Proposed Guidelines for Fixed Objects in the Roadside Design Guide (2022)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

63 Chapter 6. Application of the RAP Model for Predicting Tree- and Utility-Pole-Related Crashes The alternative approach to crash prediction for tree- and utility-pole-related crashes is adapting the RAP model for ROR crashes, which has been developed by the iRAP, the international partner of the U.S. Road Assessment Program (usRAP). The existing RAP model for ROR crashes is presented in Appendix A. This appendix develops an approach for applying the RAP model for ROR crashes in a manner that can be used in the AASHTO RDG to provide design guidance for roadside fixed objects, specifically focusing on roadside trees and utility poles. 6.1 Summary of the RAP Model for Predicting Run-Off-Road Crashes The RAP model for predicting ROR crashes, as presented in Appendix A, consists of a model that is equivalent to a SPF, combined with adjustment factors, which are termed risk factors by iRAP, and a calibration factor which should be based on local data for the agency applying the model. The structure of the model is similar to the structure of the crash prediction methods in the AASHTO HSM (AASHTO 2010), although the models have been based on research knowledge concerning the effects of roadway and roadside features on crashes, rather than on empirical models. The SPF portion of the RAP model for ROR crashes accounts for the effects of roadway AADT, traffic speed, and median traversability. The adjustment factors used in predicting ROR crashes include factors for nine roadway features and two roadside features. The roadside features are considered separately for each side of the road. The sides of the road are designated as the right and left sides of the road, typically interpreted as referring to the roadside on the outside of the roadway in the increasing milepost direction of travel and the roadside on the outside of the roadway in the decreasing milepost direction of travel, respectively. The adjustment factors representing roadway features include factors for: • lane width • horizontal curvature • advance visibility of curve • grade • shoulder rumble strips • delineation • road surface condition • skid resistance • paved shoulder width The adjustment factors representing roadside features include factors for: • distance from outside edge of the traveled way to a roadside object • type of roadside object

64 6.2 Strengths and Weaknesses of the RAP Model for Application to Crash Prediction for Roadside Trees and Utility Poles The strengths of the RAP model for ROR crashes for application to provide design guidance for roadside trees and utility poles are as follows: • The model explicitly addresses the relative severity of crashes involving roadside trees and poles. • The model is directly sensitive to the offset distance from the edge of the traveled way to a roadside tree or pole. • The model is directly sensitive to roadway geometrics. • The model is directly sensitive to the speed and volume of traffic on the roadway. • The model deals separately with crash likelihood and crash severity factors. • The model can be calibrated to local conditions. The weaknesses of the RAP model for ROR crashes for application to provide design guidance for roadside trees and utility poles are as follows: • The model estimates fatal and serious injury crashes, but does not consider minor injury or PDO crashes. • The model considers roadside poles in general, which could include utility poles, nonbreakaway luminaire supports, nonbreakaway signposts, and other types of poles but does not explicitly address utility poles. • The model does not explicitly consider the number of trees or poles on the roadside within a given roadway segment. • The model does not account for the differing longitudinal extent of individual trees and tree groups. • The model addresses offset distance from the edge of the traveled way to the roadside tree or pole by grouping the offset distance into categories for specific distance ranges. • The model does not account for the slope or grading of the roadside between the roadway and the roadside object. • The model includes the capability to predict crashes on unpaved roads, but such roads are unlikely to be candidates for roadside improvement. • The model is structured to make crash predictions for crashes involving motorcycles separately from crash predictions for other types of motor vehicles. However, the crash predictions for motorcycles need estimates of motorcycle volume data which potential users do not have available for most sites. • The description of the RAP model includes some terminology that needs modification or clarification. Each of these weaknesses can be addressed as explained in the next section. Chapter 7 presents a modified version of the RAP model, recommended for application in a future edition of the AASHTO RDG, which has been updated to address many of the listed model weaknesses.

65 6.3 Approach to Addressing Weaknesses of the RAP Model for Application to Roadside Trees and Utility Poles The weaknesses in the RAP model for application to roadside trees and utility poles can be addressed as explained below. Adapting the RAP Model to be Estimate All Injury Crashes and Then Breaking That Estimate Apart into Crash Severity Levels Many U.S. users will likely be interested in estimating crash frequencies for all injury crashes, rather than limiting the estimate to fatal and serious injury crashes. To accomplish this, the RAP models can be reorganized and calibrated to estimate total fatal-and-injury crashes related to trees and utility poles and then to break down that estimate into estimates of fatal crashes and injury crashes of specific injury severity levels. In addition, this approach can be adapted to estimating persons killed and persons injured with specific injury severity levels in crashes involving trees and utility poles. The distributions of crashes and persons injured by severity level have been estimated from the Kentucky and Washington databases whose development is described in Chapter 4. Table 34 presents the distribution of crash severity levels for trees based on statewide data for rural nonfreeway tree crashes on state highways in Kentucky and Washington. Statewide data for rural tree crashes in Kentucky and Washington were used for this table, rather than just the data for the Kentucky and Washington study sites because the detailed site-specific characteristics data gathered for this study were not needed in developing this table. Table 35 presents estimates for persons killed and persons injured per-tree crash for rural nonfreeway crashes on state highways in Kentucky and Washington. Tables 36 and 37 present comparable data to Tables 34 and 35 based on statewide data for rural utility pole crashes in Kentucky and Wisconsin. The portions of Tables 34 through 37 for Kentucky and Washington data combined are recommended for use to provide default values for crash severity distributions in a modified RAP model (see Chapter 7). Tables 34 through 37 can be modified using data for tree- and utility-pole-related crashes from an agency’s own statewide crash database for specific roadway types to make these tables more appropriate for an agency’s local conditions. This can be accomplished quite readily without the need to collected additional data on roadside objects. Default values for the CF used with the RAP model are presented in Section 7.2. An agency can calibrate the RAP model to their local conditions by determining values of the calibration factor for trees and utility poles applicable to their own roadways. Quantifying agency-specific calibration factors would require the agency to assemble a database similar to the databases assembled for sites with trees and utility poles in Kentucky and Washington as described in Chapter 4. Thus, quantifying agency-specific calibration factors involves substantially more effort than customizing the crash severity distributions in Tables 34 through 37.

66 Table 34. Crash Severity Distribution for Use as Default Values for Estimating Tree Crashes on Rural Nonfreeways with the Modified RAP Model Crash Severity Levela Kentucky Washington Combined Number of Tree Crashes (5 years) Percentage of Crashes Number of Tree Crashes (5 years) Percentage of Crashes Percentage of Crashesb Fatal 266 7.8 18 7.7 7.8 Incapacitating injury 624 18.2 40 17.1 17.7 Nonincapacitating injury 1,158 33.8 97 41.4 37.6 Possible injury 1,382 40.2 79 33.8 36.9 TOTAL 3,430 100.0 234 100.0 100.0 a most severe injury resulting from the crash. b computed giving equal weight to each state. Table 35. Deaths Per Fatal Crash and Persons Injured Per Crash for Use as Default Values for Estimating Tree Crashes on Rural Nonfreeway Crashes with the Modified RAP Model Crash Severity Levela Deaths Per Fatal Crash Persons nonfatally injured per crash Incapacitating Injury Nonincapacitating Injury Possible Injury KENTUCKY Fatal 1.05 0.03 0.06 0.08 Incapacitating injury -- 1.05 0.09 0.10 Nonincapacitating injury -- -- 1.14 0.17 Possible injury -- -- -- 1.31 WASHINGTON Fatal 1.11 0.05 0.13 0.21 Incapacitating injury -- 1.07 0.18 0.30 Nonincapacitating injury -- -- 1.16 0.13 Possible injury -- -- -- 1.13 COMBINEDb Fatal 1.08 0.04 0.10 0.15 Incapacitating injury -- 1.06 0.14 0.20 Nonincapacitating injury -- -- 1.15 0.15 Possible injury -- -- -- 1.33 a most severe injury resulting from the crash. b computed giving equal weight to each state. Table 36. Crash Severity Distribution for Use as Default Values for Estimating Utility Pole Crashes on Rural Nonfreeway with the Modified RAP Model Crash Severity Levela Kentucky Washington Combined Number of Pole Crashes (5 years) Percentage of Crashes Number of Pole Crashes (5 years) Percentage of Crashes Percentage of Crashesb Fatal 33 3.9 0 0.0 2.0 Incapacitating injury 118 13.8 4 1.7 7.8 Nonincapacitating injury 330 38.7 26 42.6 40.7 Possible injury 371 43.6 31 55.7 49.5 TOTAL 852 100.0 61 100.0 100.0 a most severe injury resulting from the crash. b computed giving equal weight to each state.

67 Table 37. Deaths Per Fatal Crash and Persons Injured Per Crash for Use as Default Values for Estimating Utility Pole Crashes on Rural Nonfreeways with the Modified RAP Model Crash Severity Levela Deaths per Fatal Crash Persons Nonfatally Injured Per Crash Incapacitating injury Nonincapacitating injury Possible injury KENTUCKY Fatal 1.03 0.01 0.02 0.03 Incapacitating injury -- 1.05 0.15 0.18 Nonincapacitating injury -- -- 1.16 0.18 Possible injury -- -- -- 1.32 WASHINGTON Fatal 1.00 0.00 0.10 0.00 Incapacitating injury -- 1.02 0.43 0.55 Nonincapacitating injury -- -- 1.04 0.04 Possible injury -- -- -- 1.06 COMBINEDb Fatal 1.02 0.01 0.06 0.02 Incapacitating injury -- 1.04 0.29 0.37 Nonincapacitating injury -- -- 1.10 0.11 Possible injury -- -- -- 1.19 a most severe injury resulting from the crash. b computed giving equal weight to each state. Calibrating the RAP Model for Trees and Utility Poles and Adapting the RAP Model to be Specifically Applicable to Utility Poles The existing RAP model has a single crash severity factor for roadside object type that is applicable to all types of poles including utility poles, nonbreakaway luminaire supports, nonbreakaway sign posts, and other types of poles. A calibration process has been performed using the Kentucky and Washington databases described in Chapter 4 to adapt the model to U.S. conditions. In this calibration, the data used for poles was specific to utility poles, and did not include crashes with other types of poles. A calibration for trees was also performed. The calibrations were performed by applying the RAP procedures to the Kentucky and Washington sites for which data were assembled in this project and summing the predicted annual crash frequencies for all sites. Observed crash frequencies for the same sites were obtained for a recent five-year period from state crash data. Calibration factors were then computed (separately for tree-related crashes and utility-pole-related crashes) as follows: 𝐶𝐹 = × (22) where: CF = Calibration factor The resulting values of the calibration factors were 1.02 for trees and 0.80 for utility poles. These values are very close to 1.00 indicating good agreement between the RAP model and actual crash data. The calibration factor indicates that collisions with trees are more frequent than collisions with utility poles. This may be because trees consist of both individual isolated trees and continuous tree groups. A reportable crash may be more likely when an errant vehicle strikes

68 multiple trees in a tree group than when it strikes an individual tree. By contrast, utility poles generally consist of individual isolated objects, rather than groups of poles close together. Section 7.2 shows how the calibration factor values are used in the crash prediction model. Section 7.6 shows how agencies can develop calibration factors from their own data. Adapting the RAP Model to Consider the Number of Trees or Utility Poles Within a Road Segment The RAP model estimates roadside severity factors based on the single most severe object type on the roadside on each side of the road within roadway segments with a length of 100 m or 0.062 mi. No consideration is given in the existing RAP model to whether there is one tree or pole or many trees and poles within the 327-ft (or 0.062-mi) roadway segment. The Kentucky and Washington database described in Chapter 4 has been used to calibrate the RAP model so that it can be applied to estimate crash frequencies and persons injured on a per-tree or per-pole basis. This will enable the model to be applied independent of RAP’s 327-ft or 0.062-mi segment length assumption. Distinguishing Between the Effects of Individual Trees and Tree Groups on Crash Frequency A challenge in adapting the RAP model was how best to handle tree groups (i.e., continuous lines of trees that extend essentially continuously for a longitudinal distance along the roadside). With a 327-ft (or 0.062-mi) roadway segment, tree groups on one or both sides of the road may extend longitudinally along the roadside ranging from 10 ft in length to 327 ft (equivalent to the full length of a roadway segment in the Chapter 4 database). A theoretical model was used to develop a tree group equivalency factor (TGE) for use in the model to represent a tree group as an equivalent number of individual trees. The TGE factor is used to establish the equivalency between a tree group (a continuous row of trees, characterized by its length in feet within a given roadway segment) and individual, isolated trees (characterized by the number of trees within a given roadway segment). This equivalency factor was quantified using the roadside departure envelope used in the RSAP, illustrated in Figure 2.

69 Figure 2. Roadside Departure Envelope from RSAP [adapted from Mak and Sicking (2003)] By simple trigonometry, the length along the roadway of the roadside departure envelope (Lenv) can be computed as: 𝐿 = 𝐿 + 𝐿 (23) 𝐿 = ( ) + ( ) (24) 𝐿 = 𝐿 (25) where: Lenv = length of departure envelope measured along the roadway of interest (ft) L1 = length of the portion of the departure envelope associated with the width of the vehicle and the width of the roadside object (ft) L2 = length of the portion of the departure envelope associated with the length of the roadside object (ft) Lobj = length of the roadside object (ft); i.e., the dimension of the object parallel to the roadway Wobj = width of the roadside object (ft); i.e., the dimension of the object perpendicular to the roadway

70 Wveh = width of the vehicle that runs off the roadway (ft) Doffset = offset of the roadside object from the outside edge of the traveled way (ft) α = angle at which the vehicle runs off the road (degrees) An appropriate value of TGE can be inferred from a calculated value of Lenv. Roadside trees typically range from 4 inches in diameter (the smallest tree typically considered in roadside analyses) to several feet in diameter. A typical or representative tree diameter of 1 ft has been selected. This 1-ft value also appears appropriate because it is approximately equal to the typical diameter of a utility pole. (Wooden utility poles typically range in diameter from 8 to 14 inches.) Therefore, for this investigation, Lobj and Wobj in Equations (24) and (25) will be set equal to 1 ft. The most common vehicle that runs off the road is a passenger car, typically about 6 ft in width. Therefore, Wveh will be set equal to 6 ft. A representative encroachment angle for vehicles running off the road is 8 degrees. Therefore, α will be set equal to 8 degrees. Using these representative values, Lenv is calculated as: 𝐿 = ( ) + ( ) + 1 = 51.23 𝑓𝑡 (26) This indicates that for a typical tree, the length of the departure envelope along the road is 51.23 ft. Thus, any standard passenger car running sufficiently far off the road (i.e., a distance at least equal to Doffset shown in Figure 2) within a 51.23-ft interval is likely to strike the tree. A 51.23-ft interval represents approximately 0.97 percent of a mile (51.23/5280 = 0.0097). Thus, an appropriate preliminary value of TGE for use in the model (see Section 7.2) would be 0.0097. This also implies that if roadside trees were spaced more closely together than 51.23 ft, then a single vehicle running off the road readily could strike more than one tree. Treating Offset Distance to Roadside Objects as a Continuous Rather Than a Categorical Variable The RAP model treats the distance from the edge of the traveled way to a roadside object as a categorical variable with four categories: • 0 to 3 ft • 3 to 15 ft • 15 to 30 ft • more than 30 ft These categories could be retained for application of the RAP model in the U.S., but the research team believes that the factors shown in Table A-11 should be modified to a piecewise linear function over the range of offsets from 1.5 to 40 ft, as shown in Figure 3.

71 Figure 3. Factor for Distance from Traveled Way to Roadside Object Expressed as a Piecewise Linear Function Effect of Slope or Grading of the Roadside Between the Roadway and a Roadside Object The RAP model does not account for the effect on crash likelihood or severity of the slope or grading of the roadside between the roadway and the roadside object. The crash data gathered for Kentucky and Washington appear to be too sparse to enable development of such a factor and we have not found a suitable factor in any other source. This appears to be an effect that simply cannot be incorporated in the modified RAP model at present. Elimination of References to Unpaved Roads As part of the pavement skid resistance factor, the RAP model includes the capability to predict crashes on unpaved roads. However, roadside improvements are unlikely to be priorities on unpaved roads. Therefore, this factor will be limited to paved roads only. Elimination of Need for Motorcycle Volume Estimates The existing RAP model makes predictions for crashes involving motorcycles separately from crashes involving other types of motor vehicles. This approach needs motorcycle volume estimates as input data. However, potential users are unlikely to have motorcycle volume data for most sites. Therefore, the RAP procedure has been modified to assume a typical vehicle mix of 3 percent motorcycles and 97 percent other motor vehicle types for all sites. The adjustment factors for horizontal curvature, advance visibility of curve, road surface condition, and skid resistance in Chapter 7 have been modified accordingly.

72 Updates to Terminology in the Predictive Method Some changes have been made to terminology in the modified predictive method. For example, the term operating speed, which in the RAP model implies mean motor vehicle traffic speed, has been changed to design speed to better match U.S. practice as presented in the AASHTO RDG. The design speed has been estimated as 5 mph higher than the mean speed of traffic. The term risk factor used in the RAP model has been changed to adjustment factor for consistency with recent AASHTO practice in safety-related documents. To clarify its meaning, the name of the adjustment factor, quality of curve, has been changed to advance visibility of curve, and its levels have been renamed as substantial and limited.