Estimating Winter Weather Road Restoration Time using Outsourced Traffic Data

Estimating Winter Weather Road Restoration Time using Outsourced Traffic Data: Three Case Studies in Maryland

ABSTRACT
The objective of this study is to examine the I-95 Vehicle Probe Project (VPP) data to determine if the speed and travel time data can be used as a basis to calculate winter weather road restoration performance measures, specifically by identifying the time required to restore the roadway to normal operating conditions.  A candidate algorithm based on reduction of speed and change in confidence score within the VPP has been proposed and tested.  The algorithm is evaluated during three known snow events in the state of Maryland to determine if it successfully identifies the onset and clearance of hazardous road conditions.

Winter Road Restoration Time Algorithm
Based on the analysis of the data from hurricane Sandy in this case study, an algorithm for assessing a Winter Road Restoration Time performance measure was proposed and tested for effectiveness on this data set. This algorithm is based on the speed and confidence score reported in the Vehicle Probe Data (VPP) in one-minute intervals:

  • Establish a speed threshold based on normal operating conditions for the time of day and day of week using historical data from the VPP – call this the reference speed.
  • Establish a confidence score threshold based on normal operating conditions for the time of day and day of week using historical data from the VPP – call this the reference confidence score.
  • Use a 15 minute rolling horizon for calculation of speed and percentage of confidence score 30 for each minute. The last 15 minutes should be used and the speed calculation should only be based on reliable speeds. This means that if the confidence score of the speed is not 30 it should be excluded from the calculation.  The average of speed for each minute should be based on real time data from the last 15 minutes.
  • At the point during a winter storm that the percentage of confidence score 30 drops below 40% of the reference confidence score during the night or below 80% of reference confidence score during the day or speed drops below 50% of the reference speed, each for at least 30 minutes, establish this time as the beginning of the winter weather event for the specified roadway segment.
  • If the drop in speed is the only cause of the beginning point of the event, further investigation should be done. The time prior to the beginning should be considered. If the speed is dropping continuously until it hits 50% of the reference speed, the beginning time of the event should be reported at the point where  speed drops below 80% of the reference speed, otherwise the previous reported beginning time is valid.
  • At the point during a winter storm that the percentage of confidence score 30 rises above 40% of the reference confidence score during the night or above 80% of the reference confidence score during the day and speed rises above 50% of the reference speed, each for at least 1 hour, establish this as the ending of the winter weather event for the specified roadway segment.
  • If at the ending of the event, the speed is continuously increasing; the ending point of the event should be reported as when the speed rises above 80% of reference speed. However if the speed does not have a similar pattern and it fluctuates, the previous reported ending time is valid.

The time difference between the beginning and end of the winter weather event establishes the restoration  time  for the roadway.  Other notable times could also be entered if available, such as salt deployment, plowing time, closure events, etc., for reference.

This slideshow requires JavaScript.

Summary and Conclusion

  • The proposed algorithm can detect unusual activity related to a winter weather, but it may be difficult to distinguish between a weather event and a major incident.
  • The reported speed when score is either ‘10’ or ‘20’ should be ignored.
  • Extension of the event into overnight hours in which volumes are typically low is problematic for the algorithm.
  • This algorithm have been used on severe storms and may not work for less severe ones.
  • The algorithm could assist in detecting onset of possible winter weather events, but the fidelity of identifying the onset of a weather event is limited. Human review or use of weather data is needed to guard against false positives.
  • Correctly assessing the beginning of a winter weather event proved most challenging. If the starting and ending of the weather situation can be provided by DOT personnel or weather sensors, then the focus of the algorithm can be simplified to determine the return to normal traffic conditions similar to the Michigan and Ohio algorithms.