Back to statistics and a look at how road risk varies with precipitation, taking the last 3 years of road accidents in Victoria against the rainfall record. One useful take-home statistic is that Melbourne has had an average of 143 days of precipitation per year, roughly 60:40 dry:wet.
Binning the rainfall record into an exponential distribution the total number of fatalities recorded in the state of Victoria on those days was divided by the number of days to produce a rate of deaths per day. The rainfall record used is from Melbourne given that the majority (70%) of the population of Victoria live there. This produces the following:
A line of best fit weighted by occurrences (days in each bin) suggests a possible negative correlation between rainfall and death rates. The general feeling by commuters is that traffic is slower and more congested on wet days. Melbourne already operates its roads close to saturation and the addition of more users displaced from walking or cycling to the more comfortable motorised transport would be expected to slow the speed of all users. This reduces risk while the wet surfaces should increase it, but the overall trend is down. To check if there is any connection with visibility the same technique is applied to the amount of sunshine (Solar Insolation) with a similar trend:
On Days with reduced sunshine the road toll might be slightly lower than average, no worries about visibility or adequate lighting on the roads. Finally the road fatality data is provided with the date and time of occurrence, while the rainfall data is reported for the 24 hours to 9am. Instead of aligning the two sets to each other if we shift the road toll data and look at the rate after rain a stronger trend appears:
Certainly looks interesting, I’ll offer a possible explanation as above that wet weather puts more users onto the roads increasing congestion and reducing average speeds. Needing to the same car for both directions of travel even after rain has ended the congestion remains and suppresses accidents. It would be very interesting to have some throughput or average speed data to confirm this against the rainfall records or more granular location data to match against rainfall measurements, but the result here is already counterintuitive and interesting alone.