Incorporating Travel Time Reliability into the Highway Capacity Manual (2014)

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.

175 This appendix provides an overview of the weather and incident data requirements needed to run a SHRP 2 Project L08 travel time reliability analysis. It also provides basic data collection and analysis guidance for data-rich agencies wishing to make their analyses more precise by including high-detail, facility-specific weather and incident statistics in their reliability analyses. Travel time reliability depends heavily on weather and inci- dent events, which must be properly taken into account in any prediction or analytical evaluation of reliability. Acquisition and Processing of Weather Data provides an overview of weather data sources (including databases compiled as part of the SHRP 2 L08 project) and a review of classifying typical weather data into Highway Capacity Manual 2010 (HCM2010) (Transportation Research Board of the National Academies 2010) weather types. Acquisition and Processing of Incident Data provides guidance on converting incident logs into HCM incident types and expanding crash records to HCM-type inci- dents. A brief review of the default incident values included in the SHRP 2 Project L08 computational engines is also included. Because the freeways and urban streets reliability method- ologies employ different measures of incidents and weather, the data processing sections clearly differentiate between them. Table D.1 shows the basic requirements and optional inputs for both weather and incident data. Weather Urban Streets The SHRP 2 Project L08 urban streets software tool does not allow for custom input of weather. The analyst must choose one of the 284 cities for which weather has been simulated. Freeways A database of 10-year average weather probabilities was com- piled by the SHRP 2 Project L08 freeways team for 101 U.S. metropolitan areas. If the database does not contain the subject freeway’s location or if the analyst prefers to use facility-specific weather, he or she can use this document to prepare and input weather data. Weather data sources and a methodology consistent with the one used to compile the database are presented below in Acquisition and Processing of Weather Data. Incidents Urban Streets The SHRP 2 L08 project urban streets software tool limits user input to annual crash rates for segments and intersections of interest. This document provides guidance on calculating these rates from crash logs or national crash prediction meth- odologies, or both. Freeways The freeway facilities spreadsheet (FREEVAL-RL) gives the analyst more opportunities for customizing incident input. This document provides guidance on extracting or estimat- ing the probability of incident occurrence by incident type by month of year. The calculation is based on the proportion of the analysis period duration that the freeway segment is sub- ject to a given incident type. For data-poor agencies, advice on how to expand crash logs into incidents, how to convert crash data into number of lanes closed, and how to customize average incident durations is provided. FREEVAL-RL also provides the option to customize free- flow speed adjustments and capacity adjustment factors— which would modify the speed–flow curves used by the methodology—but these items are outside the scope of this data collection and processing guide. A p p e n d I x d Weather- and Incident-Related Crash Frequencies

176 Acquisition and processing of Weather data Weather Data Sources The ideal weather data set would include a year’s worth of 15-min weather reports collected near the facility being stud- ied. In the likely case that 15-min weather data are not avail- able, hourly weather reports published by the Federal Highway Administration’s (FHWA) CLARUS system, the National Oceanic and Atmospheric Administration (NOAA), Weather Underground, and others can be used. The CLARUS system was developed by FHWA to compile and distribute real-time atmospheric information (Federal Highway Administration 2012). There are three ways to obtain weather data using the CLARUS system: (1) subscribing to a report that is periodically updated, (2) using latest quality- checked observations on the map interface, and (3) retrieving an on-demand report for weather observations. A major dis- advantage of the CLARUS system is the relatively small num- ber of stations, especially in the south and southeast regions (see Figure D.1). NOAA monitors weather across the United States. For purposes of hourly weather data collection, their meteoro- logical aviation reports are the most useful. They contain several data points that are of interest to HCM-type weather calculation, including temperature, visibility, wind speed, and precipitation. Weather Underground’s historical hourly weather reports—which can be downloaded freely in .csv format from www.wunderground.com—rely on the meteorological aviation reports to archive all these metrics for almost every town and city in the United States (see Table D.2). These reports were used by the SHRP 2 L08 project freeways meth- odology to develop 10-year averages of weather occurrence probabilities for 101 metropolitan areas in the United States. The L08 project FREEVAL-RL software tool contains this database and can be a valuable resource for data-poor agen- cies that do not wish to compile and analyze weather reports. A similar database of 284 urban areas was assembled for use with the urban streets methodology. Unlike the freeway facilities tool, the urban streets spreadsheet does not allow for custom input of weather. In other words, the analyst is lim- ited to the 284 urban areas included in the database. Instead of collecting and averaging hourly weather reports, the urban streets project team simulated hourly weather by using a Monte Carlo simulation based on monthly weather statistics collected from the National Climatic Data Center (2011). Guidance for data-rich agencies that wish to collect, com- pare, or analyze facility-specific weather data is included below. To be able to classify weather into HCM format, certain measurements must be part of the time-stamped weather reports. These items are illustrated in Figure D.2. Weather Data Processing SHRP 2 Project L08 Urban Street Type The urban streets spreadsheet does not allow custom input of weather. The user is limited to the 284 cities contained in the spreadsheet. This section is thus included for sake of com- pleteness and in case the analyst wishes to compare facility- specific weather with the freeway spreadsheet’s weather. If so, the facility’s weather data must be collected and classified into one of the following weather types: • Rainfall; • Snowfall; • Wet pavement, no rain; • Icy pavement, no snow; or • Clear, dry weather. If the weather reports include a column with weather con- dition (e.g., cloudy, rainy, windy), classifying weather can be Table D.1. Data Requirements Methodology Data Set Data Poor Data Rich Urban Streets Weather None (database only). None (database only) Incidents Annual crash rate for each segment and intersection. No additional options Freeways Weather None (database available). Editing of average weather event duration is optional. Probability of occurrence (duration based) for 11 HCM weather types, by month Free-flow speed adjustment factor Capacity adjustment factor Incidents None (prediction methods available). Editing of inci- dent type distribution and average incident dura- tions is optional. Probability of occurrence (duration based) for six inci- dent types, by month Free-flow speed adjustment factor Capacity adjustment factor

177 Figure D.1. Screenshot of CLARUS interactive weather data map. Table D.2. Sample of Weather Underground Hourly Weather Reports

178 as simple as setting filters and counting observations. If not, the precipitation rate column can be used to determine whether there was rainfall or snowfall. Often different condi- tions will need to be lumped into a single category. For exam- ple, thunderstorm, light rain, or scattered showers would need to be categorized together as rain. In the likely case that pavement condition is not part of the weather report, it can be assumed that pavement remains wet or icy for 1 h after a rain or snow event. If 15-min weather data are available, these drying estimates can be replaced with 30 min. After classifying the weather observations, it is possible to compute the probabilities of weather occurrence for each weather type. In 1 year, there should be 8,760 (365 × 24) hourly observations. The probability of occurrence of a weather type is simply the ratio of the number of observations of that par- ticular weather type in 1 year to 8,760. Table D.3 shows the classification of a year’s hourly observations for Springfield, Illinois. SHRP 2 Project L08 Freeway Type Unlike the urban streets spreadsheet, the SHRP 2 Project L08 freeways spreadsheet does allow for custom input of weather (see Table D.4). For this purpose, weather data must be col- lected and classified into one of the following weather types: • Medium rain; • Heavy rain; • Light snow; • Light-medium snow; • Medium-heavy snow; • Severe cold; • Low visibility; • Very low visibility; • Minimal visibility; or • Nonsevere weather. These weather types were adopted from those in the HCM capacity adjustment table. A few weather types with negligi- ble capacity reductions (i.e., high and very high wind, light rain, and cold and very cold temperature) were omitted to decrease the computational complexity of the scenario gen- erator. If these weather types are encountered, they should be counted as nonsevere weather. Classifying weather reports into these categories is slightly more time-consuming than in the Project L08 urban streets method, but it can still be done easily with the use of a spread- sheet. Data columns that include the precipitation type (i.e., snow or rain), precipitation rate (in inches per hour), tem- perature, and visibility (in miles) should be used in conjunc- tion with the thresholds in Table D.4 to classify each weather report row. In cases for which two or more weather types Figure D.2. Schematic of weather data quality. Table D.3. Sample Output for Comparison with SHRP 2 Project L08 Urban Streets Weather (Springfield, Illinois) Type No. of Hours Percentage (%) Clear, dry 7,837 89.46 Rainfall 470 5.37 Snowfall 210 2.40 Wet pavement 194 2.21 Icy pavement 49 0.56

179 apply (e.g., severe cold with light snow), the analyst should choose the weather type with the highest capacity reduction. After classifying the weather observations, it is possible to compute the probabilities of weather occurrence for each weather type. In 1 year, there should be 8,760 (365 × 24) hourly observations. The probability of occurrence of a weather type is simply the ratio of the number of hourly observations of that weather type to 8,760. Table D.5 shows an example from the HCM2010 of capacity adjustment factors for weather con- ditions in Iowa. HCM Type This section is included here for sake of completeness only, since none of the reliability methodologies use the full breadth of the HCM weather capacity adjustments (for an example, see Table D.5). Should the analyst wish to conduct a thorough accounting of weather, he or she should not omit any weather type. In other words, Project L08 freeway weather should be augmented to include the following: • Light rain; • Cold temperature; • Very cold temperature; • High wind; and • Very high wind. Adding these weather types would increase the number of possible classifications from 11 to 16. Table D.4. SHRP 2 Project L08 Freeways Weather Input Table D.5. CAFs for Weather Conditions Source: Highway Capacity Manual 2010 (Transportation Research Board of the National Academies 2010).

180 Acquisition and Processing of Incident Data Both SHRP 2 Project L08 computation engine spreadsheets include default values for incident statistics as part of their meth- odology, but their approach differs. The urban streets methodol- ogy simply asks for an annual crash rate and uses hard-coded default values, but the freeways methodology allows the user to substitute defaults with facility-specific values. This section pro- vides guidance on the input of basic incident or crash rates and the calculation of facility-specific values. Incident Data Source The ideal outcome of the incident data collection effort is an annual incident rate and a table with the percentage fre- quency and average duration of incidents categorized by • Type (e.g., accident, breakdown, debris); • Severity (e.g., property damage only, injuries); • Lane closure effect (e.g., shoulder closed, one lane closed, two lanes closed); and • Location (i.e., intersection or segment). This ideal outcome can only be achieved with high-detail, incident-by-incident logs. For the purposes of this research, agencies across the United States were contacted and suitable logs were obtained on a case-by-case basis. Because of the complexity involved in obtaining these data, the L08 project freeways methodology includes default incident probabilities and prediction methodologies for estimating crash rates. The urban streets methodology simply requires annual crash rates. Figure D.3 provides a schematic of incident data quality. For customization of a Project L08 reliability analysis, dif- ferent incident methodologies can be used depending on the incident data quality illustrated in Figure D.3. In general, each agency is considered either data rich or data poor. Data-rich agencies are those agencies with an active traffic management center that monitors and archives incident data for freeways and arterials on a daily basis. Data-poor agencies are those agencies without traffic management center operations or those with no access to incident archives. In the absence of incident-log data, those agencies will require a methodology for estimating the number of incidents and default values, which is provided later in this guide. Collision records, which only contain crashes, are a more easily obtainable source of data. Most state departments of transportation and highway patrol agencies collect crash data on public roadway facilities. However, most agencies only publish monthly or yearly summaries of collisions, which are appropriate for deriving a yearly crash rate but not for class- ifying collisions by severity type, location, and lane closure effect. Examples of these systems include California’s SWITRS (California Highway Patrol) and Kentucky’s Collision Analy- sis for the Public (Kentucky State Police). If collision data cannot be obtained, it is possible to use one of the crash prediction methodologies developed by FHWA’s Highway Economic Requirements System (HERS) (Federal Highway Administration 2005) or the Highway Safety Manual (HSM) (American Association of State Highway and Trans- portation Officials 2010). Both methodologies predict the col- lision rate based on the roadway’s geometry, number of access points, daily volumes, and so forth. The freeways software tool (FREEVAL-RL) has the ability to run HERS prediction meth- odology on the input geometries and volumes. Through techniques explained in the following sections, collision data can then be expanded to incident data. The SHRP 2 Project L08 freeways spreadsheet allows input of both crash-only and incident data. If crash data are entered, the spreadsheet expands it by using a national default of 4.9 incidents per crash or a user-input ratio. Expanded crash data can be further categorized into incident types, converted into number of lanes closed, and categorized into crash sever- ity type. Incident Data Requirements Depending on the evaluation type (e.g., existing or future facilities, urban streets or urban freeways), data-rich agencies and data-poor agencies should follow different procedures and methodologies to estimate and identify incident occur- rence probabilities and incident duration by incident type and lane closure type. Default values, which were compiled from incident data sets from various geographical locations, are provided in case of lack of data.Figure D.3. Schematic of incident data quality.

181 SHRP 2 Project L08 Urban Street Type Similar to its treatment of weather events, the SHRP 2 L08 proj- ect methodology has a simple, user-friendly approach to crash data input. In this case, the only requirement is yearly crash fre- quency by segment. In case the analyst wishes to account for higher or lower crash rates during work zones and events, fields for adjustment factors are provided. Table D.6 shows the urban streets crash data input screen. Agencies with “best” or “better” incident data quality will be able to count the number of crashes in a year for each segment and input the rates into the Project L08 urban streets methodol- ogy spreadsheet (see Incident Data Processing below). Agencies with “good” incident data quality may use the urban arterials HSM method or the HERS arterial crash prediction (see the urban arterials portion of Table D.6). If the HERS arterial crash prediction method is chosen, the analyst must convert from crashes per 1,000,000 vehicle miles traveled (1MVMT) to crashes per year by using the segment length and annual average daily traffic (AADT). If better data are lacking, the approximate national proportion of crashes occurring at intersections (40%) (Federal Highway Administration 2009) should be used to sep- arate the rates into the segment and intersection columns. The annual crash rate can be computed by using Equation D.1: = ∗ ∗ ∗  Annual crash rate rate per MVMT segment lengthAADT 365 10 (D.1) 6 The output of the urban arterials HSM method is already in crashes per year and thus requires no conversion. SHRP 2 Project L08 Freeway Type The SHRP 2 Project L08 freeways methodology lets the user choose among different levels of data quality. Option A—for data-poor facilities—prompts the user either to run the HERS model or directly input crash or incident rates. If the agency wishes to input custom rates, it should follow the instructions under Incident Data Processing below and convert them from crashes per month or incidents per month to crashes per 100 MVMT (100MVMT) or incidents per 100MVMT. This conversion, which can be calculated using Equa- tion D.2, will require knowing the segment length in miles (L) and the AADT where the crashes or incidents occurred. Number of days corresponds to the time span of the crash or incident data set. Rate per 100MVMT Number of crashes or incidents 10 AADT Number of days (D.2) 8 L( )= ∗ ∗ ∗ Option A also gives data-poor agencies the option of speci- fying a facility-specific crashes-to-incident conversion factor (the national default is 4.9) and custom annual incident type distribution and durations (see Table D.7). Option B—for data-rich agencies—expands on these items and allows the user to enter month-by-month incident occur- rence probabilities for each incident type (see Incident Data Processing below). Furthermore, if facility-specific free-flow speed adjust- ment factors and capacity adjustment factors are known, they can be entered in place of the HCM defaults. These values will modify the speed–flow curves used by the methodology. These adjustments will not be found in incident logs and are thus outside the scope of this guide. Incident Data Processing While the previous sections describe sources of incident data and minimum data quality requirements, this section focuses on data processing. The following sections provide guidance on preparing incident data for entry into the SHRP 2 Project L08 Table D.6. SHRP 2 Project L08 Urban Streets, Crash Data Input Screen

182 Urban Streets Computational Engine and Freeways Computa- tional Engine. Figure D.4 represents the overall process for both existing and future facility evaluations and for both data-rich and data- poor agencies. A delineation can be made between an existing evaluation—which is designed for a “before” or opening day analysis stage—and a future evaluation, which is more appro- priate for a long-term analysis stage. For an existing facilities evaluation, data-rich agencies have the option to convert their incident-log data to a compatible format for input in the Project L08 analysis procedures. Data- poor agencies typically lack local incident data but have access to local crash data, which are identified as one of the common incident types. These agencies can populate incident frequency based on crash data. A procedure to estimate incidents from crash data is provided later in this guide. For planned or future facilities, both data-rich and data- poor agencies would have to perform extra steps in order to estimate facility-specific incident frequency. When sufficient traffic and geometry information is available, crash frequency for arterials can be estimated using the crash prediction pro- cedures available in the HSM. Alternatively, in a situation in which only planning-level parameters (e.g., traffic forecast and length of facilities) are known, incident frequency for either urban freeways or arterials can be estimated using HERS or other crash prediction methodologies. Details for each calculation and suggested default values for both evalu- ation types are provided later in this guide. Unfortunately, there is no consistency across agencies’ inci- dent data recording systems. Some agencies simply record the incident duration and number of lane closures without regard to the roadway shoulder. However, the HCM freeway incident classification categorizes shoulder accident, shoulder disable- ment, and lane closures separately. Most of the incident data- bases show that shoulder closures are more frequent than lane closures. Consequently, shoulder closures should represent a significant share in the incident type distribution. The following procedures are recommended for both data-rich and data-poor agencies to prepare and process their incident data or estimate incidents in a compatible format with the reliability analysis developed under the SHRP 2 L08 project. Best Data Quality: Local Incident Data This option is for data-rich agencies wishing to evaluate exist- ing facilities. For any analysis period, incident occurrence can be estimated through a simple probability calculation. At least one full year of detailed incident data will be required. To customize the incident input tables (see Table D.8), the data must have information on incidents’ duration and lane closure effect. If multiple years of incident data are available, the analyst should rely on the entire data set in the analysis. To calculate the probabilities of incident occurrence in each month, the analyst should first separate the recorded incidents by type and by the month in which they occurred. Table D.7. SHRP 2 Project L08 Freeways, Incident Data Input Screen

183 Figure D.4. Incident data processing schematic. Table D.8. Incident Probabilities in SHRP 2 Project L08: Urban Freeways Methodology

184 Table D.9. National Defaults for Incident Data from SHRP 2 Project L08, Freeways Incident Type Incident Type Distribution (%) Expected Duration (min) SD of Duration Shoulder closure 75.40 32 15 One-lane closure 19.60 34 14 Two-lane closure 3.10 53 14 Three-lane closure 1.90 69 22 Four-lane closure 0.00 69 22 Note: SD = standard deviation. Table D.10. Crash-to-Incident Factors Facility Type Crash-to-Incident Factor Freeways Range 2.4–15.4 Average 4.9 Median 6.5 Arterials Range 2.8–3.3 Average 3.0 Median 2.9 A detailed explanation of categorizing incidents by the five lane closure effects is provided in this appendix. For each month of the year (m = 1, 2, 3, . . . , 12), the analyst should calculate the probability of occurrence of each of the five inci- dent types (i = 1, 2, . . . , 5). The calculation is based on the proportion of the analysis period that the facility was subject to each type of incident as shown in Equation D.3: i m nn n Ni m∑ = = = Probability Incident duration Length of analysis period (D.3), 1 , where Ni,m is the number of incidents of type i in month m, and the length of analysis period is the total duration of the analysis period (e.g., a 3-h p.m. peak analysis period should have a total monthly duration of [number of weekdays in month i × 3 h]). For the probability of a shoulder closure in January (p.m. peak analysis), for example, the analyst may replace the default value with the sum of the durations of all shoulder closure inci- dents occurring in January during the p.m. peak, divided by the length of the p.m. peak times the number of weekdays in January. If incident duration is not known, the analyst may use the product of the monthly incident counts and the average dura- tions in place of the summation of incident durations. Table D.9 shows national defaults for incident data. Better Incident Data Quality: Expansion of Crash Data into Incident Data This option is appropriate for data-poor agencies in the eval- uation of existing facilities. Data-poor agencies are those that do not routinely collect incident data or have no access to incident data. In absence of complete incident data, these agencies can use crash data for estimating incidents within the same facility. Then, they can use average durations to compute monthly incident probabilities. Total number of crashes, regardless of crash severity type, can be used for estimating a frequency of total incidents for the same facility and time period. A crash-to-incident factor was developed based on crash proportions in various incident data sets. On average, crashes account for approximately 20.5% of all incidents or, stated another way, the number of incidents is about 4.9 times the number of crashes. These factors are shown in Table D.10. The crash-to-incident factors can be used to populate total incident frequency for the evaluated facility type by directly multiplying them by number of crashes for the evaluated facility. For flexibility of application, a range, average, and median are provided. When the local crash option is used to estimate total inci- dent frequency, the analyst can further identify the probabili- ties of occurrence of the estimated incidents for a given time period and day. Good Incident Data Quality: Incident Prediction Based on Predicted Crash Frequency This option is appropriate for data-poor agencies for the evaluation of existing facilities. It can also be used by both data-poor and data-rich agencies for the evaluation of future facilities, when there is no crash history at a site. In the absence of crash records, an expected total number of crashes can be estimated using one of the following avail- able prediction tools: • HSM crash prediction models (American Association of State Highway and Transportation Officials 2010); • HERS crash prediction models (Federal Highway Admin- istration 2005); or • Crash rate (crashes per MVMT).

185 Each of the methods above requires a different set of inputs for traffic and geometry information. Depending on the facil- ity type being evaluated, the HSM or HERS methods may be applicable. At this time, only the HERS method can be used to predict crash rates in freeways. In other words, the HSM methods provide crash prediction models for arterials only, but HERS does so for both freeways and arterials. A freeway model is currently being developed by the HSM, and should be considered when available. Crash rates can also be considered for both freeways and arterials. The rates are available from the ITS Deployment Analysis System (IDAS) for crash prediction by crash sever- ity type for freeways and arterials based on the known volume-to-capacity (v/c) ratio (Cambridge Systematics 2003). Alternatively, total crash frequency for freeways can be obtained from the crash rates based on the known scale for traffic speeds (Yeo et al. 2013). Table D.11 provides a list of related crash prediction mod- els for both facility types. More details for these prediction methods can be found in the references provided in each of the prediction tools. Similar to the better incident data quality example, the expected number of crashes from the prediction methods can be factored up to estimate the total number of inci- dents by using the crash-to-incident factors provided in Table D.10. However, it is recommended that no further Table D.11. Crash Prediction Methods Facility Type Crash Prediction Tool Comments Urban Freeways Option 1: Arterial crash prediction using HSM models: Go to Chapter 12 of HSM, apply Equation 12-8. Estimate number of crashes per 100 million VMT (100MVMT) Option 2: Freeway crash prediction by crash rates Crash Severity Type V/C or Traffic Conditions Crash Rate (Crashes/MVMT) Fatal 0.09–1.00 0.0066 Injury 0.09–0.69 0.4763 0.70–0.89 0.5318 0.90–0.99 0.6770 1.00 0.7060 Property damage only (PDO) 0.09–0.69 0.6171 0.70–0.89 0.7183 0.90–0.99 0.8365 1.00 0.9192 All severity types FF 0.72 BN 4.43 BQ 4.48 CT 4.55 Source: Adapted from Tables B.2.10 through B.2.12 of IDAS User’s Manual (Cambridge Systematics 2003) and Yeo et al. (2013). Note: FF = free-flow (upstream and downstream speeds >50 mph); BN = bottleneck (downstream speed >50 mph, upstream speed <50 mph); BQ = back-of-queue (downstream speed <50 mph, upstream speed >50 mph); and CT = congested (upstream and downstream speeds <50 mph). Estimate number of crashes per million VMT (MVMT) Urban Arterials Option 1: Arterial Crash Prediction by HSM Models (Chapter 12) Use HSM Equation 12-8 Estimate number of crashes for the period of interest Roadway segment crashes are a combination of the following predictions: Multiple-vehicle nondriveway crashes Use HSM Equation 12-10 (continued on next page)

186 Table D.11. Crash Prediction Methods Facility Type Crash Prediction Tool Comments Urban Arterials Single-vehicle crashes Use HSM Equation 12-13 Multiple-vehicle driveway-related crashes Use HSM Equation 12-16 Vehicle–pedestrian crashes Use HSM Equation 12-19 Vehicle–bicycle crashes Use HSM Equation 12-20 Intersection crashes are combination of the following predictions: Vehicle–vehicle crashes for intersections Use HSM Equation 12-21 Use HSM Equation 12-20 Vehicle–pedestrian crashes for signalized intersections Use HSM Equation 12-28 Use HSM Equation 12-29 Vehicle–pedestrian crashes for stop-controlled intersections Use HSM Equation 12-30 Vehicle-Bicycle Crashes Use HSM Equation 12-31 Option 2: Arterial Crash Prediction by HERS Models (Chapter 5) Urban Multilane Highway Crashes HERS Equation 5.34: Crash rate  A  AADTB  NSIGPMC where Crash rate = number of crashes per 100 million vehicle miles on urban multilane highways; NSIGPM = number of signals per mile (0.1 ≤ NSIGPM ≤ 8); and A, B, and C = coefficients provided below. Type of Section A B C Two-way with left-turn lane 95.0 0.1498 0.4011 One-way, or two-way with a median 82.6 0.1749 0.2515 (1) wider than 4 ft, (2) cubed, or (3) a “positive barrier” Otherwise 115.8 0.1749 0.2515 Source: HERS Table 5-9 (FHWA 2005). Estimate number of crashes per 100 million VMT (100MVMT) on urban multilane highways Urban Two-Lane Highway Crashes HERS Equation 5.36: Crash Rate  19.6  ln(AADT)  7.93  (ln[AADT])2 where crash rate = number of crashes per 100 million vehicle miles on two- lane streets. Estimate number of crashes per 100 million VMT (100MVMT) on urban two-lane highways Option 3: Arterial Crash Prediction by Crash Rates Crash Severity Type V/C or Traffic Conditions Crash Rate (Crashes/MVMT) Fatal 0.09–1.00 0.0177 Injury 0.09–1.00 1.6991 Property damage only 0.09–1.00 2.4736 Source: Adapted from Tables B.2.10 through B.2.12 of IDAS User’s Manual (Cambridge Systematics 2003). Estimate number of crashes per million VMT (MVMT) (continued)

187 Table D.12. Incident Occurrence Probabilities for Each Month Month Freeway Having Incidents No Incidents January 0.9032 0.0968 February 0.8839 0.1161 March 0.9194 0.0806 April 0.9000 0.1000 May 0.8871 0.1129 June 0.8750 0.1250 July 0.9113 0.0887 August 0.8871 0.1129 September 0.9000 0.1000 October 0.7903 0.2097 November 0.8417 0.1583 December 0.8790 0.1210 Total 0.8815 0.1185 Table D.13. Incident Occurrence Probabilities by Peak Period Month Freeway Arterial 6–11 a.m. 3–8 p.m. Off Peak Total 6–11 a.m. 3–8 p.m. Off Peak Total January 0.2991 0.3870 0.3139 1.0000 0.2737 0.4031 0.3235 1.0000 February 0.2509 0.4514 0.2977 1.0000 0.2754 0.4233 0.3009 1.0000 March 0.2643 0.4400 0.2958 1.0000 0.2742 0.4402 0.2857 1.0000 April 0.2593 0.4124 0.3283 1.0000 0.2770 0.3803 0.3427 1.0000 May 0.2411 0.4456 0.3132 1.0000 0.2788 0.3772 0.3444 1.0000 June 0.2548 0.4185 0.3268 1.0000 0.3268 0.3756 0.2975 1.0000 July 0.2569 0.4455 0.2978 1.0000 0.3146 0.4272 0.2583 1.0000 August 0.2402 0.4203 0.3396 1.0000 0.2592 0.4260 0.3149 1.0000 September 0.2581 0.4227 0.3192 1.0000 0.2312 0.3869 0.3820 1.0000 October 0.2386 0.4522 0.3093 1.0000 0.3331 0.3784 0.2881 1.0000 November 0.2382 0.4538 0.3079 1.0000 0.3037 0.3878 0.3083 1.0000 December 0.2256 0.4441 0.3302 1.0000 0.4429 0.0498 0.5072 1.0000 treatment—including custom occurrence probabilities, free-flow speed adjustment, or capacity adjustments—be made on HERS estimates. Estimation of Incident Probabilities When the total incident frequency is estimated using the methodologies in either the better or good data-quality option, the analyst can further identify probabilities of incident occurrence by using the suggested probabilities provided in Table D.12 and Table D.13. Table D.12 provides the probabilities of having or not having incidents for each analysis month. Table D.13 provides the probabilities of incident occurrence within each peak period for each month. The probabilities provided in Table D.12 can be used for different purposes. They can be directly applied to estimate incident frequency for each peak period in a given month. They can also be used in conjunction with Table D.13 to pop- ulate the probability of incident occurrence for a given peak period and a given month. The level of analysis for incident occurrence should be peak periods. The incident probabilities in Table D.13 are provided for the three time periods: 6 to 11 a.m., 3 to 8 p.m., and off-peak periods. Thus, the probabilities provided are for the estimate of total incidents occurring within 5 h for the a.m. peak period, 5 h for the p.m. peak period, and 14 h for the off-peak period. The following example illustrates the use of Table D.13 to estimate incident frequency for a given peak period of each month. When incident frequency is identified in number of incidents per month, a number of incidents broken down by peak period can be obtained using the probabilities in Table D.13. For example, if a freeway facility is projected to have 20 incidents per month, the probability of incident occurrence from 3 to 8 p.m. in January is approximately

188 Table D.14. Average, Median, and Standard Deviation of Incident Type Probabilities Facility Type Incident Distribution (%) Crash Breakdown Debris Other Freeway Range 6.5–41.4 45–88.6 1.1–13.2 0.4–7.5 Average 20.5 69.7 5.9 3.9 Median 15.4 74.6 5.8 4.2 SD 13.1 15.6 4.3 3.1 Sample Size 18,206 76,758 7,813 737 Arterial Range 30.2–35.6 27.3–58.8 5.2–7.8 4–37.8 Average 32.9 45.3 6.4 15.5 Median 35.0 52.7 6.5 5.8 SD 2.9 16.7 1.8 19.0 Sample Size 1,733 1,757 205 958 20 (incidents) × 0.3870 (Table D.13, Freeways, January, 3–8 p.m.) = 7.74, or eight incidents. To identify the probability of incident occurrence for each peak period and for each month, the analyst can cal- culate a product of the probabilities from Table D.12 and Table D.13. For example, the probability of having inci- dents on a freeway facility from 3 to 8 p.m. in January is 0.9032 (Table D.12, Freeways, January) × 0.3870 (Table D.13, Freeways, January, 3–8 p.m.) = 0.3495. These can further be refined by incident type with the help of Table D.14. Determine the Type of Lane Closure Effect The lateral lane closure is considered in HCM2010 as one of the parameters that has a direct effect on freeway capacity. Closing more lanes due to incidents significantly decreases the service capacity of freeways. The majority of recorded incidents close the shoulder lane more frequently than travel lanes. The likelihood of the lane closure due to incidents declines with the number of lanes. For example, incident data analysis shows that approximately 4% of all incidents close two lanes. The proportion of lane closure for three and more lanes is approximately 2%. If lane closure information is further stratified by incident types, the lane closure proportion for three and more lanes would be even lower. Five lateral lane closure effects are recommended in the SHRP 2 Project L08 freeways reliability analysis. These include • Shoulder closure; • 1-Lane closure; • 2-Lane closure; • 3-Lane closure; and • 4-Lane closure. Urban streets methodology only considers three lane- closure types: shoulder, one lane, and two or more lanes. Data-Rich Agencies Similar to the incident occurrence analysis, the analyst can determine incident lane closure types in two ways. The first option is by turning incident-log data into a cumulative dis- tribution function by lane closure for each incident type and randomly sampling from it. Alternatively, it can be done by turning incident logs into a frequency distribution by lane closure type for each incident type and identifying a percentage distribution across all lane closure types. Data-Poor Agencies In the absence of local incident-log data, data-poor agencies can use the default probabilities of incident occurrence by lane closure type provided in both the urban freeways and urban streets tool. Common lane closure types found in various incident data sets include shoulder, one lane, two lanes, and three or more lanes. These are consistent with the lane closure types used in the HCM. Some agencies, such as the Washington State Department of Transportation, record multiple lanes without separately specifying two-lane or three-lane. Table D.15 provides a greater detail of the default incident distribution by lane closure types in the urban freeways and urban streets tools. The analyst should be aware that the urban freeways methodology will require only incident dis- tribution by lane closure type, while the urban streets meth- odology will require incident distribution by incident type or crash severity. In the urban streets tool, the incident is broken down by crash- and noncrash-related type. The analyst can use the default values provided in the urban streets tool, which is presented in greater detail in Table D.16, to identify the lane closure type for property-damage-only and injury or fatal crash types.

189 Table D.15. Lane Closure Type Distribution by Incident Type Facility Type Incident Type Statistic Lane Closure Distribution (%) Shoulder 1 Lane 2 Lanes 3 or More Lanes Freeway Crash Range 12.3–79 15.6–44.4 0.7–25.1 2–18.2 Average 55.8 27.8 9.4 7.0 Median 59.1 27.0 8.3 5.5 SD 19.8 8.1 7.5 5.2 Sample size 6,825 5,749 2,512 1,736 Breakdown/disabled/stalled Range 3–98.3 1.6–92.8 0.1–1.3 0–3.5 Average 81.0 17.9 0.5 0.7 Median 91.7 7.9 0.2 0.2 SD 30.2 28.9 0.4 1.3 Sample size 63,292 8,308 353 214 Debris Range 0.9–96.2 3.3–88 0.3–21.4 0–8.8 Average 40.3 51.2 6.1 2.5 Median 28.5 66.6 3.1 1.7 SD 36.5 32.1 7.1 3.0 Sample size 1,326 3,011 439 156 Other Range 43.2–95.9 3.5–45.8 0.4–4.2 0.2–9.7 Average 66.0 27.5 2.2 4.2 Median 63.4 30.8 2.2 3.5 SD 22.2 18.3 1.7 4.2 Sample size 376 150 15 28 Arterial Crash Range 37.4–50.5 34.4–51.3 0–9.6 1.7–15.1 Average 44.0 42.9 4.8 8.4 Median 44.0 42.9 4.8 8.4 SD 9.3 12.0 6.8 9.5 Sample size 360 275 11 97 Breakdown/disabled/stalled Range 2.8–77.2 22.8–85.2 0–0.3 0–11.6 Average 40.0 54.0 0.2 5.8 Median 40.0 54.0 0.2 5.8 SD 52.6 44.1 0.2 8.2 Sample size 199 842 3 108 Debris Range 0–45 50–89.1 2.2–5 0–8.8 Average 22.5 69.5 3.6 4.4 Median 22.5 69.5 3.6 4.4 SD 31.8 27.6 2.0 6.2 Sample size 9 132 5 12 Other Range 13.6–15.5 50–59.2 0–27.3 9.1–25.4 Average 14.5 54.6 13.6 17.2 Median 14.5 54.6 13.6 17.2 SD 1.3 6.5 19.3 11.5 Sample size 14 67 10 21

190 Table D.16. Lane Closure Type by Crash Severity Facility Type Crash Severity Distribution (%) Shoulder 1 Lane 2 Lanes 3 or More Lanes PDO I  F PDO I  F PDO I  F PDO I  F Freeway Range 85.7–96.6 3.4–14.3 41.7–100 0–58.3 0–100 0–100 58.9–100 0–41.1 Average 90.0 10.0 74.4 25.6 58.9 41.1 77.4 22.6 Median 89.5 10.5 75.7 24.3 72.2 27.8 75.3 24.7 SD 0.04 0.04 0.23 0.23 0.38 0.38 0.18 0.18 Sample size 2,563 247 3,104 482 1,158 382 710 448 Arterial Range 88.0–97.4 2.6–12.0 67.1–92.9 7.1–32.9 58.3–72.7 27.3–41.7 100.0 0.0 Average 92.7 7.3 80.0 20.0 65.5 34.5 100.0 0.0 Median 92.7 7.3 80.0 20.0 65.5 34.5 100.0 0.0 SD 0.07 0.07 0.18 0.18 0.10 0.10 na na Sample size 317 39 197 75 50 33 1 0 Note: PDO = property damage only; I + F = injury and fatal. Determine Incident Duration An unplanned incident typically begins at the time it is entered into the system and formally ends when all involved vehicles are off the shoulder or when the last related activity is recorded in the system. Some agencies may consider inci- dent duration from different clearance stages, such as when the obstruction is removed, when the lanes are reopened, or when traffic returns to its normal stage. Data-Rich Agencies Similar to the previous analyses, the analyst can determine incident durations by using two options. The first option is by turning incident-log data into a distribution function by incident duration for each incident type and lane closure type and randomly sampling from it. Alternatively, it can be done by identifying average incident durations by incident type and by lane closure type from the incident-log data. If incident duration is sampled from distributions, it is rec- ommended that the distribution be truncated on both ends to avoid very small (e.g., 1-min) or very large incident durations. The advantage of using a formal probability distribution (rather than creating a customized cumulative distribution function from the field data) is that the effect of incident manage ment strategies can be tied to the mean duration, thus not requiring that the entire cumulative distribution function be retuned. Depending on the methodology used for either urban free- way or street, the same incident types and lane closure types as suggested in the previous analyses should be considered in identifying incident durations. The lane closure type used in the urban freeways methodology include shoulder closure, one-lane closure, two-lane closure, three-lane closure, and four-lane closure. The urban streets methodology considers three types of lane closure, including shoulder closure, one- lane closure, and two-lane closure. Data-Poor Agencies In the absence of incident-log data, it is recommended that the incident durations by incident type and lane closure be estimated using the default values provided in the urban free- ways and urban streets tools. Table D.17 provides greater detail of the default incident durations in the urban freeways and urban streets tools. The analyst should be aware that the urban freeways methodology will require only incident duration by lane closure type, while the urban streets methodology will require incident duration by lane closure type and incident type. In the urban streets tool, the incident duration is broken down by severity type. The analyst can use the default values provided in the urban streets tool, which is presented in greater detail in Table D.18, to identify the duration for property- damage-only and injury or fatal crash types.

191 Table D.17. Incident Duration by Incident Type and Lane Closure Type Facility Type Incident Type Statistic Incident Duration (min) Shoulder 1 Lane 2 Lanes 3 or More Lanes Freeway Crash Range 20.5–69.5 32.9–59.3 31–73.7 30.7–97.4 Average 43.1 45.1 58.9 71.9 Median 40.4 38.9 66.2 75.8 SD 14.7 10.5 15.0 22.7 Sample size 6,099 5,476 2,424 1,685 Breakdown/disabled/stalled Range 7.2–54.1 16–58.1 23.5–72 16.8–263.1 Average 29.7 30.1 46.1 73.5 Median 27.4 27.8 42.1 40.0 SD 15.1 13.7 14.7 94.1 Sample size 59,606 8,043 339 211 Debris Range 8.1–76 6.7–53 14.1–83.6 15.4–54 Average 35.7 25.5 40.8 32.1 Median 35.7 21.5 32.1 26.5 SD 25.0 14.6 25.7 16.8 Sample size 1,242 2,925 431 156 Other Range 4.5–38.6 16.5–76.9 12–146.2 2.3–61.8 Average 26.0 46.8 77.3 40.8 Median 30.4 46.9 73.6 58.3 SD 15.1 24.7 67.2 33.4 Sample size 290 119 8 17 Arterial Crash Range 18.8–39.9 30.6–47 44.3–44.3 43.2–45.2 Average 29.3 38.8 44.3 44.2 Median 29.3 38.8 44.3 44.2 SD 14.9 11.6 N/A 1.4 Sample size 360 275 11 97 Breakdown/disabled/stalled Range 18.8–32 10.2–24.4 17–17 26.4–26.4 Average 25.4 17.3 17.0 26.4 Median 25.4 17.3 17.0 26.4 SD 9.3 10.1 N/A N/A Sample size 199 842 3 108 Debris Range 57.7–57.7 16.2–25.3 1.1–22 32.6–32.6 Average 57.7 20.8 11.6 32.6 Median 57.7 20.8 11.6 32.6 SD N/A 6.5 14.8 N/A Sample size 9 132 5 12 Other Range 41.9–47 51.4–70.8 123.9–123.9 9.2–65.8 Average 44.4 61.1 123.9 37.5 Median 44.4 61.1 123.9 37.5 SD 3.6 13.7 0.0 40.0 Sample size 14 67 10 21 Note: N/A = not applicable.

192 References AASHTO. Highway Safety Manual. American Association of State Highway and Transportation Officials, Washington, D.C., 2010. Cambridge Systematics, Inc. IDAS User’s Manual. Oakland, Calif., 2003. Federal Highway Administration. Highway Economic Requirements System: State Version: Technical Report. Washington, D.C., August 2005. Federal Highway Administration. ITS Research Success Stories: CLARUS System. http://www.its.dot.gov/CLARUS/. Accessed March 28, 2012. Table D.18. Incident Duration by Crash Severity Type Facility Type Duration by Crash Severity Type (min) Shoulder 1 Lane 2 Lanes 3 or More Lanes PDO I  F PDO I  F PDO I  F PDO I  F Freeway Range 18.1–65 42–94 27.4–61 40.9–52 31–78 32–72 29–76 29–76.9 Average 38.1 60.7 42.3 46.4 56.4 54.7 57.8 45.0 Median 36.0 48.0 37.8 48.0 58.4 53.5 63.1 29.0 SD 17.1 22.2 14.0 4.4 22.7 16.8 20.4 27.7 Sample size 2,563 247 3,104 482 1,158 382 710 448 Arterial Range 17–38 30–57 30–46 42–50 31–46 39–55 22–22 N/A Average 27.5 43.5 38.0 46.0 38.5 47.0 22.0 N/A Median 27.5 43.5 38.0 46.0 38.5 47.0 22.0 N/A SD 14.8 19.1 11.3 5.7 10.6 11.3 N/A N/A Sample size 317 39 197 75 50 33 1 0 Note: N/A = not applicable. Federal Highway Administration. The National Intersection Safety Prob- lem. FHWA-SA-10-005. Washington, D.C., November 2009. Highway Capacity Manual 2010. Transportation Research Board of the National Academies, Washington, D.C., 2010. National Climatic Data Center. Comparative Climatic Data for the United States Through 2010. National Oceanic and Atmospheric Administration, Asheville, N.C. http://www.ncdc.noaa.gov. Accessed Sept. 21, 2011. Yeo, H., K. Jang, A. Skabardonis, and S. Kang. Impact of Traffic States on Freeway Crash Involvement Rates. Accident Analysis and Prevention, Vol. 50, January 2013, pp. 713–723.