Scaling traffic variables from sensors sample to the entire city at high spatiotemporal resolution with machine learning: applications to the Paris megacity

Bonnemaizon, Xavier; Ciais, Philippe; Zhou, Chuanlong; Arous, Simon Ben; Davis, Steven J; Megel, Nicolas · 2024 · DOAJ

DOI: 10.1088/2634-4505/ad6bbf

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of estimating high-resolution traffic patterns in urban areas where sensor coverage is sparse. Road transportation is a major source of carbon dioxide and nitrogen oxide emissions in Paris, yet existing estimates often lack the spatiotemporal resolution required to inform decarbonization policies or assess local air quality. The authors aim to upscale point-scale measurements from a limited network of magnetic road sensors to generate hourly traffic flow and occupancy data for the entire city’s main road network. This bottom-up approach seeks to provide a robust framework for monitoring traffic variables and deriving emission inventories, particularly in the context of shifting mobility patterns due to the COVID-19 pandemic and infrastructure changes. The researchers utilized data from 2,086 magnetic sensors across Paris covering the period from 2018 to 2022. The raw data, which measured vehicle flow and occupancy, underwent rigorous preprocessing to remove outliers using normal and Gumbel distribution fits and to fill missing timestamps. Random Forest regression models were employed for gap-filling, incorporating temporal features such as hour, weekday, and COVID-19 stringency indices. To predict traffic variables on unmonitored road segments, the authors trained a Random Forest model using 11 temporal features and four spatial features derived from OpenStreetMap, including lane count, speed limits, road category, and betweenness centrality. The model was optimized via grid search and evaluated using normalized root mean squared error and symmetric mean absolute percentage error. The results demonstrate that the machine learning model successfully captured traffic patterns with a symmetric mean absolute percentage error of 37% for flow and 54% for occupancy. The analysis revealed distinct traffic regimes characterized by fundamental diagrams, where flow increases linearly with occupancy in free-flow conditions but stabilizes or decreases during congestion. The model effectively identified the relative importance of various attributes, showing that temporal factors like holidays and crisis stringency, alongside spatial infrastructure characteristics, significantly influence traffic variables. The study generated a comprehensive dataset covering 6,846 road segments, providing hourly insights into traffic dynamics across the city. The significance of this work lies in its provision of a versatile, high-resolution tool for sustainable urban monitoring. By generating detailed maps of traffic flow and occupancy, the methodology enables more accurate assessments of transportation emissions, as these variables allow for the derivation of mean vehicle speeds and subsequent emission factors. The approach is not limited to Paris and can be applied to other urban centers with similar data availability. This framework supports policymakers in developing targeted strategies for emission reduction and air quality improvement by offering granular, evidence-based insights into how traffic patterns respond to temporal perturbations and spatial infrastructure.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction