UTRAC Active Mode Shift Potential Research & Data

This research develops and evaluates a new Traveler Alignment  tool  (please click to access actual tool) for approximating car trips traveling along existing or proposed network segments that may reasonably be shifted to active modes by assessing the proximity and parallelism of "origin-destination" (OD) desire lines to segments for analysis. The tool functionality allows for it to operate on new and existing connections, essentially replicating a select-link analysis while avoiding the high resource costs of traditional travel demand model runs.

The purpose of this research is to understand which active transportation facilities have a high potential for mode shift, which will be a potent metric of success for funders seeking to make high impact transportation investments. Recently, Utah Department of Transportation (UDOT)'s Research & Innovation Division (UTRAC) funded research that investigated the development and validation of a traveler alignment analysis tool that examines the orientation and magnitude of short trips in OD data, in order to evaluate mode shift potential. This Story Map presents the data and methods that were used to conceptualize and develop this tool.

Traditionally, functional classification - the process by which streets and highways are grouped into classes according to the type of service they are intended to offer - provided an understanding of the tradeoffs between facilities serving mobility and accessibility needs. Functional classification generally terms roadways as arterial, collector, and local (Figure 1). However, higher order functional classifications are generally more apt for motorized vehicular travel and pose barriers to active transportation modes.

Therefore, this research was initiated to explore conceptual frameworks such as mobile trace or modeled data that might help in developing an active transportation functional classification system. In order to address the need for a more actionable outcome, the research team at Alta Planning+Design and UDOT collaborated on the development of a proof-of-concept mode shift potential tool for active modes to model travel behavior across the State of Utah, based on facility trip distances and trip alignment to proposed facilities. This tool could be a building block for identifying whether there are capturable short trips aligned with UDOT facilities and prioritize Transportation Investment Fund (TIF) active funding for high impact facilities.

There are four objectives of this research:

1) To review functional classification and its relationship to active transportation. 2) To identify what components accompany multimodal functional classification systems with special attention to how travel behavior and the built environment influence their contextual application. 3) To develop a mode shift potential tool that can associate desire lines derived from an OD matrix. 4) To review emerging and existing data sets and whether they have utility for future UDOT TIF active projects including StreetLight Data and Replica Places.

What types of trips are often used for active transportation (Figure 2)?

- Trips that are short and destinations are near. - Where options for safe and comfortable travel exists.

Considering the estimated active transportation mode shift potential from the tool output, within a functional classification framework, may help contextualize a proposed facility within the larger transportation network. The type of proposed facility should be suitable for serving active transportation trips from volume and trip characteristic standpoints.

The tool presented in this research creates a mode shift potential index on existing or new facilities that may reasonably be converted to active modes using trip data represented by OD desire lines.

Implemented as an ArcGIS Pro tool, the workflow determines the minimum distance and difference in bearing between each desire line and input network link for evaluation. It then applies weights based on defined proximity and angle thresholds to estimate the fraction of trips represented by each desire line that may be attributed to the network link.

Next, it considers the average trip distance of all trips represented by the desire line and applies an additional weighting procedure to estimate which trips attributed to the network link may reasonably be converted to active modes, acknowledging the role of trip distance in mode choice.

Central to estimating mode shift potential is identifying what trips may reasonably be assumed to be converted from vehicle to active transportation modes. This requires a fundamental understanding of the complicated and competing factors at play when a person decides on their mode of travel for any given trip, such as trip purpose, escort trips, and trip-chain. For the purpose of this research, three factors have been selected as tool parameters: trip distance, alignment (parallelism) of desire lines, and proximity of desire lines (Figure 3).

Accessibility is an important factor in influencing non-automobile travel. Specifically, shorter trips and travel in places with greater pedestrian access or shorter distances to non residential destinations were more likely to be made by walking. These built environment relationships became formalized into the multiple “D” variables in which distance was embedded directly into the “destination accessibility” variable as a measure of proximity and ease of access to destinations, and into the “distance to transit” because one can think of public transit as a way to extend a walk trip over a longer distance.

The latest travel survey data (Figure 4 - 2017 National Household Travel Survey (ORNL, n.d.)) and (Figure 5 - the 2012 Utah Travel Study (RSG, 2013)) showed that in general, both walking and bicycling mode shares decline with increasing trip distances, but the shapes and thresholds are different for each form of active transportation.

To adjust mode shift potential estimates to the network link scale, only trips taken along the link of interest were considered. Desire lines represented by OD pairs were used to assemble corridors of demand, with the goal of capturing maximum demand. Trips taken along a desire line were assigned to a corridor based on the relative directional alignment, for which the maximum angle between the corridor axis and desire line was postulated to be 22.5 degrees (Figure 6).

Active trip mode choice is highly sensitive to trip distance. Trips that are parallel but not very close to the corridor are unlikely to use a given network link because of the distance costs associated with traveling to and from the corridor. The influence width, which refers to the zone buffer around a potential demand corridor in which trips taken along a desire line will route through the corridor (Figure 7), depends on the trip mode. For active travel modes, pedestrian or cycling corridors having estimated widths of up to 100 meters, although some studies suggest using approximate widths, rather than fixed, to maintain flexibility in trip assignment.

Emerging active transportation data sources include OD data can now be obtained from location-based services data sets and providers like StreetLight Data (2020) and Replica (2020). The relative ease of mass location-based services data collection allows for greater sample sizes and more up-to-date information on changes in mobility patterns. Additionally, because these emerging transportation data sources rely on the same travel demand modeling framework as traditional travel demand models, vendors like StreetLight Data can provide detailed select-link analyses for specific existing network links that may be used in calibrating and evaluating tool performance.

The need for anonymization and spatial aggregation of OD data poses challenges for understanding current or future potential levels for active transportation. Post-processing efforts are needed to disaggregate OD data into network volumes on transportation facilities; unfortunately, this process is especially difficult for short walk and bicycle trips which can be misrepresented as zone sizes increase. One potential solution is through “jittering,” which introduces controlled randomness to the disaggregation process (Figure 8). A jittering process could be used to disaggregate OD data for a city, region, or state (and for different time periods) and generate link-specific estimated volumes of short trips that are or could be made by walking or bicycling.

This analysis relied on two key data sources to test and validate the performance of the developed tool: Replica and StreetLight. The methodology requires OD pair desire lines, which are available as an intermediate output from travel demand models, and produces a link-level trip estimate similar to a select-link analysis. The OD data was derived from Replica’s activity-based travel demand model, and tool performance was validated by cross-checking with select-link analysis performed on 25 segments within the StreetLight Data platform.

Replica is a data vendor that produces travel data analytics (Figure 9). It uses mobile location, census, and land use data that are fed into an activity-based travel demand model (Replica Places) to create a synthetic population and simulate trip-taking behavior. Model results are calibrated with ground truth data to produce a micro-simulation of all movements within the region.

For instance, data was simulated for a representative travel weekday during fall 2019 and filtered to include trips with destinations in the state of Utah. Replica then produced a trip table with an individual record for each modeled trip containing information such as the Traffic Analysis Zone (TAZ) in which the trip originated and terminated, mode, duration, distance, and purpose (Table 1).

In post-processing, the raw trip table (Table 1) was processed in Python to convert it to an OD matrix (Table 2). Trips with common origin and destination TAZs were aggregated into OD pairs with the count of associated trips taken by each mode and the average distance of all associated trips. The OD matrix was then joined to Utah TAZs, represented spatially in ArcGIS Pro. Each record in the OD matrix was represented by a straight line connecting the geographic centroid of the origin TAZ to the geographic centroid of the destination TAZ. Trips with origins outside Utah were excluded.

Trip details, including TAZ data, used country-wide building footprint data by Microsoft Maps to identify potential trip generations and attractions for implementation of OD line disaggregation via jittering. Jittering allows for intrazonal trips to be treated in the same manner as interzonal trips by assigning a random building within the TAZ to each the origin and destination and representing the OD pair as a straight line between the two (e.g. Figure 10).

Each TAZ was required to have at least two points within for trip origin and destination assignment. Thirty one of the 2,731 TAZs had one or fewer buildings reported and were supplemented by building points assigned to the geographic centroid of the TAZ. All buildings were weighted evenly, meaning they had an equally likely chance of being randomly selected as an origin or destination point in the jittering disaggregation process. The original OD lines were jittered with three different thresholds, providing increasing levels of disaggregation. The disaggregation threshold indicates the maximum number of trips that will be represented by a single line, so higher thresholds indicate lower levels of disaggregation. This research investigates OD lines jittered with thresholds of 100, 50, and 10 trips.

Similar to Replica, StreetLight Data processes anonymized smart phone and navigation device location data to produce transportation analytics, particularly at the link level. This research used select-link analysis performed by StreetLight Data to determine summary metrics about all vehicle trips taken through the network link during an average travel weekday in fall 2019. The individual links and analysis settings are set in the StreetLight Data online interface (Figure 11). The raw output provided by StreetLight Data contains information like the total volume of trips on the link, average travel time, average trip distance, and percentage of total trips in various distance bins.

Select-link analysis was performed on 25 links, each about 1 mile long and spanning urban, suburban, and rural land-use types across Utah. Links were selected from road segments with planned active transportation facilities as part of the statewide bike plan to ensure a diversity of functional classification, land-use context, and geographic coverage. The roadway functional classification and land use context were recorded for each link using roadway data provided by UDOT, points of interest from OpenStreetMap, and aerial imagery, respectively.

The Traveler Alignment tool was designed to identify the number of vehicle trips taken along a link that have the potential to be shifted to active modes. A select-link analysis of 25 StreetLight Data zones provided calibration data to compare model results to and to refine input parameters. Parameter settings were adjusted using a qualitative gradient descent method to optimize weighted thresholds for OD line trip distance, parallelism, and proximity.

Once calibrated to unaltered OD lines, the desire lines were disaggregated via jittering at three different levels to assess potential improvements in tool performance. Estimated potential conversion trips identified by the Traveler Alignment tool and through select-link analysis from StreetLight Data were compared to calculate estimation error using root mean square error (RMSE) and mean absolute error (MAE) metrics, and checked for correlation using Pearson and Spearman tests.

Statewide trip characteristic summaries were derived from the Replica Trip Table for all modeled trips within the state of Utah for a typical Thursday during the fall of 2019. Analysis of short trip-making behavior confirmed the potential for active trip conversion at a statewide level. Figure 12 shows the percentage of all trips taken by walking modes as compared to the percentage of all trips that are less than 1 mile in distance, summarized by UDOT transportation district. In all districts, over half of trips less than 1 mile are currently made by walking. District 4 in the southernmost part of the state has the greatest percentage of total trips less than 1 mile, but the lowest percentage of total trips made by walking. District 2, home to Salt Lake City, has the smallest percentage of trips less than 1 mile, but nearly double the number of total trips of any other UDOT district (Table 3).

Examining the percentage of trips made by bike compared to trips less than 3 miles and 5 miles (Figure 13) revealed a huge opportunity for converting short trips to active modes. Specifically, trips less than 3 miles are considered to have Active Trip Potential (ATP), meaning the possibility of being taken by walking or biking. Statewide, less than 5% of trips less than 3 miles are made by biking, and no district has a bike mode share greater than 2%. The map presented in Figure 4.3 shows ATP at the TAZ-level statewide.

Pockets near city centers tend to have higher ATP than suburban or rural land uses, though the size of the city does not appear to matter. Between 55% and 65% of trips ending in TAZs at the heart of the smaller cities of Junction and Fillmore are less than 3 miles, consistent with downtown Salt Lake City and Ogden (Figure 14).

When considering the density of ATP trips per square mile, bigger cities have dramatically higher trip densities, given the presence of more trip takers in areas with increased population density. The following map (Figure 15) shows the density of ATP trips per square mile, which is the highest near high-density urban cores.

State- and district-wide summaries highlight an overall potential for shifting short car trips to active modes, but stop short of identifying where those trips occur. While static maps and tables can communicate quantitative trip behavior, dynamic visualizations can provide greater detail by allowing viewers to zoom in and see both the origin and destination of trips, rather than a TAZ-level aggregation of only trip ends. The flow map uses census block group level OD data from Replica to map high active trip potential trip pairs (private auto trips less than 1, 3, or 5 miles), and all walk and bike mode trips in Utah.

Vehicle trips identified in the StreetLight Data select-link analysis were determined to have active mode conversion potential by applying the same distance thresholds and weights as the parameter set being evaluated. This number was compared to the number of active mode conversion potential trips as identified by the developed Traveler Alignment tool to assess tool performance. A sensitivity analysis employed a qualitative gradient descent method to alter tool parameters and assess the impact on tool performance.

Tool performance may be directly evaluated by comparing the number of trips identified by the Traveler Alignment tool to the comparable trips identified by the select-link analysis. Each parameter was varied individually relative to a baseline parameter set to produce a qualitative gradient descent process designed to identify the optimal tool parameter settings. This sensitivity analysis was performed and evaluated on the original, non-jittered OD lines to determine the best performing parameter set, and then those parameters were applied to jittered OD lines to assess how disaggregation impacts tool performance.

Each tool run produced a segment level estimation of the number of trips with active mode conversion potential based on the applied parameters described previously. The results were compared with the number of trips on the segment as identified through the StreetLight Data select-link analysis with the appropriate trip distance thresholds and weights applied. For example, in the baseline parameter set, the number of trips identified through the select-link analysis includes 100% of trips less than 3 miles, 30% of trips between 3 miles and 5 miles, and 10% of trips between 5 miles and 10 miles. This would compare each set of mode shift potential trip distance thresholds from the alignment analysis tool to its corresponding index derived from StreetLight Data’s pass-through zone analysis.

The difference in trip counts identified by the tool and the select-link analysis as potential active mode conversion trips was evaluated using RMSE and MAE estimation error metrics. Both RMSE and MAE measure the average magnitude of error in the estimated value, treating trip volumes from the select-link analysis as ground truth. RMSE is more sensitive to outliers because the errors are squared before they are averaged, whereas MAE treats the magnitude of error linearly.

Estimated trips with active mode conversion potential are compared to the number of trips identified via select-link analysis in a joint plots (Figures 16) for the baseline and enhanced parameter sets applied to the original OD lines and those that have been jittered to a disaggregation threshold of 10.

The histograms on each axis show the distribution of estimated and actual trips by land use context. Select links in rural land use contexts generally have lower volumes of trips than suburban and urban links. In both the baseline and enhanced tests, the tool identifies the fewest number of trips with active mode conversion potential on links in rural areas, but the distributions of estimated trips are almost evenly spread with no apparent peaks for segments in suburban or urban land use contexts.

Results from each tool parameter set were also assessed against StreetLight Data selectlink analysis results using Pearson and Spearman correlation. In all parameter sets, there were moderately strong positive correlations between the estimated number of active mode conversion potential trips as identified by the Traveler Alignment tool and the StreetLight Data select-link analysis. Spearman correlation coefficients were generally lower than the Pearson coefficients, and tended to vary less between parameter sets. The final tool parameter settings produced results with a strong Pearson correlation score of 0.67, indicating a statistically significant positive correlation between the number of active mode conversion trips identified by the tool and by the select-link analysis.

This research developed and evaluated a quick-response Traveler Alignment tool that provided estimates of short trips aligned with on-street or off-street facilities for the purposes of identifying mode shift potential. The tool functionality allows for it to operate on new and existing connections, essentially replicating a select-link analysis while avoiding the high resource costs of the traditional travel demand model. The proposed tool operates on OD desire lines, which are a more readily available intermediate output of travel demand models that represent trip-making behavior via geographic aggregation. These desire lines were evaluated for their relative parallelism and proximity to the provided corridors, and trips were filtered based on literature-guided values of active mode trip distances.

While this analysis methodology can help provide quick-response understanding of which facilities are likely to align with existing short trips, it does not replace the fidelity and utility provided by an actual select-link analysis from a travel demand model. Select-link analysis is more likely to capture aggregate network effects and route choice dynamics that this tool would not capture (Castiglione et al., 2015; Brustlin et al., 2012). This tool’s key advantage is that it can provide comparable assessments between on-street and off-street facilities without the network editing that is required to reflect changes in a travel demand model approach. However, this comes with the drawback that the results from this tool are effectively an index rather than a measurement of trips that might be possible with a model or use of mobile data–derived metrics.

The tool can be used in prioritizing investments by UDOT or its partners to evaluate how proposed facilities align with potential demand.

The tool can also be used to inform future planning by integrating arbitrary desire line data derived from OD data. This data can be derived from different travel demand models and provide insights into how future land use might influence future “corridors of demand” to have mode shift potential.

The analysis conducted to tune the parameters that influence the Traveler Alignment analysis tool’s outputs have a few limitations.

Firstly, the sample size for this analysis was relatively small at 25 zones. While this is a larger number of zones to use for an analysis tool calibration using mobile data, it is very small relative the extent of Utah’s road network.

Secondly, the small sample size required a non-random selection of street segments to vary the distribution of data points across different facility types and land use contexts.

Thirdly, the comparability of Replica Places data to StreetLight Data is not entirely understood because the data sets are derived from different sources with one being sourced from a model and another being a mobile data derivative. The comparison provides rough indicators into which parameters are suitable as part of the calibration rather than exact measurements.

Fourthly, it is unrealistic to expect all short trips to be possible to convert to active transportation (TfLa, 2017; Mackett, 2003). Even if supportive infrastructure is provided, there are a number of reasons why a trip would still be made by nonactive modes, including heavy loads, travel trip type, personal preference, physical impairment, and seasonal weather. The complexity of these variables is far beyond what may be represented in an OD desire line data set, and thus these factors are not considered in this analysis.

Fifthly, The application of a jittering technique for disaggregating OD desire lines by randomly assigning trip origins and destinations to building footprints within the aggregation geography produced mixed results with regard to intrazonal trips. Jittering the data allows for intrazonal trips to be treated in the same manner as all other OD pairs; however, there is no spatial line representation of intrazonal trips in the original desire line format and trips are instead allocated based on the proportional coverage of the aggregation geography by a buffered area surrounding the segment. In rural areas where TAZs are larger and often filled with significant natural areas where no trips would reasonably happen. Intrazonal trips in rural locations are thus much less likely to have a constant trip density than in more dense urban areas where trip generators are more evenly distributed across the TAZ.

This research revealed scope for future studies. Future research may consider implementing weighted jittering of the OD flow data so that destinations with greater trip generation or attraction potential are selected more frequently than random choice. Buildings or street network segments could be weighted by area, height, land-use, or local job or population density. Another area for potential research is have a more in-depth examination of extending the concepts behind this tool to weight or filter OD desire lines to trips that have suitable demographics or trip purposes for conversion. could take the form of an additional weight assignment used to adjust the importance of an aligned trip. This would enable a score that evaluates trip flows based on the proportion of suitable demographics and trip purposes within each.

The tool can also be used to to inform modal priorities and multi-modal functional classification systems, as well as to explore possible integration with Wasatch Front Regional Council (WFRC)'s bike model or similar modeling tools.

Bahbouh, K., Wagner, J. R., Morency, C., & Berdier, C. (2017). Travel demand corridors: Modelling approach and relevance in the planning process. Journal of Transport Geography, 58, 196–208. https://doi.org/10.1016/j.jtrangeo.2016.12.007

Cambridge Systematics, Vanasse Hangen Brustlin, Gallop Corporation, Bhat, C. R., Shapiro Transportation Consulting, & Martin/Alexiou/Bryson. (2012). Travel demand forecasting: Parameters and techniques (NCHRP Report 716). National Corporative Highway Research Program. https://doi.org/10.17226/14665

Castiglione, J., Bradley, M. A., & Gliebe, J. (2015). Activity-based travel demand models: A Primer. Transportation Research Board. Retrieved from https://www.nap.edu/catalog/22357/activity-based-travel-demand-models-a-primer

Lovelace, R., Félix, R., & Carlino, D. (2022). Jittering: A computationally efficient method for generating realistic route networks from origin-destination data. Transport Findings. https://doi.org/10.32866/001c.33873

Mackett, R. L. (2003). Why do people use their cars for short trips? Transportation, 30(3), 329- 349. https://doi.org/10.1023/A:1023987812020

Oak Ridge National Laboratory (ORNL). (n.d.). National household travel survey. Federal Highway Administration. https://nhts.ornl.gov/

Replica. (2020). Replica methodology. https://www.sacog.org/sites/main/files/fileattachments/ replica_methodology_2020.pdf?1602683559

Resource Systems Group (RSG). (2013). Utah travel study. Wasatch Front Regional Council. https://wfrc.org/MapsData/UtahTravelStudy/UtahTravelStudy_FinalReport_130228.pdf

Transport for London (TfL). (2017a). Analysis of walking potential 2016. https://content.tfl.gov.uk/analysis-of-walking-potential-2016.pdf

StreetLight Data. (2020). Bike and pedestrian metrics: Methodology, data sources, and validation (Version 4.0). StreetLight Data. https://learn.streetlightdata.com/bike-and pedestrian-methodology-validation

For access to the mode shift potential tool and the materials associated with its use, please contact Muna Shah at mshah@utah.gov. Details for each of the files available for access are provided below:

• UTRAC_Toolbox.tbx – This ArcGIS toolbox contains the Mode Shift Potential Tool in the form of a script. Please note that the toolbox should be accessed through Esri’s ArcGIS suite, preferably ArcGIS Pro.

• Mode Shift Potential Tool User Guide.docx – This document explains how to use the tool in ArcGIS Pro, supplemented by textual descriptions and screen captures.

• UTRAC Mode Shift Tool Training Video.mp4 – This is a video demonstration of how the tool is run in ArcGIS Pro, and includes a detailed explanation of the input data that is incorporated into the tool.

• UTRAC_Mode_Shift.gdb.zip – This is a geodatabase that contains sample data for running the Mode Shift Potential Tool. Each data file is described in detail in both user guide document and the training video.