Modeling Weather Conditions Consequences on Road Trafficking Behaviors

Jean-Noël Kien, Guillaume Allain, Fabrice Gamboa, Philippe Goudal, Jean–Michel LoubesarXiv 2012 · DOI · arXiv

Abstract: We provide a model to understand how adverse weather conditions modify traffic flow dynamic. We first prove that the microscopic Free Flow Speed of the vehicles is changed and then provide a rule to model this change. For this, we consider a thresholded linear model, corresponding to an application of a MARS model to road trafficking. This model adapts itself locally to the whole road network and provides accurate unbiased forecasted speed using live or short term forecasted weather data information.

Introduction

Quantifying the impact of adverse weather conditions, such as rain or snow, on traffic flow is a significant challenge in predictive modeling. Traffic flow is inherently non-stationary, influenced by time of day, road infrastructure, and varying driver behaviors. Existing deterministic models often struggle with parameter sensitivity to weather, while empirical studies face data heterogeneity issues, making it difficult to isolate weather as a single variable. This research aims to bridge these gaps by developing a flexible, data-driven model capable of forecasting vehicle speeds across diverse road networks using real-time meteorological feeds.

Methodology

The proposed approach utilizes Multivariate Adaptive Regression Splines (MARS), which employs piecewise linear regressions to identify "hinge points" where traffic behavior shifts. Two specific models were evaluated:

  1. Model $M_1$ (Standard MARS): A highly flexible model that adapts locally to individual road links. While accurate, its lack of a homogeneous structure prevents application to roads with sparse data.
  2. Model $M_2$ (Linear Thresholded Model): A constrained version aligned with traffic flow theory. It posits that weather has negligible impact on traffic moving below a specific "critical speed" threshold, but induces a linear decrease in speed for vehicles attempting to travel faster than that threshold.
Figure 3: Visualization of the piecewise linear thresholded model behavior, showing the relationship between speed under normal and adverse conditions.

A critical component of this methodology is normalization. By dividing speed parameters by the road’s Free Flow Speed (FFS)—the typical speed in the absence of congestion—the researchers created a "globalized" model. This allows a single set of parameters to be applied across different Functional Road Classes (FRCs), from motorways to local streets.

Datasets and Experimental Setup

The study analyzed a 107-day period (November 2009 – February 2010) in Toulouse, France, integrating three data sources:

  • Floating Car Data (FCD): Over 515 million GPS positions and speeds from mobile sensors.
  • Meteorological Data: Geo-tagged updates every 15 minutes from Météo-France, covering categories such as soft rain, heavy rain, and mixed precipitation.
  • Road Network: 2,070 links categorized as FRC 0 to 2 (motorways and major roads).

Key Results and Insights

The research identified a distinct threshold effect: weather impact only triggers once traffic exceeds approximately 66% of the Free Flow Speed. Below this threshold, congestion dominates weather as the primary factor slowing traffic.

Figure 2: Comparison of Root Mean Squared Error (RMSE) between the standard MARS model (M1) and the linear thresholded model (M2) across different road links.

While the unconstrained MARS model ($M_1$) achieved a lower local Root Mean Squared Error (RMSE) of 10,359.8 compared to the thresholded model’s ($M_2$) 11,795.27, the latter proved far more robust for network-wide deployment. The "globalization" of the model—moving from link-specific to network-wide parameters—resulted in a minimal error increase of only 6.07%.

In a practical validation on a freeway with a 130 km/h FFS, the model predicted a reduction to 106.6 km/h during rain. This closely aligns with French legal regulations that reduce speed limits to 110 km/h in wet conditions, providing a strong sanity check for the model's real-world applicability.

Conclusion and Future Work

The linear thresholded model offers a stable and generalizable framework for predicting weather-related traffic slowdowns. Future research will address data sparsity issues and expand the model to include more extreme or rare weather conditions, such as heavy icing. Additionally, while the global model performed well in France, larger geographic regions may require climate-zone-specific parameters to maintain accuracy.

Summary updated

Abstract

We provide a model to understand how adverse weather conditions modify traffic flow dynamic. We first prove that the microscopic Free Flow Speed of the vehicles is changed and then provide a rule to model this change. For this, we consider a thresholded linear model, corresponding to an application of a MARS model to road trafficking. This model adapts itself locally to the whole road network and provides accurate unbiased forecasted speed using live or short term forecasted weather data information.
BibTeX
@misc{kien2012modeling,
  author = {Jean-Noël Kien and Guillaume Allain and Fabrice Gamboa and Philippe Goudal and Jean–Michel Loubes},
  title = {{Modeling Weather Conditions Consequences on Road Trafficking Behaviors}},
  year = {2012},
  eprint = {1210.2294},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1210.2294},
}
Figure