Diurnal variation mapping of urban NO₂ concentrations at high spatial resolution using mobile phone signaling data

Atmospheric NO₂ remains one of the primary pollutant gases in urban areas. However, current techniques such as ground-based measurements, remote sensing, and atmospheric chemistry models have persistently faced challenges in monitoring sub-daily NO₂ concentrations at high spatial resolution. Mobile phone signaling data holds significant potential for characterizing human activity intensity, offering new opportunities to track near-surface NO₂ concentrations with refined spatiotemporal precision. This study designed a simple algorithm to extract population mobility patterns from mobile phone signaling data and developed a novel deep learning model tailored for environmental variable modeling, marking the first implementation of continuous NO₂ concentration estimation at a 250-meter/1-hour spatiotemporal resolution driven by mobile phone signaling data. Application results in Ningbo City demonstrated that the derived dynamic population mobility information effectively captured local traffic activity patterns, including morning/evening rush hours and nighttime variations. The proposed deep learning model exhibited excellent robustness and spatial generalizability, with a 5-fold cross-validation R² of 0.815 and a spatial prediction validation R² of 0.774, significantly outperformed traditional machine learning models (e.g., RF, XGBoost, and DNN). Interpretability analysis revealed the dominant role of dynamic traffic emissions in NO₂ concentration estimation (22.59 %), along with auxiliary influences from wind speed (16.45 %), boundary layer height (9.07 %), and rainfall (7.24 %). The high-resolution NO₂ concentrations output by the model showed strong spatial consistency with Tropospheric Monitoring Instrument (TROPOMI) observations during satellite overpass times, while their diurnal cycle distributions displayed significant variability across agricultural, urban, and suburban areas. This research highlights the potential of leveraging mobile phone signaling data to advance high-resolution air pollutant modeling and tracking.