Abstract Land use regression models (LUR) associate observed air pollution concentrations with surrounding land use characteristics for air pollution modelling. This technique is common in urban landscapes focused at a city-wide spatial scale. Our study tested the applicability of LUR modelling at a local scale, defined as multiple air monitors within a neighbourhood. The study area was 15.4 km of an urban transportation corridor in Mississauga, Canada. Nitrogen dioxide (NO2) was sampled at 112 sites during the summer in 2018 and observations ranged from 5.8 ppb to 19.65 ppb. A linear regression LUR model explained 69% of the variation in NO2 concentrations at this local scale, with estimated prediction errors less than 1.61 ppb, which were calculated by three cross-validation methods. Traffic volume, major and minor road lengths were key determinants among the predictor variables, and park area and distance to the nearest major intersection were the only variables with negative coefficients in the local-scale model. Extending the linear model approach with regression kriging improved the model's explanatory ability with a coefficient of determination at 0.91; however, smaller improvements were observed during cross-validation. Leave-one-out cross-validation for the linear model LUR model (RMSE = 1.44 ppb and a R2 = 0.64) and the regression kriging LUR model (RMSE = 1.34 ppb and a R2 = 0.69) were similar. Model performance remained stable when 10-fold cross-validation was performed with the regression kriging LUR model (regression kriging, R2 = 0.68 and RMSE = 1.36 ppb). The predicted air pollution levels ranged from 4.5 ppb to 25.6 ppb. This study demonstrates the ability of LUR modelling to perform well for local scale modelling in transportation dominated local urban environments.

中文翻译：

---

城市交通走廊中使用回归克里金法的邻域尺度二氧化氮土地利用回归模型

摘要 土地利用回归模型 (LUR) 将观测到的空气污染浓度与周围土地利用特征联系起来，以进行空气污染建模。这种技术在专注于全市空间尺度的城市景观中很常见。我们的研究在本地范围内测试了 LUR 建模的适用性，定义为一个社区内的多个空气监测器。研究区域是加拿大密西沙加城市交通走廊的 15.4 公里。2018 年夏季，在 112 个地点对二氧化氮 (NO2) 进行了采样，观测值范围为 5.8 ppb 至 19.65 ppb。线性回归 LUR 模型解释了该局部范围内 69% 的 NO2 浓度变化，估计的预测误差小于 1.61 ppb，这是通过三种交叉验证方法计算得出的。交通量，主要和次要道路长度是预测变量中的关键决定因素，而公园面积和到最近主要十字路口的距离是局部尺度模型中唯一具有负系数的变量。用回归克里金法扩展线性模型方法提高了模型的解释能力，决定系数为 0.91；然而，在交叉验证期间观察到较小的改进。线性模型 LUR 模型（RMSE = 1.44 ppb 和 a R2 = 0.64）和回归克里金 LUR 模型（RMSE = 1.34 ppb 和 R2 = 0.69）的留一法交叉验证是相似的。当使用回归克里金 LUR 模型（回归克里金，R2 = 0.68 和 RMSE = 1.36 ppb）进行 10 倍交叉验证时，模型性能保持稳定。预测的空气污染水平介于 4.5 ppb 至 25.6 ppb 之间。