This study aimed to investigate how land-use pattern affects crash frequency at traffic analysis zone (TAZ) level. Traffic, road network, land use, population and crash data were collected from Los Angeles County, California in 2014. K-means clustering analysis was first conducted to divide land use at each TAZ into five different patterns. Geographically weighted Poisson regression (GWPR) models were then developed to investigate the associations between crash counts and land-use patterns. The elasticity was calculated to compare the safety effect of each explanatory factor across different patterns. The results of this study indicated that land use combinations at TAZs can be divided into different patterns using land-use mix and proportions of different land use types, and that each land use combination can be assigned with a certain safety level. The effects of contributing factors on crash frequency are different across different land-use patterns. The results suggest that proper combinations of different land uses can improve safety performance at the urban and road network planning stage.

这项研究旨在调查土地利用模式如何影响交通分析区域（TAZ）级别的撞车频率。2014年，从加利福尼亚州洛杉矶县收集了交通，道路网络，土地利用，人口和崩溃数据。首先进行了K均值聚类分析，将每个TAZ的土地利用划分为五种不同的模式。然后开发了地理加权的Poisson回归（GWPR）模型，以研究事故数量与土地利用模式之间的关联。计算弹性以比较每个解释因素在不同模式下的安全效果。这项研究的结果表明，可以根据土地利用组合和不同土地利用类型的比例将TAZ的土地利用组合划分为不同的模式，并且可以为每个土地使用组合分配一定的安全级别。在不同的土地利用模式下，影响因素对崩溃频率的影响是不同的。结果表明，不同土地用途的适当组合可以改善城市和道路网络规划阶段的安全绩效。