A multi-agent system for neighborhood-scale carbon simulation: Integrating intelligent multi-source sensing with interaction-driven modeling

Excessive carbon emissions pose a major barrier to sustainable urban development; however, neighborhood-scale assessments are often hindered by data scarcity and low resolution, limiting the ability to explain how human activities and environmental factors jointly shape emissions. This study develops a neighborhood-scale carbon emissions simulation framework that integrates unmanned aerial vehicle-based remote sensing, computer vision, geographic information system techniques, and field surveys to acquire multi-source data. A multi-agent system was constructed to represent human behavior, built environments, and climate conditions as interacting agents with heterogeneous attributes and behavioral rules, enabling interaction-driven emissions estimation. A case study in Nanjing, China, estimated annual carbon emissions of 2.61 × 10 8 kg CO 2 e, corresponding to a land-based emission intensity of 106.53 kgCO 2 e/m 2 . The results reveal detailed spatiotemporal emission patterns and pronounced seasonal variability: buildings dominate emissions, mobility contributes 1.6%, and open spaces act as a net sink, offsetting over 2% of neighborhood emissions. Seasonally, summer emissions were approximately 40% higher than those in the transitional seasons, primarily driven by extended cooling operations and climate-sensitive behavioral shifts. The proposed approach advances explanatory power and computational efficiency in neighborhood-scale carbon accounting, providing a practical basis for low-carbon neighborhood management and sustainable urban development. • A multi-agent system simulates interaction-driven neighborhood-scale carbon emissions. • A framework integrates UAVs, computer vision, and GIS for high-resolution data. • Results reveal spatiotemporal distributions and seasonal variability. • The simulation offers a data-driven basis for low-carbon neighborhood management.