Weather has a substantial influence on people’s travel behavior. In this study we analyze if meteorological variables can improve predictions of hourly traffic counts at 1400 stations on federal roads and highways in Germany. Motorbikes, cars, vans and trucks are distinguished. It is evaluated in how far the mean squared error of Poisson regression models for hourly traffic counts is reduced by using precipitation, temperature, cloud cover and wind speed data. It is shown that in particular motorbike counts are strongly weather-dependent. On federal roads the mean squared error is reduced by up to 60% in models with meteorological predictor variables, when compared to models without meteorological variables. A detailed analysis of the models for motorbike counts reveals non-linear relationships between the meteorological variables and motorbike counts. Car counts are shown to be specifically sensitive to weather in touristic regions like seaside resorts and nature parks. The findings allow for several potential applications like improvements of route planning in navigation systems, implementations in traffic management systems, day-ahead planning of visitor numbers in touristic areas or the usage in road crash modelling.
