随着工业化加速和经济快速发展,PM2.5引起的空气污染日益严重,对环境和人类健康造成严重影响。本研究采用Adaboost机器学习方法优化土地利用回归模型(LUR),利用2015年中国PM2.5监测数据及多源遥感数据,模拟中国PM2.5的空间分布,并评价...随着工业化加速和经济快速发展,PM2.5引起的空气污染日益严重,对环境和人类健康造成严重影响。本研究采用Adaboost机器学习方法优化土地利用回归模型(LUR),利用2015年中国PM2.5监测数据及多源遥感数据,模拟中国PM2.5的空间分布,并评价模型拟合效果。结果显示,Adaboost优化后的LUR模型拟合精度显著提高,R2从0.241提高至0.62 (春)、0.69 (夏)、0.60 (秋)、0.67 (冬)和0.65 (年),并通过SPSS软件识别出28个与PM2.5浓度相关的变量。研究发现,PM2.5浓度具有季节性变化,冬季最高,夏季最低,且存在明显的空间自相关性,表现为高–高集聚以及低–低集聚。本研究为PM2.5浓度精确预测提供了新方法,对公共健康保护和空气质量管理具有重要意义。With the acceleration of industrialization and rapid economic development, the air pollution caused by PM2.5 is becoming more and more serious, causing serious impacts on the environment and human health. In this study, the Adaboost machine learning method was used to optimize the land use regression (LUR) model to simulate the spatial distribution of PM2.5 in China by using the 2015 Chinese PM2.5 monitoring data and multi-source remote sensing data, and to evaluate the model fitting effect. The results showed that the fitting accuracy of LUR model optimized by Adaboost was significantly improved, R2 increased from 0.241 to 0.62 (spring), 0.69 (summer), 0.60 (autumn), 0.67 (winter) and 0.65 (year). 28 variables related to PM2.5 concentration were identified by SPSS software. It was found that PM2.5 concentration has seasonal variations, with the highest in winter and the lowest in summer, and there is an obvious spatial autocorrelation, which is manifested as high-high concentration as well as low-low concentration. This study provides a new method for accurate prediction of PM2.5 concentration, which is important for public health protection and air quality management.展开更多
文摘随着工业化加速和经济快速发展,PM2.5引起的空气污染日益严重,对环境和人类健康造成严重影响。本研究采用Adaboost机器学习方法优化土地利用回归模型(LUR),利用2015年中国PM2.5监测数据及多源遥感数据,模拟中国PM2.5的空间分布,并评价模型拟合效果。结果显示,Adaboost优化后的LUR模型拟合精度显著提高,R2从0.241提高至0.62 (春)、0.69 (夏)、0.60 (秋)、0.67 (冬)和0.65 (年),并通过SPSS软件识别出28个与PM2.5浓度相关的变量。研究发现,PM2.5浓度具有季节性变化,冬季最高,夏季最低,且存在明显的空间自相关性,表现为高–高集聚以及低–低集聚。本研究为PM2.5浓度精确预测提供了新方法,对公共健康保护和空气质量管理具有重要意义。With the acceleration of industrialization and rapid economic development, the air pollution caused by PM2.5 is becoming more and more serious, causing serious impacts on the environment and human health. In this study, the Adaboost machine learning method was used to optimize the land use regression (LUR) model to simulate the spatial distribution of PM2.5 in China by using the 2015 Chinese PM2.5 monitoring data and multi-source remote sensing data, and to evaluate the model fitting effect. The results showed that the fitting accuracy of LUR model optimized by Adaboost was significantly improved, R2 increased from 0.241 to 0.62 (spring), 0.69 (summer), 0.60 (autumn), 0.67 (winter) and 0.65 (year). 28 variables related to PM2.5 concentration were identified by SPSS software. It was found that PM2.5 concentration has seasonal variations, with the highest in winter and the lowest in summer, and there is an obvious spatial autocorrelation, which is manifested as high-high concentration as well as low-low concentration. This study provides a new method for accurate prediction of PM2.5 concentration, which is important for public health protection and air quality management.