Amtrak Rail Trespasser Analysis using a GIS Space-Time Approach

Abstract

This research explores Amtrak trespass incident data from 2011 to 2019 using a GIS spatiotemporal process. The objective is to evaluate incident characteristics based on space, time, incident factors, and statistical significance. Incidents were first analyzed at the megaregional level, revealing Northern and Southern California as the highest trespassing risk in the country, followed by the Northeast and Great Lakes megaregions. A new standardized point density approach was applied to reveal incident clusters representing high-risk localities. Then, the optimized and emerging hot spot methods were applied to the top four megaregions. The results showed four Amtrak corridors as hot spots, including three along coastal California railways and the Philadelphia region. Trends for incident report factors were analyzed (e.g., pre-crash activity, time of day, location of impact). “Walking” prior to impact, occurrence in the “afternoon,” and crash location “on the tracks” were found to be the most prominent incident characteristics for those factors.

Keywords

railway trespass safety megaregion GIS

Introduction

Railroad trespass incidents continue to be a high priority issue for rail authorities throughout the nation as more people lose their lives on railroads accidentally due to trespassing more than any other cause (Sumwalt, 2019). In particular, trespassing on railroad rights-of-way is the leading cause of those railroad deaths (approximately 44% of all railroad-related fatalities) (Laffey, 2019). The Federal Railroad Administration (FRA, 2011b) defines trespassers as “persons who are on that part of railroad property used in railroad operation and whose presence is prohibited, forbidden, or unlawful”; however, the trespasser action can be either intentional or accidental. Railroads, such as Amtrak, are required to report injuries and fatalities each month to the FRA using standardized codes, (FRA, 2011b). Previous studies that focus on FRA data are limited as they do not include suicide incidents (Topel, 2019a) and this is presented as a warning when evaluating and summarizing FRA fatality data. More recently suicide data has been displayed in aggregate form by state (FRA, 2020b). Therefore, exploring all types of incidents as well as spatial and temporal relationships of these incidents across the nation is valuable to taking action and preventing future incidents.

This research explores Amtrak trespass incident data (inclusive of all trespassing incident types including suicides) from 2011 to 2019 using a GIS spatiotemporal process. Based on the trespassing definition provided above, trespassing can be in various forms (vehicular, pedestrian, at a crossing, on open track, etc.) therefore, all trespassing incident types are included in the dataset and are formatted and structured based on the standard FRA (2010) Injury/Illness report 6180.55a. The primary objective is to evaluate incident characteristics over time, space, incident factor as well as statistical significance based primarily on megaregional boundaries. This research expands on Oswald Beiler et al. (2019) as it provides an additional 2 years of data and includes the temporal dimension in the analysis. Through the process, a new GIS method is developed to reflect geographically the distribution of the incidents without imposing jurisdictional boundaries (such as county boundaries) that may not be truly reflective of incident clustering or rail network connectivity. Also, statistical significance is used to present the factor prominence by time and by location. Lastly, planning and policy recommendations are provided as a way to connect the findings to methods for improvement and increased oversight at various governance levels from national to local organizations.

Background

This section includes a literature review on rail trespassing including national trends and countermeasure advancements. In addition, GIS and spatial techniques are investigated in terms of how they have been used previously to railroad data as well as their relevance to this spatiotemporal study.

Rail Trespassing in the United States

Railroad lines and related rail property are inherently hazardous and pose life-threatening risk to trespassers (DaSilva et al., 2019). Anyone who is on railroad property without permission is considered a trespasser. In terms of deaths, general members of the public are more likely to be killed rather than passengers or employees (Topel, 2019b). Based on FRA data, total annual trespassing related pedestrian fatalities have increased 18% from 2012 to 2017 (FRA, 2018). FRA reporting is required by all railroads based on the current FRA Guide (FRA, 2011b) however, the dataset does not capture a full comprehensive understanding of trespassing activities that do not result in injury or death (Searcy et al., 2019). As these incidents continue to increase, further investigation into spatial, temporal, and incident level characteristics are required to understand the extent of trespasser activity.

Previous studies have emphasized highway-rail grade crossing safety (Khattak & Luo, 2011; Park & Saccomanno, 2005), however, additional emphasis on pedestrian trespassing causes and activities is needed as well in order to understand the influence of drug-related factors (Craig, 2017) as well as more recent social media technology (Blizzard & Cudahy, 2017).

National trends

In the United States, 6,204 people were killed on railroad property between 2012 and 2017 (Laffey, 2019). As previously mentioned, crossing deaths of pedestrians is growing compared to vehicular occupants from approximately 10% in the late 1970s to 35% in the middle 2010s out of total crossing deaths (Topel, 2019b). When looking more closely at locations, the megaregional level can be used to analyze incident density. Given the large magnitude of railway infrastructure, the megaregional scale is appropriate for analysis. A megaregion is a large interconnected region defined by connected networks (Gottmann, 1961). Currently, the Regional Plan Association (2008) has defined 11 megaregions. Based on Oswald Beiler et al. (2019), Northern and Southern California megaregions, the Great Lakes, and the Northeast were the four megaregions with the highest Amtrak trespass incident rate. Those four megaregions included 81% of the suicides that fall into a megaregional boundary across the entire country (Oswald Beiler et al., 2019). Further analysis on how changes over time as well as the correlation to incident characteristics such as time of day, location, pre-crash action, etc. can be explored with regard to the megaregional scale. Knowing specifically when, where, and how these incidents are occurring can provide insight into specific countermeasure recommendations at the more regional level.

Countermeasure/Technological advances

As trespass incidents continue to be a national safety concern, the opportunity to develop and implement countermeasure techniques continues to be a focus of railroad organizations. Efforts to reduce rail deaths at rail crossings with regard to employees and passengers have been effective however, efforts to reduce railroad trespassing and suicide rates have not be as successful (Havarneau & Topel, 2019). Due to the inherent differences in how trespassers enter the track, countermeasure development is challenging and requires a separate investigation of improvements for suicides than for other trespassing actions. Ryan (2018) provides insight into specific behaviors of those at risk for suicide and developed a framework with five classes (emotion, appearance, posture/movements, activities, and interactions) that help to identify potential suicide activity. Also, Havarneau and Topel (2019) have created a chain of events leading to railway suicides versus accidental incident which separates out the differences in pre-crash actions. They developed RESTRAIL (Reduction of Suicides and Trespasses on RAILway Property) as a countermeasure selection guide (Hararneau & Topel, 2019). Gabree (2015) provides specific recommendations on types of countermeasures that can be implemented and Oswald Beiler et al. (2019) provide insight into the countermeasure decision-making process based on location of incident (station versus open track) as well as pre-crash actions. All of these provide insight into possible methods to begin countermeasure selection and implementation.

More recently, technology has played a role in countermeasure implementation. Video analytics as well as elaborate detection algorithms may lead to more effective detection of trespasser activity (Havarneau & Topel, 2019). Also, overall wider availability of video data through closed circuit television cameras can provide easier access to trespasser data (Zaman et al., 2019). Artificial intelligence algorithms are used to watch for and recognize trespassing events in real time and the more access to this recognition, the more opportunity for live sound alert systems as well as other intervention methods (call systems or lighting). This study focuses more on the existing data with regard to incidents but does provide recommendations at the megaregional level with regard to specific incident characteristics (such as pre-crash action) and changes over time in order to highlight frequent and consistent trespasser incident factors that can be addressed.

GIS and Statistical Methods

Given the spatial nature of railroad data, GIS serves as a critical spatial analysis and data modeling tool in evaluating and analyzing rail trespass incidents. ArcGIS (ESRI, 2016) includes unique spatial tools that are useful in understanding both where and when the incidents occur. The more basic approach of the traditional hot spot analysis using Getis-Ord Gi* (Getis & Ord, 1992) was applied in Oswald Beiler et al. (2019) in order to explore rail trespass density across megaregions. However, with the new ArcGIS Pro (ESRI, 2019a) platform, an emergent hot spot analysis is more readily accessible and combines the traditional hot spot analysis (ESRI, 2016) method with the creation of time-space cubes in order to explore the temporal dimension. This emergent hot spot analysis has been applied to previous roadway studies such as Cheng et al. (2019) and Kang et al. (2018), however, there is the opportunity to apply this method specifically to rail trepassing incidents. This study focuses on this opportunity as a way to further explore trespass incidents using a three dimensional approach.

Research Methodology

In order to explore Amtrak trespass data using a space time approach, a five step process was applied, as shown in Figure 1. Step 1, Data Cleaning, refers to the organization and filtering of the data for accuracy and consistency provided by Amtrak. Step 2, National Investigation by Megaregion, is the first analysis step which involves exploring the trespass incidents across the megaregional boundaries using a standardized point density approach. Step 3, Space-Time Analysis, focuses on the addition of the temporal dimension to the spatial data using an emerging hot spot analysis approach. Step 4, Temporal Factor Analysis, includes the application of temporal analysis to specific incident characteristics (also referred to as factors) for high priority megaregions. Lastly, step 5, Reflection, includes a reflection on the trends in trespass incidents based on incident factors and a connection to recommendations.

Figure 1.

Research Methodology Flowchart.

These five steps are applied to an Amtrak trespass dataset which includes trespass incident types as of August 2019. Amtrak data (formatted and structured on standard FRA Injury/Illness report 6180.55a) was used for this study as it is comprehensive as it includes suicide incidents, which is not included in the Federal Railroad Administration publicly available dataset. The following section explains in detail the application of these steps to the data.

GIS Space-Time Application to Trespass Incidents

The five-step research method was applied to the Amtrak trespass incident dataset. Each subsection below includes a detailed description of the step, the process followed, and the results of the application. This application expands upon previous work completed by Oswald Beiler et al. (2019) as it emphasizes a new temporal dimension. Understanding the temporal trends as to when and how frequent over time the incidents are occurring as well as trends in incident factors over time provides the opportunity for improvements at the regional as well as local levels. Also, this research emphasizes the role in GIS in exploring spatial-temporal relationships with regard to incidents so that agencies can apply similar analyses at their own jurisdictional level.

Step 1—Data Cleaning

The first step in the data exploration process was cleaning the data provided from Amtrak. The Amtrak dataset provided in September 2019 included 5,442 incidents in the United States (lower 48) on the Amtrak railway system and associated rail networks from as early as 1990. Associated networks include segments of the Caltrain Commuter Railroad Company system, Connecticut Department of Transportation, and the Maryland Area Regional Commuter Train Service. For each trespass incident provided, detailed information was provided in reference to the FRA (2011a) coding requirements such as location, time of day, time of year, and pre-crash action.

In order to explore the data in a consistent and comprehensive method, the following data cleaning steps were followed (as shown in Table 1). In general, the data was cleaned based on two specific location criteria:

Table 1.

Amtrak Railroad Trespassing Data Cleaning Process.

Number of incidents (%)	Cleaning process
5,442 (100)	Total incidents in Amtrak database as of 9/5/2019 (from as early as 1990)
−3,531 (64.9)	Incidents lacking latitude and longitude data (note May 2011 lat/long FRA regulation change)
−68 (1.2)	Incidents outside 8/1/2011 – 8/1/2019 range
−14 (0.3)	Incidents within 5 km linear distance of North American Rail Lines
−7 (0.1)	Incidents in Canada
=1,822 (33.5)	33.5% of original dataset used in analyses

Spatial accuracy—Incident latitude and longitude is within 5 km radius of a known North American rail line (Bureau of Transportation Statistics, 2019c).

Temporal accuracy—Incident occurred between August 1, 2011 and August 1, 2019 in order to have eight full years of data compared (over nine calendar years of comparison for ease, keeping in mind that the calendar years of 2011 and 2019 have fewer days). August 1, 2011 was selected as the lower bounding date for our study to reflect the regulation change in FRA Latitude/Longitude reporting style, initiated in May 2011.

The final dataset after the cleaning process resulted in 1,822 points (33.5% of total) as shown in Table 1.

For the spatial accuracy step, the boundary threshold of a 5 km radius from North American rail lines is a replicable metric for including/excluding points. Multiple radii were tested to determine 5 km to be the most comprehensive radius while still keeping within a reasonable distance from viable railroad tracks. It is recommended that for improvements in data reporting, a more accurate georeference of the latitude/longitude is provided to include more data as well as provide more detailed description of the rail line location.

To highlight the difference between the total number of people involved in railroad incidents and the actual number of collisional events, including single and multi-person collisions, the data was reported with respect to two categories: person incidents and event incidents. Event incidents include the time and location that a train struck an individual, a vehicle, or a group of people, whereas person incidents are reported individually for each person involved. Although FRA requires trespassing incidents always be reported at the person incident level, it is important to understand the relationship between single and multi-person events and other trespassing factors. After cleaning, between August 1, 2011 and August 1, 2019, 1,822 person incidents and 1,590 event incidents were recorded by Amtrak. From this, 12.7% of impacted trespassers were injured with at least one other person, and 10.5% of train impacts to trespassers injured more than one person.

Step 2—National Investigation by Megaregion

After the data was cleaned, an understanding of the spatial distribution of the incidents was explored using GIS mapping (Figure 2). All person trespassing incident points, megaregion boundaries and the Amtrak rail lines were mapped. Given the large scale of rail systems, the megaregional level was used to identify regions of high trespasser incidents. This reflects the initial mapping analysis provided by Oswald Beiler et al. (2019) with an addition of two more years of data. Providing a similar megaregional comparison of the data an understanding of changes over time can be explored, as discussed later in step 3.

Figure 2.

Amtrak Trespasser Incidents by Megaregion (source data from Amtrak, 2019; Bureau of Transportation Statistics, 2019a; ESRI, 2019b, 2019d; and Regional Plan Association, 2008).

Of the 1,822 person incidents, 1,531 occurred within a megaregion boundary, which is 84% of the cleaned dataset. Table 2 includes the breakdown of trespass incidents by megaregion. Due to an overlap in the Great Lakes and Texas Triangle megaregions, there are four points that are counted twice in the table, so although the values sum to 1,826, the actual total number of incidents is 1,822. Similar to the results of Oswald Beiler et al. (2019), the highest concentration of points in the US occurs within the Northern and Southern California megaregions, followed by the Northeast and Great Lakes megaregion.

Table 2.

Proportions of Clean Dataset by Distance from North American Rail Lines.

Rank	Megaregion	Incidents (% of total)	Incidents per 10,000 square miles of megaregion area	Fatal/suicide incidents (% of megaregion incidents)
1	Northern California	338 (18.6)	152	216 (64)
—	Not in a megaregion	291 (16.0)	1	146 (50)
2	Southern California	277 (15.2)	74	185 (67)
3	Northeast	257 (14.1)	48	187 (73)
4	Great Lakes	211 (11.6)	13	118 (56)
5	Piedmont Atlantic	189 (10.4)	31	105 (56)
6	Florida	98 (5.4)	26	53 (54)
7	Cascadia	86 (4.7)	26	59 (69)
8	Gulf Coast	42 (2.3)*	9	21 (50)
9	Texas Triangle	28 (1.5)*	7	13 (46)
10	Front Range	6 (0.3)	1	5 (83)
11	Arizona Sun Corridor	3 (0.2)	2	3 (100)
	Total	1822 (100)*

Four incidents are found in both the Gulf Coast and Texas Triangle megaregions which overlap, so the total is 1,826−4 = 1,822.

Event incidents were also mapped and analyzed within each megaregion. The ranking of megaregions by number of incidents in Table 2 applies to event incidents as well. The Northern California and Piedmont Atlantic megaregions show a higher proportion of multi-person event incidents than other large megaregions. Inversely, the Northeast megaregion show a significantly higher percentage of only single person events than any other large megaregion.

In addition to simply mapping the incidents, understanding the local jurisdiction and trends in frequency can be valuable when providing recommendations. Therefore, county level trends were explored using two different spatial methods. The first method is a traditional mapping by number of trespass incidents within a geographic boundary (in this case, the county level) (Figure 3).

Figure 3.

Number of Trespass Incidents within County Boundary (source data from Amtrak, 2019; US Census Bureau, 2019).

In step 3, another geographic boundary method (grid) is used in order to explore spatial-temporal relationships. Using jurisdictional boundaries has two fundamental disadvantages for comparing prevalent locations for railroad incidents:

Non-uniform boundaries like states, counties, etc. (which are represented by local governments capable of making policy changes and locally managing any issues) are commonly not of same size scale—not standardized for comparing number of points within boundaries. This causes places with high point counts and large boundaries (e.g., Southern California) to stand out at a much higher scale than areas of the country with smaller boundaries (e.g., the Northeast Corridor).

Grids do not represent a constant distance along the rail line—this is the same issue of non-standardization because essentially trespassing incidents are linear data and not areal. Any comparisons of point densities within grid cells is not a perfect representation of how dense the points are along a standardized length of track.

Therefore, to address these concerns, a second, and perhaps more descriptive and applicable, method is one developed for this particular research, referred to as the “standardized point density” approach. The technique followed is used to standardize trespass incident point density and/or clustering estimation along lengths of track. This is potentially the beginning of developing linear hot spot analysis tools across a two dimensional space. The method involves estimating relative density of points based on a fixed area defined by where the actual points are located—circular area of 5 km radius out from a point. This causes nearby points to be captured within another area, thus increasing the density above 1:1, as shown in equation (1).

1 p o i n t : 1 u n i t o f a r e a = π * {(5 k m)}^{2}

(1)

Highly clustered points will fall within another area and show high density. These generated areas are contained within county boundaries (US Census Bureau, 2019), and the standardized relative density of points for the county is displayed graphically as a statistic for just the county. County density measurement is based on the number of incidents within the county divided by the sum of local standardized area, generated around all points and clipped within the county. This new method is used to create a map of the United States that highlights the unnoticed high clustering densities throughout the east, Mississippi River valley, and Piedmont Atlantic megaregion (Figure 4).

Figure 4.

Counties Colored Based on Standardized Point Density (Source data from Amtrak, 2019 and US Census Bureau, 2019).

When comparing to Figure 3 which shows the point count within each county, California (specifically San Diego, Alameda, Contra Costa, and Fresno) is highlighted due to the skewed/anomalously high number of trespassing incidents. Figure 4, standardized incident density/clustering for the United States more accurately and visually compares the counties with local problem zones (high density/clustering) of California and the rest of the country since it does not use the county size/boundary as a basis for comparison. Therefore, in addition to highlighting cities in California, it also highlights cities including Philadelphia, Pennsylvania, Niagara Falls, New York, Greensboro-Raleigh, North Carolina, Portland Oregon, St. Louis, Missouri, El Paso, Texas, Tampa and Jacksonville, Florida, and Birmingham, Alabama, for example. In general, this method provides a more realistic understanding of where the high incident areas are throughout the country. Also, this method can be applied to space-time analysis by creating density maps for each year in order to see how density/clustering is changing through time.

Step 3—Space-Time Analysis

Assessing and mapping trespassing incidents involves two degrees of freedom: spatial distribution and temporal trends. Since Figures 2 to 4 show an inequal distribution of incidents or dense, high-risk areas across the nation, it is important to spatially explore hot spot areas within high-priority megaregions. Additionally, the number of annual trespassing incidents changes over time unequally across all megaregions or local urban areas. Figure 5 shows the trend of the Amtrak dataset from 2012 to 2018 for the top four megaregions. Data from complete years only are plotted in Figure 5. The overall trend suggests an average national increase of approximately 19 person incidents per year (6.5–10% annual increase). This increase suggests within the hot spots, local increases and decreases within each identified hot spot should be explored.

Figure 5.

Number of Person Incidents by Megaregion between 2012-2018.

In order to highlight areas in the country that have a statistically significant clustering of incidents, hot spot analyses were used in ArcGIS Pro to map high-priority megaregions (Northern and Southern California, Northeast, and Great Lakes). The Optimized Hot Spot Analysis (OHSA) was constrained to a 5-km buffered area along North American rail lines (California) or Amtrak main lines (Northeast and Great Lakes) to extract a higher precision result, versus using the entire megaregion boundary. An optimized square grid is created within the constraining buffered area polygons. The Getis-Ord Gi* statistic (Getis & Ord, 1992) is calculated for each grid cell based on a local spatial autocorrelation analysis (Cheng et al., 2018; Kang et al., 2018). These z-scores are used to determine the confidence level for the presence of a hot spot in a given grid cell. In the spatial OHSA, all trespassing person incidents are included; as a result, these hot spots report statistically-significant areas where the trespassing activity is occurring over an 8-year period. The hot spot maps for the top four megaregions are shown in Figure 6. Although the method for calculating statistical significance is the same for all megaregions, the exact parameters utilized to generate a hot spot grid differ between each megaregional analysis; therefore, hot spots are not directly compatible across all panels of Figure 6 and are not representative of similar numbers and densities of trespassing incidents.

Figure 6.

Person Incident Hot Spots Identified in Top Four Megaregions (source data from Amtrak, 2019; Bureau of Transportation Statistics, 2019a, 2019c; ESRI, 2019b, 2019d; and Regional Plan Association, 2008).

To apply a temporal dimension to the hot spot analyses, trespassing incidents can be grouped and compared across several time intervals. The space-time cube feature in ArcGIS Pro uses a square grid to spatially sort trespassing incidents, and then organizes them temporally by the date the incident occurred. Figure 7 shows a three dimensional visualization of incidents grouped into space-time cubes each covering a 6-month interval. Two grid sizes were used for analysis: a 20 km grid (for higher precision) and a 50 km grid (for coarser granularity). Each time-slice of the grid cells, represented by a single layer of space-time cubes, is a plane in which a local spatial autocorrelation analysis can be performed. Using the Emerging Hot Spot Analysis (EHSA) tool in ArcGIS Pro, a Getis-Ord Gi* statistic can be generated for each space-time cube (Cheng et al., 2018). Additional parameters including Neighborhood Distance and Neighborhood Time Step are used to customize the resulting Getis-Ord Gi* statistic (Gudes et al., 2017). The EHSA tool categorizes each grid cell based on specific patterns of statistical significance over the 8-year studied interval. The hot spot pattern types identified in megaregion analyses are described in Table 3. An EHSA was performed for each of the four top megaregions; however, only the two California megaregions are shown in Figure 8.

Figure 7.

Visualizations of Space-Time Cubes along Amtrak Rail Lines in Northern California and Southern California Megaregions (source data from Amtrak, 2019; Bureau of Transportation Statistics, 2019a; ESRI, 2019b, 2019d; Regional Plan Association, 2008; and US Census Bureau, 2019).

Table 3.

Summary of Emerging Hotspots for the California Megaregions.

Hot spot type	Northern California	Southern California	Great Lakes and Northeast	Hot spot definition (ESRI, 2019a)
Consecutive hot spot	Oakland Area (Hayward to Richmond)	Los Angeles area (LA to Anaheim); Santa Barbara; Ventura/Oxnard; Simi Valley		A single uninterrupted run of statistically significant hot spot bins based on the final time-step intervals. This hot spot has never been statistically significant prior to the final hot spot run and < than 90% of all bins are statistically significant hot spots.
Intensifying hot spot		Oceanside to San Diego coastal corridor	Philadelphia area (Trenton, NJ to Wilmington, DE)	A statistically significant hot spot for 90% of the time-step intervals (including the final time step). Also, the intensity of clustering of high counts in each time step is increasing and is statistically significant.
Sporadic hot spot	Fresno		Chicago, IL; Baltimore, MD	An on-again then off-again hot spot where < than 90% time-step intervals have been statistically significant hot spots and zero time-step intervals have been statistically significant cold spots.
New hot spot	Merced			A statistically significant hot spot for the final time step but was never a statistically significant hot spot before.

Figure 8.

Emerging Hot Spot Analysis Applied for Northern and Southern California Megaregions (source data from Amtrak, 2019; America 2050, 2008; Bureau of Transportation Statistics, 2019a; ESRI, 2019b, 2019c, 2019d; and US Census Bureau, 2019).

For the EHSA, several grid sizes, Neighborhood Distance, and Neighborhood Time Step were tested. Resulting hot spot classifications for each combination did not differ greatly; however, smaller grid sizes and larger Neighborhood Distance and Time Step values increased the potential for false interpretations, where empty grid cells would be marked with a hot spot type unique to surrounding hot spots. Figure 8 shows this challenge and compares a 20 and 50 km grid scale. Urban areas identified in Table 3 were determined when the locality exhibited a consistent hot spot classification for more than 50% of the parameter combinations tested. These hot spot type classifications, although representative of a statistically-significant and documented pattern, should merely be used as guides to relatively compare different hot spots within the same megaregion. They should not be taken to represent the true character of a hot spot without further analysis into individual incidents and their spatiotemporal characteristics.

Both hot spot analyses reveal that Northern and Southern California are host to extensive and consistent railroad trespassing. Although Table 3 lists all of the general urban areas captured by hot spots, it is the California urban coastal rail network that appears to be of primary concern. The metropolitan area surrounding Oakland, city centers from Santa Barbara to Simi Valley, and the coastal Oceanside to San Diego corridor are all identified to be consistent or intensifying hot spots and collectively contain the majority of trespassing incidents across the entire state of California.

Step 4—Temporal Factor Analysis

For each megaregion, specific details of the incidents (organized by incident factors shown in Table 4) were analyzed based on four dimensions (time—year, location—megaregion, incident characteristic—Federal Railroad Administration coding, and “prominence”—statistical significance).

Table 4.

Summary of Factor Categories.

Factor	Categories and descriptions
Season	Winter (12/21 – 03/20), Spring (03/21 – 06/20), Summer (06/21 – 09/20), Fall (09/21 – 12/20)
Time of day	Morning (5:00 a.m. – 10:59 a.m.), Afternoon (11:00 a.m. – 6:59 p.m.), Evening (7:00 p.m. – 10:59 p.m.), Night (11:00 p.m. – 4:59 a.m.)
Day of the week	Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday
Pre-crash trespasser action	Walking, Driving, Riding, Lying down, Standing, Running, Sitting, Jumping from, Jumping onto, Climbing over/on, Crossing over, Crossing between, Sleeping, Ascending, Other
Injury type	Fatality (non-suicide), Suicide, Amputation, Bruise, Concussion, Cut or laceration, Electric Shock, Fracture, Internal injury, Rupture, Strain, Other, No injury type
Gender	Male or Female
Age	Within the following age range categories: < 16, 17–29, 30–39, 40–54, 55–64, > 65, or none reported
RR incident location type	On track, On highway-rail crossing, Beside track, Between tracks, On bridge/trestle, On pole/signal mast, In tower, In/operating vehicle, Other location

This exploration builds on previous work from Oswald Beiler et al. (2019). There are three key differences between the Oswald Beiler et al. (2019) study and this updated study. First, this study includes a temporal aspect. In addition to knowing the number of incidents that occurred over time, the characteristics of these incidents and the trends occurring over time is valuable. Second, two more additional years of data are included. Lastly, this study includes statistical significance as well to provide information as to the level of prominence of a “most-common” category versus all the others within each individual factor.

Data from trespassing person incidents within each megaregion were first organized into a series of Pivot Tables (Microsoft, 2020) that provided point counts for all factor categories for each year of data. The most common factor category is reported overall and for each year in Table 5. The proportion of points counted for the most common category was compared against an equal-distribution proportion in a one-proportion z-test (see equations (2)–(4)). Categories were only counted for the equal-distribution proportion if they held greater than 10% of the total point counts. This procedure was repeated for the total and yearly most common categories. Prominence is indicated in Table 5 with special formatting for p-values less than 0.01 and 0.0001.

Table 5.

Statistical Factor Analysis by Megaregion for 2011 to 2019.

	Factor	2011	2012	2013	2014	2015	2016	2017	2018	2019	Total
Northern California	Number of incidents	16	31	27	36	35	57	52	57	27	338
	Season^†	X	Fall	Fall	Spring	Winter	Spr/Sum	Winter*	Winter*	X	Winter*
	Incident time of day	Morning	Afternoon**	Afternoon*	Afternoon*	Afternoon*	Afternoon**	Afternoon**	Morn/Aft*	Afternoon**	Afternoon**
	Day of week	Friday	Tuesday	Sunday	M/T/Th	Friday	Monday	Tuesday	Tue/Wed	Sun/Wed	Tuesday
	Pre-crash action	Walking	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**
	Injury type	Fatality	Fatality**	Fatality**	Fatality**	Fatality**	Fatality**	Fatality**	Fatality**	Fatality	Fatality**
	Gender	Male*	Male**	Male*	Male**	Male*	Male*	Male	Male*	Male*	Male**
	Victim age^††	17–29	40–54	40–54	17–29	40–54	30–39	40–54	30–39	40–54	40–54*
	Reported RR zone	On track	On track	On track	On track*	On track	On track*	On track	On track	On track	On track **
Southern California	Number of incidents	8	25	29	27	32	38	47	47	24	277
	Season^†	X	Summer	Wint/Spr	Fall	Spring	Fall	Spring	Fall	X	Fall
	Incident time of day	Morn/Aft	Afternoon*	Afternoon	Afternoon**	Afternoon*	Afternoon**	Afternoon*	Afternoon*	Afternoon	Afternoon**
	Day of week	Wed*	Saturday	Wed	Friday	Wed	Friday	Sunday	Sat/Thu	Tue/Thu	Friday
	Pre-crash action	Walking	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**	Walking**
	Injury type	Fatality*	Fatality	Fatality	Fatality**	Fat/Suic	Fatality	Fatality**	Fatality**	Fatality**	Fatality**
	Gender	Male	Male	Male*	Male*	Male	Male	Male*	Male	Male*	Male**
	Victim age^††	40–54*	40–54	40–54	17–29	17–29*	17–29	30–54	30–54	40–54	40–54*
	Reported RR zone	On track	On track	On track	On track*	On track*	On track*	On track*	On track*	On track*	On track **
Northeast	Number of incidents	15	27	24	31	34	36	44	26	20	257
	Season^†	X	W/Sp/Su	Wint/Spr	Summer*	Summer*	Fall	Fall	Spring	X	Summer
	Incident time of day	Afternoon	Afternoon**	Afternoon	Afternoon*	Afternoon	Afternoon*	Afternoon*	Morn/Aft	Afternoon*	Afternoon**
	Day of week	Sun/W/Th	Thursday	W/Th	Sun/M/W	Wed	Friday	Friday	Monday	Sunday*	Wed
	Pre-crash action	Walking*	Walking*	Walking*	Walking	Walking**	Walking**	Walking**	Walking*	Walking*	Walking**
	Injury type	Fatality	Fatality*	Fatality*	Fatality**	Fatality*	Fatality**	Fatality**	Fatality	Fatality*	Fatality**
	Gender	Male*	Male*	Male*	Male	Male*	Male*	Male**	Male*	Male	Male**
	Victim age^††	40–54	17–29/40–54	17–29*	17–29/40–54	17–29	17–29	17–29	17–29	>65	17–29**
	Reported RR zone	On track*	On track*	On track**	On track**	On track**	On track**	On track**	On track*	On track	On track**
Great Lakes	Number of incidents	11	24	20	25	26	30	28	31	16	211
	Season^†	X	Winter	Summer	Fall	Spring	Summer	Summer	Fall	X	Wint/Sum
	Incident time of day	Morning	Evening	Afternoon*	Afternoon*	Morn/Aft	Afternoon	Evening*	Afternoon	Evening	Afternoon*
	Day of week	Saturday*	Friday	Sunday*	Sunday	Friday	Sunday	Tuesday	Monday*	Sunday**	Sunday
	Pre-crash action	Walking	Driving	Walking*	Walking	Walking*	Walking*	Driving**	Driving*	Driving*	Driving**
	Injury type	Fatality	Fatality*	Fatality*	Fat/Bruise*	Fatality*	Fatality**	Other**	Other	Fatality**	Fatality**
	Gender	Male	Male*	Male*	Male*	Male*	Male	Male*	Male	Male	Male**
	Victim age^††	40–54	17–39	17–29	17–29*	40–54	17–29	17–29	17–29*	17–29/40–54	17–29**
	Reported RR zone	On track	On track	On track	Hwy cross	Hwy cross	Hwy cross	Hwy cross	Hwy cross	Hwy cross	Hwy cross

Note. *p-value < 0.01 = 99% confidence that most common characteristic is preferred/more likely than other characteristics.

**p-value < 0.0001 = 99.99% confidence that most common characteristic is preferred/more likely than other characteristics. The cells that fall into this category are highlighted by bold text.

†

2011 and 2019 results excluded since incomplete year skews most common season result – total calculated from 2012 to 2018 data.

††

Victim age occasionally not reported – these values are excluded from most common category analysis and one-proportion z-test.

X = not applicable due to the dataset either beginning or ending in August of that particular year.

p_{o} = \frac{# i n c i d e n t s e q u a l s h a r e}{# i n c i d e n t s t o t a l}

(2)

\hat{p} = \frac{# i n c i d e n t s f o r p r o m i n e n t c a t e g o r y}{# i n c i d e n t s t o t a l}

(3)

Z s c o r e = \frac{\hat{p} - p_{o}}{\sqrt{\frac{p_{o} (1 - p_{o})}{# i n c i d e n t s t o t a l}}}

(4)

The results from the temporal factor analysis reveal the categories of occurrence in the “afternoon,” a pre-crash action of “walking,” an injury type of “fatality,” a trespasser gender of “male,” and a location “on track” are all statistically significant (p < 0.0001) prominent characteristics of trespassing incidents for the top three megaregions: Northern California, Southern California, and the Northeast (Table 5). The Great Lakes factor profile shows similarities in occurrence time, injury type, and victim gender yet exhibits distinguishing elements. The most prominent pre-crash action in the Great Lakes is “driving,” with a p-value less than 0.0001. Similarly, although not statistically significant, the most prominent incident location is at highway crossings. This distinction suggests trespassing incidents occur as a result of similar processes across Northern California, Southern California, and the Northeast and different processes in the Great Lakes megaregion.

The temporal dimension of this factor analysis reveals the inherent heterogeneity in trespass incident characteristics within a single megaregion across several years. This is particularly useful when comparing most prominent age categories. Although the highest statistical significance is reached for 40 to 54 years (California) and 17 to 29 years (NE and GL), there appears to be strong bimodality for both of these categories across all megaregions (Table 5). This result is reflected in the research of Topel (2019a) where similar age distribution peaks of US trespassing fatalities were seen in data from 2008 to 2013.

Step 5—Reflections

The results based on Amtrak trespassing incident data from 2011 to 2019 show many similarities to initial analyses of total FRA data, including both Amtrak passenger and commercial railways (despite FRA excluding suicide incidents). FRA data can be viewed and accessed publicly on the Trespass and Suicide Dashboard (FRA, 2020b).

A similar overall increase in the number of annual trespassing incidents is observed over the 8-year interval in the total FRA dataset (FRA, 2020b). This suggests there is a need for further analytical investigations and mitigation efforts to understand and eliminate this dangerous trend. Several of the counties in the United States with the highest number of total FRA trespassing incidents also are in California, similar to the analysis of Amtrak data (FRA, 2020b). However, the FRA highest incident county list also includes Florida and Illinois counties, which did not compare to the disproportionately high incident count for California counties in the Amtrak analysis (FRA, 2020b). This could be explained by a disproportionately higher usage of rail lines in California for Amtrak passenger trains in comparison to the rest of the country.

Cities with high incident

Table 6 shows a summary of urban areas that are highlighted by different analyses for high-incident count. The top four urban areas we have identified that are of primary concern for trespassing along Amtrak rail lines include the Oakland, CA corridor (stretching from Hayward to Richmond), two Southern California coastal corridors (one from Santa Barbara to Simi Valley; two from Carlsbad to San Diego), and the Philadelphia, PA corridor (from Trenton, NJ to Baltimore, MD).

Table 6.

Summary of High-Incident Urban Areas Identified through Analyses.

Megaregion	Urban areas with potential high point density (from Step 2)	Urban areas highlighted by optimized hot spot analysis (from Step 3)	Urban areas highlighted by emerging hot spot analysis (from Step 3)
Northern California	Oakland area (Hayward, Oakland, Richmond, San Pablo); Davis; Sacramento; Stockton; Fresno	Oakland area (Hayward, San Leandro, Oakland, Berkeley, Richmond); Fremont; Davis, Fairfield, Martinez, Sacramento (northwest); Salinas, Fresno, Merced, Stockton	Oakland area (Hayward to Richmond); Fresno; Merced
Southern California	Santa Barbara; Oxnard; Anaheim; Oceanside; Carlsbad; San Diego	Santa Barbara; Oxnard; Simi Valley; Anaheim; San Clemente; Encinitas; Carlsbad; Oceanside; San Diego	Los Angeles area (LA to Anaheim); Santa Barbara; Oxnard; Simi Valley; Oceanside to San Diego coastal corridor
Northeast	Philadelphia, PA; Wilmington, DE; Baltimore, MD	Philadelphia area (Trenton, NJ to Wilmington, DE); Baltimore, MD	Philadelphia area (Trenton, NJ to Wilmington, DE); Baltimore, MD
Great Lakes	Niagara Falls, NY	Detroit suburbs, MI; Kalamazoo, MI; Chicago suburbs, IL	Chicago suburbs, IL – spread out
Piedmont Atlantic	Birmingham, AL; Durham, NC; Greensboro NC
Florida	Jacksonville, Orlando, Tampa
Cascadia	Portland, OR
Other	El Paso, TX;

Although the hot spot analyses (Step 3) limit our view to locations with the highest risk for total recurring trespassing incidents, the standardized point density approach (Step 2) highlights smaller areas across the country with unusually high geographic clustering of trespassing incidents. There are several of these localities outside of the top four megaregions, and they have been recorded in Table 6. To understand the significance of these localities, smaller scale investigations are necessary to determine whether or not these reflect a fundamentally high-incident prone area.

Connection to Policy and Planning

The following section includes reflections on the findings and direct connections to policy and planning implications. In addition, limitations and future research efforts are included as well, in order to continue the investigation of this critical safety issue.

Policy Implications and Planning

After a review of the trespassing incidents at various levels (megaregional as well as metropolitan) a connection to policy and planning efforts is critical to addressing this safety issue. Currently, there is an increasing level of effort in planning efforts toward trespass incidents at the federal level (FRA, 2020c) as well as at the private national level through agencies such as Amtrak (2018) directly. Support for updated trespassing data, statistics, as well as resources including public service announcements and virtual events for emergency responders and law enforcement agencies, are provided. FRA also provides a Community Trespass Prevention Guide for local, state, and national partnerships to implement steps to reduce trespass related deaths (FRA, 2011a).

Although there are resources for local agencies to use, more detailed and directive policy and planning efforts are needed to support and address trespass incident trends. Therefore, the following recommendations are provided, (1) federal level division oversight of megaregions, (2) local level investigation of high incident areas, and (3) policy efforts that include incident thresholds and limits.

Federal level oversight of megaregions

Currently, the Federal Rail Administration’s Office of Safety has sixteen different divisions, including the Grade Crossing and Trespasser Outreach Division. Within this division, more effort to provide concentrated investigation at the megaregional level first, and then even more locally is recommended. National approaches are effective but as the research shows, there are concentrated areas which require more attention. Providing personnel that oversee the megaregional changes and can allocate outreach to those growing metropolitan corridors can be helpful. Outreach would include direct assistance with applying for existing grants (FRA, 2020a), identifying need for additional grant types, coordination within megaregional corridor needs, and coordination with private rail agencies, such as Amtrak, within each megaregion.

Local investigation of high incident areas

In order to provide more directive measures, a step-by-step investigation of local high-density incident areas is needed and requires a true partnership between local and federal level entities. A step-by-step investigation should include clear spatial analysis steps that can guide local planners to analyze not only the incident frequency but also incident characteristics (similar to the factors in this study).

National level oversight is needed to execute a more local level investigation for three primary reasons. First, to ensure that as high-density areas change, new localities are involved in the investigation process. Second, to provide consistency among steps and the data collection process. Third, to determine the overall effectiveness of existing countermeasures at the track level as well as and relevance of potential countermeasures that could be implemented. Given that each incident area is different, a local lens is needed for effective investigation of incidents. Identifying and addressing repeat incident areas is important but also being proactive in identifying locations with similar design/safety concerns that perhaps do not have any incidents yet would be useful to maintain a proactive (versus retroactive) approach to preventing trespasser deaths.

Figure 9 shows an example of a local level investigation for four high-priority California metropolitan areas. Generated from a point-density buffering analysis similar to that employed in Step 2, it displays five discreet areas along rail lines that contain anomalously high densities of trespassing incidents when compared to all urban/suburban localities across the country. Further investigation of these analyses in order to identify the local segments of rail with the largest history of repeated trespassing has the potential to increase the efficiency with which preventative measures are employed and causality factors are explored.

Figure 9.

Local Investigation of Four Example California Metropolitan Areas (source data from Amtrak, 2019; Bureau of Transportation Statistics, 2019a, 2019b, 2019c; ESRI, 2019b, 2019c, 2019d; ESRI et al., 2020).

Understanding similar local level concerns and being able to connect to local planning factors such as adjacent land use, zoning, intersection design and spacing, traffic flow, signal timing, and existing rail safety infrastructure, is essential to addressing trespassing incidents. Each one of these local planning factors requires data that should be collected and compared across high and low incident areas within a metropolitan area. The local agencies can provide data with high levels of detail and tailor the dataset to the needs of a particular high incident area as well as potential future areas of concern based on similar characteristics.

Policy efforts that include incident thresholds and limits

In addition to the coordinated planning efforts at multiple levels, having a clear policy program that targets “zero” deaths is recommended, similar to “Vision Zero” efforts. Vision Zero is a strategy that focuses on eliminating traffic related incidents (fatalities and severe injuries) (Vision Zero Network, 2020). This approach has been effective at various levels including state and metropolitan regions. Although “Operation Lifesaver” has been implemented at the national level, a more defined and bold approach to zero trespass deaths could be effective. This would require a partnership approach between FRA and local communities to educating and communicating this goal. Also, having thresholds that signal a “warning” for individual municipalities to require coordination with the Federal Rail Administration’s Office of Safety and development of strategies/implementation of a step-by-step local level investigation is suggested. Having the threshold and limit be conservative, a proactive approach could be adopted to prevent future fatalities.

Limitations and Future Work

Throughout the research, limitations were found but were noted as assumptions or identified as future work. The following is a list of assumptions in order to analyze the data provided using GIS:

1. Time-space cube and EHSA grids were evaluated at several scales and produced consistent yet slightly variable results; using a rectangular grid to evaluate a curvilinear data-space is inefficient and inhibits accuracy.

2. The hot spot analysis process is relative to the geographic size and shape of the boundary used (5 km rail line buffer), so all megaregional findings are specific to that region and should not be compared.

3. Trespassing incident reports are generated for each involved trespassing individual, yet several events include multiple individuals; these data are thus analyzed as two datasets: event incidents (i.e., event defined by same location, date, and time) and person incidents (including each reported incident).

4. Data was provided for August 1, 2011 through August 1, 2019, and therefore, yearly data is complete for all years except 2011 and 2019.

5. Data could be separated out by type prior to analysis (vehicular, pedestrian, etc.) as well as location (crossing or along track) to fully understand the types of trespassing incidents that are of high priority and to try to understand intentional versus accidental incidents.

The following recommendations for future work are included in order to further the study and provide more local/case study results:

Repeat these analytical methods with FRA data (not including suicides) to compare Amtrak and commercial railway hot spots and trends.

Analyze data at other geographical levels such as metropolitan statistical areas (MSAs) in order to provide a context directly linked to resource allocation funding.

Within each megaregion, explore local data including emerging trends specifically with regard to action/injury type in order to connect to existing countermeasure in place and potential improvements.

Conduct time-space cube (EHSA) for all megaregions, beyond the top four completed in this study.

Continue to gather temporal data beyond 2019 and compare changes over multiple decades.

Analyze data factors more in depth such as hour of the day, month of the year, etc.

Develop tools for evaluating spatiotemporal hot spots along discrete lengths of rail lines, instead of with a rectangular grid.

Enhance the standardized point density analysis to isolate discrete urban regions unusually prone to high levels of repeated trespassing. For example, localized case studies on specific regions such as California can be explored. These studies could involve urban planning and Amtrak popularity considerations, such as spacing of crossings, proximal zoning juxtapositions, human and vehicle traffic flow, and frequency and speed of Amtrak train activity.

Conclusion

With railroad trespassing as the leading cause of railroad fatalities, the need to explore trends geographically as well as temporally has never been more important (Laffey, 2019). This study explores rail trespassing incidents (including suicides) using Amtrak data from 2011 to 2019. An investigation of the incidents at the megaregional level is explored using space, time, incident factor, as well as statistical significance. Northern and Southern California were first identified as the highest centers of trespassing risk in the country, followed by the Northeast and Great Lakes megaregions. Using GIS, a new standardized point density approach was applied to determine representing high-risk localities. Then, a spatial analysis (optimized hot spot method) and a spatiotemporal analysis (emerging hot spot method) were applied.

Through this process, the coastal California region and the Philadelphia area were identified as areas of concern. Trends for incident report factors were analyzed across the 2011 to 2019 study interval which revealed “walking” prior to impact, occurrence in the “afternoon,” “male” trespassers, and crash location “on the tracks” were found to be prominent factor characteristics. Policy and planning recommendations are provided including specifying federal level divisional oversight of megaregions, encouraging and supporting local level investigation of high incident areas, and introducing policy efforts that include incident thresholds and limits. By exploring spatial and temporal relationships of trespass incidents across the nation, there is the opportunity to encourage proactive planning and support for effective trespass prevention at all governance levels.

Footnotes

Acknowledgements

The authors would like to thank Don Varley, Manager Analytics at Amtrak, for contribution of the data toward the project.

Declaration of Conflicting Interests

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

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Michelle R. Oswald Beiler

Author Biographies

Michelle R. Oswald Beiler, PhD, PE, LEED AP, is an associate professor in the Department of Civil and Environmental Engineering at Bucknell University. She has completed a Doctoral degree and Master’s in Civil Engineering, and a Master’s degree in Urban Affairs and Public Policy, from the University of Delaware, specializing in transportation planning. She received a Bachelor of Science in Civil Engineering from Lafayette College.

Evan Filion graduated with a Bachelor of Science in Environmental Geosciences from Bucknell University. He specializes in clastic sedimentology and rail transportation topics.

References

Amtrak. (2018). Operation clear track goes virtual for largest rail safety program in the US and Canada. https://media.amtrak.com/2020/09/operation-clear-track-goes-virtual-for-largest-rail-safety-program-in-the-u-s-and-canada/

Amtrak. (2019). Trespass incident data, provided by Amtrak on September 4, 2019. Amtrak.

Blizzard

Cudahy

(2017). Uncle of teen in train death: People should stop posting this stuff. My Dayton Daily News, http://www.mydaytondailynews.com/news/uncle-teen-train-death-people-should-stop-posting-this-stuff/hWRGfhVR0aFhqXPclpf0gK/

Bureau of Transportation Statistics. (2019a). Amtrak routes. US DOT. Retrieved October 15, 2019, from https://data-usdot.opendata.arcgis.com/datasets/amtrak-routes

Bureau of Transportation Statistics. (2019b). Amtrak stations. Retrieved October 15, 2019, form https://data-usdot.opendata.arcgis.com/datasets/amtrak-stations

Bureau of Transportation Statistics. (2019c). North American rail lines. Retrieved November 3, 2019, from https://data-usdot.opendata.arcgis.com/datasets/north-american-rail-lines

Cheng

(2018). Traffic crash evolution characteristic analysis and spatiotemporal hotspot identification of urban road intersections. Sustainability, 11(160), 1–17.

Craig

(2017). Philly, rail company will start cleanup of open-air drug market on train tracks. The Inquirer-Philly.com. http://www.philly.com/philly/health/addiction/A_hidden_heroin_hellscape.html

DaSilva

Grizkewitsch

Jacobini

F. B.

(2019). Development, implementation, and evaluation of a community-based trespass prevention model. TR News, July-August 2019, Transportation Research Board.

10.

ESRI. (2016). ArcGIS pro-how hot spot analysis (Gets-Ord Gi*) works. https://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-statistics-toolbox/hot-spot-analysis.htm

11.

ESRI. (2019a). ArcGIS pro-emergent hot spot analysis. https://pro.arcgis.com/en/pro-app/tool-reference/space-time-pattern-mining/emerginghotspots.htm

12.

ESRI. (2019b). USA county boundaries. Retrieved November 3, 2019, from https://www.arcgis.com/home/item.html?id=a00d6b6149b34ed3b833e10fb72ef47b

13.

ESRI. (2019c). USA major cities. Retrieved November 3, 2019, from https://hub.arcgis.com/datasets/esri::usa-major-cities

14.

ESRI. (2019d). USA state boundaries. Retrieved November 3, 2019, from https://hub.arcgis.com/datasets/esri::usa-major-cities

15.

ESRI, Maxar, Earthstar Geographics, CNES/Airbus DS, USDA FSA, USGS, Aerogrid, IGN, IGP, and the GIS User Community. (2020). World imagery. Retrieved October 26, 2020, from https://services.arcgisonline.com/ArcGIS/rest/services/World_Imagery/MapServer

16.

FRA. (2010). Railroad and injury summary report, 6180.55a. https://safetydata.fra.dot.gov/officeofsafety/publicsite/forms.aspx?itemno=3.05

17.

FRA. (2011a). Community trespass prevention guide, office of safety. https://railroads.dot.gov/elibrary/community-trespassing-prevention-guide

18.

FRA. (2011b). Guide for preparing accident/incident report. Office of safety. http://safetydata.fra.dot.gov/officeofsafety/publicsite/publications.aspx?itemno=3.06

19.

FRA. (2018). Report to congress: National strategy to prevent trespassing on railroad property. US DOT. https://railroads.dot.gov/elibrary/national-strategy-prevent-trespassing-railroad-property

20.

FRA. (2020a). Grade crossing safety and prevention grants. https://railroads.dot.gov/divisions/highway-rail-crossing-and-trespasser-programs/grade-crossing-safety-and-trespass

21.

FRA. (2020b). Trespass and suicide dashboard. Retrieved June 3, 2020, from https://explore.dot.gov/t/FRA/views/TrespassandSuicideDashboard/TrespassOverview?:iid=1&:isGuestRedirectFromVizportal=y&:embed=y

22.

FRA. (2020c). Trespass prevention. https://railroads.dot.gov/highway-rail-crossing-and-trespasser-programs/trespassing-prevention/trespass-prevention

23.

Gabree

(2015). Prevention of railroad trespass casualties: Accidental and intentional. U.S. Department of Transportation: John A. Volpe National Transportation Systems Center. https://www.ntsb.gov/news/events/Documents/2015_trespassing_FRM_Pres25Gabree2Volpe.pdf

24.

Getis

Ord

(1992). The analysis of spatial association by use of distance statistics. Geographical Analysis, 24(3), 189–206.

25.

Gottmann

(1961). Megalopolis, the urbanized Northeastern seaboard of the U.S. MIT Press.

26.

Gudes

Varhol

Sun

Meuleners

(2017). Investigating articulated heavy vehicle crashes in Western Australia using a spatial approach. Accident Analysis and Prevention, 106(2017), 243–253.

27.

Havarneau

Topel

(2019). Rail trespassing and suicide: What can be done to improve safety? TR News, July–August 2018, Transportation Research Board.

28.

Kang

Cho

Son

(2018). Spatiotemporal characteristics of elderly population’s traffic accidents in soul using space-time cube and space-time kernel density estimation, PLoS One, 13(5), e0196845.

29.

Khattak

Luo

(2011). Pedestrian and bicyclist violations at highway–rail grade crossings, transport. Research Record: Journal of the Transportation Research Board, 2250, 76–82.

30.

Laffey

(2019). Railroad fatalities in the United States 2012-2017. TR News, July–August 2019, Transportation Research Board.

31.

Microsoft. (2020). Creating a Pivot table to analyze worksheet data. https://support.microsoft.com/en-us/office/create-a-pivottable-to-analyze-worksheet-data-a9a84538-bfe9-40a9-a8e9-f99134456576?ui=en-us&rs=en-us&ad=us

32.

Oswald Beiler

Miller

Varley

. (2019). Railway trespass prevention: Spatial analysis of incidents to connect to countermeasures. Journal of Transportation Engineering, 145(2), 04018086-1–04018086-15.

33.

Park

Saccomanno

(2005). Evaluating factors affecting safety at highway-railway grade crossings. Transportation Research Record: Journal of the Transportation Research Board, 1918, 1–9.

34.

Regional Plan Association. (2008). US megaregions smooth polygon. Retrieved November 3, 2019, from http://www.america2050.org/maps/

35.

Ryan

(2018). Developing a framework of behaviours before suicides at railway locations. Ergonomics, 61(5), 605–626. https://doi.org/10.1080/00140139.2017.1401124

36.

Searcy

Cunningham

Vaughan

Coble

Avr

(2019) Rail corridor trespass severity assessment (NCDOT Project 2017-15). Institute for Transportation Research and Education, North Carolina State University,.

37.

Sumwalt

(2019). Toward railroad trespassing solutions. TR News, July–August 2019, Transportation Research Board.

38.

Topel

(2019a). FRA’s total fatalities figures do not count all deaths. TR News, July–August 2019, Transportation Research Board.

39.

Topel

. (2019b). Scope and trend of U.S. rail trespassing and suicide fatalities. TR News, July–August 2019, Transportation Research Board.

40.

US Census Bureau. (2019). 2019 TIGER/line shapefiles: Counties (and equivalent). Retrieved November 3, 2019, from https://www.census.gov/geographies/mapping-files/time-series/geo/tiger-line-file.html

41.

Vision Zero Network. (2020). What is vision zero? https://visionzeronetwork.org/about/what-is-vision-zero/

42.

Zaman

Ren

Liu

(2019). Artificial intelligence aided automated detection of railroad trespassing. Transportation Research Record, 2673(7), 25–37.