, carnivore, or mongoose and livestock or wildlife and used these as dependent variables as well.
To address the question of whether livestock presence hindered or enhanced the use of underpasses by wildlife, livestock was used as an explanatory variable and all other species and groups were used as dependent variables. In addition, when lion was used as dependent variable, plains zebra, African buffalo and livestock were included as independent variables because these are preferred prey species [44][45][46][47][48] . When leopard, and spotted hyena were employed as dependent variables, antelope, zebra, and livestock were included as independent variables, as these are these are some of their preferred prey species [49][50][51] . For all carnivore species (classi ed and unclassi ed), we used antelopes as an independent variable.
GLMMs were performed using the glmmTMB package 52 . We used AIC model selection to distinguish among a set of possible models describing the relationship between infrastructure design and ecological factors and mammalian use of SGR underpasses in the TCA. Conditional effects of covariates were visualized using the software package ggeffects 53 . For each model we included a comprehensive list of all independent variables the best set of covariate in uencing the likelihood of using the underpass was evaluated using AIC in the MuMIn package 54 . All the software packages are part of the R software for statistical computing 43 .

Differential utilization of the SGR underpass by various wildlife species
Thirty-three species of medium-to large-sized mammals were observed using the SGR underpasses, including unidenti ed species grouped in general categories of carnivore, mongoose, and antelope. Livestock (e.g., cattle, goats, sheep, donkeys, and camels) were also frequently observed to use the SGR underpasses (Table S2, Table 1). The top 5 wildlife species that utilized the underpass bridges (percent of observations) were elephants (30%), plains zebra (20%), baboons (12%), buffaloes (8%), and dik-diks (8%) (Table S2). Similarly, the top species using culverts were elephant (4%) and baboon (3%). Livestock used the culverts more than any wildlife species (11%). Hyena was the top identi ed carnivore utilizing the underpasses, but the top ve carnivores using the underpasses frequently, includes leopard, lion, black-backed Jackal, and civet (Table 1). Table 1 The frequency and percentage use of underpass by medium-to large-sized mammals in the Tsavo Conservation Area. The most abundant wildlife species sighted from monthly road counts over a seven-year period were elephants, Grant's gazelle, Kirk's dik-dik, plains zebra and impala. Black-backed jackal and lion were the most sighted carnivores, but species such as hyena and leopard were less sighted during road counts (Table S3,   Table 2). Some species frequently sighted among the top 20 during road counts were also observed to be among the most frequent users of the SGR underpasses. These includes the savannah elephant, African buffalo, Kirk's dik-dik, impala and yellow baboon. However, there were also some species observed frequently during road counts that were observed infrequently using the SGR underpasses. These infrequent underpasses users were Maasai giraffe, Coke's hartebeest, common eland, common hippopotamus, and Grant's gazelle (Table S2, Table S3). Generally, the frequency of underpass use by different species was not dependent on their corresponding monthly road count frequencies (χ 2 17 = 45698, P < 0.0001; Fig. 3a). This was also the case when we separately tested whether underpass use by carnivores and herbivore species was expected based on their abundance from road sighting (herbivores: χ 2 11 = 7265.7, P < 0.0001, Fig. 3b; carnivores: χ 2 4 = 3429.4, P < 0.0001, Fig. 3c).

Effect of fencing on underpass and embankment utilization by wildlife and livestock
Many species used embankments to cross the SGR, but most of them stopped using embankments following the installation of an electric fence (median of times crossed before = 4 and median of times crossed after = 0, V = 91, P = 0.0016, Wilcoxon test for matched pairs), reducing the risk of wildlife mortalities from train-wildlife collisions. The exceptions were elephants, and leopards, but even for these, electric fencing dramatically reduced their use of embankments. Electric fencing increased underpass use by most species, except for elephant, black-backed jackal, caracal, and waterbuck ( Table 2, Table S4) where fencing reduced the rate of underpass usage (Table 3) and leopards and civets, for which fencing exhibited no discernable effect. Among the underpass design factors, height was a more important factor than either type (bridge or culvert) or width because it was selected in nearly all the models whereas bridge type or width were selected in 5 and 3 models respectively (Table 3). Speci cally, there was a positive relationship between underpass height and the probability of underpass use by mammals (see coe cients in Table 3). Underpass type was important for some mammalian species, but its effect was weak. Generally, culverts were used to a lesser extent relative to bridges and this effect was stronger for elephants ( Table 3).
The distance of the underpass to the Nairobi Mombasa highway had varied effects on different species. Baboons and livestock preferred to use underpasses closer to the highway whereas buffalo and antelopes preferred to use underpasses farthest from highways. Proximity of the underpass to the highway did not in uence their crossing by most carnivores considered in the analyses (Table 3).

Effects of ecological factors on wildlife and livestock utilization of underpasses
Underpasses located in areas with higher NDVI were more likely to be used by buffalo, livestock and hyenas and these species were also more likely to use underpasses in proximity to water sources. In contrast, baboons, dik-diks and antelope avoided to utilize underpasses with high NDVI (Table 3). However, the plains zebra preferred underpasses farthest from perennial water sources (Table 3).
Livestock presence or use of underpasses reduce the likelihood of underpass use by most wildlife species except baboons and most carnivore species (Table  3). The utilization of underpasses by lions, leopards, hyenas, and unclassi ed carnivore species was in uenced by presence of their prey species. Lions were more likely to use underpasses where zebras buffaloes and livestock (i.e., their key prey species) were present. Leopards, hyenas, and other carnivores used underpasses where their prey species, antelopes, were also present ( Table 3).

Discussion
In this study we determined that many wildlife species utilize the SGR underpasses, but species such as the Maasai giraffe, common warthog, impala, Coke's hartebeest, common eland, and Grant's gazelle had a lower propensity to use underpasses, whereas most carnivores and baboons had a higher propensity to use underpasses. The rarity of underpass use by giraffes, and the strong positive correlation between underpass use by most wildlife species and underpass height observed, highlights the limitation giraffes face in using underpasses. Giraffes with their long neck and legs, have an extended viewing horizon to maintain vigilance 55 , and may even view bridges as obstacles with the small difference between their height and that of the bridges. The average height of giraffe is about 5.5 m for males and 4.3 m for females 56 and the modal height of bridges in this study is 6 m (see discussion on the in uence of underpass height below). Most culverts are inaccessible to giraffes due to their low heights (3-4 m).
For other species, predator-prey interactions may explain their less than expected use of the underpasses. Indeed, the use of underpasses by lions, leopards, and hyenas, was positively in uenced by presence of buffalo, zebra, and antelope, which are their key prey species 46,49,51 . This suggests that predator-prey interactions were important factors in underpass use by the major carnivores in the TCA. These ndings are similar to those found for coyotes, Canis latrans in California, where coyotes favored underpasses with high presence of their main prey items, rodents and lagomorphs 29 . Generally, Mata et al. 57 concludes that wildlife crossing structures will be less effective for prey if their use is adversely in uenced by predator-prey interactions. A similar nding has been echoed regarding the use of underpasses by brown bandicoots and their fox predators 58 . Such may have been the case for species such as common warthogs, impalas, grant gazelles, dik-diks, elands, waterbucks, and lesser kudu, which all used underpass less than expected. While at the same time their main predators, leopards, and hyenas, used the underpass more than expected even when potential bias is adjusted by considering cryptic carnivore alone.
Generally, electric fencing reduced the rate of SGR crossing by wildlife at the embankments while increasing the use of underpasses. The exception being elephants, where the use of underpasses with electric fences was reduced. This suggests that while fencing can help reduce wildlife mortalities from trainwildlife collisions by helping to funnel wildlife through underpasses, electric fencing may also reduce elephant connectivity. Other studies have found that fencing has a funneling effect that directs larger animals toward culverts 35 . Underpasses when combined with fencing have been shown to reduce large mammal-vehicle collisions by 86% on Highway 93 in Montana, United States, while also maintaining wildlife connectivity across roads 71 . The effectiveness of crossing structures is signi cantly enhanced when combined with fences, and both measures are usually best implemented together 17 .
However, fencing although effective, may reduce overall permeability of landscapes traversed by roads. For example, we observed a reduction in underpass use by elephant and black-backed jackal following the erection of electric fencing along the SGR. Other species that reduced underpass use include the caracal, leopard, and waterbuck. These ndings suggest that species with traditional migratory routes or those that defend territories may be adversely affected by electric fencing. For example, elephant families have traditional movement routes 22,72,73 and if underpass structures do not take this into consideration many elephants may fail to cross the SGR. Its therefore important that the underpass structures take into consideration of the traditional elephant routes to ensure. It has been observed, for example, that underpasses placed at identi ed panther (Felis concolor cory) crossing points, using prior knowledge of panther movements, were more likely to be used by panthers 74 . Similarly, deer underpasses placed without regard to traditional paths failed, irrespective of addition of fences, to direct deer to those crossings 74,75 . Fencing has been shown to reduce underpass vehicle moose collisions, but also to reduce the use of underpasses 76 . These ndings underscore the need for extensive wildlife pathway monitoring prior to road construction, so that underpasses can be located along natural wildlife trails.
Black-backed jackals, leopards, caracals and waterbucks are all territorial species 77,78 that had reduced underpass use following fencing. Our results suggest that fencing along the SGR, like along highways, may impose arti cial home range boundaries on some territorial species See: 79 . This suggest a need to examine more explicitly the impact of road and rail infrastructures on home range displacement and abandonment using modelling tools detailed by Sells and Mitchell 80 .This study also revealed that livestock presence reduces the likelihood of underpass use by most wildlife species except baboons and most carnivore species. Several studies report spatial segregation between cattle and wildlife [81][82][83] . For example, Hibert et al. 82 found spatial avoidance between cattle and wildlife grazers including elephants. Another study found that elk (Cervus elaphus nelsoni), mule deer (Odocoileus hemionus hemionus), and cattle which frequently co-occur in the northwestern United States 83 were spatially segregated and avoided each other. The occurrence of livestock also suggests the presence of humans and many studies have shown wildlife avoidance of human presence (sources).
Among the underpass design factors, this study shows that height is more important than either type of underpass (bridge or culvert) or width in affecting use by wildlife and livestock. This result agrees with several ndings in the literature. For example, Donaldson 84 found that in Virginia, USA underpasses with a minimum height of 4m were successful in facilitating the passage of white-tailed deer and other wildlife species. Our result also concurs with a study in Banff National Park, Alberta where crossing structures that are high, wide, and short in length strongly in uence grizzly bear (Ursus arctos horribilis) passage 24 . Structural designs are the main determinant in species' use of wildlife passages along roads and highways. It is thought that animals using an underpass require an unobstructed view of the habitat or horizon on the far side of the underpass to ensure safety 74 . Moreover, some studies (sources) have used the openness index ratio, a measure of an animal's ability to see into the other side of the underpass. It has been suggested that this feature is probably is more important than the exact width and height of the underpass as it integrates both height width and length 85 . Wildlife overpasses which offer a better view of where animals are moving to were utilized more than underpasses by mule deer 20 . In studies examining the use of underpasses and overpasses by wildlife it was found that large mammals exclusively use overpasses while small mammals used underpasses 86 . For livestock being herded into the park, it is not surprising that height or width were not statistically signi cant variables in uencing their use of the underpasses because they are not making decisions on crossing based on predation risk.
This study also revealed that parallel infrastructure had a varied in uence on underpass use by wildlife and livestock. Most railways and roads are usually coaligned in the same corridor 10 . This creates a challenge for wildlife as they must cross multiple infrastructure impediments when moving from one side to another. Indeed, we observed in this study that buffalo and unclassi ed antelope preferred underpasses that were farthest from the highway. On the contrary, baboons and livestock used underpasses that were near roads. This was not surprising for baboons as they often scavenge on human food leftovers 87 .
Along the infrastructure corridor which includes the SGR, food leftovers are often thrown out of cars, particularly in areas close to human habitation. This likely to attract baboons towards roads and their greater use of wildlife corridors near roads. Moreover, baboons' proclivity for using wildlife corridors near roads can be further explained by their use of roads as an e cient method of travel, where groups of baboons can move faster and in a more directed manner 88 .
Livestock use of underpasses closer to roads has to do with the easy access to the protected resources within the national parks. Livestock are frequently illegally grazed and watered within the Tsavo National Parks. Some of the underpasses used by livestock in this study were the ones closest to areas where parklands interface with human settlements and communities, often where the highway and the SGR are in closest proximity to one another.
Several mammalian species are known to track seasonal changes in primary productivity, measured as changes in NDVI, and water availability to occupy areas that have lush vegetation or regular access to drinking water [89][90][91][92][93][94] . In this study, we observed that underpasses located in areas with higher NDVI and near water sources were more likely to be used by African buffalo, livestock, and spotted hyenas. On the contrary yellow baboons, Kirk's dik-dik, lesser kudu, mongoose and antelope preferentially utilized underpasses with low NDVI. Predation or anthropogenic disturbances have been linked to avoidance of optimal habitats and locations with drinking water sources by wildlife 95 . The preferential use of underpasses that are not near greener areas (i.e., lower NDVI) and far from water sources, might suggest this as a strategy to minimize predation. Some ungulate species use open areas, which are less green, and avoid bushy areas, which have higher greenness 96 . These species are selecting to use these underpasses as passages routes, rather than ones close to foraging areas, because they afford greater safety from predators.
This study demonstrates that to ensure the effectiveness of underpasses it is critical to consider species' predator-prey interactions, behaviors, and foraging needs in relation to underpass design and location. With sound wildlife corridor placement and design in areas where railways traverse habitats with diverse wildlife, underpass use by all species should increase. This will become increasingly important in mitigating against habitat fragmentation and guaranteeing safe wildlife passage as transportation infrastructure continues to expand and support greater volumes of tra c.

Conclusion
Rail and road infrastructure is essential for economic growth and development but can cause a gradual loss in biodiversity and degradation of ecosystem function and services. We assessed in uence of underpass dimensions, fencing, proximity to water and roads, Normalized Difference Vegetation Index (NDVI), presence of other species and livestock on underpass use by large and medium-sized mammals. Results revealed hyenas and leopards used the underpasses more than expected whereas giraffes and antelopes used the underpasses less than expected. Generalized linear mixed effects models (GLMMs) revealed that underpass height in uenced their use by wildlife, with several species preferring to use taller underpasses. Electric fencing increased underpass use by funneling species towards underpasses, except for elephants and black-backed jackal for which it reduced underpass passage. GLMMs also revealed that the use of underpasses by livestock reduced the probability of their use by nearly 50% of wildlife species. Carnivore species were more likely to cross underpasses used by their prey. Buffalo, livestock, and hyenas used underpasses with higher NDVI and near water sources while baboons, dik-diks and antelope avoided underpasses with high NDVI. The ndings suggest a need for diverse, and comprehensive approach for mitigating the negative impacts of rail on African wildlife. Figure 1 The Tsavo Conservation Area (TCA) in south-eastern Kenya (inset). The bold dashed black line indicates the Mombasa-Nairobi highway infrastructure corridor (MNHIC) adjacent to Tsavo National Parks (A) and the transport infrastructures that cuts through the Tsavo Conservation Area in southeastern Kenya (B).

Figure 2
Wildlife crossings were classi ed as bridges (A), culverts (B), and embankments (C) along the standard gauge railway in Tsavo National Parks, Kenya.

Figure 3
The percent deviation of SGR underpass utilization by medium-to large-sized mammals when herbivores and carnivores are simultaneously considered (A) and when herbivores (B) and carnivores (C) are independently considered. The expected frequency was calculated based on the frequency of their sighting on TCA roads.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.