The Effects of Road and Trailbuilding on Primate Populations in the Yasuni National Park, Eastern Ecuador

Jonathan Greenberg
Ecology Graduate Group
17 December 2000

Abstract
In the early 1990s, an oil-access road and drilling station, as well as a scientific research station, were constructed through the Yasuni National Park in the rainforests of eastern Ecuador. The road passed through the homelands of two groups of hunter-gatherers, who have since become sedentary and joined the regional market economy. Bushmeat, previously hunted only for sustenance, is now sold at local markets, resulting in dramatic reductions in animal densities. Although local researchers implicate the road as the sole cause of the animals’ decline, trails systems built by researchers have not been examined as a factor in the hunters’ increased access. Here, I hypothesize that by using roads and trail systems; hunters have significantly greater access to the monkeys and other animals on which they make their livelihood, an effect that has been seen in black bear populations in North Carolina (Powell et al. 1996). Furthermore, I suggest that through careful planning, including the use of GIS and individual based models (IBMs), researchers can better shape their field sites to minimize their effects. I employed GIS modeling techniques to determine the changes in hunter ranges possible with the building of the roads and the trail systems and the effect on the population levels of the Woolly monkey (Lagothrix lagothricha). The models show that the road more than tripled the access of the hunters to the woolly monkeys as compared to a pristine forest, from 159 hectares with an hour’s walk to 446 hectares. The inclusion of the trail system increases this area to 545 hectares. This study also suggests that the ARC/INFO GIS environment is not suited to individual based modeling (IBMs), a technique that was attempted to provide another insight into the hunter’s behavior.

Introduction
Yasuni National Park (Parqué Yasuni) is located in the lowland rainforests of eastern Ecuador. Like most tropical rainforests, the area has an extremely high diversity of animal species, including at least 11 species of primates known to reside within the park’s boundaries. In the early 1990s, an oil company, Maxus Energy Ecuador, was allotted rights to drill in the region and built a road into the heart of the park for the transportation of equipment and workers. The road, which connects the Rio Napo and the Rio Yasuni passes directly through the homelands of two indigenous groups, the Quichua and the Huaorani (Holmes 1996). The Huaorani, historically mobile hunter-gatherers, were uncontacted until about 50 years ago. Since then, some of these people have become sedentary, constructing permanent living quarters along the road
As part of the agreement for allowing oil extraction to occur in such a diverse ecosystem, a scientific research station was established in the park. Although concerned with preserving the area’s diversity, some of the biologists conducting research in the park may have inadvertently increased hunting pressure upon the animals they study, particularly through the effects of Huaorani hunters using the trails that the researchers have built. In this paper, I ask the question: do roads and trails facilitate prey depletion by significantly increasing access to prey species? I approach the problem through determining both a hunter’s maximum theoretical range and by examining an individual hunter’s behavior in a forest that has been modified through the building of roads and researcher’s trails.

Methods
Trail Map Generation
The map of the researcher trails and the Maxus road within the Yasuni National Park (Figure 1, originally published in Di Fiore 1997) was scanned and then digitized in vector format using ARC/INFO on a 500 mhz. Alpha platform. The map was then converted to a raster format using the smallest feature as a basis to set the scale. In this case, the trails are only 1 meter wide, so the cell size was set at 1 meter. This feature grid was then reclassified to values denoting the amount of time it takes a hunter to cross that pixel in centiseconds (Table 1, Di Fiore, pers. comm.) resulting in an integer grid.
To test the hypothesis that roads and trails increase a hunter’s access to the forest and therefore the fauna, we created three grids: TERRAINTIME in which the hunter has access to both the roads and trails, NOTRAILTIME in which the trails are removed from the environment, and NOROADTIME in which both the trails and the roads are removed. The Huaorani hunters are known to typically begin hunting at a specific prominent trailhead location (Dew, pers. comm.), so a 1 x 1 grid with this starting location (STARTGRID) was created. Appendix 1 summarizes the ARC/INFO commands used to create these grids.

Hunter ranges
I used the most vulnerable and easily depleted local prey species, the woolly monkey (Lagothrix lagothricha) as the model prey organism. Due to constraints on processing time and disk space, I chose an hour-long hunting day (360,000 centiseconds) as opposed to the more realistic 8-hour day (Lu 1999). Through the use of ARC/INFO’s COSTDISTANCE function (see Appendix 2), maps of the time necessary to reach a given pixel within the TERRAINTIME, NOTRAILTIME and NOROADTIME environments were generated and the total accessible area and number of woolly monkeys (at 0.31 woollies/ha, Di Fiore 1997) was calculated.

Random Walk Model: HunterMove v. 1.0
An ARC Macro Language (AML) model was constructed to simulate an individual hunter moving within the TERRAINTIME, NOTRAILTIME and NOROADTIME environments following optimal foraging rules to search for the monkeys (Stephens and Krebs 1986). In this model, the hunter spends one hour hunting. We assume there are no monkeys actually along the road but only in forest and trail pixels (pers. obs.) and all of these pixels contain equal probability of woolly monkeys. Furthermore, once the hunter has moved through a pixel, that pixel then becomes devoid of monkeys. The hunter chooses which direction to go by maximizing his capture per unit time as influenced by the changes in walking speed across different terrains, and repeats this process until he runs out of time. Appendix 3 describes the initialization requirements and AML code for this model.

Results
Hunter ranges
In the area near the researcher trail system, the oil company built 10.2 hectares of road and researchers built 4.5 hectares of trails. A pristine forest, with unaltered prey densities and no roads or trails allows a hunter access to 159 hectares, or approximately 49 woolly monkeys within one hour’s walk (Figure 2). With the increased forest access provided by the road, a hunter has access to 446 hectares of forest or 138 woolly monkeys (Figure 3). By adding the researcher trail system into the equation, a hunter’s access grows to 545 hectares or 169 woolly monkeys (Figure 4). Following the construction of the road, the hunters experience a 180% increase in prey accessibility. The trail system provides an additional 62% increase in accessibility (Figure 5).

Random Walk Model
The results of the random walk model, while not informative to our question, do shed insight into the usefulness of applying ARC/INFO to cell-based modeling. While the model did perform as designed, each run required over twelve hours on a fast 500 mhz Alpha with 640 MB RAM and always ran into memory overrun problems. ARC/INFO’s high disk I/O requirements was most likely the cause of the speed issues. Shorter runs were successful, but to yield meaningful results the model should run many times under each of the terrain conditions for the full 1 hour of hunting time.

Conclusions
The efficacy of using ARC/INFO for individual based modeling
ARC/INFO, while a formidable tool for creating maps, was found to be lacking in modeling power. The latest version (8.0.2) was used on a relatively fast 500 mhz Alpha. Considering the statistical requirements for minimal sample size, examining individual hunter behavior in this programming environment is extremely difficult. A larger cell size and therefore smaller file sizes would have increased the execution speed, but the cell size was chosen appropriately, and could not be altered for this particular study. I would recommend that those interested in applying an individual based model to a geographic area do so through lower level programming languages or through modeling packages designed for this purpose such as SWARM and MATLAB.

The effects of road and trails on hunter ranges
The forests of the Yasuni Park, until the intrusion of the large oil companies, were largely inhabited only by hunter-gatherers who hunted to feed their family and community. Once the Maxus Road was constructed, the Huaorani began to transport hunted animals up the Napo River to sell at the markets as bushmeat. Nearly every restaurant in Coca, the largest town in the region, carries some form of bushmeat as part of their menu (pers. obs.). Not only does the road allow quick transportation of the bushmeat out of the forest, it also allows for greater access to the prey.
Beyond the interests of oil companies, this study shows the importance of a well-planned research site for the viability of a hunted species. Researchers, while their intentions are benign, need to be aware of the long-term effects of their studies. While the road caused an 180% increase in accessibility and is the primary cause of increase, the construction of the trail system, in this study, caused a sizable 62% increase in local accessibility to the woolly monkeys. Issues such as the number of trail-road intersections, distance from hunter homes, and shape of the trail system are all important issues to consider before beginning a field system.

References
Di Fiore, A. F. 1997. Ecology and behavior of lowland woolly monkeys (Lagothrix lagotricha poeppigii, Atelinae) in Eastern Ecuador. Thesis (Ph.D.). University of California, Davis, Davis.
Holmes, B. 1996. The low impact road. New Scientist 151:40.
Lu, F. E. 1999. Changes in subsistence patterns and resource use of the Huaorani Indians in the Ecuadorian Amazon. University of North Carolina at Chapel Hill, Chapel Hill.
Powell, R. A., J. W. Zimmerman, D. E. Seaman, and J. F. Gilliam. 1996. Demographic analyses of a hunted black bear population with access to a refuge. Conservation Biology 10:224-234.
Stephens, D. W., and J. R. Krebs. 1986. Foraging theory. Princeton University Press, Princeton, N.J.

Tables and Figures
Table 1: Time required for a hunter to traverse 1 pixel (1 meter).

Figure 1: Researcher trail system and Maxus Road (from Di Fiore 1997). The red circle denotes the hunter’s starting position.

Figures 2-4: Area accessible by a hunter under different terrain conditions. The labels denote the time in minutes needed to reach a particular pixel. The red circle denotes the hunter’s starting position.

Figure 2: Neither road nor trail system present.

Figure 3: Road present, but no trail system.

Figure 4: Road and trail system present.

Figure 5: Range of hunters under different stages of road and trail development and area of road and trail systems.

Appendices

Appendix 1: Trail map to TERRAINTIME conversion
1. HUNTERARCS was created by scanning Di Fiore's (1997) map as a TIFF file and then digitizing it with ARCTOOLS. Known waypoints were used to georeference the map. The arcs were given values of 1 to signify a trail and 2 to signify a road.
2. HUNTERGRID = LINEGRID(HUNTERARCS,#,#,#,1,ZERO)
This creates a grid with a cell size of 1 meter and fills the non-trail and road pixels with the value of ZERO.
3. HUNTEREXP = EXPAND(HUNTERGRID,8,LIST,2)
The road is 16 meters wide, so we expand all values denoting the road (2) by 8 meters on a side.
4. TERRAINTIME = RECLASS(HUNTEREXP,TERRAINTIME.RMP,NODATA)
The time to cross each pixel is substituted for the identification values from a file called “TERRAINTIME.RMP”. These values are in 10-2 seconds, which creates an integer grid instead of a larger floating point grid. TERRAINTIME.RMP contains the following data:
0 : 480
1 : 240
2 : 90
5. NOTRAILTIME = CON(TERRAINTIME == 240, 480, TERRAINTIME)
This replace all transverse time values associated with a trail with values from the forest, therefore removing the trail.
6. NOROADTIME = CON(NOTRAIL == 90, 480, NOROAD)
This removes the transverse time values associated with the road with values from the forest, therefore removing the road.

Appendix 2: Generation of maps of hunters’ range within 1 hour of the starting point.
1. HUNTERRANGE = COSTDISTANCE(STARTGRID,TERRAINTIME,#,#,360000)
2. NOTRAILRANGE = COSTDISTANCE(STARTGRID,NOTRAILTIME,#,#,360000)
3. NOROADRANGE = COSTDISTANCE(STARTGRID,NOROADTIME,#,#,360000)
4. ArcView was used to add legends, cardinal directions and map scale to the resulting outputs.

Appendix 3: HunterMove 1.0 AML code
1. Required grids for program initialization:
a. STARTGRID
b. MONKEYS
i. OPTIMAL = 1 / TERRAINTIME
ii. MONKEYS = CON(TERRAINTIME == 90, 0, OPTIMAL)
2. &RUN HUNTERMOVE
/* HunterMove version 1.0
/* Jonathan Greenberg, UC Davis
/* Initialize some variables
&set elapsedtime := 0
&set maxdailytime := 360000
setcell startgrid
&set cellsize := [extract 1 [show setcell]]
newlocation = startgrid
hunterpathold = newlocation
/* The main program loop, repeats until the hunter runs out of time
&do &while %elapsedtime% < %maxdailytime%
&type %elapsedtime%
/* Display the hunter's path as he moves
/* gridshades hunterpathold
/* Set the current hunters location as the analysis window
setwindow newlocation
/* Get the current hunter location
&set xmin := [extract 1 [show setwindow]]
&set xmax := [extract 3 [show setwindow]]
&set ymin := [extract 2 [show setwindow]]
&set ymax := [extract 4 [show setwindow]]
/* Expand the zone of interest to one cell surrounding the hunter
&set xexpmin := %xmin% - %cellsize%
&set xexpmax := %xmax% + %cellsize%
&set yexpmin := %ymin% - %cellsize%
&set yexpmax := %ymax% + %cellsize%
setwindow %xexpmin% %yexpmin% %xexpmax% %yexpmax%
/* We need to prevent the hunter from testing the current square
remcentmp = con(isnull(newlocation), 1, 0)
huntertrav = con(isnull(hunterpathold), 1, hunterpathold > 0, 0, 1)
/* The region of interest is the local 3 x 3 monkeys excluding
/* the hunter's current and past locations
localtmp = monkeys * remcentmp * huntertrav
/* We need a 3 x 3 grid of random numbers in case the hunter has
/* a choice of pixels to move to
localrnd = rand()
/* We now create a 3 x 3 grid of the maximum hunting return in the area
/* excluding the current and past locations
maxtmp = focalmax(localtmp,rectangle,5,5) * remcentmp * huntertrav
/* All cells with satisfy the maximum condition are assigned a
/* random number
testtmp = con(localtmp == maxtmp, localrnd, 0)
/* This is a good time to clean up the temporary grids except for TESTTMP
kill newlocation all
kill huntertrav all
kill localtmp all
kill remcentmp all
kill localrnd all
kill maxtmp all
/* Now we have our hunter move. First, we extract a the maximum
/* random value in TESTTMP
docell
maxval }= testtmp
end
&set maxval [show maxval]
/* We now need to step through the 9 locations the hunter can move
/* to determine his new location
/* NOTE: this routine was used because ARCINFO doesn't seem to be able
/* to crop the 3 x 3 grid in which all but one value are NODATA
/* &set xmax := %xexpmax%
/* &set ymax := %yexpmax%
&do xtestlo := %xexpmin% &to %xexpmax% &by %cellsize%
&do ytestlo := %yexpmin% &to %yexpmax% &by %cellsize%
/* Create the one pixel analysis window
&set xtesthi := %xtestlo% + %cellsize%
&set ytesthi := %ytestlo% + %cellsize%
setwindow %xtestlo% %ytestlo% %xtesthi% %ytesthi%
/* Extract the pixel value
docell
maxtest }= testtmp
end
&set maxtest [show maxtest]
/* If the pixel is the largest value in the 3 x 3 grid,
/* set it as the new hunter location (value = 1) and update the time
/* elapsed.
&if %maxtest% = %maxval% &then
&do
newlocation = 1
docell
movetime {= terraintime
end
&set movetime [show movetime]
&type %movetime%
&set elapsedtime = %elapsedtime% + %movetime%
&end
&end
&end
/* We now need to build our past movement grids. First, we'll set
/* the analysis window to include past and current locations.
setwindow maxof
hunterext = hunterpathold + newlocation
setwindow hunterext
/* Now, we'll remove the pesky NODATA values in the grids by replacing
/* them with 0.
hunterpathemp = con(isnull(hunterpathold),0,hunterpathold)
newlocationz = con(isnull(newlocation),0,newlocation)
/* Our new paths should include past and current hunter locations.
hunterpathnew = hunterpathemp + newlocationz
/* Now, the present becomes the past, and we create a past grid
kill hunterpathold all
setwindow maxof
hunterpathold = hunterpathnew
/* Finally, we clean up our temporary grids and repeat the loop
kill hunterext all
kill hunterpathemp all
kill newlocationz all
kill hunterpathnew all
kill testtmp all
&end
&return