Evaluation of Strategies to Mitigate Culvert-Involved Crashes

State-of-the-Practice Survey

In addition to reviewing the extant literature, a questionnaire was sent to staff involved in hydraulic and geometric design at state DOTs across the U.S. to identify current practices related to the implementation of safety treatments for roadside culverts. The questions in the survey were related to culverts installed perpendicular or diagonal to the highway (excluding culverts parallel to the highway such as those under driveways or side road crossings, as this was beyond the scope of this study). The survey was conducted using an electronic survey platform and the study design was approved as exempt by the Institutional Review Board (IRB) at Iowa State University. Complete responses were received from 18 state DOTs.

Agencies were asked about the methods used to reduce the risks associated with culvert-involved crashes, as well as the contextual factors that are generally considered in selecting among these alternatives. Figure 1 shows that the most commonly used alternatives to limit the risk of ROR crashes are: 1) to shield edge drops with steel guardrail; 2) to extend the length of the culverts to provide recoverable side slopes; 3) to install traversable culvert grates; and 4) to shield edge drops with an approved bridge rail system. Examples of each of these treatments are shown in the photographs in Figure 2. While additional comments indicated the preferred method would be to locate the culvert drop off outside the clear zone, it was noted that is not possible in many situations. In that case, shielding the culvert is preferred.

OPEN IN VIEWER

Figure 1.

Methods for limiting risk of culvert-involved crashes.

OPEN IN VIEWER

Figure 2.

Example applications of culvert safety treatments.

Agencies were also asked about the factors that are used to determine when and where culvert treatments are applied. Twelve of the 18 responding agencies indicated they had some type of written policy that indicates when to provide ROR protection for culverts. Most of these policies are included in state highway design manuals. Figure 3 shows the prevalence of various screening factors that respondents indicated affected their selection of treatment alternatives. The most commonly used factors included design speed/speed limit, lateral offset from the edge of the traveled way to culvert opening, traffic volume, embankment slope, crash history, culvert size, and embankment height.

OPEN IN VIEWER

Figure 3.

Factors affecting selection of treatment alternatives.

Overall, the guidance provided by agencies tended to be very general in nature. To this end, the principal objective of this study was to evaluate the effectiveness of various safety treatments for culverts under several design scenarios. This approach allows for more detailed insights as to the circumstances under which selected treatments provide advantages or disadvantages as compared with other alternatives.

Roadway Database

The Geographic Information Management System (GIMS) is a database maintained by the Iowa DOT, which provides information on roadway characteristics and traffic information. Within this database, there are three datasets that are commonly used in safety analysis. First, a traffic information file contains information on the annual average daily traffic (AADT) and percentage of vehicles by type (e.g., trucks, motorcycles, buses). A roadway information file includes details of geometric design characteristics such as the number of lanes, lane width, and presence/type of median. Finally, a lane inventory dataset outlines information such as the presence of rumble strips, shoulder widths, posted speed limit, and other features. A unique identifier is used to link information from each of these three files for each road segment in the GIMS database. The final analysis dataset included information from culverts that were located along 5,909 mi of two-lane 55-mph highways and 630 mi of four-lane 70-mph divided highways.

Culvert Database

A culvert dataset was obtained from the Iowa DOT that contained information pertaining to culverts on the primary (i.e., state-maintained) road network. This inventory file provides information as to the placement status (e.g., parallel versus transverse), culvert dimensions, shape, material, and culvert location (latitude and longitude). This database initially included a total of 29,019 culverts. A quality assurance (QA) process was conducted to determine the percentage of roadway system in which reliable culvert location data existed. Approximately 29 percent of the entries included missing information as to the culvert dimensions (either length or width). About 27 percent of the entries did not have placement information as to the culvert orientation (e.g., crossing, median, or ramp culvert), with these errors typically occurring along long, contiguous stretches of roadway. As it was uncertain whether these were crossing culverts or not, cases where culvert placement was unknown or could not be verified using aerial photographs were removed from the dataset. As the emphasis of this study was on perpendicular culverts, traverse culverts were also removed. Collectively, these filters reduced the dataset to 11,045 culverts. Using information from the culvert dataset and GIMS dataset, all of these culverts were spatially linked to the nearest road segment. This allowed for integration of culvert and roadway information for analysis purposes.

Barrier Database

The barrier dataset consists of detailed information on the installation locations of beam guardrail, concrete barrier, cable barrier, and crash cushions. These data were used to identify those culverts that are being shielded by some type of barrier. A QA review was conducted for these data using aerial photographs. Ultimately, this dataset was found to be largely incomplete with a substantive amount of inaccuracies. Thus, a manual review was conducted using satellite imagery to determine whether any form of protective barrier was present. Approximately 6.2 percent of culverts were found to be shielded by some form of barrier.

Crash Database

In this study, crash data from Iowa DOT were utilized from between January 2007 and August 2017, the duration over which such information were available. The crash data consists of information pertaining to each crash, from the crash level to the person level. For the purposes of this study, culvert-related crashes were identified using two methods:

Integration of Crash and Culvert Data

Attempts were also made to match the crash data with information for the related culvert that was involved in the crash. Consequently, each crash was spatially linked to the nearest culvert. It is important to note there is significant imprecision with respect to the location information in both the culvert and the crash databases. For example, some of the crashes were found to be more than 1 mi from the nearest culvert. In this study, all crashes that were more than 1,640 ft (500 m) from the nearest culvert were removed, reducing the number of crashes to 937. This threshold distance was selected among various alternatives to allow for conditions in which the vehicle would have struck the culvert and still continued to travel for some distance before coming to a stop.

Since this study only focused on the crossing culverts, parallel culverts were removed from the dataset, which resulted in 547 crossing culverts that had experienced a crash during the study period. As a result, the total number of crashes was reduced from 937 to 568 in this study. After removing those cases in which some pertinent information about the crash, the culvert, or the associated roadway segment was missing, the number of crashes was reduced to 500. Table 1 provides descriptive statistics for the entire sample of 11,045 crossing culverts where complete data were available, as well as details for the subset of 481 culverts that were linked to these 500 crashes.

OPEN IN VIEWER

Table 1.

Descriptive Statistics for All and Crash-Involved Crossing Culverts

		All culverts (n = 29,019)		Crash-involved (n = 547)
Variable	Category	Count	%	Count	%
Culvert shape	Round	19,563	67.4	358	65.5
	Box	7,743	26.7	164	30.0
	Arch	360	1.2	5	0.9
	Round/box	954	3.3	15	2.7
	Box/arch	152	0.5	2	0.4
	Unknown	247	0.9	3	0.6
Culvert length (ft)	<75	4,642	16.0	181	33.1
	75–150	4,772	16.4	226	41.3
	150–225	1,767	6.1	91	16.6
	≥225	1,029	3.5	38	7.0
	Unknown	16,809	57.9	11	2.0
Culvert size/width (ft)	≤4	18,908	65.2	429	78.4
	4.01–8	2,586	8.9	71	13.0
	>8	1,473	5.1	47	8.6
	Unknown	6,052	20.9	0	0.0
Speed limit (mph)	<45	962	3.3	18	3.3
	45–50	2,049	7.1	43	7.9
	55–60	17,349	59.8	278	50.8
	65	4,833	16.7	99	18.1
	70	3,826	13.2	109	19.9
Number of lanes	<4	16,865	58.1	283	51.7
	4 or 5	11,149	38.4	242	44.2
	≥6	1,005	3.5	22	4.0
Roadway classification	Interstate system	5,692	19.6	158	28.9
	U.S. system	11,430	39.4	207	37.8
	State system	11,897	41.0	182	33.3
Culvert offset from center line (ft)	<40	5,316	18.3	213	38.9
	40–80	4,439	15.3	203	37.1
	80–120	1,602	5.5	88	16.1
	≥120	853	2.9	32	5.9
	Unknown	16,809	57.9	11	2.0

Note that this study assumed that the center of the crossing culvert lies on the centerline of the roadway. Thus the offset of the culvert from the centerline of the roadway is calculated by dividing its length by two. For the purpose of this study, four different types of culverts were analyzed based on their sizes and shapes: small pipe (diameter less than 4 ft), medium pipe (diameter between 4 ft and 10 ft), medium box (width between 4 ft and 10 ft), and large box (width greater than 10 ft).

Negative Binomial Regression Model

The number of culvert-involved crashes is comprised of a series of non-negative integers. As such, negative binomial regression models have emerged as a preferred statistical method for the analysis of such data. Within the context of this study, the probability of culvert i experiencing y_i crashes during the analysis period can be expressed as shown by

P ( y i ) = Γ ( ( 1 / α ) + y i ) Γ ( 1 / α ) y i ! ( 1 / α ( 1 / α ) + λ i ) 1 / α ( λ i ( 1 / α ) + λ i ) y i ,

(1)

where

Γ (.) is a gamma function,

α is an overdispersion parameter, and

λ i is equal to the expected number of crashes at culvert i.

The λ i parameter is related to a series of site-specific characteristics as shown by

λ i = EXP ( β 1 X 1 + β 2 X 2 + … + β k X k + ε i ) ,

(2)

where

X₁ to X_k are a series of independent variables (i.e., traffic volumes, speed limit, culvert characteristics, etc.),

β₁ to β_k are a series of parameters estimated from the regression model, and

EXP(ε_i) is a gamma-distributed error term with mean equal to one and variance of α.

The negative binomial model is estimated using standard maximum likelihood estimation methods using NLOGIT version 6.0.

Roadside Safety Analysis Program

The RSAP uses an encroachment-based method to perform benefit–cost analyses on various roadside design alternatives. It allows for the determination of the “best” alternative based on a series of benefit–cost analyses that compare all alternatives. The program estimates the frequency, severity, and societal cost of roadside crashes using a conditional encroachment–collision severity approach for each alternative design in the analysis. Each alternative requires the agency costs, which include construction and maintenance costs, to be included in the analysis (default value is $ 0). The “best” alternative will be selected based on the biggest reduction in costs (benefits) compared with the agency costs for improvement. Equation 3 shows how the program calculates the benefit–cost ratio of an alternative. The numerator represents the reduction in crash costs, and the denominator represents the cost of the improvement. The program considers two factors for reduction in crash costs (CC): crash frequency and the severity of each crash. For the costs of the improvement, on the other hand, it considers the installation cost (I), maintenance cost (M), and repair cost (RE). The subscripts in the equation refer to two different alternatives (BCR₂₁ is the benefit–cost ratio of alternative 2 versus alternative 1).

BC R ji = C C i − C C j ( I j + M j + R E j ) − ( I i + M i + R E i )

(3)

Each analysis in RSAP is based on four conditional probability modules. First, the program will estimate the expected number of encroachments based on traffic information and roadway characteristics (encroachment prediction module). Next, if an encroachment occurs, the program will determine the likelihood of that encroachment resulting in a crash (crash prediction module). Subsequently, if that encroachment has a high probability of resulting in a crash, the program will evaluate the severity of that crash (severity prediction module). Finally, the software will convert the severity of the crash into a dollar amount to calculate the benefit–cost ratio of that alternative (benefit–cost analysis module). Detailed explanations of each module can be accessed from the RSAP Update developed as a part of NCHRP Project 22-27 ( 12 ).

For the purpose of this study, two different scenarios were created to simulate the most common types of sites in Iowa: (1) two-lane undivided highways with posted speed limit of 55 mph; and (2) four-lane divided highways with posted speed limit of 70 mph. The study also considered two types of cross-drainage culverts, which included crossing culverts (culverts that cross under all lanes of travel) and median culverts (culverts crossing under one direction of travel along divided highways). For crossing culverts, four different sizes were analyzed: small pipe, medium pipe, medium box, and large box culverts; while for median culverts, only small pipe culverts were analyzed.

Table 2 shows the characteristics of different variables that were included in the RSAP analysis. The design life of a culvert, as well as the rate of return, are the values suggested by the Iowa DOT ( 13 ). For the value of statistical life (VSL), the RSAP user’s manual defines it as “the average comprehensive crash cost of a fatal crash” ( 12 ). For this study, $4.5 million was used for this variable, based on the cost per fatality recommended by the Iowa DOT (13, 14). Sensitivity analyses were conducted with different values and the results were generally insensitive across the ranges that were considered.

OPEN IN VIEWER

Table 2.

Project Characteristics Used in Roadside Safety Analysis Program Analysis

Characteristics	Value
Design life	20 years
Construction year	2020
Rate of return a	4%
Gross domestic product deflator a	7%
Value of statistical life	$4.5 million
Encroachment adjustment a	1
Decision point benefit–cost ratio	2.0
Traffic growth rate a	1%
Terrain a	Flat
Annual average daily traffic	Mid-life
Percentage of traffic in primary direction a	50%
Segment length	600 ft
Shoulder width a	6 ft (undivided highway)
	6 ft for outside and 10 ft for inside shoulders (divided highway)
Median width (for divided highway only)	50 ft
Lane width	12 ft
Cross section used	1V:6H

a

Default value from the program.

The AADT used in this study was based on the minimum and maximum AADT values from the GIMS files. The scenario for two-lane undivided highways had AADT ranging between 1,000 and 9,000 vehicles per day (vpd) in 2,000-vpd increments; while for the four-lane divided highway scenario, the AADT ranged from 10,000 to 50,000 vpd with an increment of 10,000 vpd. In addition, the offset distance to the culvert was also varied between the scenarios. This value is measured from the edge line of the traveled way. Five different values were analyzed for both scenarios: between 8 and 32 ft with an increment of 6 ft for two-lane undivided highways, and from 14 to 34 ft with an increment of 5 ft for four-lane divided highways. Note that the maximum offsets for both scenarios represent the limits of the clear zone. These values were calibrated using Highway Design Tools (Roadside Barriers) from the FHWA ( 15 ).

RSAP allows designers to simultaneously analyze a maximum of five alternatives. This study examined four alternatives for both scenarios: do nothing (alternative 1), safety grate installation (alternative 2), guardrail installation (alternative 3), and culvert extension outside the clear zone (alternative 4) as shown in Figure 4. The do-nothing alternative served as a baseline and did not include any safety treatments. The only cost associated with this alternative was the annual maintenance cost of $600 (initial amount for all alternatives), obtained from previous studies (16, 17).

OPEN IN VIEWER

Figure 4.

Roadside Safety Analysis Program alternatives on four-lane divided highway.

Safety grates were the first alternative safety treatment considered. The construction cost for this treatment varied between $500 and $6,000 ( 18 ), depending on the size of the culvert. For the annual maintenance cost, this alternative required an additional $200 for grate cleaning and debris removal (twice a year) ( 19 ). The guardrail alternative required long lengths of steel guardrail to be installed along the travel lanes to keep drivers from crashing into the culvert. The length of the guardrail depended on the offset of the culvert, the design speed of the roadway, and the associated grade, which was calculated using a spreadsheet provided by FHWA ( 15 ). The construction cost of this safety treatment depended on the guardrail length, end anchor, and tangent end terminal. For the annual maintenance cost, an additional $1,000 was required for this alternative, based on Iowa DOT figures. The last alternative examined in this study was a culvert extension outside the clear zone. For this alternative, the construction cost depended on the size of the culvert and the clear zone. As for the annual maintenance cost, there was no additional cost required for this alternative.

RSAP Evaluation

In total, 225 different RSAP models were developed in this study. These models were developed based on different AADTs, culvert offsets, and types of culverts. The threshold used for the benefit–cost ratio in selecting the best alternative in each model is 2. The following sections present the cost-effectiveness analyses. The alternative with the highest incremental benefit–cost ratio is selected as the preferred alternative under each scenario.

Scenario 1: Two-Lane Undivided with 55 mph Speed Limit

Figure 7 shows the RSAP results for two-lane 55-mph undivided highways where all the culverts in this scenario are crossing culverts. The 32-ft offset (outside clear zone) models only include the first three alternatives as the culvert is assumed to be located outside the clear zone.

OPEN IN VIEWER

Figure 7.

Base case alternative for two-lane 55-mph undivided highways.

For the small pipe culvert, culvert extension is the preferred alternative when the distance between the edge of the traveled way and the culvert end is less than 20 ft for all ranges of traffic volumes (except for AADT less than 1,000 vpd for the 20-ft offset). As the culvert offset increases beyond 20 ft for the small pipe, the best alternative changes from culvert extension to either installing safety grates or doing nothing. This is because as the offset increases, the length of the extension to the culvert will also increase, which will result in higher installation cost compared with other alternatives.

The best alternatives for both the medium pipe culvert and the medium box culvert for all ranges of AADTs and offsets are similar, as shown in Figure 7. The primary reason for this trend is that in the RSAP analysis, the only difference between these two types of culvert is the culvert installation costs. Other variables such as the width/diameter of culvert, safety grate costs, and guardrail costs remain the same. Thus, the benefit–cost ratio does not change significantly. For these two types of culverts, most cases suggest that installing safety grates will be more effective than other alternatives (except for the cases with AADT less than 1,000 vpd, and culvert outside the clear zone).

For the large box culvert, Figure 7 shows that in most cases, the do-nothing alternative is preferable. This result is obtained because the installation cost for safety grates is significantly higher for this type of culvert than for others. Similarly, for culvert extension, the cost of installing a large box culvert is higher than for the medium box, and small and medium pipe culverts.

The benefit–cost ratio for installing guardrails is negative in all cases, indicating that the crash costs associated with the guardrails are always higher than the other alternatives. The findings suggest that while the guardrail may reduce the probability of a vehicle striking the culvert in a way that may result in serious injury, it will in turn increase the probability of a vehicle being involved in a PDO crash because of the presence of a guardrail.

Scenario 2: Four-Lane Divided with 70 mph Speed Limit

Figure 8 shows the RSAP results for four-lane 70-mph divided highways. Only small pipe culverts were modeled in RSAP for median culverts because these were the only culverts that were seen to be present in the culvert database. Similar to two-lane undivided highways, cases with offset outside the clear zone consider only the first three alternatives.

OPEN IN VIEWER

Figure 8.

Best case alternatives for four-lane 70 mph divided highway.

For small pipe crossing culverts, culvert extension can be observed to be the optimal choice when the offset value is less than 14 ft for all ranges of traffic volumes. As the culvert offset increases beyond 14 ft, the installation of safety grates is shown to be more economical when compared with other alternatives because of the low installation cost. Note that for traffic volumes less than 10,000 vpd, if the culvert is located outside the clear zone, none of the safety treatments prove to be cost-effective.

Similar to the two-lane undivided highway analyses, both medium pipe crossing culverts and medium box crossing culverts have the same trend for all cases. As mentioned previously, the only difference between these two types of culverts is the cost of installation, which for medium box culvert is twice that for medium pipe culvert. The results suggest that for these two types of culverts the greatest economic benefits will be obtained when safety grates are used when AADT is greater than 20,000 vpd.

For large box crossing culverts, Figure 8 shows that safety grates are preferred when the culvert offset is less than 24 ft, except when AADT is less than 10,000 vpd. This safety treatment is preferred mainly because of the lower installation cost compared with other treatments. However, as the culvert offset increases beyond 29 ft for AADT less than 40,000 vpd, none of the safety treatments show positive impacts in the form of economic benefits.

For small pipe median culverts, the same trend can be observed as in the small pipe crossing culvert case. Culvert extension is preferred when the culvert offset is less than 14 ft. As the distance between the edge of the traveled way and the culvert end increases, safety grates become the preferred option over other alternatives.

Limitations

There are several limitations that can be addressed through future work or changes in the manner in which highway agencies maintain their culvert inventory data. One of the main limitations of this project was the degree of missing or incomplete information in the culvert database, which required an extensive quality assurance review and some manual investigation to fill in missing data where possible. Ultimately, approximately 10 percent of the culvert sizes were missing from the analyzed data, which resulted in a limited sample for specific categories of culverts.

Another limitation of this study is that the crash information provided for this study was based on information in police crash reports. There were likely numerous cases where a crash occurred with a culvert, but was not reported. RSAP predicts crashes based on the encroachment and vehicle trajectory data and, as such, may be expected to provide a more accurate estimate of the number of culvert-involved crashes. Generally, these unreported crashes tend to be less severe and, as such, the number of crashes predicted under various scenarios using RSAP is higher than what was shown by the in-service evaluation.

The installation cost provided by the Iowa DOT for safety grates was a general figure that was not associated with a specific size of grate. The costs for different sizes of safety grates was found from an online source. The maintenance costs for culverts and safety grates were found through literature review; however, these costs did not have a size associated with them either. Therefore, the same maintenance costs were used for all culverts and all safety grates of all sizes.

Another limitation is related to the RSAP software and the underlying data on which the program is based. The ROR crash frequencies generated by RSAPv3 are based on the encroachment data collected by Cooper ( 20 ). These data were collected in the 1970s in Canada and there are some ranges of volume and geometric conditions in which data are sparse. An ongoing NCHRP study (NCHRP 17-88) is aimed at updating these data, which may provide improved predictive capabilities.

In the analyses performed in this study, it was assumed that the maintenance costs for culverts, safety grates, and guardrails remained the same for varying lengths and sizes. With a better dataset having the accurate installation and maintenance costs with varying sizes for culverts and safety grates, it would be interesting to see how these results vary. Currently, the culverts were combined into groups based on highway classification, speed limit, number of lanes, median type, and culvert sizes. As a future research work, each culvert from the list of those 547 culverts can be modeled separately in RSAP. This way the simulations will give accurate results and safety treatments can be chosen thereafter based on the individual results.