Does rainfall actually contribute to increase of accident rates?

A 2016 GovHack Project


We are Team Bit Byte Bit from Perth, Western Australia. At the 2016 Australian GovHack competition, we have taken the direction of crunching data obtained from Main Roads Western Australia, Bureau of Meteorology and Insurance Commission of Western Australia. Our aim is to produce an analysis report on Western Australia's motor vehicle accident rate based on traffic volume, weather conditions and motor injury insurance data.

WA Traffic Incidents Time Lapse Map

Correlation between Rainfall and Traffic Incidents

Due to the large amount of rainfall data and the complexity of matching it to traffic incidents, we attempted to simplify this issue by defining rainfall amount as No Rain (<10mm) or Heavy Rain (>10mm). Thus, using CartoDB, a data visualization tool, we can observe the WA traffic incidents in the No Rain and Heavy Rain conditions, to see the effect rainfall has on traffic incidents.

The time lapse visualization map on the left shows WA traffic incidents over the years from 2011 to 2015.

WA Traffic Incident with No Rain

Using CartoDB, we attempted to analyze the data obtained from Main Roads Western Australia on WA Traffic Incidents with No Rain (0mm) condition. As visualized at the map on the left, the red areas indicate traffic accidents over the period of 02/04/2012 to 10/06/2016, with a rainfall of 10mm. The darker shade of red also illustrates the intensity of traffic accident occurrences over the years.

WA Traffic Incident with Heavy Rain

Next, we analyzed the WA traffic incidents under the Some Rain (>0mm) condition, over the same period of 02/04/2012 to 10/06/2016 as well. Surprisingly, from the visualization map on the right, a smaller area of red and lighter shade of red can be observed, which could be due to our data not being normalized. However, this observation could also mean that contrary to popular belief, heavy rainfall doesn't necessarily contribute to a higher rate of traffic accidents. However, further data analysis and research has to be conducted to affirm this hypothesis.


Kwinana Freeway Analysis

We decided to narrow our research and analysis on just a section of the Kwinana Freeway, as this freeway is considered one of the busiest roads in Perth. We used data obtained from Main Roads WA to observe the average road occupancy on Kwinana Freeway. In traffic congestion studies, the road occupancy is defined as the ratio of the number of cars given a length of road. This means that a high road occupancy rate implies a high congestion of cars. We plotted the Road Occupancy vs. Time for Kwinana Freeway Northbound and Southbound on both No Rain and Heavy Rain conditions. The results are illustrated below.



As expected, during Heavy Rain condition the road occupancy appears to be greater than No Rain condition. Additionally, the heaviest traffic congestion can be seen to be during 8:45am, which is expected since most people are driving north towards the city to go to work during this hour.



The Southbound chart also shows that during Heavy Rain condition, the road occupancy is greater than No Rain condition. Opposite to the Northbound chart, the heaviest traffic congestion is observed to be 5:30pm evening, which makes sense as people are leaving the city to get home from work.

Summary of Results & Conclusion


By summarizing our data findings in a bar chart, we can see that there is a 3.16% decrease in traffic incidents in Heavy Rain condition compared to No Rain condition. Additionally, analyzing the data from Motor Injury Insurance, a bar chart is generated and it shows a 11.3% decrease in traffic incidents in Heavy Rain condition compared to No Rain condition as well. This further reinforces our hypothesis that heavy rain doesn't necessarily contribute to an increase in traffic accidents.


This finding could mean that people are actually driving safer and with more alertness during Heavy Rain conditions.

Future improvements

We could further improve our data analysis on additional roads to further affirm the validity and accuracy of our hypothesis.

For example, normalization of number of incidents is required. The data has to be adjusted to the volume of cars, to result a bias free rate of traffic incidents.  Only then, a correct visualization of the heat map can be achieved.

Furthermore, a better measure of traffic "congestion" is required, as occupancy rate is only an implication but not a true measure of congestion. For further information, please see https://onedrive.live.com/view.aspx?resid=C1B6684C6B6C0C19!1439&ithint=file%2cdocx&app=Word&authkey=!AL_vWMM3DeW_fcs. Moreover, the data presented here only take account one road link (Kwinana FWY). The data can be scaled further on other roads. To visualize the data better, a time lapse animation of congestion heat map can be produced.

SIMBLAWebsite Builder