Multiple datasets and repositories were utilized to accurately predict locations best suited for HFST. The datasets provided and collected by KYTC and WDM USA include pavement friction data, crash data, highway information systems (HIS), and resurfacing data. These datasets were combined into one central location through a combination of Microsoft Excel and Geographic Information Systems (GIS). Upon combining the datasets, the sites were efficiently prioritized based on highest risk locations to set up a desktop review and maximize KYTC's traffic safety return on investment. The datasets combined are shown in Figure 2 and described in further detail throughout the data subtopics below.

Upon collecting and combining all the data, a desktop review was conducted to identify sites for field visit. The data was combined spatially through a GIS exercise with the pavement friction data segments being the base segmentation through route unique, begin mile point, and end mile point for location identification.

Crash modification factors (CMFs) were used to estimate the number of crashes saved by applications of HFST. The CMF for curves and signals used was 0.37, ID 10327 (Merritt et al., 2020), and the CMF for ramps was 0.21, ID 10341 (Merritt et al., 2020). The CMFs applied to the five-year crash data provided an estimate of crashes saved. Due to the CMF applying to all crash severities, the USDOT estimated crash costs to be an average of $124 130 (Harmon et al., 2018). The projected number of crashes saved was multiplied by the average crash cost to get an estimate of crashes saved. The crashes saved, or the benefit, was then divided by the estimated HFST cost for each segment to provide an estimated five-year crash return on investment (ROI).

The sites were first sorted by highest friction scores and then potential five-year crash return on investment, where ‘5’ indicates the lowest MSC and MPD range and a higher ROI would suggest a strong benefit-to-cost ratio. The top sites were checked through the desktop review to utilize Google Maps, Google Earth, and the KYTC Photo Log viewer to confirm the data and to decide whether a field visit is required. The desktop review identified some other roadway issues, such as pavement condition or drainage issues, which would reduce the benefits of HFST.

Field reviews were conducted for the top sites identified after the completion of the desktop review. Field reviews looked at both objective and subjective items, including drainage issues, pavement type and condition, vehicle speeds, hard braking events, utilizing the ‘shoe test’ for friction confirmation, and other noteworthy items not seen in the data or the desktop review.

The ‘shoe test’ is a confirmation of the pavement friction. This is completed by the field employee rubbing their shoe on dry on the shoulder or off of the wheel path, rubbing their shoe in the wheel path in the direction of travel and perpendicular to the direction of travel, and finally placing water in the wheel path to repeat rubbing their shoe in the wheel path parallel and perpendicular to the direction of travel to subjectively report on whether the friction is poor or not. Figure 4 shows the shoe test being conducted during a field visit. Upon confirmation of the shoe test, pavement condition, and other noteworthy considerations, the field employees would suggest start and end locations of HFST from the field to begin desktop finalization and site selection.

The desktop finalization brought together the final sites from the field visit and data driven process to update the segment locations for HFST placement. The desktop finalization utilized Google Earth to put the suggested bounds for HFST, recombine crash data for the new bounds, read crash narratives if applicable, estimate the cost for the site, and readjust the five-year crash ROI. Based on the adjusted data and the field visit notes, sites were rated as excellent, good, maybe, and poor candidates for HFST to prepare for a final project team review and implementation.

Formulas were used for the three different location types identified, including ramps, intersections, and segment curves. HFST on ramps started HFST one hundred feet before the curve, went through the curve, and ended one hundred feet after the curve, all between the lane markings. HFST at signalized intersections utilized AASHTO Green Book Stopping Distance calculations for the start of the HFST from the point where typical queue lengths end. Intersection HFST placement ended ten feet beyond to stop bar. Segment curves utilized information shown in Table 2 from Texas Technology Institute (Brimley & Carlson, 2012) for HFST placement:

The final project team meeting before site selection brought together each site for the project team to decide on whether to move forward with a site for HFST application. The meeting utilized pictures and notes from the field, data collected, and priority based on the five-year ROI and cost. Upon confirmation of sites, the sites were packaged up and sent out for contract bids. To reduce costs and mobilization, sites were grouped in bidding by location proximation and site type.

At the beginning of the study, data mile points were not accurate to the location, but the geolocation was. The process was fixed by the next round and tied to KYTC's roadway GIS data appropriately. Due to the learning curve, the WDM USA model is able to be limited to less sites for final processing, hence why the data went from 10 000 sites to 5 000 sites between the two rounds.

The KYTC District resurfacing data was not initially understood. The data presented a list of corridors, but only certain corridors would become projects during the construction season. The KYTC Districts began providing a more detailed list of projects most likely to happen, and if the projects are not planned for the upcoming year, the chances are high for the following year.

Pavement condition became an issue on some sites. During the field visits, pictures were taken to show the pavement condition to the full project team. The field visit teams were analyzing pavement condition for excess cracking, potholes, and rutting. Some locations passed during the field visit, but between the field visit and project construction, pavement condition deteriorated. Traffic counts and truck traffic counts were added to the evaluation metrics for further supplemental data as this may impact pavement condition and lifespan of HFST projects.

Post-construction had some issues, specifically after snow events. One location had a pothole, most likely due to a mixture of pavement condition, snow/ice, and snow plows due to higher friction. This lesson learned is still under evaluation across all constructed sites.

The three different treatment location types included freeway ramps, signalized intersections, and segment curves. HFST along freeway ramps have been constructed over time through other projects due to the high results along similar projects. HFST placed at signalized intersections is newer and aims to have positive results in the future. Segment curves have been used in locations as a countermeasure. Some locations in this study were placed with lower five-year crash ROI's, specifically at segment curves in rural locations. These locations had a correlation of fatal and suspected serious injury crashes, paired with evidence of roadway departures upon the field visit. Although the five-year crash ROI is lower, this is due to the crash costs associated with all crashes and not specific to fatal and suspected serious injury crashes, which typically have a higher crash cost. This project was accepted for implementation to be proactive and studied. An example of a roadway segment curve with HFST application is shown in Figure 5.

HFST has a high ROI for safety, but it is still a higher cost than alternate treatments and is usually placed along shorter roadway segments. Other pavement friction or texturing treatments exist and are being evaluated by the project team for future development into the Kentucky pavement texturing program. These results aim to be reported upon completion as an add on to the HFST program.