Suppr超能文献

整合可解释人工智能与因果推断以揭示中国区域空气质量驱动因素。

Integrating explainable AI and causal inference to unveil regional air quality drivers in China.

作者信息

Fu Zhiyuan, Yang Xiao, Ma Yike, Sun Yuhang, Wang Tianlian

机构信息

School of Resources Environment and Tourism, Anyang Normal University, Anyang, 455000, China.

College of Mechanical Engineering, Quzhou University, Quzhou, 324000, China.

出版信息

J Environ Manage. 2025 Aug;390:126270. doi: 10.1016/j.jenvman.2025.126270. Epub 2025 Jun 20.

Abstract

Air pollution poses a pressing global public health challenge, demanding a comprehensive understanding of its causes and evolving dynamics to inform effective control strategies. In China, significant spatial heterogeneity complicates the national air quality improvement process. Different regions exhibit varying pollution drivers, which makes uniform governance approaches less effective. Addressing this complexity requires a framework capable of capturing localized causal mechanisms. This study introduces the CADEPT (Causal Analysis-Detection-Explanation-Prediction-Threshold) multi-scale causal inference framework. Using nationwide air quality monitoring data from 2014 to 2022, the study integrates urban, socio-economic, and climatic datasets. It systematically investigates the driving forces and future evolution of the Air Quality Index (AQI) through four major stages: spatial heterogeneity detection, causal and interpretable inference, threshold identification, and scenario-based prediction. The results reveal the following key findings: (1) Local Moran's I detects significant spatial clustering of AQI, while spatial heterogeneity analysis uncovers region-specific influences from climate, emissions, and industrial structures. (2) Interpretability analysis, based on SHAP and TabPFN, uncovers nonlinear and region-specific contributions of key variables to AQI, highlighting climate's mitigating role and industry's aggravating effect. (3) Causal inference quantifies the true impacts of dominant factors, confirming that increases in temperature and precipitation improve air quality, while SO emissions and industrial expansion worsen it. (4) Threshold analysis identifies critical response intervals, highlighting a synergistic amplification effect between meteorological conditions and emission factors. (5) Scenario simulations suggest that promoting low-carbon transitions and coordinated emission reductions are essential for achieving sustained nationwide air quality improvements.

摘要

空气污染构成了紧迫的全球公共卫生挑战,需要全面了解其成因和动态变化,以为有效的控制策略提供依据。在中国,显著的空间异质性使全国空气质量改善进程变得复杂。不同地区呈现出不同的污染驱动因素,这使得统一的治理方法效果不佳。应对这种复杂性需要一个能够捕捉局部因果机制的框架。本研究引入了CADEPT(因果分析-检测-解释-预测-阈值)多尺度因果推理框架。该研究利用2014年至2022年的全国空气质量监测数据,整合了城市、社会经济和气候数据集。它通过四个主要阶段系统地研究了空气质量指数(AQI)的驱动因素和未来演变:空间异质性检测、因果和可解释推理、阈值识别以及基于情景的预测。结果揭示了以下关键发现:(1)局部莫兰指数检测到AQI存在显著的空间聚类,而空间异质性分析揭示了气候、排放和产业结构对特定区域的影响。(2)基于SHAP和TabPFN的可解释性分析揭示了关键变量对AQI的非线性和特定区域贡献,突出了气候的缓解作用和产业的加剧影响。(3)因果推理量化了主导因素的真实影响,证实温度和降水的增加改善了空气质量,而二氧化硫排放和产业扩张则使其恶化。(4)阈值分析确定了关键响应区间,突出了气象条件和排放因素之间的协同放大效应。(5)情景模拟表明,促进低碳转型和协同减排对于实现全国空气质量的持续改善至关重要。

文献AI研究员

20分钟写一篇综述,助力文献阅读效率提升50倍。

立即体验

用中文搜PubMed

大模型驱动的PubMed中文搜索引擎

马上搜索

文档翻译

学术文献翻译模型,支持多种主流文档格式。

立即体验