Rakesh Malhotra, Paul Johns, Marguerite Madden, and Gary Wein

Deer-Vehicle Collisions: Is There a Pattern?

Deer-vehicle collisions are a common occurrence on highways of Eastern United States. This study uses aerial photographs and GIS to analyze deer-vehicle collisions at the Savannah River Site (SRS), South Carolina. Vegetation, topography, presence of water bodies, and road conditions are analyzed to better understand their impact on such collisions. The spatial correlation between these factors, within 1000m buffered zones around collision sites, are compared to the correlation at random points along roads. The results are then used to create a spatial model to identify conditions conducive to deer-vehicle collisions. It is hoped that by understanding these factors, management practices can be used to reduce such incidents.


INTRODUCTION

    The Savannah River Site (SRS) is a nuclear production facility of the U.S. Department of Energy and occupies approximately 80,000 hectares of land in west central South Carolina (Figure 1). In 1951, Dr. Eugene P. Odom and other researchers from The University of Georgia were invited by the Atomic Energy Commission (AEC) to conduct a census of plants and animals in this area. This research led to the Savannah River Ecology Laboratory (SREL) being established at SRS in 1961.
    When land was acquired from the public by AEC in 1950, most existing agricultural fields were converted to pine plantations and public hunting of deer was suspended. This led to a dramatic increase in the deer population and by 1965, deer-vehicle collisions problem had become so widespread that public hunting was re-introduced on the property. During the annual public hunts, 1,000 to 1,200 deer out of a total estimated population of 7,000 deer are harvested. This problem was not studied further until 1990 when SREL personnel started collecting information about deer-vehicle collisions. Since then, data have been collected on over 900 collisions that have taken place on the SRS. The restricted access to SRS makes it unique as most deer-vehicle collisions are reported and recorded.
    The study uses geographic information system (GIS) and remote sensing techniques to collect data on and analyze deer-vehicle collisions to evaluate the impact of vegetation, elevation, and water bodies on this phenomenon. Based on the analysis, a risk model will be created that can be used by the managers at SRS to warn drivers of potential high risk sites or modify land management policies to reduce the propensity of such accidents. Such management practices may help in reducing the attraction that the deer might have for roadways.
 

Figure 1.  Savannah River Site Figure 2. Deer-Vehicle Collisions at SRS in Fall, 1995

Figre 1. Savannah River SiteFigure 2. Deer-vehicle Collisions at SRS in Fall, 1995
 

HYPOTHESIS

    Since the 1970s, yearly deer-vehicle collisions in the United States have increased from approximately 200,000 to an estimated 500,000 in 1995. The National Safety Council Report estimated that in 1995 such collisions cost over $125 million in medical expenses, over $940 million in vehicle repairs and over $350 million in wildlife loss. Apart from being economically costly, deer-vehicle collisions are also a threat to humans traveling on roads (Huges et al., 1996). Various solutions offered to this problem such as the construction of underpasses, reflectors, chemical repellents, and noise creating devices are either very expensive or have not been conclusively shown to reduce collisions. However, data on deer-vehicle collisions at SRS offer a unique opportunity to study the interactions between deer movement and vegetation.

    The two hypotheses that are tested as part of this study are:

    Hypothesis I : Deer-vehicle collisions are not random occurrences in space but are clustered in areas which can be identified using spatial analyses.

    Hypothesis II : The occurrence of deer-vehicle collisions can be related to physical attributes such as vegetation and slope of the areas where collisions occur and these attributes can be used to predict future collisions
 

Figure 3. Vegetation along the roads at SRS Figure 4. A 500m (radius) buffer around a collision
 


 

Figure 5. Legend

METHODS

    SREL personnel, who are called to deer-vehicle collision sites, collect data on the animal (age, sex) and the location (date, time, and position) of the accident. This information was entered into a GIS database to create a point coverage of the deer-vehicle collisions. Each collision was identified by the year and given a unique identification number. This process was started in 1990 and continues till date.Figure 2 shows a map of deer vehicle collisions that occurred on SRS in Fall 1995. The next step was to use ArcView to extract paved roads traversing the site from an existing roads coverage and buffer them 500 meters on both sides. We then used ArcView Image Analysis extension to display true color orthophotos (1:16,000), that are available for March 1996 at SREL, to interpret and delineate vegetation within this buffer zone (Figure 3). The deer accidents were chosen for Fall 1995 because this was the closest date to air-photo acquisition date of Spring 1996. The various vegetation categories created for this study are listed in Table 1. Two factors that helped determine the vegetation classification were - classification schemes used for previous vegetation studies at SRS and habitat suitability of each category for deer.
    In order to test our hypotheses we carried out the analysis for fifty-one collisions that occurred in Fall, 1995. We took each collision point and buffered it 500 m (1000 m diameter) and then used this buffer to clip the vegetation cover and extract the amount of vegetation (of each type) that was present within this area (Figure 4). As this process had to be repeated several times, we automated the process by creating an Avenue script. For each collision, we also selected 4 random points along the roads and extracted the same information. Fifty-one deer-car collisions and 204 random points were buffered  and vegetation information was extracted. We then created an average buffer area by summing up the area for each vegetation type and dividing this number by the number of points used for collisions and random locations respectively. The data collected on vegetation and water bodies was then tabulated and analyzed. This entire process was also repeated for a 250m (500 m diameter), and 125m (250 m diameter) buffer around the points. This was done in order to find out whether the vegetative characteristics around the points varied with the buffer size.

RESULTS

    The results shown in Table 2  summarize the average total area (in square meters) of each vegetation category contained within a buffer around each point. The comparison between collision points and random points shows that collisions tend to be spatially associated with fewer buildings, water bodies, and clear-cuts but are more commonly associated with open-grass and hardwood areas. This relationship was also observed for the smaller 250m and 125m buffer radii. For all the three radii, the amount of vegetation in collision points is 1.35 times and 1.5 times for hardwood areas and open-grass areas respectively when compared to random points (i.e. there is approximately 50% more open-grass area around collisions than around random points). Similarly, the amount of area covered by buildings around collision points is about a third of the area around random points. The relationship for water bodies is more complex. At the 500m radius buffer, the area covered by water around collisions is half the area covered by them around random points but this figure falls to a fourth at the 250m radius and then rises back to a third for the 125m radius. Although there appears to be less water around collisions, the ratio of water around collision points and random points does not remain constant as we reduce the buffer radius.
    Most collisions probably occur away from buildings because deer avoid built-up areas and also because motorists tend to slow down when approaching buildings. Fewer water bodies around collisions may be explained by the observation that as roads approach water bodies, the embankment gets steep and it is harder for deer to cross the roads at such points. However, it is harder to intuitively explain why there are fewer collisions near clear-cuts.
    Relative to random points, collision points were surrounded by more hardwood areas. This may be explained by the fact that as most of SRS is covered with planted pine without any understory growth, deer might find forage in hardwood patches. Collision points also were associated with open-grass areas than random points. This may counter the prevalent belief  that wide patches of open grass along roads reduce the possibility of accidents by giving drivers ample visibility along the road shoulders. In fact, as the animals stay hidden in the pine forests during the day, they may be utilizing the open-grass areas along roadsides more during dusk, dawn, and night thereby increasing the risk of deer-vehicle collisions.

FUTURE STUDIES

    This is a preliminary report and we plan to expand the study to incorporate collision data from other years (e.g. collisions from 1994 and 1996). This longer term study will include the analysis of two other important factors; edge and elevation. It is important to consider the impact of these two characteristics may have on deer-vehicle collisions because they are closely associated with vegetation. The analysis presented above uses only area as a measure of the presence or absence of a vegetation type. Further analysis will be carried out that compares the number of polygons of each vegetation type in collision and random points. Vegetation maps for SRS are being created using small scale (1:48,000) aerial photographs and satellite (Landsat TM) images. These maps will be used to determine if image resolution plays a role in identifying potentially dangerous sites. Finally, we intend to carry out categorical modeling and non-parametric statistical analysis to build a model that can help predict the probability of deer-vehicle collisions.

CONCLUSIONS

    Deer-vehicle collision analysis is important because the increasing number of cars and deer will result in more damage in the coming years. The problem has been exacerbated by the desire of several city residents to move into the suburbs. This not only leads to more travel time but also enhances the chances of a dangerous encounter between cars and deer by bringing the people in close proximity to deer habitat. From the above study preliminary results demonstrate that deer-vehicle collisions do not occur at random but may be influenced by the surrounding vegetation. The study also helps in identifying open-grass and hardwood areas as vegetation cover that may be contributing to such accidents. As the analysis of a single year of data only provides preliminary spatial patterns, this study will be extended to other years. As the results show the influence of surrounding vegetation, it may be surmised that other factors such as edge, and elevation also play a role in such collisions and hence they should be further investigated.

ACKNOWLEDGMENTS

We wish to acknowledge the assistance provided by Savannah River Ecology Laboratory, operated by the University of Georgia and supported by Financial Assistance Award Number DE-FC09-96SR18546 from the U.S. Department of Energy to the University of Georgia Research Foundation and the Center for Remote Sensing and Mapping Science (CRMS), Department of Geography, University of Georgia.
 
 

Table 1. Vegetation classification
Vegetation Description
Buildings Buildings and built up area
Roads All paved roads
Water Bodies Lakes, ponds, and rivers wider than 20 meters
Open Bare Open ground with no vegetative cover
Open Grass Open ground covered with grass (minimum width 15 meters)
Clear-cut Clear-cut pine stands with characteristic signature with trees harvested within a year
Pine Hardwood Mixed forest where pines and hardwoods cannot be separated but there are more pines than hardwoods
Hardwood Pine Mixed forest where pines and hardwoods cannot be separated but there are more hardwoods than pines
Pine - high density Dense pine forest where no vegetation is visible below the canopy
Pine - medium density Pine forest where some vegetation is visible below the canopy
Pine - low density Pine savannah with 2-5 trees in a 20x20 m area
Young pine / Open Pine tress planted within the previous 2-4 years and they have open spaces between them
Hardwood Hardwood forest
Scrub/shrub bushes, shrubs, etc. 
Edge The boundary between two vegetation types
Table 2. Percentage of vegetation type in a circular area around collisions and random points.
Vegetation Type Area (sq. meters) of vegetation in a 500m radius buffer around points Area (sq. meters) of vegetation in a 250m radius buffer around points Area (sq. meters) of vegetation in a 125m radius buffer around points
Collision Random Collision Random Collision Random
Buildings 18491.8 55688.4 3973.7 11237.8 882.7 2842.1
Water Bodies 5128.3 10907.8 533.0 2184.1 104.2 382.3
Open Grass 119595.1 80460.9 34685.4 23382.7 10733.7 6897.3
Open Bare 1942.3 2076.8 0.0 353.4 0.0 72.3
Clear-cut 6320.0 10764.8 438.2 2216.8 0.0 335.1
Pine Hardwood 59706.1 83341.9 18464.0 23481.8 5074.7 5792.4
Hardwood Pine 37449.2 39857.2 9475.1 10562.7 2401.1 2590.2
Pine-high density 259706.8 263349.1 59696.1 66266.2 13068.8 15935.7
Pine-med density 130447.5 119543.6 31249.1 26938.7 6780.2 6256.3
Pine-low density 7011.9 6927.0 1503.8 1358.5 332.9 287.6
Young pine / Open 14303.5 12010.6 4262.7 3724.7 1298.7 1018.9
Hardwood 103946.1 79978.1 23362.7 17332.2 5068.1 3658.9
Scrub/shrub 3297.2 5371.6 1446.3 1209.1 229.2 325.1
Roads 14071.0 11139.1 6264.2 5105.4 2864.2 2444.2

REFERENCES

Huges, W.E., A.R. Saremi, and J.F. Paniati, 1996. Vehicle-Animal Crashes: An Increasing Safety Problem, ITE Journal, August:24-28.



Rakesh Malhotra
Graduate Student
204 GGS Building
Department of Geography
University of Georgia
Athens, GA 30602
(706)-542-2856
rakesh@uga.edu

Mr. Paul E. Johns
Savannah River Ecology Laboratory
University of Georgia
Drawer E,
Aiken, SC 29802
johns@srel.edu

Dr. Marguerite Madden
Associate Director for Environmental Studies
Center for Remote Sensing and Mapping Science
Department of Geography
University of Georgia
Athens, GA 30602
mmadden@crms.uga.edu

Dr. Gary Wein
Savannah River Ecology Laboratory
University of Georgia
Drawer E,
Aiken, SC 29802
wein@srel.edu