Spatial Analysis of Weather Crash Patterns in Wisconsin

1
Spatial Analysis of Weather Crash Patterns in
Wisconsin
Mid Continent Transportation Research Forum 2006

• Ghazan Khan
• Xiao Qin Ph.D., P.E
• David A. Noyce Ph.D., P.E
• Department of Civil and Environmental Engineering
  
• University of Wisconsin-Madison
• August 17, 2006

2
Outline

• Background
• Problem Statement
• Objectives
• Data Collection
• Data Analysis
• Results
• Conclusions and Recommendations

3
(No Transcript)
4
Background

• Oct. 2002, I-43 Sheboygan County, Deadliest crash
  in Wisconsin-10 fatalities, 39 injuries
• 1,023,000 (16) Weather-related crashes in the US
  in 2003
• WisDOT considering Road Weather Safety Audit
  implementation

5
Problem Statement

• Where should Road Weather Safety Audits (RWSA) be
  conducted
• What are the locations of weather-prone areas
• How to quantify road weather safety performance
• How to utilize weather data in a proactive
  approach towards road safety

6
Objectives

• Identification of weather-prone locations on a
  county level
• Identification of weather-prone locations on a
  road segment level
• Establishment of a procedure for the utilization
  of crash and weather data collectively

7
Data Collection

• Weather-related crashes in Wisconsin 2000-2002
  for fog, rain, snow
• Relative crash rate defined
• RWIS Data
• Winter Severity Index 2000-2002
• National Weather Service Cooperative Observing
  Network (COOP) data from National Climatic Data
  Center (NCDC)
• Automated Weather Observing System (AWOS)

8
Data Analysis

• Spatial Statistical procedure for Identifying
  spatial patterns of clustering, Getis-Ord Gi
  statistic
• H0 Weather-related crashes display no clustered
  patterns and are randomly distributed hence not
  affected by weather
• Ha Weather-related crashes show clustered
  patterns signifying that they are affected by
  weather
• The location of clustered patterns are
  weather-prone areas

9
County Level Analysis of Weather Related Crashes

• Relative crash rate for each county in Wisconsin
  (fog, snow, and rain crashes)
• Definition Ratio of weather to all crashes
• Statistically significant spatial patterns of
  clustering at various location
• Temporal consistency of spatial patterns
• Comparison with weather data

10
County Level Spatial Patterns of Fog-Related
Crashes
11
County Level Spatial Patterns of Snow-Related
Crashes
12
County Level Spatial Patterns of Rain-Related
Crashes
13
Weather Data Analysis (point..)
14
Grid Level Analysis

• Maps of Wisconsin divided into 5 by 5 km grid
  sections
• Relative crash rate for weather-related crashes
  and weather parameter calculated for each grid
• Statistically significant spatial patterns of
  clustering at various locations
• Relationship between weather-related crashes and
  weather parameter
• Example Snow-related crashes

15
Relative Crash Rate for Snow-related Crashes by
Grid Locations in Wisconsin 2000 2002
16
Statistically Significant Spatial Patterns for
Snow-related Crashes using Getis-Ord Gi
Statistic 2000 2002
17
Relationship between Snow Crashes and Intensity
18
Results Summary

• Weather related crashes on a county / grid level
  show statistically significant and consistent
  spatial patterns of clustering
• Weather prone locations are areas of
  statistically significant high crash rate
  clusters
• Comparisons with weather data verify and validate
  the accuracy of the results

19
Contributions

• Sites of promise identified through
  statistically significant results on
  Macro/Micro-level
• Unique procedure quantifying road weather safety
  by incorporating both crash and weather data
• Address the weather problem both reactively and
  proactively
• Procedures can help in the assessment of
  countermeasures implemented
• Evaluation of multiple sources of weather data
  identified to supplement RWIS data

20
Recommendation

• Road Weather Safety Audits should be conducted to
  assess the actual impacts of weather at locations
  identified
• Input of winter maintenance staff is very
  important
• Procedures should be applied to current data

21
Questions?