Affordable Access

Using Site-Level Factors to Model Areas at High Risk of Deep-Vehicle Collisions on Arkansas Highways

Authors
  • Philip A., Tappe
  • Donald I. M., Enderle
Type
Published Article
Publication Date
May 19, 2007
Source
Road Ecology Center John Muir Institute of the Environment
Keywords
License
Unknown
External links

Abstract

Deer-vehicle collisions (DVCs) are increasing across the United States, including Arkansas. These collisions involve risk of human injury and fatality, property damage, and loss of wildlife. The annual number of DVCs in Arkansas may be as great as 18,000 (13.6% of the reported legal deer harvest in 2005) with an estimated property damage of $35 million. Numerous studies have examined the impacts and causes of DVCs; however, few studies have utilized a state-wide approach to increase understanding of the factors involved. We evaluated the influence of site-level factors on the number of DVCs reported during 1998-2001 along all state and federal highways in Arkansas. Site-level factors included landcover patterns, landcover characteristics, and number of stream/highway intersections within 400, 800, and 1200 m of collision sites; landcover crossing types and maximum topographic relief within 100 m of collision sites; and distances to nearest forest and to nearest water. A total of 3,170 DVC locations were compared with an equal number of randomly-chosen highway locations based upon proportions of DVCs within each physiographic region of Arkansas. Logistic regression analysis was used to develop and test a state-wide model and six physiographic region models to identify high risk areas along highways. Akaike information criterion values were used to select the best model for the entire state and for each physiographic region. We randomly selected 25% of the DVC sites and randomly-located highway sites to exclude from model development in order to test the predictive ability of each model. Over 1,000 variables were considered prior to model development. However, exclusion of intercorrelated variables and variables that did not differ between collision and random sites reduced the variable set to 31. These 31 variables revealed a variety of differences between known DVC locations and randomly-selected locations. Twenty-six variables were more associated with known DVC locations than with random locations. Five variables were more associated with randomly selected highway locations than with known DVC locations. The state-wide model had an overall correct classification rate of 62%. Most models developed for individual physiographic regions performed as well or better than the state-wide model: Arkansas River Valley (62%), Boston Mountains (69%), Gulf Coastal Plain (59%), Mississippi River Delta (70%), Ouachita Mountains (67%), and Ozark Mountains (60%). Almost all variables included in the state model were also included in at least one physiographic region model, and most variables of each physiographic region model were also found in the state model. Five groups of factors that were strongly correlated with DVC locations were apparent in all models: (1) the presence and amount of water in terms of distance to the nearest source, number of streams intersecting within 400 m, and amount of water within 1200 m; (2) the presence of a diverse association of land cover types in close proximity to a highway; (3) the amount and size of urban area within 1200 m; (4) forested area (deciduous and/or coniferous) in close proximity to a highway, particularly in terms of higher density of coniferous forest and greater size and irregularity of deciduous forest patches; and (5) the density of pasture and crop patches, and the density of pasture edge in particular, within 1200 m of a highway. These results and models may be used to produce maps indicating potential segments of highways at high risk for the occurrence of DVCs. Additionally, they may aid in planning and road construction. Finally, these results provide a foundation for future research in examining more specific deer-vehicle interactions, and can aid in the evaluation of appropriateness and effectiveness of proposed methods to reduce DVCs in Arkansas.

Report this publication

Statistics

Seen <100 times