Spatial And Temporal Variation Characteristics Of Urban Air Pollution And Air Quality Prediction

In recent years,China has witnessed sustained and rapid development,accelerated urbanization and industrialization,and substantially increased energy consumption.At the same time,it brings a variety of urban air pollution problems to Chinese cities,which makes the severity of air pollution in China remain high in the world.It has brought serious health problems to urban residents,such as cardiovascular and respiratory diseases.Many scholars have studied the spatial and temporal characteristics of urban air pollutant concentration,and found that the air pollution inside the city is often related to the natural conditions,economic development,traffic conditions and population density of the area.In order to better measure the heterogeneity of air pollutants in urban areas and provide basis for epidemiological research,land use regression model?LUR?was proposed.The traditional LUR model only considers the effects of fixed factors,such as land use type,road density and population density,on pollutants.Moreover,the model is based on multiple linear regression,and can not capture the non-linear change characteristics.In recent years,scholars have tried to introduce time factors,such as adding virtual variables representing seasons,months and dates,or constructing independent prediction models for each time period.But such research is rare in China.At the same time,machine learning has attracted the attention of scholars for its excellent performance in non-linear prediction.Some scholars began to introduce machine learning algorithm into land use regression model,and compare its effect with traditional methods.However,up to now,the number of such studies is very small,which shows its great research potential.Therefore,this paper hopes to make an attempt in this field by using the data of Shanghai,China.The virtual variables and time-varying meteorological data representing months are introduced into LUR model,and two popular machine learning algorithms are applied.Before the model was constructed,the spatial and temporal variation characteristics of air pollution in Shanghai were analyzed.Taking NO₂ and PM_2.5 as examples,it is found that the concentration of pollutants varies significantly with month and season,and has a trend of high in the west and low in the east and high in the north and low in the south.This characteristic is related to the geographical location of Shanghai.Clean sea breeze in the southeast direction can promote the evacuation of pollutants in the city,while pollutants transmitted from other provinces in the northwest inland direction will lead to the deterioration of air quality in the northwest of Shanghai.Therefore,factors including geographical location,meteorological factors and land use types were selected to explore their impact on pollutant concentration.After calculating Spearman correlation coefficient,the factors with high correlation with pollutant concentration were selected to enter the LUR model.The most important 20 factors were selected to enter the random forest and Addaboost model after the factor importance was ranked automatically by random forest.By comparing R² of the three models,it can be concluded that stochastic forest is better than the other two models for the prediction of NO₂ concentration,and traditional LUR model has advantages for PM_2.5 prediction.The difference in model performance may be due to over-fitting or artificial selection of influencing factors.Therefore,this should be explored in future research.