Forecasting effects of transport infrastructure on endangered tigers: a tool for conservation planning

Study area

We applied our model to Chitwan National Park, situated in the Himalayan lowlands in south central Nepal (Fig. 1, Sharma, 1990). The park was established in 1973 and is approximately 950 km². It is located in a river valley basin along the flood plains of the Rapti, Reu, and Narayani Rivers with an elevation range of 150–815 m. A buffer zone surrounding the park was established in 1996 and is about 750 km². The park and buffer zone consist of natural vegetation communities including sal (Shorea robusta) forest, khair (Acacia catechu) and sissoo (Dalbergia sissoo) riverine forests, and grasslands dominated by species of the genera Saccharum, Themeda, and Imperata (Chaudhary, 1998; Carter et al., 2013). These vegetation types in the park support a number of large herbivorous species, including spotted deer (Axis axis), barking deer (Muntiacus muntjak), wild boar (Sus scrofa), sambar (Rusa unicolor), hog deer (Axis porcinus) and gaur (Bos gaurus) (Seidensticker & McDougal, 1993)—all of which tigers prey upon. Other large predators in the park include the common leopard (Panthera pardus) and dhole (Cuon alpinus). The climate in the park is subtropical with a cool dry winter and a summer monsoon season from mid-June to late-September.

Many existing roads in the lowlands region are going to be upgraded and expanded in the near future to support greater traffic volumes and speeds (NPC, 2020). For example, the 1,795 km long Postal (Hulaki) highway—oldest highway in Nepal—is being resurfaced. The 1,028 km East-West (Mahendra) highway is being expanded from two to four lanes. Both of these highways pass through forests, habitat corridors, and protected areas including Chitwan (Van Merm & Talukdar, 2016; Quintana et al., 2022). Development and expansion of municipal roads that pass through Chitwan—such as the Thori-Madi-Bharatpur road—are also ongoing or proposed (Van Merm & Talukdar, 2016). Furthermore, the Government of Nepal is planning cross-country railway projects, including the 1,205 km long East-West electrified railway (NPC, 2020). Construction of this railway has begun in eastern Nepal. Currently, the railway is projected to intersect the Barandabhar corridor in Chitwan’s buffer zone (Fig. 1); however, the exact alignment and design elements through Barandabhar and other sensitive forested areas are still under discussion.

Model parameterization

A long-term study began in the park core in the 1970s to collect field measures on many aspects of tiger behavior and ecology, including interactions with conspecifics and their environment, as well as population vital rates, such as survival and mortality (Sunquist, 1981; Smith, McDougal & Sunquist, 1987; Smith, 1993; Kenney et al., 2014). Those extensive field measures supplemented with those from other studies were directly used as parameters in our model (Table 1) instead of fitting the full model to data, that is, by calibration. In previous studies (Carter et al., 2015; Carter, Levin & Grimm, 2019), we conducted a sensitivity analysis and assessed model fit using a pattern-oriented modeling approach (Grimm et al., 2005), which compares multiple patterns observed in real systems to those generated by the model. Through those previous analyses, we determined that a range of key model outputs, such as reproduction rates, dispersal distances, resource selection, and land tenure, matched closely with observed patterns of the real tiger population in the park (Carter et al., 2015; Carter, Levin & Grimm, 2019). Furthermore, the overall response of the position and size of territories to different prey and tiger densities was realistic (Carter, Levin & Grimm, 2019). We are therefore confident that the model performs well and serves as an excellent foundation to address our central research question about the future impacts of transportation networks on tigers. In the following, we outline major components of the model but direct readers to Carter et al. (2015) which includes a detailed description of the model following the Overview, Design concepts, and Details (ODD) protocol (Grimm et al., 2006; Grimm et al., 2010) for describing ABMs.

Mapping prey biomass and transport infrastructure

Previous research in the Nepal lowlands used a Poisson regression to relate land cover to tiger prey abundance and then mapped the fitted abundance estimates across the region at a resolution of 250 × 250 m (Shrestha, 2004). As our model time step was one month, we converted the abundance estimates to a monthly rate of prey biomass production per cell (0–10.46 kg per month, average = 3.78 kg per month) by scaling the abundance estimates to empirical rates of average daily prey consumption by females and information on female territory sizes in Chitwan (Table 1; Fig. 1). We simplified the model by holding prey biomass in each cell constant over time. Although seasonal change and predation can affect prey biomass over time (Karanth et al., 2004), we ignore those dynamics here to focus instead on transportation-induced changes in prey density.

We defined a road as “primary” if it was maintained year-round and provided access between large towns and “secondary” if it was not regularly maintained but allowed access between villages or into forested lands. Primary roads were paved, allowing for high-speed traffic, whereas secondary roads were either paved or hard-packed dirt. We collected existing data on the geographic locations of roads in Chitwan from OpenStreetMap, which provided the most comprehensive and accurate road data. Specifically, we used the R package osmdata to select roads in Chitwan and assign them as primary or secondary roads based on OpenStreetMap’s tags and our knowledge of the area. Roads with key “highway” and values “motorway” were classified as primary roads. Those with values “primary”, “secondary”, or “tertiary” were classified as secondary roads. Finally, the location (as of 2018) of the proposed railway through Barandabhar forest corridor was digitized from hardcopy maps provided by the Nepal Department of Transportation. Most recent plans indicate the proposed railway will still be constructed through Barandabhar forest, though the exact location is uncertain. We assumed the railway has the same effect on tigers as a primary road. We chose this assumption because railways are major linear infrastructure; however, we lacked empirical data to differentiate effects of railways from primary roads. In total, there were approximately 198 km of primary roads including the railway and 723 km of secondary roads in the model landscape.

Simulating tigers and their territories

In the model, tigers follow a full life cycle—born, grow, reproduce, disperse, establish and modify territories, and die (Sunquist, Karanth & Sunquist, 1999). Other tigers will disperse to and establish territories in the gaps left by dead tigers. Animals at different age and stage classes—cub, juvenile, transient, and breeding—have distinct behaviors associated with them in the model (Karanth & Stith, 1999). Cubs (0–1 years old) and juveniles (1–2 years old) stay in their natal territory and die if their mom dies. Individuals enter the transient stage (2–3 years old) immediately following the juvenile stage and are no longer associated with their natal range. Transients therefore do not die if their mom dies; however, they do not have an explicit movement process until they reach the breeding stage. Once entering breeding stage, animals disperse in order to establish a territory.

Breeding females (3+ years old) seek to acquire a given amount of prey resources, whereas breeding males seek to acquire a given number of females (Sunquist, 1981; Smith, 1993; Ahearn et al., 2001). An adult female will add prey resources to her territory until she can maintain basal metabolic rate (number of calories an individual burns to perform basic life-sustaining functions) of 76 kg/month (Miller et al., 2014). Female territories do not overlap; however, they can adjust their size and shape to encompass the highest possible prey resources in the area. For example, a female can adjust her territory to encompass some or all of a territory left vacant by a female that recently died. Furthermore, neighboring females can occasionally compete over territorial borders but that competition rarely leads to mortality, unless a female’s territory is already so small to begin with that losing parts of her territory to other females could cause her to starve.

Dispersing males will challenge breeding males for access to his females, with the probability of the dispersing male winning based on age (Kenney et al., 2014). If he wins, he gains access to all the females of his opponent and there is a probability that any dependent cubs and juveniles of those females will die via infanticide by the new male (Pusey & Packer, 1994). Whereas, if the dispersing male loses the challenge there is a probability that he (i) dies or (ii) moves on to another location where he will continue to search for females or challenge males for access to females (Kenney et al., 2014). Breeding males that lost a challenge to a dispersing male and did not die will become dispersing males again.

Experiments

We created two configurations for transport infrastructure in Chitwan National Park: (i) the existing road network consisting of primary and secondary roads; and (ii) the combination of the existing road network and the alignment for a proposed electrified railway through the park’s buffer zone just north of the park core. These configurations allow us to assess the effects of existing road network and its synergistic effects with the proposed railway.

For each of the two configurations we experimentally tested the effects of different scenarios, including (i) transportation-induced increases in tiger mortality, (ii) transportation-induced depletion of prey abundance, and (iii) the interaction between increased mortality and prey loss (Fig. 2). To parameterize the tiger mortality experiments, we relied on previous work in Russia that found that female tiger survival declines by 67% when primary roads bisect their 50% minimum convex polygon home ranges and by 10% when bisected by a secondary road (Kerley et al., 2002). Thus, in the model, breeding females with any of their territory intersecting a primary road/railway had their annual survival rate decreased from 0.9 to 0.297 and if intersecting a secondary road their annual survival rate decreased from 0.9 to 0.81 (Fig. 2). If spanning both, then her annual survival rate was decreased by the larger of the two (i.e., primary road/railway). We are not modeling the encounter rates between tiger agents and roads but rather enforcing a reduction of survival based on the presence of a road within a female’s territory. We also assumed no transportation-induced mortality of males, as we had no empirical estimates with which to parameterize that effect and we anticipate population dynamics will be strongly affected by changes in the mortality of reproducing females.

For the prey depletion experiments, we assumed that prey abundance was zero along a road or railway and then increased linearly as distance from the road or railway increased. For primary roads and the railway, prey was depleted up to 5 km and for secondary roads up to 2 km (Fig. 2). Prey abundances were unaffected by roads or the railway past those distances. These distances were derived from two sources. The first source was a global meta-analysis showing that mammal abundances declined up to about 5 km from roads (Benítez-López, Alkemade & Verweij, 2010). We therefore set 5 km from primary roads as the max distance at which prey are depleted. The paper indicated a non-linear effect; however, we used a linear function for simplicity and ease of interpretation, especially since the specific shape of the non-linear function is unknown for tigers and their prey. The second source was a study that calculated a relative cost of movement for tigers in areas with varying road densities (Rathore et al., 2012). Areas with medium road densities were considered 2.5 times more costly to move on than areas with low road densities, assuming that an area having a high density of roads would be avoided compared to an area with few or no roads (Rathore et al., 2012). Although not an exact proxy, we used this ratio to approximate the road effect zone for secondary roads. Thus, we set 2 km from secondary roads as the max distance at which prey are depleted.

We also tested the interaction between tiger mortality and prey depletion for each transportation configuration. These three experiments—tiger mortality, prey depletion, and tiger mortality and prey depletion—were compared against a baseline experiment in which roads were present but they did not affect tiger mortality rates or prey abundances. This comparison allowed us to determine how many tigers with roads/railway intersecting their territories died due to those roads/railway on top of those that would have died from natural causes such as old age or starvation in the baseline.

The models were initialized with 14 females and 7 males, numbers chosen to avoid large cycles in population size before activation of the experiments. The model runs for 600 time steps (months), approximately 15 tiger generations, without effects from roads or the railway to give the tiger population enough time to reach quasi-stationary dynamics. Then, transportation effects occur instantaneously across the landscapes at time step 601. Although instantaneous changes to prey and tiger mortality are unlikely, we assessed their effects by running the model for 240 times steps (20 years) to give the model enough time to reach a new quasi-stationary state while also being a reasonable time period. The model code implemented in the NetLogo program (Wilensky, 1999) is available online at https://github.com/nhcarter/tiger_abm.

Data analysis

Experiments were replicated 28 times to take into account variation due to model stochasticity (Carter et al., 2015). For each experiment, we plotted through time the mean and 95% confidence intervals (based on quantiles of the t-distribution) across 28 replications of monthly population sizes of adult tigers, breeding females, and dependent offspring (i.e., cubs and juveniles). We also plotted annual mortality rates along transport infrastructure as well as summed total deaths over the 20-year simulation for each experiment. This way we could examine how the experiments altered population sizes and mortality rates along both primary roads/railway and secondary roads relative to a baseline (i.e., road present but no impacts on tigers or their prey). In addition, we calculated the average population size of different groups—adult tigers, breeding females, and dependent offspring—for the last time step of each of the experiments to assess how they affected tiger numbers after 20 years relative to the baseline. We used unpaired two-sample Wilcoxon tests to test for statistical differences in the means between experiments for (1) the adult tiger population sizes at the last time step and (2) total tiger deaths summed over the entire simulation. We tested differences in means within each road scenario as well as between the two scenarios.

To illustrate the spatial effects of transport infrastructure, we plotted female occupancy per cell as the proportion of all time steps (240) that any breeding female had her territory on a cell. We calculated female occupancy for all four experiments—(i) baseline, (ii) tiger mortality, (iii) prey depletion, and (iv) tiger mortality and prey depletion—using the existing road configuration. We ran 10 replications for each experiment to calculate mean female occupancies per cell over 20 years.

Tiger populations

All three experiments—mortality, prey depletion, and prey depletion and tiger mortality—ended with tiger populations that were significantly lower than the baseline (Table 2). The population of adult tigers rapidly decreased in experiments with increased probability of mortality when tiger territories intersect primary or secondary roads. Using the existing road configuration, the average size of the adult tiger population in the mortality experiment decreased, after 20 years, 39% compared to the baseline, from 133 to 81 animals (Table 2, Fig. 3). The decrease in tiger populations was greater when adding the proposed railway: the average size of the adult tiger population in the mortality experiment decreased 62% (from 133 to 51 animals) compared to the baseline. Moreover, mortality associated with the railway alone contributed to more than 20% reduction in tiger numbers (81 to 51 individuals) over 20 years compared to the mortality experiment without the railway (Table 2, Fig. 3).

In contrast to the mortality experiments, the population of adult tigers did not respond as severely to the experiments in which prey densities are diminished near primary and secondary roads. Using the existing road configuration, the average size of the adult tiger population in the prey depletion experiment decreased 7% compared to the baseline, from 133 to 123 animals (Table 2, Fig. 3). Adding the railway decreases adult tiger numbers an additional 3% (10% total) due to loss of prey. Interestingly, combining prey depletion and tiger mortality from roads did not additively decrease tiger populations but in fact reduced the effects of considering mortality alone. Using the existing road configuration, the average size of the adult tiger population in the experiment combining prey depletion and tiger mortality from roads decreased 10% compared to the baseline, from 133 to 119 animals (Table 2, Fig. 3). However, adding the railway to the existing road configuration nearly doubles the loss of tigers over 20 years (133 to 110) from the combination of prey depletion and tiger mortality (Table 2, Fig. 3). The changes to population sizes of adult females and dependent offspring are approximately proportional to those of the total adult population.

Patterns of mortality

In the mortality experiment, the death of tigers with territories crossed by existing roads was on average 10% of all mortality (Fig. 4), compared to only 6.5% in the baseline (i.e., tigers with territories crossed by existing roads that died from natural causes). The proposed railway increased this average proportion to 12.3%, nearly doubling the rate in the baseline. For most experiments, the proportion of deaths on secondary roads was approximately 2 to 3 times greater than those on primary roads/railway (Fig. 4). However, in the mortality experiments including the railway, a greater proportion of tigers died along primary roads (a category encompassing the railway) than on secondary roads. In the mortality experiments, the mortality rates are highest in the first few years, declining somewhat and stabilizing after 7 or 8 years once the population density decreased following the initial deaths of those tigers with territories crossed by roads. In the prey depletion experiment, the mortality of tigers intersecting roads is lower than all other experiments through time. Likewise, although higher in the first 3 to 4 years, the proportion of tiger deaths along roads or the railway becomes lower in the combined prey depletion and tiger mortality experiment than in the baseline. Prey depletion appears to dampen the effects of mortality rather than amplify it.

About 171 tigers with territories crossed by roads died over 20 years in the mortality experiment (Fig. 5), which was 46 more than those that died in the baseline indicating those additional deaths were due to the increased likelihood of mortality caused by roads. The proposed railway directly added about 10 more tiger deaths over 20 years (Fig. 5). In the baseline, the proportion of adult females and dependent offspring that died were approximately equal. The same was true of the prey depletion experiment, although a few more adult females died than dependent offspring. Furthermore, the total numbers of adult females and dependent offspring that died over the 20-year simulation were significantly less in the prey depletion experiment than in the baseline. In contrast, in the mortality experiments, the proportion of dependent offspring that died along transport infrastructure was greater than the proportion of adult females that died along transport infrastructure. However, the causes of those disparities are different. In the mortality experiments, existing roads and the proposed railway are causing more adult females and dependent offspring to die than those that died from natural causes in the baseline (13–19% more), with greater impacts on dependent offspring (60–68% more, Fig. 5). In the combined prey depletion and tiger mortality experiments, 24 fewer adult females die than in the baseline, whereas approximately the same number of dependent offspring die (Fig. 5). That is, the chances of adult females with territories crossed by roads dying are lower than those in the baseline.

Patterns of occupancy

The average occupancy per cell, measured as proportion of timesteps occupied by a female tiger, was 0.64 for the baseline. In other words, on average, a cell in the model was occupied by a female tiger for about 13 out of the 20 years in the simulation. The average occupancy for the mortality experiment was 0.4, nearly 40% lower than the baseline and equivalent to only 8 out of the 20 years. Experiments that depleted prey were less severe. In the combined prey depletion and tiger mortality experiment, the average occupancy was approximately 15% lower (0.55) than the baseline. The prey depletion experiment had little effect, overall, on average occupancy, which was about 4% (0.61) lower than the baseline.

The locations of roads affected tiger occupancy patterns and differences between experiments. Here, occupancy is defined as the proportion of months within 20 years in which a grid cell was part of a female tiger territory. Female occupancy was lower in cells near roads (i.e., 2 km from secondary roads and 5 km from primary roads, 0.57) than far from roads (0.68) in the baseline (Table 3). The baseline pattern of female occupancy corresponds to the distribution of prey across the landscape, where the mean prey resources of cells near roads was 2.89 and 4.3 far from roads. Compared to the baseline, mortality experiments equally decreased female occupancy (∼1/4 of total) in cells near (0.57 to 0.32) and far (0.68 to 0.44) from roads (Table 3, Fig. 6). That is, road-induced mortality decreased the average female occupancy across the entire landscape by 5 years in the 20-year simulation. In contrast, reductions in female occupancy for experiments that depleted prey largely concentrated the changes near roads, compared to the baseline (Fig. 6). The prey depletion experiment did not substantially change female occupancy patterns far from roads compared to the baseline, but decreased the occupancy by nearly 20% in cells near the roads (Table 3). The reduction in female occupancy near roads compared to the baseline was even greater (40%) in the combined prey depletion and tiger mortality experiment (Table 3, Fig. 6). In this experiment, tigers only occupy cells near roads about 6.6 years of the 20-year simulation, compared to 11.4 years in the baseline.