GPS Data and Machine Learning Tools, a Practical and Cost-Effective Combination for Estimating Light Vehicle Emissions

Rivera, Néstor; Ramírez, Blanca Arenas; Sanz, José Luis Muñoz; Jiménez, Edisson · 2024 · OpenAlex-citations

DOI: 10.3390/s24072304

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 accurately estimating light vehicle emissions under real-world driving conditions, particularly in regions like Latin America where standard models (such as IVE and COPERT) may fail to account for local variations in driving styles, fuel quality, and environmental factors. The authors propose a practical, cost-effective methodology that combines GPS data with machine learning techniques to estimate emissions without the need for expensive, long-term Portable Emissions Measurement Systems (PEMS) campaigns. The research focuses on the three most sold vehicle categories in Ecuador: sedans, SUVs, and pickups. The experimental design involved conducting Real Driving Emission (RDE) tests on two standardized routes in Cuenca, Ecuador, using the three vehicle types. Data were collected using a GPS-enabled data logger and a PEMS (Brain Bee AGS-688 gas analyzer) to measure CO2, CO, NOx, and HC emissions. Route 1 data were used to train the models, while Route 2 data served for validation. The methodology included estimating gear usage via K-means clustering and classification trees, determining the relative importance of driving variables using Random Forest techniques, and training Artificial Neural Networks (ANNs) to predict emissions. The trained ANNs were then validated against a large dataset comprising 324, 300, and 316 km of random driving for each vehicle type, respectively. The results demonstrated that the machine learning models, fed with GPS-derived variables (speed, gradient) and vehicle-specific parameters (mass, engine displacement, aerodynamic coefficients), produced emission estimates comparable to those from the IVE model and an OBD-based model. The ANNs proved robust across different traffic conditions due to training on extensive random driving data. The study confirmed that this approach effectively estimates pollutant emissions without requiring the continuous mounting of PEMS equipment for long test drives, offering a viable alternative for large-scale emission inventories. The significance of this work lies in providing a low-cost, scalable tool for estimating vehicular emissions that accounts for real-world operational factors often ignored by traditional models. By leveraging widely available GPS data and machine learning, the method enables more accurate environmental impact assessments and policy evaluations in regions with distinct driving and environmental characteristics. This approach facilitates better monitoring of urban air quality and supports the development of targeted strategies to reduce vehicular pollution.

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.

• exposure measurement