Non-linear impact mechanisms of multi-modal urban traffic on air quality: An interpretable machine learning study for sustainable policy making.

Hou, J; Wang, J; Guo, X · 2026 · PubMed Central

DOI: 10.1371/journal.pone.0350301

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 investigates the non-linear mechanisms linking multi-modal urban traffic to air quality, specifically focusing on Nitrogen Dioxide (NO2) concentrations at the Bundaran HI intersection in Jakarta, Indonesia. Motivated by the limitations of traditional linear emission models that fail to capture micro-scale "stop-and-go" dynamics, the research aims to quantify how mixed traffic flows—comprising motorcycles, private cars, and heavy vehicles—drive pollution spikes under tropical conditions. The authors seek to identify specific congestion thresholds and evaluate the environmental utility of targeted policy interventions to support sustainable urban planning. The researchers utilized a high-resolution, year-long longitudinal dataset (January–December 2023) containing 8,760 hourly observations of traffic flow, air quality, and meteorological data. They employed a Random Forest (RF) regression model, optimized via grid search and 5-fold cross-validation, to predict NO2 levels. To ensure interpretability, the study integrated Permutation Importance and Partial Dependence Analysis (PDP) to decipher feature contributions and identify non-linear thresholds. Data preprocessing included anomaly detection using the Interquartile Range method and temporal k-Nearest Neighbor imputation to handle sensor noise while preserving diurnal patterns. Key findings reveal that private car volume and the Volume-to-Capacity (V/C) ratio are the primary drivers of NO2 spikes, outweighing the direct contribution of heavy vehicles. Crucially, the study identified a distinct non-linear threshold: when the V/C ratio exceeds 0.65, NO2 concentrations undergo a regime shift, rising exponentially from approximately 38 ppb to 44 ppb. This "elbow" effect corresponds to the transition from laminar flow to a stop-and-go regime, where frequent acceleration and idling generate disproportionate emissions. Meteorological factors, particularly temperature, were found to be the most sensitive predictors, exacerbating pollution through a synergistic "heat-and-pollution bubble" created by the high thermal inertia of Stone Mastic Asphalt (SMA) road surfaces. Policy simulations indicated that restricting heavy vehicles yielded a 6.9% reduction in mean NO2, whereas a 30% reduction in private car volume resulted in a negligible 0.6% increase in NO2 due to compensatory traffic turbulence. The significance of this work lies in its demonstration that traditional linear mitigation strategies significantly underestimate pollution risks during saturated traffic states. The findings advocate for a shift from generic traffic bans to precision-based management strategies, such as threshold-based adaptive signaling to keep V/C ratios below 0.65 and targeted restrictions on heavy vehicles during peak thermal hours. By integrating interpretable machine learning with traffic physics, the study provides a data-driven framework for urban planners to optimize traffic flow and pavement materials, aligning with UN Sustainable Development Goal 11 and principles of cleaner production.

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.

• traffic density