Land use regression models to estimate the annual and seasonal spatial variability of sulfur dioxide and particulate matter in Tehran, Iran.

The Middle Eastern city of Tehran, Iran has poor air quality compared with cities of similar size in Europe and North America. Spatial annual and seasonal patterns of SO2 and PM10 concentrations were estimated using land use regression (LUR) methods applied to data from 21 air quality monitoring stations. A systematic algorithm for LUR model building was developed to select variables based on (1) consistency with a priori assumptions about the assumed directions of the effects, (2) a p-value of <0.1 for each predictor, (3) improvements to the leave-one-out cross-validation (LOOCV) R(2), (4) a multicollinearity index called the variance inflation factor, and (5) a grouped (leave-25%-out) cross-validation (GCV) for final model. In addition, several new predictive variables and variable types were explored. The annual mean concentrations of SO2 and PM10 across the stations were 38 ppb and 100.8 μg/m(3), respectively. The R(2) values ranged from 0.69 to 0.84 for SO2 models and from 0.62 to 0.67 for PM10 models. The LOOCV and GCV R(2) values ranged, respectively, from 0.40 to 0.56 and 0.40 to 0.50 for the SO2 models; they were 0.48 to 0.57 and 0.50 to 0.55, respectively, for the PM10 models. There were clear differences between the SO2 and PM10 models, but the warmer and cooler season models were consistent with the annual models for both pollutants. Although there was limited similarity between the SO2 and PM10 predictive variables, measures of street density and proximity to airport or air cargo facilities were consistent across both pollutants. In 2010, the entire population of Tehran lived in areas where the World Health Organization guidelines for 24-hour mean SO2 (7 ppb) and annual average PM10 (20 μg/m(3)) were exceeded.