GEOG 855
Spatial Data Analytics for Transportation

7.1 Highway Safety


Highway safety is an important area of focus for state DOTs and the USDOT. There are few groups within the USDOT who are focused on improving highway safety. The first is the Office of Safety. The Office of Safety is comprised of two units. The Technologies Unit deals with safety-related highway design considerations and technologies which can be used to improve highway safety performance. The Programs Unit oversees federal and state safety programs. One of the key programs they administer is the Highway Safety Improvement Programs (HSIP). HSIP is a federal-aid program designed to provide funding to states for projects aimed at reducing fatalities and serious injuries on qualifying roadways. In 2016, the program provided about 2.2 billion dollars to the states for safety projects.

Figure 1 - Highway Safety Improvement Program Funds
Fiscal Year 2016 2017 2018 2019 2020
Estimated Funding* $2.226 B $2.275 B $2.318 B $2.360 B $2.407 B

Reference: FHWA Website accessed 12/31/2016

To qualify for HSIP funds, a state is required to develop and maintain a Strategic Highway Safety Plan (SHSP). An SHSP is designed to guide the investment of funds to projects which have the greatest potential to reduce fatalities and serious injuries. To qualify for HSIP funds, states are also required to identify their priorities using a Data-Driven Safety Analysis (DDSA).

The second group within USDOT which is responsible for highway safety is the National Highway Traffic Safety Administration (NHTSA). NHTSA is an administration within USDOT whose mission is to reduce crash fatalities and injuries. We’ll take a close look at NHTSA later in this lesson.

State DOTs commonly collect and use crash data to identify areas of their roadway networks where there are unusually high crash rates. However, looking at crash data alone can be misleading and result in a less than optimal use of available state and federal dollars. To address this problem, AASHTO, in conjunction with the FHWA, developed the Highway Safety Manual (HSM), a document which many consider the definitive reference on highway safety. The HSM offers a comprehensive and balanced approach and set of tools which consider operations, the environment, and the cost of construction alongside safety considerations. A good overview of the HSM can be found here. The approaches provided in the HSM go beyond traditional approaches to identifying priority locations for safety improvements which rely solely on crash history data.

There are two fundamental problems associated with using crash data alone. First, crashes are statistical events and as such don’t occur at regular predictable intervals. Consequently, crash data alone can sometimes lead an agency to falsely identify sections of a roadway as high risk and, conversely, sometimes overlook a risky section. The second problem of looking solely at historic crash data is that it disregards the dependence of crash frequency on traffic. As traffic levels increase on a section of the roadway due to changing travel patterns, crash rates can increase. To overcome these limitations, it is necessary to look not only at historic crash frequencies but also at expected crash frequencies based on roadway characteristics and traffic data.

Tools have been developed which implement the approaches defined in the HSM. These include AASSHTO’s Safety Analyst and FHWA’s Interactive Highway Safety Design Model (IHSDM). However, states often lack much of the data required to effectively use these tools, such as horizontal and vertical curve data. Horizontal curves are roadway curves that turn to the left or right, and vertical curves are roadway peaks/hills and valleys. For my Capstone Project, I used roadway centerline data to extract horizontal curvature data from Pennsylvania’s roadways. I gave a lightning talk on the project at Penn State in November 2016 for GIS day. My presentation was just under 10 minutes in length (embedded video below). 

My Lightning Talk 
Click here for a transcript

Host: JD are you ready?

Host: Alright! JD Kronicz will be our next speaker and he will talk about using GIS to identify and characterize horizontal curvature. JD has worked as a consultant for over 25 years serving clients in the scientific, environmental, transportation, and manufacturing industries. For the past 15 years, he’s been primarily focused on GIS software applications for transportation. He works closely with the Pennsylvania department of transportation and he also teaches in Penn States online geospatial program.

JD: Good afternoon. Today I have the pleasure of talking to you briefly about some work I did using GIS to identify the horizontal curvature in the roadways.

So, curves, as one might guess are an important feature of roadways when it comes to highway safety. Surprisingly, many state DOT’s, departments of transportation, don’t have good inventories of their roadway curves. Often if they have curvature information it’s embedded on engineering diagrams or other plans and really not in a format that is readily accessible for highway safety analysis. So, in this project I basically focused on trying to extract that information and producing an inventory of Pennsylvania’s horizontal curves from roadway centerline data which is data all state DOT’s have.

Now, it’s a priority of state DOTs to make sure that their roads are safe and to constantly try and improve safety. Generally, is what they’re trying to do is their trying to use the limited amount of safety dollars to apply to those roads that would best benefit, where they could make the biggest impact and get the biggest bang for the buck. The question is, how do they identify their priorities? And there’s really two different ways you can approach the problem.

The first, is by doing crash analysis. States all have good data on their crashes, and you can use the crash data to identify areas where there are high crashes and, in that manner, identify your priority sections of roadway. That’s a reactive approach and it may be erroneous because it could just be an anomaly, it might not represent a section of roadway that inherently has safety issues.

The other approach is to do a systemics analysis of the roadways where you’re looking at the characteristics of the roadways and your calculating the expected crash rates based on features of the road such as curvature features.

Either way once you identify the priority sections of roadway you want to improve safety on you can implement any one of many many counter measures or safety improvements. So here I’ve just shown a few and I could probably put together 100 but for example, center-line rumble strips, high friction surface treatments are just a couple.

Ok, so before I get into what approach I used, just a little bit on the geometry of the horizontal curve, this is kind of a complex figure but really all we’re interested in is 3 parameters when it comes to horizontal curves. We’re interested in the radius of the curve; we’re interested in the length of the curve and we’re interested in the central angle of the curve: the number of degrees in which the curve turns. So those are the 3 parameters we’re interested in and given any 2 of those parameters we can derive the 3rd.

So, the approach that I used was I started with roadway centerline data for Pennsylvania and I basically took each road feature and deconstructed into its ordered series of vertices which is basically what it represents and then for each pair of vertices I determined the straight line that went through those vertices and determined the baring angle, the angle between that straight line and the positive x-axis.

And then I continued to do that for each pair of ordered vertices in sequence. Basically, looking at how that baring angle changed and anytime the change in baring angle exceeded a certain threshold value I threw up a flag and said we’re in a curve. So that’s how we determined the start of the curve and continuing that process stepping through the vertices and when we get to the point where we drop below that threshold value, we know the curve ended. Any by aggregating the change in baring angle we can calculate the central angle of the curve; we can calculate the length and from those two we can derive the radius.

Again, if we had a little bit more time, we could get a little bit more detail on what the approach was but at the end of the day, by using this technique we can establish curve features that have attributes of radius, central angle, length, etc.

Now to do this sort of process manually would be extremely time consuming so I derived a, I created a program in python, implemented that as a custom toolbox in ArcGIS and named it curve detective. Essentially, it automates that algorithm I just kinda walked you through.

So, here’s an example of an output of that tool. You can see super imposed or layered on top of the roadway network we have this new curve feature class that this tool created. Each curve in red, labeled according to central angle and radius.

I then went ahead and processed, once I established it worked okay, I processed all the state roadway in Pennsylvania. So, in Pennsylvania, the state actually owns roughly 45 thousand miles of roadway. There’s a lot more roadway that’s local roads but the state actually owns and maintains about 45,000 miles of roadway. In order to process this roadway, it took the tool about 2 hours and it ultimately identified 170,000 or so horizontal curves.

Then I went ahead and I wanted to make sure that the output of the tool was legitimate, that it was accurate and precise so I went ahead and found locations that had been survey in the field and presumably had good data on the curves and then compared those engineering diagrams which is where that data is embedded to the results of the curve detective. Without getting into any great detail with you I found that the tool was very accurate and reasonably precise.

Ok, so at this point I had identified, or created an inventory of horizontal curves in Pennsylvania, and I wanted to basically see what could I learn by using the crash data we have in Pennsylvania and combining the two, just to see how crash rates differ on curves and I created a little tool on Microsoft access where I brought these two data sets together that allowed me to perform a bunch of analyses and I did conduct a number of analyses, I’ll just walk you through a couple briefly.

So in the first one I basically just looked at all the road sections in Pennsylvania that have a horizontal curve based on the output of the curve detective and I looked at the number of crashes, the crash rates on those sections of roadway and compared them to crash rates on straight sections of roadway and what I found is that one curves the crash rate is about 2.3x higher than on straight sections of roadway. I then went ahead and limited the crashes I was looking at just to crashes that involved fatalities. And when we just limit it to fatal crashes, we see that the crash rate is 2.8x times higher on the curved sections of roadway than it is on straight sections of roadway. So, not only are crashes more frequent on curves, they’re generally more serious.

I also looked at, I wanted to see the relationship between the central angle of the curve and the radius of the curve and the crash rate so I ran a series of analyses at various central angles, various radii and what I saw was there was little or no relationship between central angle and crash rate which was kind of intuitively surprising to me but there was a very strong relationship between radius and crash rate. So, as the radius of the curve got smaller, especially as it went below 1000 feet in radius the crash rate dramatically increased. So, on the vertical axis there we have crash rate, on the horizontal axis we have radius. So, you can see as the radius decreases the crash rate shoots up, each data series here corresponds to a different central angle and you can see there’s really no discernable relationship there.

So, in conclusion, this is a technique that can be implemented by any state DOT because state DOT’s all have roadway centerline data, it’s a technique that’s very rapid, it’s cost effective and it can produce a highly accurate and precise inventory of a state’s horizontal curves. In looking at the Pennsylvania crash data in conjunction with this horizontal curve data we saw that the crash rates on curved sections of roadway are a lot higher than they are on straight sections of roadway. And in addition the crashes that do occur on curves, they tend to be much more series or fatal, much more frequently than they are on straight sections of roadway and with that I will thank you for your time and if you have any questions I’d be happy to take them.

Two model frameworks have been developed to help states structure the crash and roadway data needed for highway safety analyses in a standard format. The first is the Model Minimum Uniform Crash Criteria (MMUCC). MMUCC is a list of standard crash data elements and associated definitions developed by NHSTA. While the implementation of this model is voluntary, states are encouraged to adhere to the standard in collecting and compiling crash data. Similar in concept to the MMUCC, the Model Inventory of Roadway Elements MIRE is a list of over 200 roadway and traffic data elements critical to safety management developed by the FHWA.

Collecting roadway data according to the MIRE model will not only benefit the state DOT in regards to traffic safety efforts, it will also help other core areas of transportation such as operations, asset management, and maintenance.


Once a section of roadway has been identified for needed safety improvements, an agency needs to decide which types of countermeasures would be the most effective. There are many types of safety countermeasures that could be implemented. Here’s a list of 20 proven countermeasures published by FHWA’s Office of Safety.