Browse Projects > Detailed View

Safety Effectiveness Assessment of Advanced Rail/Highway Grade Crossing Improvements


The US Department of Transportation (USDOT) originally developed its Accident Prediction and Severity (APS) models for rail/highway grade crossings[1] (HRGCs) in early 1980s. Three APS models – one each for passive, flashing lights and gated crossings – were developed by Volpe researchers. The APS models have not addressed the risk exposure to pedestrian and non-motorized transportation at all. Since the early 1980s the model coefficients have been calibrated every few years but a full re-estimation has not been done for a long time. These models are still being used in GradeDec.Net, which is a web-based application and decision support tool provided to federal, state and local authority decision makers for the identification and evaluation of HRGC upgrades, separations and closures.

As a result, the current APS model does not reflect the latest incident data, nor does it include improvements in analytical methods such as machine learning[2] and does not reflect the full range of HRGC treatments now available, including many new HRGC technologies that have been developed since the 1980s. Such newer technologies include but are not limited to quad gates, PTC integration, remote health monitoring for crossings, loop detectors and other approaches for detecting stranded vehicles, ITS-DSRC radios for alerting oncoming vehicles, Variable Message Warning (VMS) signs, GPS integration[3] to cellphone maps to alert drivers of upcoming rail crossings, WAZE beacons and other new grade crossing treatments might effectively mitigate some of these risks.

The existing APS modeling framework provides absolutely no support for assessing the safety benefits of these kinds of advanced HRGC treatments, nor does it address needs in the area of pedestrian and non-motorized transportation safety. Filling this gap and expanding the assessment to include new technologies will definitely add significant value to the current practice and help with making better decisions.Updating evaluation tools for HRGC improvements is listed as one of the 22 high-priority recommended actions as reported and described by working groups at the fourth Grade Crossing Research Needs Workshop held by the Federal Railroad Administration (FRA), in August of 2017 in St. Louis, MO[4].

  • “Modernize both the Accident Prediction and Severity model and GradeDec.net; grade crossing technology and the railroad operating environment and available data have changed; help ensure that grade crossing resources are directed to the areas of greatest benefit and risk reduction—thereby, saving lives. It may even end up justifying an increase in spending on crossing improvements.”

As many of the most egregious HRGC deficiencies have been mitigated over the past 30 years, crossing incidents and death rates have declined. But recently, grade crossing incident rates have plateaued and even started to rise again[5]. Some of this increase has been attributed to the rise of “distracted driving”, that is the increasing use of mobile devices while driving. Regardless of the cause, these recent increases suggest that an “inflection point” may have been reached – the old investment strategies are not working to reduce grade crossing incidents as effectively as they have in the past. If these recent unfavorable trends are to be reversed, a change in approach may be needed.

In a recent report (November 2018) from the US. Government Accountability Office (GAO), it was pointed out that DOT should evaluate whether the federal program that helps states fund rail crossing projects have provided states flexibility to address ongoing challenges. The program’s requirements “favor certain projects, such as gates — but most of the incidents since 2009 have happened at crossings with gates, largely because of impatient drivers who drive through or go around. Officials said that the requirements may be preventing them from using more innovative approaches.” [6]

(1) This model is described in the document Summary of the DOT Rail-Highway Crossing Resource Allocation Procedure-Revisited, Office of Safety, Federal Railroad Administration, June 1987, Report No. DOT/FRA/OS-87/05.

(2) For example, see https://www.researchgate.net/profile/ZijianZheng3/publication/321807368AccidentPredictionforHighway-RailGradeCrossingsaModelComparisonofDecisionTreeandNeuralNetwork/links/5b195c99a6fdcca67b63654f/Accident-Prediction-for-Highway-Rail-Grade-Crossings-a-Model-Comparison-of-Decision-Tree-and-Neural-Network.pdf)

(3) Source: http://www.cnn.com/2015/06/29/politics/google-alert-drivers-car-prevent-death-railroad/

(4) 2017 Grade Crossing Research Needs Workshop_, Federal Railroad Administration. https://www.fra.dot.gov/Elib/Document/18253

(5) Ten Year Accident/Incident Overview by Calendar Year. Federal Railroad Administration. https://www.fra.dot.gov/Elib/Document/18253

(6) https://www.gao.gov/products/GAO-19-80


The current APS models focus only on the benefits associated with upgrading from passive cross bucks, to flashing lights, to basic gates. These were the prevalent technologies at the time when the APS models were created and calibrated in the 1980’s – but in the intervening 30 years, technology and the railroad operating environment have both changed.

As a result, the research objective is bringing the APS model up-to-date. In doing this, the updated APS model should reflect the latest data, modern modeling approaches and availability of modern grade crossing treatments. The research should correct issues with illogical and inconsistent behavior [7] of some of the existing APS models, add additional variables as predictors as appropriate, and re-estimate the models based on the best available current data. Also, the impact of grade crossing improvements on safety of pedestrians and non-motorized transportation modes should also be assessed.

(7) When train count starts to exceed a very low threshold, some of the model calibrations are known to be highly insensitive to increased traffic. Also the calibrations of the three models are inconsistent – when upgrading a device type (e.g., passive to lights) the model sometimes will predict more and more severe accidents even as crossing equipment is upgraded. These kinds of issues can cause the State DOT users to lose confidence in the tool and its underlying methodologies. A recalibration and enhancement of the APS safety model is needed to eliminate the inconsistent behavior and to ensure that the model always gives meaningful results throughout its entire range of calibration.


The most important benefit of making improvements in rail grade crossing technology is that it is a proven way to save lives. However, since this Research Needs Statement focuses on safety assessment, the primary benefit will be to help ensure that whatever grade crossing resources are available are directed to the areas of greatest benefit and risk reduction – thereby saving as many lives as possible.

Since the long-improving safety trend at highway-rail grade crossings has plateaued and even reversed in the past few years, this indicates that a change in the way resources are allocated may be appropriate. Completion of the proposed research would facilitate the best possible allocation of resources – by clarifying the potential benefits of such investments to railroads and to society as a whole, it would also help decision makers to better understand the value of spending on rail crossing safety improvements.

Related Research:

There is a large body of knowledge of grade crossing safety, both in North America and internationally, since a significant amount of research has been completed over the past 30 years. Furthermore over the same time frame, many equipment suppliers have innovated and devised new means for improving crossing safety.

Unfortunately, the problem is that the tools that are frequently used for evaluating the benefits of grade crossing investments have lagged these developments. These tools have not been brought up to date, or in line with this large body of knowledge, for many years. The research proposed here would accomplish this much-needed update.

There are a few ongoing projects that focus on improving rail crossing safety models [8].

  • Florida DOT sponsored a project that aims to develop the incident prediction model for Florida’s rail crossings (began in September 2018).
  • A recent research report published by North Dakota State University compared different incident prediction modeling frameworks, including Generalized Linear Models and data mining algorithms, in predicting HRGC incidents. They concluded that “data mining can serve as great alternative modeling tools to perform incident forecasting with relatively accurate forecasting power and strong ability to model non-linear relationships.”
  • Another ongoing project, conducted by the University of Nebraska-Lincoln, pointed out that the state DOT currently uses the state-specific Accident Prediction Model for rail crossings to identify and rank crossings that need subsequent safety improvements, however, these models were developed in 1999 and used 5-year data from 1993 through 1998. The models have over-prediction issues and new prediction models are required.
  • Other state DOTs (e.g., Iowa) are in the process of updating their HRGC incident prediction models as well. Sperry et al. (2017) summarized current issues related to HRGC hazard-ranking and project development. They stated that due to missing of certain key factors in the US DOT APS models, many states faced the challenge of selecting a few projects for improvements from a list of hundreds of crossings that the models show little difference in incident experience. Such factors are sometimes considered in state-specific models and should be incorporated into the DOT APS models.

As can be seen, multiple states have begun the effort of re-estimating incident prediction models for HRGCs using modern statistical and data mining methods and data that are more recent. However, the methodology, data, and even conclusions varied in different ways. On the other hand, comprehensive research is lacking on assessing the effectiveness of new HRGC safety technologies and their influences on the APS models. A national effort should take place to establish uniform, advanced, and practical guidance and readily available models that can be adopted by different states.

Follow On and Implementation Activities - Starting in 2003 the U.S. Federal Railroad Administration packaged the APS models into GradeDec [9], a web based evaluation tool, to help support investment decisions in highway-rail grade crossings GradeDec adds to the safety analysis a full set of standard benefit cost metrics including the value of highway delay savings, air quality improvements, and highway vehicle operating costs.[10]

The tool is intended to assist state and local transportation planners in identifying the most efficient grade crossing investment strategies and to help them identify and prioritize the most effective projects. One of the most important limitations of GradeDec is that it has been constrained by the overly simplified limitations of the APS model. By packaging an enhanced APS within an expanded GradeDec the usefulness and capability of GradeDec can be greatly enhanced. In addition, the economic criteria supported by GradeDec are not quite comprehensive, most notably:

  • Rail carriers’ own business benefits through reduction in train delays and rail equipment damage are not included [11]

  • Impacts on emergency vehicle operations should be made explicit, and* The benefits of noise reduction from quiet zone treatments are also not estimated or included in the current model.

These limitations should be addressed at the same time as the new APS model is incorporated into GradeDec. Therefore, an effort to develop an enhanced and upgraded Accident Prediction and Severity (APS) models should be promptly followed up by the effective delivery of these results to state DOT users. This will give state decision makers a better ability to decide whether to focus funding into high-risk urban crossings or to continue to invest in conventional crossing treatments at rural crossings, with diminishing marginal returns.

(8) https://trid.trb.org/Results?txtKeywords=rail%20crossing%20safety%20prediction&txtTitle=&txtSerial=&ddlSubject=&txtReportNum=&ddlTrisfile=&txtIndex=&specificTerms=&txtAgency=&txtAuthor=&ddlResultType=&chkFulltextOnly=&language=1%2C2%2C4%2C8&subjectLogic=or&dateStart=&dateEnd=&rangeType=emptyrange&sortBy=publisheddate&sortOrder=DESC&rpp=25#/View/1538235

(9) Source: http://decisiontek.com/Solutions/GradeDecNet

(10) Source: https://www.fra.dot.gov/Page/P0337

(11) See for example slide #7 from http://railtec.illinois.edu/GLXS/presentations/C/10C1-GLXS2014-1037-REZVANI.pdf ; the categories of Train Idling Cost, Train Crew Cost and_ Rail Supply Chain Cost_ contributed $24,122 or 46.8% of the overall accident cost of $51,564 in this example. This shows that the “Business Benefits” of grade crossing investment to the railways and to their customers can be significant.

  1. Identify a set of contributing factors, special high-risk conditions, and additional grade crossing technologies which should be considered as candidate predictors for inclusion in the new APS model.
  2. Assess the availability of incident history, risk and exposure data that will be needed to estimate an updated Accident Prediction and Severity (APS) model for both motorized and non-motorized transportation and for pedestrians.
  3. Assess which technologies or special conditions have enough data available to assess the effectiveness of the technology. For those technologies which do not have enough or the right kind of data, the research should suggest changes in the data collection procedures so that the analysis can be expanded in a future update.
  4. Develop a new APS methodology that employs modern statistical techniques including data mining, machine learning and rare event prediction methods. The new methodology should be able to predict both total incidents and incidents by different severity levels.
  5. Summarize and package the new APS model in the form of mathematical formulas, which can be effectively delivered to State DOT and other users. One way of doing this could be by updating the GradeDec software with the most up-to-date models and calibrations, another way could be by embedding the algorithms in spreadsheet packages so that the modeling results can be made available to users. In either case, there should be an explicit user manual delivered to introduce how to use the models and the tools. The modeling methods should be practical enough that the methods can be readily incorporated into current practices.
  6. Validate the model by developing a number of case studies. Provide evidence showing how the new methodology outperforms the former APS model.

This research is anticipated to be completed within 24 months, with a final report detailing the calibration, results, data gap recommendations, new calibrated model, and case study results. The estimated funding requirement for the requested research is approximately $375,000.

KEYWORDS: Rail, safety, highway grade crossings, risk, positive train control


States DOTs, FRA, and Railroads

Sponsoring Committee:AR030, Railroad Operating Technologies
Research Period:12 - 24 months
Research Priority:High
RNS Developer:Edwin R "Chip" Kraft
Source Info:AR030
Date Posted:09/27/2018
Date Modified:02/26/2019
Index Terms:Grade crossing protection systems, Railroad grade crossings, Railroad safety, Safety assessment, Improvements,
Cosponsoring Committees:AR080, Highway/Rail Grade Crossings; AR040, Freight Rail Transportation
Data and Information Technology
Safety and Human Factors

Please click here if you wish to share information or are aware of any research underway that addresses issues in this research needs statement. The information may be helpful to the sponsoring committee in keeping the statement up-to-date.