交通运输工程与信息学报

【背景】街景图像等多源空间数据为交通安全建模提供了精细化环境观测基础，但现有研究多停留在统计关联或预测贡献层面，空间横截面数据下的因果识别仍受非均匀空间结构与局部扰动影响。【目标】构建适应局部非均匀空间结构并降低模式识别敏感性的空间因果识别方法，在估计总体因果强度的同时区分其作用方向构成。【方法】提出自适应地理模式因果方法（Adaptive Geographical Pattern Causality， AGPC），在收敛交叉映射框架下以局部协方差结构构造各向异性邻域，并通过模式分布一致性度量空间因果强度及其方向构成。【数据】以美国迈阿密都市区2 km × 2 km 规则网格为分析单元，利用街景图像语义分割提取绿视率，并结合交通事故频数数据开展实证检验。【结果】绿视率与总交通事故频数之间的空间因果结构以暗因果为主要组成，在可明确判定方向的成分中负向占优，负向符号比例达0.947，且主要因果构成在多项扩展与稳健性检验中保持稳定。【结论】AGPC可为非均匀空间结构下交通安全问题的因果解释提供稳健方法支撑；在满足交通安全与道路功能要求的前提下提升高交通活动区域街景绿化品质，具有潜在的交通安全改善价值。

[Background] Street-view imagery and other multisource spatial data provide a refined basis for environmental observation in traffic safety modeling. However， existing studies have largely focused on statistical associations or predictive contributions， while causal identification from spatial cross-sectional data remains affected by non-uniform spatial structures and local perturbations. [Objective] This study aims to develop a spatial causal identification method that adapts to locally non-uniform spatial structures and reduces the sensitivity of pattern identification， while estimating overall causal strength and distinguishing its directional composition. [Method] An Adaptive Geographical Pattern Causality (AGPC) method is proposed. Within the convergent cross-mapping framework， AGPC constructs anisotropic neighborhoods based on local covariance structures and measures spatial causal strength and its directional composition through pattern distribution consistency. [Data] Taking 2 km × 2 km regular grids in the Miami metropolitan area， USA， as the analytical units， the green view index was extracted from street-view imagery using semantic segmentation and combined with traffic crash frequency data for empirical analysis. [Results] The spatial causal structure between the green view index and total traffic crash frequency was dominated by dark causality. Among the components with identifiable directions， negative causality was predominant， with a negative sign proportion of 0.947. The main causal composition remained stable across multiple extension and robustness analyses. [Conclusion] AGPC provides robust methodological support for causal interpretation of traffic safety issues under non-uniform spatial structures; enhancing streetscape greenery quality in high-traffic areas， while satisfying traffic safety and road function requirements， holds potential value for traffic safety improvement.