2025-09-14 中国科学院 (CAS)
This photo taken on Dec. 27, 2023 shows a car being automatically transported into a stereo garage at Yantai Port in Yantai, east China’s Shandong Province. (Xinhua/Lan Hongguang)
＜関連情報＞
- https://english.cas.cn/newsroom/cas_media/202509/t20250915_1054624.shtml
- https://www.sciencedirect.com/science/article/abs/pii/S2210670725006043
都市交通ネットワークにおける昼間動態CO2増加予測のためのパノプティックAIと多源観測の統合 Integrating panoptic-AI and multi-source observations for daytime dynamic CO2 increment prediction in urban traffic networks
Yonglin Zhang, Tianle Sun, Li Wang, Bo Huang, Shiguang Xu, Xiaofeng Pan, Xiangyun Xiong, Wanjuan Song, Guoxu Li, Zheng Niu
Sustainable Cities and Society Available online: 15 August 2025
DOI:https://doi.org/10.1016/j.scs.2025.106730
Highlights
- Achieve precise daytime hourly predictions of ∆CO2 at a resolution of 30 m.
- A new multimodal mobile platform oriented towards road ∆CO2 was developed.
- Panoptic-AI and multimodal fusion can support road ∆CO2 prediction.
- IML revealed “meteorology-traffic-landscape” nonlinear contributions of road ∆CO2.
Abstract
Urban expansion and population mobility have driven a continuous rise in road CO₂ emissions, posing significant challenges to local climate regulation, public health, and carbon neutrality. The development of precise monitoring methods for performing multi-factor analysis of on-road CO₂ levels is considered of great importance for their effective reduction. Along these lines, in this work, focusing on Shenzhen, a global megacity, a hybrid framework combining Panoptic-Artificial Intelligence (Panoptic-AI) and mobile observation framework was developed to predict the hourly 30-meter resolution of the on-road CO₂ concentration and its increment-∆CO₂ (DTCO₂). Using interpretable machine learning (IML), the nonlinear mechanisms of “meteorology-traffic-landscape” dynamic driving forces were uncovered. The key findings of our work include: 1) The proposed IML model achieved high accuracy in daytime DTCO₂ prediction (R² > 0.93 MAE < 1.3 ppm), addressing spatiotemporal resolution challenges. 2) Meteorological and traffic factors significantly affect DTCO₂, with average traffic density (>7 vehicles/30 m) from static snapshots increasing the CO₂ concentration by ∼40 %. 3) DTCO₂ hotspots cluster around transportation hubs on weekdays and diffuse to suburban areas on weekends. A closed-loop “observation-prediction-decision” framework for high-resolution road CO2 monitoring has also been introduced, offering precise, interpretable, and intelligent support for urban traffic CO2 emission reduction. Our work provides valuable insights into the underlying nonlinear mechanisms, which could be useful for applications including landscape optimization and ventilation corridors, thereby advancing low-carbon urban resilience. It provides a novel perspective for the enhanced management and reduction of CO2 emission in urban road traffic by leveraging multi-source information fusion and AI-driven methods.
