Safety Prediction Methodology and Analysis Tool for Freeways and Interchanges (2021)

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.

372 APPENDIX B: DATABASE ENHANCEMENT As described in Chapter 4, the data maintained by three state DOTs provide the foundation for the database used for model calibration. These data describe road inventory information for the state highway system in each state. The highways are represented as a series of consecutive road segments with a homogenous cross section and a length ranging from 0.05 to 1.0 mi. The state data were acquired from the Highway Safety Information System (HSIS). This appendix describes the process used to enhance the state data. Data enhancement consisted of using supplemental data sources to acquire additional data for each segment in the database. This appendix consists of two parts. The first part describes the procedural steps of the enhancement process. The second part summarizes the findings from a data verification activity that was undertaken to assess the consistency between data in a state database and that acquired from aerial photographs. DATA ENHANCEMENT ACTIVITIES The data enhancement activities focused on two tasks. One task was the development of data to describe for each segment the proportion of hours per day that are congested. Supplemental data were collected and used to derive this proportion for freeway segments. Automatic traffic recorder (ATR) data for the station nearest to each freeway segment were used for this purpose. These ATR data were acquired from the appropriate state agencies. The second task of the data enhancement activity was the use of aerial photography to collect additional data for each segment or ramp terminal. These photographs were obtained from Google Earth. The data collected include the width of key cross section elements, barrier presence and location, horizontal curvature, ramp configuration, turn bay presence, and median type. Data Extraction Process A process was developed to partially automate the extraction of data from aerial photographs. The process is based on the digitization of roadway design elements using aerial photographs that are keyed to a geodetic coordinate system. Using this process, the road alignments, the road cross sections, the barrier pieces, and the speed-change lanes are all manually digitized. Software is used to process the digitized locations and compute the desired quantities. Using this process, a technician digitizes the photograph of each segment, saves the measurement locations in a file, and submits the file to a computer program for error checking. At a later time, an engineer reviews the technician’s measurement locations by viewing them on the digitized photo. He or she then submits the file to a second computer program that computes the desired variable values. This process is described in more detail in the following sections.

373 This process is used to digitize cross sections and alignments. When applied to alignments, the engineer digitizes the horizontal alignment of a length of roadway and saves the file of digitized locations. He or she then submits the file to a computer program that computes (1) the coordinates of each segment begin milepost and (2) the geometry of each horizontal curve. Digitized locations along an alignment are saved in an “alignment file.” Digitized locations along a cross section are saved in a “segment file.” Details of these two files and the process by which they are assembled are provided in the next two sections. Digitized Alignment File Google Earth is used to develop the digitized alignment file. The start and end of an alignment is predetermined to include one or more target road segments, as well as the nearest interchange. Figure B-1 illustrates a digitized alignment for a 2.2-mile section of Interstate 5 in California. The alignment shown in Figure B-1 consists of 29 “placemarks” located along the alignment. A placemark represents a point of known latitude and longitude. Each placemark is shown using a push-pin symbol. For this reason, placemarks are hereafter referred to as “pins.” The digitizing process consists of three steps. During the first step, pins are located along the roadway reference line in the direction of increasing milepost. For freeway segments, this line is defined by the inside edge of traveled way for the increasing milepost direction. The pins are placed at the start of the alignment, the end of the alignment, and at a short spacing along each horizontal curve located between the start and end points. The maximum distance between pins along a curve is intentionally short such that errors in length measurement are negligible. When the alignment file is complete, it is saved in keyhole markup language (kml) format (Wikipedia, 2010b). This format is an xml-based, standardized file structure for describing geographic annotation in Internet-based Earth browsers. A more complete description of this language is provided at the Google Earth website (KML, 2010). During the second step, additional pins are placed along the alignment and saved in a second kml file. These pins correspond to known interchange gore points or intersections that are defined in the state database by milepost. Multiple gore points and intersections are located in this manner and then used to determine the milepost that corresponds to the start of the alignment file. During the third step, additional pins are placed along the alignment and saved in the kml file established in the second step. These pins correspond to the begin milepost of each road segment in the roadlog database provided by HSIS. The location of a segment begin milepost is determined by computing the difference between it and that determined to correspond to the start of the alignment file. Each begin milepost pin is located at the computed distance, as measured along the alignment reference line.

374 Figure B-1. Example digitized alignment. Steps two and three are automated using Earth Tools software developed for this project and described in a subsequent section. The geodetic coordinates (i.e., latitude and longitude) of each segment begin (and end) milepost are determined through this process. In terms of defining a feature’s relative location on a common photograph, the standard error of these coordinates is about ±0.1 ft. Thus, the standard error for a lane width or curve length measurement is about ±0.14 ft. In terms of defining a feature’s true earth location, the standard error is about ±30 ft. Digitized Segment File Google Earth is used to develop the digitized segment file. The segment file is used to describe the cross section, barrier, and speed-change lanes on a given segment. One file is created for each segment. Figure B-2 illustrates a digitized 0.273-mile segment on Interstate 5 in California.

375 Figure B-2. Example digitized segment. The segment shown in Figure B-2 consists of 59 pins located at key points along the alignment. One pin is used to define the begin milepost (i.e., 29.226) and one pin is used to define the end milepost (i.e., 29.499). One set of 14 pins are used to define the cross section elements (i.e., clear zone, outside shoulder, lane, inside shoulder, median, and median barrier widths) near the middle of the segment and another set of 14 pins are used to define the cross section at the end of the segment. A third set of pins are used to define the roadside barrier near the start of the segment (right roadbed) and that near the end of the segment (left roadbed). A fourth set of pins are used to define the gore points associated with the weaving section (left roadbed). A fifth set of pins are used to define the speed-change lane near the start of the segment (right roadbed). The digitizing process consists of locating the pins for each design element present on the segment. In practice, one technician is tasked with locating the cross section pins. Another technician is tasked with locating the barrier pins. An engineer is tasked with locating the pins for speed-change lanes (and weaving sections) due to their greater complication. All pins for a segment are saved in one kml file. The geodetic coordinates of each design feature are determined through this process. In terms of defining a feature’s relative location or length, the standard error of these coordinates is about ±0.1 ft.

376 Processing Software Computer software was developed to process both the alignment and segment files. This processing included reading the kml file, diagnosing the pin placements, and computing the desired database variables. The software was written as a Visual Basic for Applications macro in an Excel® spreadsheet. It is called Earth Tools. The welcome screen for Earth Tools is shown in Figure B-3. Figure B-3. Google Earth Calculation Tools welcome screen. The software includes a variety of tools useful to the researchers in developing or evaluating kml files. Each tool is provided its own spreadsheet, as accessed by the corresponding tab identified along the bottom of Figure B-3. Of particular note are the following tools: ● Curve Analysis ● Alignment ● Segment The purpose of each of these tools is described in the following subsections.

377 Zecef Yecef Xecef North Up East Prime Meridian Curve Analysis Tool The Curve Analysis tool is used to compute the geometry of each horizontal curve represented in an alignment file. The tool computes the radius, deflection angle, chord, length of curve on a specified segment, curve begin milepost, and curve end milepost. The tool converts the geodetic coordinates associated with each pin into earth-centered- earth-fixed (ECEF) Cartesian coordinates, and then into east-north-up (ENU) Cartesian coordinates (Wikipedia, 2010a). The ENU coordinates are desirable because they place the roadway in an x-y plane where x is east, y is north, and z represents elevation. The distance between any two pins is then computed using their x-y coordinates (the error caused by ignoring the elevation change is negligible for the distances being measured). The relationship between the ECEF and ENU coordinate systems is shown in Figure B-4. Figure B-4. ECEF and ENU coordinate systems. Curve radius is computed using an algorithm developed by Imram et al. (2006). This algorithm incorporates a non-linear regression procedure derived by Manthey (2010). The algorithm was adapted to use the pins in an alignment file (see Figure B-1). It can be used to compute the geometry of simple curves, two-centered compound curves, and three-centered compound curves. Geometric data are provided for each curve in a compound curve. A circular curve with spiral transitions is approximated as a compound curve and an average radius is computed for the spiral transition. The accuracy of the computed curve geometry was evaluated using data from one state database. The computed radii were found to be within 4 percent of the reported radii. The

378 computed deflection angles were found to be within 5 percent of the reported deflection angles. Further inspection of the data indicates that the larger deviations in either range occurs when the photograph quality is poor or when the curve can be characterized as having a short length and small deflection angle. Additional information about this evaluation is provided in the second part of this appendix. Alignment Tool The alignment tool is used to compute the coordinates of user-specified mileposts. Typically, the mileposts of interest are those representing the begin milepost of a segment. The computations require an alignment file (as described previously) that includes the segment of interest. They also require the user to input the milepost of the starting point of the alignment. The alignment tool then reads the file and defines the coordinates of the user-specified mileposts based on their distance along the reference line. Once computed, the coordinates for each user-specified milepost are exported to a new kml file. This file can be loaded into Google Earth and the computed points displayed on an aerial photograph of the roadway. This type of display is shown in Figure B-5 for a section of road comprised of five segments. Figure B-5. Display of computed begin mileposts for five segments.

379 Segment Tool The Segment tool is used to diagnose the segment file and compute the desired database variables. It includes an extensive set of routines that check the completeness and logic of the pins placed in the segment file. If a file is determined to have missing pins or illogical pin placements, then an error message is displayed indicating the nature of the error and a suggested means of correction. If a file is determined to be error-free, then the computed values for specified variables are displayed in a manner suitable for inclusion in the safety database. Table B-1 lists the variables computed from the segment files. The speed-change lane variables listed in the table are provided for up to four speed-change lanes per segment. VERIFICATION OF SELECTED VARIABLES Some of the data collected during the enhancement process were redundant to the data in the state databases. These data were used to verify the accuracy of the extracted data. The findings from this activity are described in this section. The objective of the verification process was to provide justification for using the enhancement process. It is not intended to suggest that state highway databases are inaccurate for their intended purposes. The verification process is based on the graphical and statistical comparison of selected variables. For each variable, data were extracted from aerial photographs and checked following the process described in the previous part of this appendix. These variables are referred to as “measured” variables. They are then compared to identically defined data provided in the state database. These variables are referred to as “reported” variables. Separate verification activities were undertaken for the cross section measurements and the alignment measurements. The findings from these two activities are described separately in the following two sections.

380 TABLE B-1. Variables computed from segment file Category Variable Description Roadway inc_drop-add_lanes Number of lane drops or adds on seg. for travel in increasing milepost dec_drop-add_lanes Number of lane drops or adds on seg. for travel in decreasing milepost out_shld_meas Outside shoulder width (average of both directions) lane_meas Lane width (average for all lanes in both directions) in_meas Width of both shoulders and median in_shld_meas Inside shoulder width (average of both directions) med_width_meas Width of median med_nontrav_meas Width of median barrier, if present Roadside med_type_meas Median type (1 = raised curb, 2 - barrier, 3 = depressed or unsurfaced) in_lane_barrier_len Total length of barrier adjacent to the lane in median in_shld_barrier_len Total length of barrier adjacent to the shoulder in median in_off_barrier_len Total length of barrier offset from the shoulder in median out_lane_barrier_len Total length of barrier adjacent to the lane on roadside out_shld_barrier_len Total length of barrier adjacent to the shoulder on roadside out_off_barrier_len Total length of barrier offset from the shoulder on roadside inc_clear_zone Average clear zone width for travel in increasing milepost dec_clear_zone Average clear zone width for travel in decreasing milepost Speed- Change Lane sc_design Design for speed-change lane (e.g., P=parallel, T=taper, etc.) sc_type Orientation of speed-change lane (e.g., entrance/exit, left/right side) sc_lgt_on_seg Length of speed-change lane on subject segment sc_ramp_lanes Number of lanes in the speed-change lane Weaving Section inc_A_lanes a Number of lanes on freeway and ramps before weaving section inc_C_lanes a Number of lanes on freeway and ramps after weaving section inc_D_lanes a Number of lanes on right-side entrance ramp before weaving section inc_E_lanes a Number of lanes on right-side exit ramp after weaving section inc_Lw a Length of weaving section (gore to gore) inc_wev_lgt_on_seg a Length of weaving section on segment dec_A_lanes b Number of lanes on freeway and ramps before weaving section dec_C_lanes b Number of lanes on freeway and ramps after weaving section dec_D_lanes b Number of lanes on right-side entrance ramp before weaving section dec_E_lanes b Number of lanes on right-side exit ramp after weaving section dec_Lw b Length of weaving section (gore to gore) dec_wev_lgt_on_segb Length of weaving section on segment Other ramp_exit_cnt Count of ramp exit gore points adjacent to segment ramp_ent_cnt Count of ramp entrance gore points adjacent to segment

381 1 2 1 2 4% 86%96% 14% 0% 20% 40% 60% 80% 100% Percent of Segments Measured Number of Lanes Reported Number of Lanes y = 0.3097x + 10.707 R2 = 0.0561 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Measured Lane Width (x), ft R ep or te d La ne W id th (y ), ft Cross Section Elements The findings for selected ramp cross section variables are described first. Then, the findings for selected freeway segment cross section variables are described. Data associated with ramp lane count and lane width are shown in Figure B-6. Figure B-6a indicates that ramps with one lane were correctly identified in the database as having one lane for 96 percent of the segments (4 percent of these segments were reported as having two lanes). Similarly, ramps with two lanes were correctly identified in the database as having two lanes for 86 percent of the segments (14 percent of these segments were reported as having one lane). a. Number of lanes. b. Lane width. Figure B-6. Ramp lane data comparison. Figure B-6b indicates a weak correlation between the measured ramp lane width and the lane width reported in the database. The figure indicates that there is considerable random variation in the error, and the “best fit” trend line suggests that there is some bias (e.g., ramps measured to have 10-ft lane width, tend to be reported as having a 14-ft lane width). This type of bias is particularly problematic because it translates into biased regression coefficients. From a statistical standpoint, random error in a variable can be overcome by increasing the sample size. Thus, it could be argued that using the entire state database (instead of just those segments that can be manually verified using data from aerial photographs or similar) would overcome the random variation shown in Figure B-6b. However, random variation in the independent variable of a regression model will bias the regression coefficients. Through simulation experiments, Weed and Barros (1987) found that significant variability in the independent variable causes bias in the regression model coefficients. This variability also increases the model’s residual error and makes the t-tests of model coefficients less efficient. The findings from a comparison of ramp shoulder width are shown in Figure B-7. The random error is notable, as is the bias in the data. The findings from a comparison of freeway median width and left (inside) shoulder width are shown in Figure B-8. Again, the random error and bias in the shoulder width data is notable. The median width does not appear to exhibit significant bias but the random error is large.

382 y = 0.1412x + 3.6289 R2 = 0.049 0 2 4 6 8 10 12 14 16 18 0 5 10 15 20 25 Measured Left Shoulder Width (x), ft R ep or te d Le ft Sh ou ld er W id th (y ), ft y = 0.3985x + 5.0456 R2 = 0.2336 0 5 10 15 20 25 0 5 10 15 20 25 Measured Right Shoulder Width (x), ft R ep or te d R ig ht S ho ul de r W id th (y ), ft y = 0.9101x + 20.314 R2 = 0.7223 0 100 200 300 400 500 600 5 105 205 305 405 505 Measured Median Width (x), ft R ep or te d M ed ia n W id th (y ), ft y = -0.0067x + 6.3597 R2 = 0.0002 0 2 4 6 8 10 12 5 10 15 20 25 30 Measured Left Shoulder Width (x), ft R ep or te d Le ft Sh ou ld er W id th (y ), ft a. Left shoulder width. b. Right shoulder width. Figure B-7. Ramp shoulder data comparison. a. Median width. b. Left (inside) shoulder width. Figure B-8. Freeway shoulder and median data comparison. Horizontal Curve Geometry Curve data for ramp and freeway segments in one state database were computed using the Curve Analysis tool and alignment files. The findings from the evaluation of the ramp curve data are shown in Figure B-9. Each data point shown represents one curve. The 28 curves included in this evaluation were randomly selected and rationalized to be representative. The graphs shown (and statistics cited) in Figure B-9 indicate that there is good agreement between the measured and reported curve radius and deflection angle on ramps. The trend line shown in each graph is a line of “best fit” based on a linear regression analysis. The t- statistic for the slope of the regression line in Figure B-9a indicates that the slope is not significantly different from 1.0. Also, the intercept for this line is not significantly different from 0.0. Similar results were found for the regression line coefficients shown in Figure B-9b.

383 y = 0.9799x + 18.189 R2 = 0.9677 0 500 1,000 1,500 2,000 2,500 3,000 0 1,000 2,000 3,000 4,000 Measured Curve Radius (x), ft R ep or te d C ur ve R ad iu s (y ), ft y = 1.0023x - 1.379 R2 = 0.9853 0 50 100 150 200 250 0 50 100 150 200 250 300 Measured Deflection Angle (x), degrees R ep or te d D ef le ct io n A ng le (y ), de gr ee s a. Curve radius. b. Deflection angle. Figure B-9. Ramp curve geometry data comparison. A small number of data points in Figure B-9a indicate a notable difference of several hundred feet between the measured and reported radius values. A closer inspection of these points indicated that the difference may be due to the inclusion of short spiral transitions (or compound transition curvature with a short, large-radius curve) prior to a sharp curve on the ramp. Transition curves are difficult to visually detect from photographs, especially if they have a short length and small deflection angle. If they were not detected, the measured radius would represent an average value for the combined circular curve and transition curve. A couple of the deflection angles shown in Figure B-9b indicate a notable difference of several degrees between the measured and reported values. An examination of the data indicated that the measured values from curve analysis tool were in agreement with manually measured deflections taken directly from the aerial photograph with a protractor. It is recognized that photograph quality could explain a few degrees of deviation in extreme cases, but not the several degrees found for a couple of curves. It is believed that these curves have spiral transitions and that the deflection angle reported in the state database is for the circular portion of the curve, rather than the total deflection in the alignment. The findings from the evaluation of the freeway curve data are shown in Figure B-10. Each data point shown represents one curve. The 26 curves included in this evaluation were randomly selected and rationalized to be representative. The graphs shown (and statistics cited) in Figure B-10 indicate that there is good agreement between the measured and reported curve radius and deflection angle on freeway segments. The trend line shown in each graph is a line of “best fit” based on a linear regression analysis. The t-statistic for the slope of the regression line in Figure B-10a indicates that the slope is not significantly different from 1.0. Also, the intercept for this line is not significantly different from 0.0. Similar results were found for the regression line coefficients shown in Figure B-10b.

384 y = 1.01x - 35.261 R2 = 0.9965 0 1,000 2,000 3,000 4,000 5,000 6,000 0 2,000 4,000 6,000 8,000 Measured Curve Radius (x), ft R ep or te d C ur ve R ad iu s (y ), ft y = 0.925x + 0.7028 R2 = 0.9486 0 20 40 60 80 100 0 20 40 60 80 100 120 140 Measured Deflection Angle (x), degrees R ep or te d D ef le ct io n A ng le (y ), de gr ee s a. Curve radius. b. Deflection angle. Figure B-10. Freeway curve geometry data comparison. A few of the deflection angles shown in Figure B-10b indicate a difference of several degrees between the measured and reported values. The reasons for these deviations are the same as offered in the discussion of Figure B-9b. REFERENCES Imran, M., Y. Hassan, and D. Patterson. (2006). “GPS-GIS-Based Procedure for Tracking Vehicle Path on Horizontal Alignments.” Computer-Aided Civil and Infrastructure Engineering. Vol. 21. Blackwell Publishing, Malden, Massachusetts, pp. 383-394. KML Documentation Information. (2010). Google Earth. http://code.google.com/apis/kml/documentation/. Accessed June 10, 2010. Manthey, David. (2010). “General Least-Squares - Direct Solutions and Bundle Adjustments.” http://www.orbitals.com/self/least/least.htm. Accessed June 10, 2010. Weed, R.M., and R.T. Barros. (1987). Demonstration of Regression Analysis with Error in the Independent Variable. Transportation Research Record 1111. Transportation Research Board, National Research Council, Washington, D.C., pp. 48-54. Wikipedia. (2010a). “Geodetic System.” Wikipedia Foundation, Inc. http://en.wikipedia.org/wiki/Geodetic_system. Accessed June 10, 2010. Wikipedia. (2010b). “Keyhole Markup Language.” Wikipedia Foundation, Inc. http://en.wikipedia.org/wiki/Keyhole_Markup_Language. Accessed June 10, 2010.