This paper incorporates various currently known collection mechanisms (including Brownian diffusion, interception effect, inertial impaction, thermophoresis, diffusiophoresis, and electrostatic interaction) into the calculation of the total collection efficiency to analyze their impacts on the scavenging coefficient. The turbulent effect is introduced into the parametric study of the scavenging coefficient. Combining the local raindrop size distribution and aerosol size distribution, a theoretical prediction model for multi-fraction aerosol scavenging by rainfall is established and verified and corrected with measured data. The main conclusions are as follows: For particles within the accumulation mode range, the influence of the collision efficiency needs to be carefully considered. When studying the scavenging coefficient, it is necessary to combine the locally measured raindrop size distribution and aerosol size distribution. The influence of the aerosol size distribution on the scavenging coefficient under different seasonal conditions in the same area can be neglected. When the turbulent effect is introduced, the theoretical prediction is closer to the actual situation. In comparison with the actual measured PM_2.5 values in Guangzhou City, Hefei City, and Tianjin City, the temporal variation characteristics of PM_2.5 estimated by the theoretical model exhibit a substantial degree of consistency with the trends revealed by the measurement results. Additionally, a linear correlation is discernible between the scavenging coefficients obtained from field measurements in these three regions and those calculated by the theoretical model. Specifically, the equations of the linear relationships are Λ_s = 0.498 × 10⁻⁵ + 1.025Λ_m; Λ_s = 1.035Λ_m − 0.036 × 10⁻⁵; and Λ_s = 0.903Λ_m − 1.11 × 10⁻⁵.

Rising anthropogenic carbon emissions are a major driver of climate change and pose a critical challenge to global sustainable development. As a rapidly advancing technology, artificial intelligence (AI) has shown strong potential to enhance carbon emissions management. This review provides a critical and comprehensive synthesis of recent advances in AI-enabled approaches for carbon emissions monitoring, prediction, and reduction. For monitoring, it explores the integration of satellite remote sensing, sensor networks, and machine learning (ML) algorithms, which can improve multi-scale, high-resolution, and near-real-time monitoring capabilities. For prediction, it categorizes prediction models into three groups, namely deep learning (DL), ensemble learning, and statistical learning, to facilitate the selection of appropriate technical approaches based on varying data characteristics and prediction requirements. For reduction, it examines the practical effectiveness of AI in industrial process optimization, energy structure transformation, transportation scheduling and management, construction energy efficiency improvement, and carbon capture, utilization, and storage (CCUS). We further reveal core challenges and potential solutions across the data layer, model layer, and application layer in AI deployment, including data availability and quality, model generalization and interpretability, and engineering and governance barriers that hinder the translation of AI methods into real-world applications. Furthermore, future research directions are discussed to promote the development of more reliable and scalable AI methods that can better support decision-making and practical governance in carbon emissions management. Overall, distinct from previous reviews that mainly focus on single tasks, specific model types, or sectoral applications, this review represents, to our knowledge, one of the first review-level attempts to develop a policy-relevant and interdisciplinary AI framework for carbon emissions management across the full process of monitoring, prediction, and reduction. By integrating unified evaluation metrics, evidence matrices, deployment-constraint analysis, and a technology readiness level (TRL)-based assessment, this framework links methodological performance, application readiness, and governance needs. It provides an integrated methodological foundation for fine-grained emissions sensing, predictive analysis, and emissions reduction decision support, while supporting quantifiable, verifiable, and actionable carbon balance and management.