A team from the Institute of Information and Communication Technologies (ITACA) at the Polytechnic University of Valencia (UPV) and the Institute of Corpuscular Physics (IFIC), a joint centre of the Spanish National Research Council (CSIC) and the University of Valencia (UV), has developed an innovative urban traffic prediction and early warning system based on deep learning techniques.
The new method, which makes it possible to anticipate high pollution episodes and thus facilitate the adoption of preventive measures, “is ready to be exported and help improve air quality in urban environments around the world,” the researchers emphasize.
In this project, the team from the ITACA institute and IFIC started from a premise: reducing transport emissions not only helps mitigate climate change but also directly improves air quality in cities. In Valencia, for example, traffic represents around 60% of total greenhouse gas (GHG) emissions.
To address this situation, the researchers applied in this city a system that predicts, 30 minutes in advance, whether a street segment will experience heavy traffic, thus facilitating preventive measures to reduce pollution and protect public health.
“Urban traffic is an important source of harmful air pollutants. We must not forget that air pollution is the leading environmental cause of premature deaths,” says Edgar Lorenzo-Sáez, researcher at the ITACA Institute and one of the study’s authors. Poor air quality has also been linked to diseases such as asthma, lung cancer, and cardiovascular problems, “responsible for around 300,000 premature deaths per year in the European Union,” he adds.
A precise, reliable, and scalable system
The system developed by the UPV and IFIC team has been trained with data from 1,472 traffic sensors distributed across the city of Valencia and complemented with meteorological variables (wind, rain, atmospheric pressure, etc.). The new method classifies each road segment into three alert levels and, thanks to the use of Long Short-Term Memory (LSTM) neural networks, achieves high real-time accuracy, even during rush hours.
The model has also demonstrated that traffic data can serve as a reliable indicator of NOx (nitrogen oxides) levels, one of the most harmful pollutants to health. This is particularly useful in places where there isn’t a dense network of air quality sensors. This capability could strengthen the effectiveness of Low Emission Zones (LEZ), with more localized measures tailored to the actual risk of each street, avoiding generalized restrictions with greater social impact.
“Our system is right 90% of the time when traffic is smooth and 70% when predicting heavy traffic episodes. This opens the door to quicker decisions that prevent exceeding legal pollution limits in sensitive areas,” adds Edgar Lorenzo-Sáez.
For his part, Javier Urchueguía, also an ITACA researcher, highlights: “We have found a direct correlation between traffic flows and recorded NOx levels, which allows us to generate alerts even without a complete air quality sensor network. This is a key finding for many European cities with limited resources.”
Likewise, Verónica Sanz, professor at the UV, IFIC researcher, and coauthor of the study, explains that the system’s intelligence has been developed using AI models capable of learning how the city “breathes” and anticipating changes in traffic and pollution.
“The models have been designed to be robust and adaptable to different scenarios, opening the door to their application in many other cities,” she notes. “Artificial intelligence can be a great ally to help cities breathe better. This system, developed in Valencia, is ready to be exported and help improve air quality in urban environments around the world,” she stresses.
A step toward more sustainable and resilient cities
This work represents significant progress in data-driven urban management, integrating artificial intelligence as a tool to address complex environmental challenges. According to its authors, the system could become an essential instrument for designing more dynamic, effective, and socially accepted interventions, particularly aimed at protecting vulnerable groups such as schoolchildren, the elderly, or patients with respiratory diseases.
Future development lines include creating a digital twin of the city of Valencia to simulate measures before real implementation and incorporating additional Internet of Things (IoT) sensors to improve the direct prediction of pollutants.
The study was published in the scientific journal Neural Computing and Applications and supported by institutions such as the Generalitat Valenciana and the Ministry of Science and Innovation.
Reference: Miguel G. Folgado, Verónica Sanz, Johannes Hirn, Edgar Lorenzo-Sáez, Javier Urchueguía. Methodology development for high-resolution monitoring of emissions in urban road traffic systems, Atmospheric. Neural Computing and Applications. https://link.springer.com/article/10.1007/s00521-025-11316-0