利用LUR模型模拟天津市大气污染物浓度的空间分布

针对传统监测方法无法满足对大气污染物空间分布高分辨率的要求,以Arcgis为平台,利用LUR模型模拟天津市PM10和NO2年均浓度的空间分布.选取的回归变量为1~4km半径缓冲区内的道路总长度、不同土地利用类型的面积、人口密度、风向指数及距海距离,选取3个监测点的监测数据对方程进行了验证.结果表明,对PM10年均浓度影响最大的因素是缓冲区为1km的道路总长度(R2为0.560),而对NO2年均浓度影响最大的因素是人口密度(R2为0.414).多元线性回归方程计算结果显示,PM10和NO2的R2分别达到0.946和0.691;如果考虑风向的影响,R2可分别提高到0.980和0.849.对天津市中心城区建立5km′5km网格嵌套,根据多元线性回归方程计算每个网格交点的污染物浓度模拟值.通过kriging插值得到2种污染物在天津市中心城区的空间模拟分布图.PM10年均浓度分布以研究区中心最高,向四周逐渐降低;NO2的年均浓度以研究区中心最低,向四周逐渐升高.模拟结果与实际情况相符.

Application of land use regression to simulate ambient air PM10 and NO2 concentration in Tianjin City

Land use regression (LUR) model was employed to simulate the spatial distribution of PM10 and NO2 based on environmental air quality monitoring data and Arcgis. LUR predictor variables included density of population, wind index, distance to sea shore, as well as 2 buffer indices: total length of road and areas of 5 different land use patterns within a buffer. Radii of buffer area were chosen of 1, 2, 3, 4 km, respectively. The PM10 and NO2 annual average data from 7 national ambient air quality monitoring sites were chosen as dependent variables, data from another 3 sites were chosen for LUR model validation. PM10 and NO2 were highly correlated with total length of road within a buffer of 1km (R2=0.560) and population density (R2=0.414), respectively. Top five predictor variables without wind index were chosen based on the correlation coefficients for PM10 and for NO2, respectively. The 2 multi linear regression equations were established based on above five predictor variables with wind index for PM10 (R2=0.980) and NO2 (R2=0.849), respectively. Grid was established by 5 km ′ 5 km in Tianjin central districts. Concentrations of PM10 and NO2 at intersection points were calculated using 2 equations, respectively. The spatial concentration distribution of PM10 and NO2 were interpolated by Kriging approach and showed: PM10 concentration is highest in center of the studying area, and decreased gradually from center to surrounding area; while NO2 concentration is lowest in center of the studying area, and increased gradually to surround area. The final models predicted well at study area.