Collection,Analysis,and Reporting of Kentucky Traffic Incident Management Performance

Abstract

Traffic incidents remain all too common. They negatively affect the safety of the traveling public and emergency responders and cause significant traffic delays. Congestion associated with incidents can instigate secondary crashes, exacerbating safety risks and economic costs. Traffic incident management (TIM) provides an effective approach for managing highway incidents and reducing their occurrence and impacts. The paper discusses the establishment and methods of calculation for five TIM performance measures that are used by the Kentucky Transportation Cabinet (KYTC) to improve incident response. The measures are: roadway clearance time, incident clearance time, secondary crashes, first responder vehicle crashes, and commercial motor vehicle crashes. Ongoing tracking and analysis of these metrics aid the KYTC in its efforts to comprehensively evaluate its TIM program and make continuous improvements. As part of this effort, a fully interactive TIM dashboard was developed using the Microsoft Power BI platform. Dashboard users can apply various spatial and temporal filters to identify trends at the state, district, county, and agency level. The dashboard also supports dynamic visualizations such as time-series plots and choropleth maps. With the TIM dashboard in place, KYTC personnel, as well as staff at other transportation agencies, can identify the strengths and weaknesses of their incident management strategies and revise practices accordingly.

Traffic incidents are prevalent on U.S. roadways, negatively affecting the safety of the traveling public and emergency responders and causing significant traffic delays. Congestion resulting from incidents can lead to secondary crashes, amplifying safety risks and economic costs. Traffic incident management (TIM) provides an effective approach for managing incidents on the highway system and reducing their occurrence and impact. According to FHWA, TIM “consists of a planned and coordinated multi-disciplinary process to detect, respond to, and clear traffic incidents so that traffic flow may be restored as safely and quickly as possible” ( 1 ). A robust TIM program relies on efficient data collection, analysis, and reporting to measure performance and identify where and when traffic management can be improved. However, performance management through enhanced data collection remains elusive in many agencies that either do not collect TIM data, collect data for a limited number of traffic incidents, or have not yet adopted TIM performance measurement.

Many efforts at national, state, and local levels aim to advance TIM. FHWA developed the TIM Capability Maturity Self-Assessment Survey that for nearly two decades has been used to gauge overall development of TIM programs across the country ( 2 ). The 2019 survey indicated that the overall capacity maturity score increased 45% over the baseline of 2003 to 2005 ( 2 ). However, as in previous years, scores on performance measurement-related questions remained the lowest. To help agencies enhance TIM programs, FHWA included “Using Data to Improve Traffic Incident Management” in the Every Day Counts (EDC)-4 initiative ( 3 ). Meanwhile, a series of studies were conducted by FHWA and TRB covering a variety of themes, including analyzing gaps in TIM programs ( 4 ), assessing TIM-related big data sources ( 5 ), providing processes for TIM performance measurement ( 6 ), making business cases for TIM ( 7 ), and synthesizing best TIM practices ( 8 ). These resources are valuable in providing guidance and assistance to state and local agencies.

State and local efforts mainly focus on implementation and institutionalization of TIM programs ( 9 ). According to the EDC-4 final report, less than half of the states have assessed or institutionalized the use of statewide data to improve TIM ( 3 ). Among these states, only a handful have calculated the three nationally recommended performance measures (i.e., roadway clearance time, incident clearance time, and secondary crashes) ( 10 ). For example, Arizona includes roadway clearance, incident clearance, and secondary collision indictors in the electronic collision reports to capture these three TIM measures ( 11 ). However, building on the understanding of TIM data, several states have expanded performance measurement or integrated additional data sources. For instance, Florida includes Responders Struck By crashes, which FHWA recently designated as another important TIM measure ( 12 ). Pennsylvania has incorporated real-time Waze and traffic data into its Traffic Management Centers to improve incident timelines and operations ( 13 ).

Another integral component of a TIM program, in addition to data collection and performance measurement, is analysis and visualization of performance measures (6, 14). It is critical for tracking improvements, informing strategic and tactical decision making, communicating with partners and the public, enabling management buy-in and competing for funding. For instance, the Washington State Department of Transportation publishes a quarterly performance and accountability report called The Gray Notebook, which reports on performance measures related to key agency functions, including TIM ( 15 ). The agency has found the TIM performance trend to be valuable for conveying the benefits of its established Incident Response program and justifying the program’s expansion to decision makers and the state legislature. Other good examples of reporting TIM performance measurement can be found at West Michigan Transportation Operations Center ( 16 ) and the departments of transportation of Nevada ( 17 ), Virginia ( 18 ), and Wisconsin ( 19 ).

To assist states who are developing TIM assessment programs and share with those that already have such programs established, this paper describes the experience from the Kentucky Transportation Cabinet’s (KYTC) TIM program associated with collecting, assessing, and using TIM data for performance measurement and reporting (10, 14). Of particular note is the recent development of a TIM website dashboard and the use of big data to support the calculation of metrics. The Kentucky TIM program is part of FHWA’s EDC-4 initiative. The program promotes better TIM data collection with the goal of increasing transparency, improving operations, and facilitating better outcomes in program performance, resource management, and future planning. An Incident Management Task Force with representatives from FHWA, KYTC, Kentucky State Police (KSP), local agencies and Kentucky Transportation Center has been formed, with a focus on identifying the resources, tools, and technologies needed to compute major TIM performance measures. The standardized performance measurement and development of an interactive dashboard from the Kentucky TIM program can provide insights to other agencies and contribute to the state of the practice of TIM.

The remainder of the paper is structured as follows: the next section introduces the five TIM measures currently calculated in Kentucky. The third section discusses the collection and assessment of four TIM-related data sources, including KSP crash data, Traffic Response and Incident Management Assisting the River City (TRIMARC) incident logs, Waze incident reports, and HERE probe speeds. The measurement and analysis of five measures are detailed in the fourth section, followed by a section delving into the development of an interactive dashboard. The final section concludes the paper by summarizing the findings and providing recommendations.

TIM Performance Measures

Five TIM performance measures are currently calculated and reported following FHWA guidelines and Kentucky TIM Committee’s input. The measures are: roadway clearance time (RCT), incident clearance time (ICT), secondary crashes, first responder vehicle crashes (which substituted responder struck by crashes owing to data constraints), and commercial vehicle crashes. Their definitions are as follows.

RCT is the amount of time that elapses between an incident report and when all lanes reopen to traffic (Figure 1, T5–T1).

ICT is the amount of time that elapses between an incident report and when the last responder leaves the scene (Figure 1, T6–T1).

A secondary crash occurs as a result of the original or primary crash, either at the crash scene or within the queue in either direction. The inclusion of the opposite direction is necessary as the original crash could cause driver distraction or traffic queues that lead to a secondary crash in this direction.

First responder vehicle crashes are crashes where a first responder, e.g., law enforcement, fire and rescue, emergency medical service (EMS), and towing, vehicle is involved.

Commercial motor vehicle crashes involve commercial motor vehicles (CMVs).

Figure 1.

Incident management timeline.

The first three are primary measures recommended nationally by FHWA. Responder struck by crashes has been recently noted by FHWA to be an important measure for evaluating and improving first responder safety. The commercial motor vehicle crashes measure was added to the list at the TIM Committee’s recommendation.

Data Collection and Assessment

A variety of data sources are currently available and relevant to the measurement and assessment of TIM in Kentucky. However, the coverage and contained information vary depending on data sources, so we first collected these data from 2015 to 2019 and evaluated them from spatial, functional, and temporal perspectives.

Data Collection

Kentucky State Police Crash Data

The KSP crash database only contains crash records that are collected from collision reports on all facility types across the state. This database provides the most complete information to calculate TIM measures. One of available items is time notified, which is the time of the first recordable awareness of the crash by the state police. In addition, there may be up to three notification times from EMS in the original collision report. This issue was brought before the TIM Committee, which decided the earliest time notified should be used in cases where more than one time exists.

Time until roadway opened is another item that can be obtained from the crash data. However, it was unclear whether this meant all lanes or at least one lane was open to traffic. Therefore, an assumption was made that it would be a close approximation to T5 in Figure 1 and is used to calculate RCT.

The time the last responder leaves the scene has not historically been collected in Kentucky. It was added as a new field to the collision report in late 2017. The field was completed on a fraction of crash reports in 2018 and became available across the state in 2019 crash data.

Secondary crash has been a required input in the collision report for a few years, but its accuracy has been a concern. A Secondary Collision Definition help button was added on August 10, 2007, which shows “a secondary collision is a crash that has occurred because of nonrecurring congestion, which should be a result of an earlier documented collision.” Additional logic was added on April 16, 2013, to make sure users see the help message each time the “Yes” indicator is selected for the Secondary Collision field. However, there is still confusion between a secondary crash and a secondary event in a crash, resulting in secondary crashes being over reported. For example, a miscoded secondary crash that was a secondary event involved a car to car collision (i.e., first event) and then an impact with a tree (i.e., second event). Although ongoing training should improve reporting accuracy, corrections will still be needed to more accurately estimate the number of secondary crashes. Therefore, a more rigorous approach based on spatiotemporal analysis and narrative review was applied in this study.

The Commercial Motor Vehicle and Vehicle Type fields have been historically collected in Kentucky and are available in the crash database. The database has other important information that is useful in subsequent analyses, including crash occurrence time, crash severity, latitude and longitude, route name (LRS_ID), and milepoint.

TRIMARC Incident Records

TRIMARC is a 24/7 traffic operation center in Louisville, KY. It monitors traffic on interstates in the Louisville and Northern Kentucky (south of Cincinnati) metro areas and provides real-time traveler information and TIM services. The incident data provided by TRIMARC include both crash and noncrash incidents that include abandoned vehicle, disabled vehicle, road work, etc. It contains basic characteristics of incidents, for instance, incident type, roadway and direction, milepoint, incident beginning date/time and end date/time, number of vehicles involved, number of lanes blocked, and so forth.

Of those items, beginning date/time and end date/time were of particular interest in this study. According to TRIMARC, beginning date/time is closely aligned with time notified. This date/time is recorded from the responder’s call or from an abnormal speed drop and then verified by a TRIMARC operator—often through camera feed. The end date/time aligns with the time the responder leaves the scene. The police are usually the last to leave a scene. However, if freeway patrol is the only responder, the end time entry is reasonably accurate because of the constant communication with the TRIMARC operations center.

TRIMARC incident records would be the only source that provides timelines of crashes and noncrash incidents. After discussion with the TIM Committee, it was determined that the scope of the study should focus only on crashes. The limitation of TRIMARC data is their availability, which is limited to freeways in the Louisville and Northern Kentucky metro areas.

Waze Incident Reports

Waze is a community-driven, crowdsourcing navigation application program that allows its users to obtain and share real-time traffic information. What is relevant to TIM performance is the incident alerts that contain crash-specific records. Those alerts are generated by users when they pass by an incident and report it via the application installed in their smartphones. For this reason, multiple reports could be generated for the same incident. Depending on the number of users, the first alert may provide a close approximation to the first recordable awareness of an incident. Waze jam alerts are generated automatically by the software when the traveling speed is below a certain threshold. The data contain length of queue, congested speed, and delay information. Waze data could be available statewide as long as users submit their reports. However, availability is subject to the number of users and often limited to heavily traveled roads.

Although Waze data offer great value for increasing situational awareness on the highway system, their use in developing TIM performance measures has not been documented. This is primarily attributable to the characteristics of Waze data, which do not contain unambiguous incident timelines or attributes such as secondary crashes or crashes involving first responders that are needed for performance measurement.

HERE Speed Data

HERE speed data are collected from probe vehicles equipped with GPS-enabled devices or smartphones. There are currently two different types of HERE speed data in Kentucky. One is archived real-time data for longer Traffic Message Channel (TMC) sections used for the GoKY traffic information website, whereas the other is historical speed data for shorter links from previous years, which has been used by KYTC for generating travel time reliability measures, identifying bottlenecks, and assisting project selection (20, 21). This study focused on the archived real-time data that are available only on the National Highway System. The speeds can be useful from the perspective of revealing the impact of crashes. For example, a sudden speed drop is often caused by a crash and a speed increase could be associated with a roadway reopening. In addition, there have been studies looking into using speed data for secondary crash identification ( 22 ).

Data Assessment

To obtain an in-depth evaluation of the consistency of these data sources, the data at specific times and locations involving two separate crashes were extracted and analyzed.

The first crash occurred on I-71 southbound near mile marker 7.3 at 21:11 on May 10, 2016. There were nine Waze accident reports that seemed to be related to this crash. The first Waze alert was generated at 21:19 whereas the last one was reported at 22:54. Figure 2 shows the crash timeline and speed pattern at the crash location based on Waze and HERE data. Black and orange triangles indicate the information on the crash from KSP and Waze, respectively. The blue line represents the speed trend based on HERE data whereas the green dots are the speeds from Waze jam data.

Figure 2.

I-71 crash timeline and speed pattern.

It can be seen that Waze can pick up the crash impact more quickly than HERE does. Also, the impact in relation to speed slowdown is more significant based on Waze than HERE. Waze speeds are indicative of instantaneous speeds at a particular time and location and this is probably the reason for the speed slowdown. In contrast, HERE speeds are aggregated into 2-min intervals over the whole TMC, which in this case is about 4 mi long. As a result, there would be a time lag before the impact of the crash is seen at the upstream TMCs. From the spatiotemporal perspective as represented by the following heatmap (Figure 3), the resulting queue also propagated to the upstream TMC. Note that the direction of travel was southbound, that is, from a higher to a lower milepoint on the heatmap. The crash was in the first third of the TMC and congestion developed behind it, creating higher average congestion across the upstream zone before the zone with the crash.

Figure 3.

I-71 speed heatmap.

The second example is a crash that occurred on Dixie Highway (US-31W) near mile marker 12.1 at 16:20 on 16 May, 2016. Ten minutes after this fatal crash, a nonfatal crash occurred about 1.2 mi upstream. It was unclear whether this was a secondary crash; verification needed further investigation. The first Waze incident report came in at 16:30 and a few more were reported at 17:03 and 17:28 near the fatal crash site. The last Waze alert was at 18:42. Figure 4 shows the EMS (red triangles) and police timeline of the crash, HERE speed trend, Waze incident alerts and jam reports (green dots). The second crash occurred on the upstream TMC, according to the coordinates provided in the crash report.

Figure 4.

US-31W crash timeline and speed pattern ( 23 ).

Figure 5 shows the heatmaps generated from HERE data. The one on the left indicates the speeds on the day of the crashes (which was a Monday), and the one on the right shows a typical eventless Monday (May 2, 2017). Since the location of the second crash was right at the beginning of the upstream TMC, it was assumed to be a secondary crash.

Figure 5.

US-31W speed heatmaps.

Based on the assessment of data sources, KSP crash data provided the most complete and consistent data source for TIM performance measurement. Other data sources identified were good supplements to these measurements, which might be useful for future work. The following lists several additional key points from the data source analysis:

There was a time lag between the Waze reported accident time and KSP collision time or notification time based on the two examples examined. This is understandable since Waze incident alerts are entirely dependent on a roadway user’s actions.

Waze speeds appeared to be more representative of actual traffic conditions immediately after a crash, when and where data were available. Based on observations, a Waze jam alert (which contains speed data) was usually generated when speeds dropped below a certain level on a roadway.

HERE speeds represented traffic conditions at the TMC level while the slowdown caused by a crash happened upstream of the crash location. As a result, the speeds on long TMCs may not immediately show the impact of crashes. However, HERE data had a coverage advantage and, based on this investigation, were able to reveal the impact zone resulting from a major crash. This will be valuable in future analyses. Using shorter link level speed data could improve the sensitivity of speeds to the impact of crashes. However, the current subscription KYTC holds with HERE data is only at the TMC level.

All the data sources were able to provide some information from different aspects, therefore developing a scheme to use them collectively is likely to enhance the explanatory ability of those data and better fulfill TIM performance assessment purposes.

Performance Measurement and Analysis

This section describes the calculation and analysis of five TIM performance measures using KSP crash data, which contain all the required data elements. Note that the performance calculation and reporting focus on crashes was only based on KYTC’s guidance owing to limited data availability for noncrash incidents. Figure 6 summarizes how each of the measures was calculated.

Figure 6.

Traffic incident management (TIM) performance measure computational logics.

Roadway Clearance Time

KSP data have information on the time incidents are reported and the time roadways are cleared. RCT is calculated based on these; however, the data have some limitations at this time. First, values for TimeNotified, TimeRoadwayOpened, or both could be missing. Second, the timelines are lacking dates, making error detection difficult. For example, according to a collision report, a crash occurred on I-65 in the Louisville area at 14:30 whereas the roadway was reopened at 13:30. It is unlikely, although possible, for the aftermath of a crash on the interstate to last almost 24 hours. If dates were available, it would be possible to determine whether this was simply a coding error or indicates the crash spanned 2 days. Another issue is that all times are coded as integers and the numbers vary from 0 to 2,359, making it difficult to directly process the data. Some logical assumptions have to be made when querying the data and calculating the measure.

Between 2015 and 2019 Kentucky recorded 803,148 crashes, of which 448,072 (55.8%), 46,333 (5.8%), 16,405 (2.0%), and 292,338 (36.4%) crashes had a positive RCT, zero RCT, negative RCT, or no TimeRoadwayOpened value, respectively. Only positive RCT values are valid for performance reporting and tracking. Table 1 provides a year-by-year breakdown of these numbers. The number of valid calculations (RCT > 0) increased during this period, while the number of crashes with no TimeRoadwayOpened value fell, indicating that reporting quality has improved over time.

Table 1.

Breakdown of Roadway Clearance Time (RCT) Data

Year	RCT > 0 (%)	RCT = 0 (%)	RCT < 0 (%)	No TimeRoadwayOpened value (%)
2015	54.4	4.3	1.8	39.6
2016	54.0	4.4	1.8	39.8
2017	55.0	5.4	1.9	37.7
2018	57.8	7.1	2.2	32.9
2019	57.9	7.8	2.5	31.8

Table 2 reports the statewide summary statistics of RCT by year. The statewide RCT generally trended downward from 2015 to 2019. For example the average RCT declined from 34.2 min in 2015 to 31.1 min in 2019—a 3.1 min or 9.1% reduction in the time elapsed from incident notification to all lanes being opened to traffic. In addition, the 25th, 50th, and 75th percentile RCTs in 2019 were 8, 20, and 41 min, respectively, which reflects a move toward faster clearances (over earlier years). This could be attributable to Kentucky’s successful TIM responder training.

Table 2.

Roadway Clearance Time Aggregated Statistics by Year

Year	Minimum	25th percentile	Median	Mean	75th percentile	Maximum
2015	1	11	26	34.2	45	1,425
2016	1	12	27	35.0	46	1,410
2017	1	11	25	34.2	45	1,430
2018	1	8	21	32.0	42	1,415
2019	1	8	20	31.1	41	1,427

Figure 7 provides another perspective by looking at average RCT values for CMV and non-CMV crashes by year. Two points are worth noting. First, it took approximately 80% longer to clear CMV crashes than non-CMV crashes. Second, similar to the trend observed for all crashes, the average RCT of CMV crashes fell over the 5-year period. Compared with 2015, the amount of time required to clear CMV crashes was down 12% in 2019.

Figure 7.

Average roadway clearance time (RCT) of commercial moter vehicle (CMV) crashes by year.

We also analyzed RCT values based on crash severity (i.e., fatal, injury, and noninjury crashes) (Figure 8). Fatal crashes had the longest RCT values—about 2.5 h on average. This was roughly three times the RCT for injury crashes and 5.5 times longer than for noninjury crashes. Average RCT values for fatal crashes declined from 167 min in 2015 to 147 min in 2017, but rebounded by 12 min in 2018 before sliding back to 152 min in 2019. Average RCT for injury crashes was roughly 50 min throughout the study period, although slight downticks were recorded in 2018 and 2019. RCT values for noninjury crashes also fell modestly, from a peak of 30.2 min in 2016 to 26.4 min in 2019.

Figure 8.

Average roadway clearance time (RCT) by crash severity.

Incident Clearance Time

The TimeLastLeft data only became available across the state in 2019, so we calculated the ICT for the year. Despite being a recent addition, most entries were complete and valid: 96.42% had positive ICT values. Just 0.22%, 2.58%, and 0.77% of crashes had missing TimeLastLeft values, negative ICT values, or ICT values equal to zero, respectively. Like RCT, only positive ICT values are valid for performance reporting and tracking.

Figure 9a shows average ICT values by time of day for CMV and non-CMV crashes. Regardless of time of day, crashes involving CMVs demanded that first responders remained on scene longer than if no CMV was involved. On average, it took 20 min longer for responders to leave crash scenes of CMV crashes during the day and 22 min at night. Year-over-year trend analysis can be done in the future once more ICT data become available.

Figure 9.

Average incident clearance time (ICT) by (a) time of day and commercial motor vehicle (CMV) versus non-CMV crashes, and (b) by time of day and crash severity.

Figure 9b charts average ICT values by time of day in relation to crash severity. ICT increased significantly as crash severity worsened. During day and night, fatal crashes required 1.7 times more time to clear than injury crashes and three times more than noninjury crashes. For crashes with the same severity level, ICT values at night were 1.1% higher than those for the day.

Secondary Crashes

Secondary crashes could be directly derived from the SecondaryCollisionIndicator field in collision reports. However, some secondary crashes are not reported and some are wrongly coded as secondary crashes. Our review on randomly selected crash samples showed 8% to 13% of crashes reported as being secondary were confirmed to be true secondary crashes in 2015 to 2017, an improvement from only 3.6% to 4.4% in 2009 and 2010 respectively ( 24 ). To more accurately ascertain the number of secondary crashes, we first implemented a spatiotemporal approach to identify candidate primary and secondary crash pairs. Several interstate and arterial corridors were studied to establish more representative spatial and temporal thresholds in Kentucky. It was found that thresholds of 2 mi and 100 min were suitable for limited access highways. For other roadways, including urban arterials and rural roadways, the distance and temporal thresholds were 0.5 mi and 40 min, respectively. To account for secondary crashes that do not occur on the same road as the primary crashes, a buffer radius of 1,000 ft was used to select additional candidate secondary crashes. We then retrieved and manually reviewed the crash narratives associated with the candidate crashes to confirm which were indeed secondary.

Table 3 shows the annual number of secondary crashes based on the SecondaryCollisionIndicator field in collision reports (i.e., Police reports column) versus manual crash narrative reviews (i.e., Crash narratives column). The number of reported secondary crashes fell throughout the period, whereas the percentage of crashes correctly identified as secondary (or Reporting accuracy) improved. This indicates secondary crash reporting quality has improved and is likely to continue improving with better training. Confirmed secondary crashes based on narratives fell in 2017 but rose by almost one-third in 2018. We recommend a more in-depth investigation at the individual secondary crash level to determine why this occurred. The number of secondary crashes in 2019 can be obtained once crash narratives become available. This will allow us to verify whether 2018 was an outlier or a harbinger of an upward trend.

Table 3.

Confirmed Secondary Crashes by Year

Year	Police reports	Crash narratives	Correctly reported secondary crashes	Reporting accuracy (%)
2015	2,064	511	178	8.6
2016	2,033	526	168	8.3
2017	1,805	451	151	8.4
2018	1,459	597	184	12.6

Figure 10 shows the spatial distribution of confirmed secondary crashes for 2015 to 2018. Louisville Metro, Northern Kentucky, and the Lexington areas tend to have more concentrated secondary crashes.

Figure 10.

Spatial visualization of secondary crashes.

First Responder Vehicle Crashes

The objective of this element of the study was to report Responders Struck By crashes, which should be derivable from the Working in Trafficway (Incident) indicator that is included as a pedestrian factor in the collision report. However, we reviewed all crash records for the 2000 to 2019 period and found zero records with this code. We also checked Kentucky Occupational Safety and Health and the U.S. Bureau of Labor Statistics’ Kentucky profile, but did not locate representative numbers. We looked at other sources (e.g., respondersafety.com) but found numbers were significantly underreported. In light of these data constraints, we recommended using crashes involving first responder vehicles as an alternative because the Vehicle Type field in the collision report distinguishes first responder vehicles (e.g., ambulance, police car, fire truck, tow truck/wrecker) from other vehicle types.

Table 4 lists the number of crashes involving first responder vehicles over the last 5 years. It also breaks down these figures by crash severity. Over the study period, we saw an increasing trend, particularly in 2019, which had 602 crashes—a significant increase over the 417 crashes recorded in 2018. In addition, there was a large jump in fatal and injury crashes in 2019, which increased from 76 in 2018 to 108 in 2019. Further investigation would be needed to understand this trend and identify corresponding measures to reduce the frequency of these crashes and improve the safety of first responders.

Table 4.

First Responder Vehicle Crashes by Year

Year	Fatal	Injury	Other	Year total
2015	2	65	289	356
2016	2	80	300	382
2017	3	74	333	410
2018	2	74	341	417
2019	5	103	494	602
Total	14	396	1,757	2,167

Figure 11 displays the spatial distribution of the crashes involving first responder vehicles for 2015 to 2019. These crashes are mostly concentrated in urban areas, especially Louisville Metro, Northern Kentucky, and Lexington.

Figure 11.

Spatial distribution of first responder vehicle crashes.

Commercial Motor Vehicle Crashes

Table 5 reports the number of crashes involving CMVs and a breakdown by crash severity. Between 2015 and 2017, CMV crashes declined; however, from 2017 through 2019 they rebounded. Figure 12 is a heatmap of CMV crashes, which demonstrates CMV crashes are common in urban areas and on rural interstates.

Table 5.

Commercial Motor Vehicle Crashes by Year

Year	Fatal	Injury	Other	Year total
2015	77	1,046	6,506	7,629
2016	81	960	6,268	7,309
2017	58	908	5,990	6,956
2018	69	951	6,411	7,431
2019	82	891	6,527	7,500
Total	367	4,756	31,702	36,825

Figure 12.

Spatial distribution of commercial motor vehicle crashes.

Kentucky TIM Dashboard

Dashboards are important visualization tools commonly used for data analytics and reporting. They distill complex data into easily understandable graphs and maps from which users can gain actionable insights. Here we present an interactive TIM dashboard, which is used to display the five Kentucky TIM performance measures. We explored three design options: JavaScript with WordPress as a front-end, Python Dash library, and Microsoft Power BI platform. Ultimately, we selected the Power BI platform because it is designed to work easily with a variety of data sources, including comma-separated values files, Excel files, and SQL Server database, and supports a suite of built-in visualization forms. Many useful third-party add-ons are available as well. KYTC has used this platform for other applications, which made it easier to transfer the dashboard to the Cabinet for publishing and future updates.

Based on discussions with the TIM Committee, some desirable features and functions were identified for the Kentucky TIM dashboard beforehand. These include

Dynamism—the dashboard will update numbers and graphs based on user-selected location (e.g., county) and time (e.g., quarter);

Having more options to view TIM measures, for example, by agency, time of day, geographical area;

Displaying time-series plots of TIM measures; and

A map that can show segments disproportionally affected by incidents.

Based on these features, we first outlined a draft dashboard and then finalized the required input tables and map files as well as their structures. For instance, to be able to display geospatial maps in Power BI, original polygon shapefiles were converted into the TopoJSON format. Relationships between input files play a central role in ensuring accurate calculation, filtering, and display of the data in Power BI. Therefore, we designed TIM tables and map files to contain unique columns that helped to establish necessary relationships with each other (Figure 13). In particular, the performance measure table (left) contained relevant information that drilled down to the level of individual crashes. Each of the other tables (right) contained a column with unique values (i.e., primary key) and the same name as is found in the performance measure table. This let the dashboard create a many-to-one relationship between the performance measure table and other tables. These tables and established relationships allowed us to add various filtering options and visualizations to the dashboard. Figure 14 shows the current design of the Kentucky TIM dashboard.

Figure 13.

Relational linkages between tables.

Figure 14.

Kentucky Ttraffic Incident Management (TIM) dashboard interface.

The dashboard is designed to address all potential user needs as previously identified. It is fully dynamic and interactive. The panel on the left provides a set of criteria that can narrow the data to a specific area and time period. Figure 14 shows the RCT for the whole state from 2015 to 2019. The user can make a selection by county, district, police post, or a particular agency. A second dashboard (not shown) with the same layout is available for state house, senate, and congressional districts. For the temporal filter, the dashboard provides two options: the user can pick a date range by specifying the beginning and ending dates or dragging the date slider; the other option is to specify a period of interest from the year, quarter, and month list. This option also allows selection of multiple periods, for example all five second quarters during 2015 to 2019.

The right-hand section consists of various visuals. The six cards across the top present a summary of the five TIM measures plus the number of crashes. The line chart tracks the quarterly trend for the selected measures over time. The bar chart summarizes the selected measure by crash severity and time of day. The three choropleth maps in Figure 14 display the measure at county, district, and police post level, enabling quick comparison among different areas. The heatmap visualizes statewide crash hotspots, which are shown in red. The user can further zoom in to pinpoint crash hotspots at the street level.

In addition, because underlying tables and map files are linked, the dashboard charts and maps can cross-filter each other. For example, if the user selects Fayette County and clicks on the bar representing injury crashes during the day from the bar chart, all the visuals automatically update to reflect the selection (see Figure 15). We can determine there were 8,204 injury crashes within the county during the daytime period of 2015 to 2019, and their average RCT and ICT were 44.3 and 60 min, respectively. Among these crashes, 25, 21, and 209 were secondary crashes, first responder vehicle crashes, and commercial motor vehicle crashes, respectively. The line chart displaying the quarterly RCT of these crashes suggests a generally upward trend over time. Meanwhile, the heatmap reveals the hot spots of these injury crashes within the county.

Figure 15.

Kentucky Traffic Incident Management dashboard display for Fayette County.

Conclusion

This study addressed critical gaps in the Kentucky TIM program by generating an expanded list of TIM performance measures using a streamlined process. These measures include RCT, ICT, secondary crashes, first responder vehicle crashes, and commercial motor vehicle crashes. The study evaluated several TIM-related data sources and found that KSP crash data were the best data source for performance measurement whereas other data sources (e.g., crowdsourced alerts, speed data) can supplement primary data by contextualizing information on traffic conditions that may instigate crashes. Detailed analysis was performed to calculate the five measures using KSP data and to locate their temporal trends as well as variability in crash type, crash severity, and time of day. An interactive TIM dashboard was then developed and populated with performance measure data. The dashboard includes various spatial and temporal filters and fully dynamic features and functions. Its clear visualizations and representation of TIM data will help KYTC personnel as well as staff at other transportation agencies derive actionable insights in relation to TIM improvements.

Other key findings with potential policy implications are summarized as follows.

The number of crashes lacking data on the time of roadway opening fell between 2015 and 2019, indicating that reporting quality has improved. For instance, 58% of crashes in 2019 had valid entries. In 2015, just 54% did.

At the state level, average RCT declined over the last 5 years—from 34.2 min in 2015 to 31.1 min in 2019, which represents a 9.1% reduction in the time between incident notification and when lanes reopened to traffic.

For both RCT and ICT, fatal crashes took significantly longer to clear than injury crashes, which in turn required much more time to clear than noninjury crashes. Crashes involving CMVs also took much longer to clear than those which did not. Lastly, crashes occurring at night took longer to clear than those that occurred during the day.

The percentage of secondary crashes that were correctly reported improved between 2015 and 2018, indicating training is increasingly effective, that familiarity has improved, or both.

First responder vehicle crashes increased significantly, jumping from 417 crashes in 2018 to 602 crashes in 2019.

Commercial motor vehicle crashes and fatal commercial motor vehicle crashes declined from 2015 to 2017 but increased between 2017 and 2019.

Based on the analysis, we recommend the following improvements. First, although reporting quality of TimeRoadwayOpened and SecondaryCollisionIndicator fields has improved over the years, there remains much room for improvement. Providing better training programs could prove beneficial for further increasing reporting quality. Second, to derive the Responders Stuck By measure, we suggest that the Working in Trafficway (Incident) indicator, which has been included as a pedestrian factor in the collision report, be reported. Third, dates should be added to all time fields in the crash database, if practicable, to improve the accuracy of calculations. Finally, more in-depth investigations are necessary to determine why secondary crashes, first responder vehicle crashes, and commercial motor vehicle crashes went up in 2019 and to identify corresponding countermeasures to improve the safety of travelers and first responders.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: R. Souleyrette, X. Zhang, E. Green, M. Chen; data collection: X. Zhang, E. Green, P. Ross; analysis and interpretation of results: X. Zhang, R. Souleyrette, E. Green, T. Wang, M. Chen, P. Ross; draft manuscript preparation: X. Zhang. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was made possible by the funding from the Kentucky Transportation Cabinet.

Data Accessibility Statement

The data used to support the findings of this study have not been made available owing to the restrictions of the data use agreement.

References

Federal Highway Administration. Traffic Incident Management. https://ops.fhwa.dot.gov/eto_tim_pse/about/tim.htm. Accessed July 30, 2020.

Brewster

Delich

2019 Traffic Incident Management Capability Maturity Self-Assessment National Analysis Report. Office of Operations, Federal Highway Administration, Washington, D.C., 2019.

Federal Highway Administration. Every Day Counts: An Innovation Partnership With States. EDC-4 Final Report. Federal Highway Administration, Washington, D.C., 2019.

Amer

Roberts

Mangar

Kraft

W. H.

Traffic Incident Management: Gap Analysis Primer. Federal Highway Administration, Washington, D.C., 2015.

Pecheux

K. K.

Pecheux

B. B.

Carrick

NCHRP Research Report 904: Leveraging Big Data to Improve Traffic Incident Management. Transportation Research Board of the National Academies, Washington, D.C., 2019, p. 207.

Pecheux

K. K.

Holzbach

Brydia

R. E.

NCHRP 07-20 Report: Guidance for Implementation of Traffic Incident Management Performance Measurement. Transportation Research Board of the National Academies, Washington, D.C., 2014.

Pecheux

K. K.

Shah

O’Donnell

Making the Business Case for Traffic Incident Management. Office of Operations, Federal Highway Administration, Washington, D.C., 2016.

Jodoin

Vasconez

Zimmer

Best Practices in Traffic Incident Management. Federal Highway Administration, Washington, D.C., 2010.

Pecheux

K. K.

Process for Establishing, Implementing, and Institutionalizing a Traffic Incident Management Performance Measurement Program. Office of Operations, Federal Highway Administration, Washington, D.C., 2016.

10.

Souleyrette

Chen

Zhang

Green

E. R.

Sagar

Improving the Quality of Traffic Records for Traffic Incident Management. Kentucky Transportation Center, Lexington, KY, 2018.

11.

Cartier

J. P.

Arnson

Brown

Barnett

, eds. Arizona Traffic Incident Management (TIM) Past–Present–Future. 2019 Arizona ITE/IMSA Spring Conference, Phoenix, AZ, February 26-28, 2019.

12.

Carrick

Next Steps in TIM Data Messaging Dashboards and Responder Struck by Data. FHWA EDC4 KY TIM Peer Exchange, Lexington, KY, July 12-13, 2018.

13.

INRIX. Pennsylvania DOT using Crowd Sourced Data to Assess and Improve Statewide Traffic Incident Management (TIM). INRIX, Kirkland, WA, 2019.

14.

Zhang

Souleyrette

Wang

Ross

Green

Chen

Traffic Incident Management Dashboard. Kentucky Transportation Center, Lexington, KY, 2020.

15.

WSDOT. The Washington State Department of Transportation Gray Notebook Edition 74. The Washington State Department of Transportation, Olympia, WA, 2019.

16.

WMTOC Monthly Report West Michigan Transportation Operations Center (WMTOC). https://www.michigan.gov/mdot/0,4616,7-151-9615_44489_44549-,00.html. Accessed July 30, 2020.

17.

Pecheux

K. K.

Pecheux

O’Donnell

Development of a Statewide Pilot Project for Standardized TIM Performance Measurement and Reporting. Nevada Department of Transportation, Carson City, NV, 2016.

18.

The Virginia Department of Transportation. Performance Reporting Systems for Projects and Programs. http://dashboard.virginiadot.org/Default.aspx. Accessed July 30, 2020.

19.

WDOT. MAPSS Performance Improvement Report. Wisconsin Department of Transportation, Madison, WI, 2013.

20.

Chen

Zhang

Green

Analysis of Historical Travel Time Data. Kentucky Transportation Center, Lexington, KY, 2015.

21.

Chen

Zhang

Rahman

Brashear

Souleyrette

R. R.

Measuring Congestion for Strategic Highway Investment for Tomorrow (SHIFT) Implementation (PL-32). Report No. 1682. Kentucky Transportation Center, Lexington, KY, 2019.

22.

Zhang

Green

Chen

Souleyrette

R. R.

Identifying Secondary Crashes using Text Mining Techniques. Journal of Transportation Safety & Security, Vol. 12, No. 10, 2020, pp. 1338–1358.

23.

Zhang

van Dyke

Erhardt

Chen

NCHRP Synthesis of Highway Practice 541: Practices on Acquiring Proprietary Data for Transportation Applications. Transportation Research Board of the National Academies, Washington, DC, 2019.

24.

Pigman

J. G.

Green

E. R.

Walton

J. R.

Identification of Secondary Crashes and Recommended Countermeasures. Kentucky Transportation Center, Lexington, KY, 2011.