With the development of industrialization and urbanization,air pollution is becoming more and more serious,and PM2.5 fine particles are the main source of atmospheric pathogens.Only by mastering the accurate temporal and spatial distribution data of PM2.5 pollutants can we deeply understand its harm to human health.However,many methods cannot meet the needs due to high costs and complex personnel operations,so the Land Use Regression(LUR)model is widely used in the study of spatial and temporal distribution of atmospheric pollutants.However,there are still two shortcomings in this method.Firstly,the nonlinear relationship between the relevant factors is ignored,which leads to low simulation accuracy and makes the predicted pollutant distribution inconsistent with the actual distribution.Secondly,there is a lack of research on the spatial correlation between land use/conversion and PM2.5.And it is impossible to determine the land use/conversion types that have the greatest impact on PM2.5,resulting in unreasonable land planning in cities.In view of this,this paper studies the temporal and spatial distribution of PM2.5 based on LUR model and its correlation with land use,and proposes a method to improve the accuracy of the model by machine learning algorithm and to study the correlation between PM2.5 and land data by GIS spatial analysis.The research contents and results are as follows:(1)Improved LUR model based on machine learning algorithm.Taking the PM2.5 data of77 stations in Liaoning Province from 2015 to 2019 as the dependent variable.The land use,meteorology,population and road data were selected as the independent variables.The bivariate correlation method and synthetic minority oversampling technology(SMOTE)were used to screen the data and unbalanced pretreatment.Four models were constructed and compared:LUR,LUR-RF(random forest),LUR-GBDT(gradient boosting tree)and LUR-XGBoost(limit gradient boosting).It is concluded that LUR-XGBoost model has the best prediction effect,and R2 is 34.4%higher than LUR model.LUR-RF model followed,the traditional LUR model is the worst.(2)Exploring the temporal and spatial characteristics of PM2.5based on the improved LUR model.The improved LUR model combined with Kriging interpolation method was used to realize the spatialization of PM2.5 data,and the distribution characteristics of PM2.5under different time and space partitions in Liaoning Province was explored.The global/local autocorrelation method was used to extract and analyze the distribution characteristics of PM2.5 concentration.The results showed that the distribution of PM2.5 in Liaoning Province showed a pattern of’high in the north and low in the south’and’high in the west and low in the winter’.The global autocorrelation difference was strengthened year by year.The local autocorrelation was clustered,the size relationship was L-L>H-H>L-H>H-L(H-high,L-low).(3)Study on spatial and temporal distribution characteristics of PM2.5 and land use correlation.Combined with the spatial and temporal distribution results of the improved LUR model and the land use data map,the spatial analysis method and correlation coefficient method were used to quantitatively analyze the relationship between PM2.5 and land.The results showed that there was a significant positive correlation between cultivated land and PM2.5,where the PM2.5 concentration increased with the cultivated area;and a significant negative correlation between forest land,grassland and PM2.5,where the PM2.5 concentration decreased with the forest and grass land.However,water and PM2.5 were different due to the particularity of the study area,and the conversion between different land types had a certain impact on the concentration of PM2.5.Among them,the conversion of cultivated land and forest land to residents and construction land and the conversion of forest land to cultivated land were the most related with the increase of PM2.5. |