基于GWR的中国地级城市SO2年均质量浓度模拟

中国城市空气污染问题已经引起广泛关注。目前相关研究很多，但是以空间位置为拟合参数，对空气质量进行回归模拟的研究较少。以2010年中国地级以上城市SO2年均质量浓度为因变量，分别应用普通线性回归和地理加权回归（GWR）模型模拟SO2年均质量浓度，其中地理加权回归方法考虑了空间位置的影响并以此作为回归参数。回归的自变量指标体系包括气象要素（多年平均温度、光照、降水）、植被覆盖（NDVI）、地形要素（坡度、坡向、起伏度）、人为因素（GDP、能源消费）几个方面。由于各指标之间存在较强的相关性，用主成分分析方法计算得到温度、日照、降水、NDVI表征的气象植被综合指标，高程、坡度、起伏度表征的地形综合指标，和GDP、能源消费表征的人为因素综合指标。用3个综合指标值作为自变量进行回归模拟。普通回归结果较差，其r^2为0.11，矫正的r^2为0.10；GWR模型模拟结果相对较好，其拟合优度显著提高，r^2为0.66，矫正的r^2为0.47。因此，地理加权回归适合进行此类拟合，普通线性回归不适合。通过对比地理加权回归模拟的各个城市的拟合优度，发现年均质量浓度数值较高的地区拟合效果较差，这些地区主要集中在中国华北和南部部分地区。与基于机理的模型相比，GWR 模型和其各具优缺点，GWR 的优势主要表现在数据及其格式化要求低，计算机软硬件条件要求低，运算速度快等。

Simulation of Annual Average SO2 Concentration of the Prefecture-level Cities in China Based on GWR Mode

The problems of city air pollution have attracted worldwide attention. There’re various kinds of researches on air pollution, while very few of them are on the air quality regression considering the space location as the fitting parameter. This research respectively applies ordinary linear regression and geographically weighted regression model （GWR） to simulate the annual average SO2 concentration of the prefecture-level cities in 2010 in China, with annual average SO2 concentration as the dependent variable. The effect of spatial location is considered and taken as a regression parameter in the GWR. The indicator system of independent variables in the research includes meteorological factors （annual average temperature, sunlight, rainfall）, vegetation cover （NDVI）, topography （slope, slope aspect e, relief） and human factors （GDP, energy consumption）. As there is a strong correlation among the indicators, the principal component analysis method is adopted to calculate these comprehensive indexes： the meteorological ＆amp;vegetation index represented by temperature, sunlight, precipitation and NDVI;the topographic index represented by the elevation, slope and relief;and the human factors index represented by GDP and energy consumption. The regression simulation is conducted with these three comprehensive indexes as independent variables. Compared with the ordinary regression model, whose r^2 is 0.11 and corrected r^2 is 0.10, the simulation result of GWR model is better with much improved fitting. Its r^2 is 0.66, and corrected r^2 is 0.47. Therefore, geographically weighted regression is suitable for this kind of fitting, while the ordinary linear regression is not. By comparing the fitting in each city, we found the cities with higher annual average SO2 concentration had poor fitting effects, which were mainly concentrated in North China and South China. Compared with the models based on the mechanism, the GWR model has its own advantages, such as lower requirements of data and formatt