Data generated by everyday mobile phone activity could offer a faster way to monitor carbon dioxide emissions across cities, according to a new international study published in The Innovation. Researchers have developed a machine-learning framework that uses human mobility patterns to estimate emissions, potentially providing climate policymakers with more timely information than conventional reporting systems.
The framework was developed by researchers from the National University of Defense Technology, the University of Hong Kong and international collaborators. Tested across China, Italy, the United States and Mexico, the approach achieved prediction accuracy with R² values reaching 0.99 in several international assessments.
Human Mobility Patterns Reveal Changes in Carbon Emissions
Traditional carbon emission estimates largely depend on statistical records and emissions inventories. Although these methods provide valuable information, they can take months to reflect changes in economic activity. Satellite-based measurements also face limitations involving geographical coverage, spatial resolution and how frequently data can be updated.
Researchers explored whether human movement could provide a faster alternative. Commuting, business travel and deliveries involve activities that consume energy and contribute to carbon emissions. Changes in these movement patterns could therefore help reveal fluctuations in emissions.
To test this approach, the team analysed anonymised cellular signalling data from China Unicom, covering approximately 318 million active users. The researchers converted the information into daily intercity mobility networks, representing cities as nodes and movement between them as connections.
They then extracted network characteristics, including node centrality and clustering, and combined these with population and geographical information. Machine-learning models were trained using the GRACED daily emissions dataset to estimate carbon emissions associated with different types of transportation.
Machine-Learning Framework Maintains Accuracy During COVID-19 Disruptions
The researchers tested their approach using data from January and February 2020, a period marked by Spring Festival travel and the initial COVID-19 lockdowns in China.
Despite the sudden changes in travel and economic activity, human mobility remained closely associated with carbon emissions. National daily mobility and CO2 emissions recorded a correlation coefficient of 0.89, while the relationship between mobility and emissions at the city level reached 0.49.
Across 366 cities and 60 days, the LightGBM machine-learning model achieved R² values of 1.00 for multi-source emissions, 0.99 for domestic aviation and 0.98 for ground transportation.
The researchers also tested whether the framework could work beyond China. Models trained using the same approach were applied to Italy, the continental United States and Mexico.
Using the GRACED emissions dataset, the framework achieved R² values of 0.91, 0.95 and 0.99 for the three countries, respectively. Comparisons against the independent ODIAC emissions inventory produced R² values ranging from 0.90 to 0.99.
These results suggest that mobility data could provide a useful indicator of transportation-related emissions across different geographical and economic settings.
Mobility Networks Could Help Improve Regional Climate Planning
Beyond estimating emissions, the researchers found that mobility networks could reveal functional clusters of cities. Unlike conventional administrative boundaries, these groupings reflect how people actually travel and interact across regions.
The emission patterns of these city clusters were also associated with their positions within mobility networks. This could help policymakers identify interconnected urban areas and develop coordinated regional strategies for reducing emissions.
However, the researchers caution that mobility data should complement existing emissions inventories rather than replace them. Human movement represents only a partial indicator of carbon-producing activities and cannot capture every source of emissions.
The Chinese analysis also covered an unusual two-month period, while the mobile data primarily captured movement between cities rather than travel within them. Changes in working arrangements, transportation systems and economic activity could also alter the relationship between mobility and emissions over time.
The research team recommends combining mobility information with satellite observations, energy-use data and meteorological measurements to improve future estimates.
Read More :