Rain reconstruction from various weather-related data sets using logistic regression: methodology and applications
SAINT-PIERRE ; ARON ; BERGEL ; VIOLETTE
Type de document
COMMUNICATION AVEC ACTES INTERNATIONAL (ACTI)
Langue
anglais
Auteur
SAINT-PIERRE ; ARON ; BERGEL ; VIOLETTE
Résumé / Abstract
In order to compute the effect of rainfall on the occurrence of injury accident and on the driver's behaviour, knowledge of rain exposure is necessary. Different meteorological data sources, each of which has advantages and drawbacks, have to be matched before being used for that purpose. A statistical approach (the logistic regression technique) has been chosen, the aim being to reconstitute the relevant information related to rainfall, even at places which are remote from the meteorological measurement stations. Various meteorological sources, both human and sensor-based, are collected and combined. The analysis is based on a six-minute time scale, rather than on the usual hour time scale. The available weather information is used, after a learning phase, to model the probability of rainfall during each six-minute period. This methodology has been applied to estimate the risk of injury accident due to rain, on the French main and country roads in Haute-Normandie. The risk is computed during rainfall and is compared with the risk under non rainy conditions. During rainfall the risk is estimated at 21.9 accidents per 100 million vehicles whereas during normal weather conditions it is estimated at 10.4. Therefore, the average added risk due to rain is estimated at 2.1, and 2.4 in the case of bends, and these results are consistent with other related results.
Editeur
TRB
