Traffic crashes are geographical events, and their spatial patterns are strongly linked to the regional characteristics of road network, sociodemography, and human activities. Different human activities may have different impacts on traffic exposures, traffic conflicts and speeds in different transportation geographic areas, and accordingly generate different traffic safety outcomes. Most previous researches have concentrated on exploring the impacts of various road network attributes and sociodemographic characteristics on crash occurrence. However, the spatial impacts of human activities on traffic crashes are unclear. To fill this gap, this study attempts to investigate how human activities contribute to the spatial pattern of the traffic crashes in urban areas by leveraging multi-source big data. Three kinds of big data sources are used to collect human activities from the New York City. Then, all the collected data are aggregated into regional level (ZIP Code Tabulation Areas). Geographically Weighted Poisson Regression (GWPR) method is applied to identify the relationship between various influencing factors and regional crash frequency. The results reveal that human activity variables from multi-source big data significantly affect the spatial pattern of traffic crashes, which may bring new insights for roadway safety analyses. Comparative analyses are further performed for comparing the GWPR models which consider human activity variables from different big data sources. The results of comparative analyses suggest that multiple big data sources could complement with each other in the coverage of spatial areas and user groups, thereby improving the performance of zone-level crash models and fully unveiling the spatial impacts of human activities on traffic crashes in urban areas. The results of this study could help transportation authorities better identify high-risky regions and develop proactive countermeasures to effectively reduce crashes in these regions.

交通事故是地理事件，其空间格局与道路网络、社会人口学和人类活动的区域特征密切相关。不同的人类活动可能对不同交通地理区域的交通暴露、交通冲突和速度产生不同的影响，从而产生不同的交通安全结果。以往的研究大多集中在探索各种道路网络属性和社会人口特征对事故发生的影响。然而，人类活动对交通事故的空间影响尚不清楚。为了填补这一空白，本研究试图通过利用多源大数据来研究人类活动如何影响城市地区交通事故的空间格局。三种大数据源用于收集纽约市的人类活动。然后，将所有收集的数据汇总到区域级别（邮政编码列表区域）。应用地理加权泊松回归（GWPR）方法来识别各种影响因素与区域碰撞频率之间的关系。结果表明，来自多源大数据的人类活动变量显着影响交通碰撞的空间格局，这可能为道路安全分析带来新的见解。进一步进行比较分析以比较考虑来自不同大数据源的人类活动变量的 GWPR 模型。对比分析结果表明，多个大数据源在空间区域和用户群体的覆盖范围上可以相互补充，从而提高区域级碰撞模型的性能，充分揭示人类活动对城市地区交通事故的空间影响。这项研究的结果可以帮助交通部门更好地识别高风险地区，并制定积极的应对措施，以有效减少这些地区的撞车事故。