Estimating PM_2.5 Concentration Using The INLA-SPDE Method In Henan Province

With the rapid development of China’s social economy and the continuous acceleration of urbanization,the problem of air pollution is becoming increasingly prominent.PM_2.5is the main pollutant of air pollution,causing serious negative impacts on public health.In recent years,China has experienced frequent occurrences of heavily polluted weather,and air quality issues have attracted unprecedented attention from society,becoming the focus of public opinion and academic research.At present,ground monitoring is the main source of PM_2.5concentrations data and the data foundation for PM_2.5exposure health risk assessment.However,PM_2.5monitoring stations in China are still sparse,and regional and urban-rural imbalances are evident.The limited ground monitoring data cannot effectively reflect the changes in PM_2.5concentrations at different scales,which has become a bottleneck affecting the accuracy of PM_2.5concentrations estimation and pollution exposure assessment,seriously restricting the research on PM_2.5pollution exposure risk.The spatiotemporal statistical modeling method has good spatial interpretation ability and unique advantages in depicting the spatiotemporal changes of geographical phenomena.Based on spatiotemporal statistical modeling methods,conducting spatiotemporal estimation and simulation research on PM_2.5concentrations,obtaining continuous spatiotemporal PM_2.5concentrations data,can make up for the shortcomings of existing PM_2.5monitoring data,and is of great significance for grasping the spatiotemporal distribution of PM_2.5concentrations at small spatiotemporal scales and accurately evaluating the long-term and short-term health effects of air pollution.This article is based on the monitoring data of 119 air quality monitoring stations in Henan Province and its surrounding areas from October 2019 to March 2020,as well as remote sensing inversion product data,to explore the spatiotemporal changes of PM_2.5concentrations in Henan Province.The effects of land use,altitude,road density,and meteorological factors on the spatiotemporal distribution of PM_2.5concentrations are analyzed.On this basis,geographic variables such as elevation,road density,and land use are integrated as spatial covariates,while meteorological variables such as precipitation,air pressure,relative humidity,temperature,and wind speed are integrated as spatiotemporal covariates.The nested Laplace approximation with stochastic partial differential equation（INLA-SPDE）method is used,Construct a spatiotemporal statistical model with spatial correlation and first-order autoregressive structure for estimating PM_2.5concentrations.Finally,the estimation accuracy and effectiveness of the model were evaluated through cross validation and comparison with the results of the Spatio Temporal Kriging（STK）method.The main research conclusions are summarized as follows:（1）The change in PM2.5 concentration shows a significant temporal variationFrom the daily mean time series of PM_2.5concentrations,from October 2019 to January 2020,PM_2.5concentrations showed a significant fluctuation and overall upward trend.Starting from February 2020,the fluctuation of PM_2.5concentrations decreased and showed an overall downward trend.The highest daily average concentrations of PM_2.5occurred in January 2020,and the lowest value occurred in February 2020.On a monthly scale,the variation of PM_2.5concentrations shows an inverted"U"shape,meaning that PM_2.5concentrations continues to rise rapidly from October to December,and after reaching a high value in January 2020,PM_2.5concentrations continues to decline.（2）The overall spatial distribution pattern of PM_2.5concentrations is high in the north and low in the southThe PM_2.5concentrations at the northern monitoring stations in the study area is relatively high,while the PM_2.5concentrations at the southern and southwestern stations is relatively low.The station with the highest PM_2.5concentrations is located in Anyang City,with a concentration of 104.94μg/m³.From the perspective of urban scale,there is a significant difference in PM_2.5concentrations between cities.Overall,cities in northern Henan such as Anyang,Puyang,and Hebi have higher concentrations of PM_2.5,with Anyang having the highest concentrations of 89.94μg/m³;The concentrations in Xinyang,Zhumadian,and other cities in the southern region of Henan are relatively low,with Xinyang having the lowest concentrations of 56.43μg/m³.（3）Meteorological factors,land use,and altitude have important impacts on the spatiotemporal changes of PM_2.5concentrationsThe results of the GTWR model indicate that the influence of various meteorological factors on PM_2.5concentrations is spatiotemporal non-stationary.For example,in time,the wind speed coefficient is mostly negative,indicating that it mainly has a negative impact on PM_2.5concentrations.The precipitation coefficient is negative from 1 to 10 weeks（before December 9,2019）,indicating that it mainly has a negative impact on PM_2.5concentrations during this period,while it is positive at other times,showing a positive impact.In space,rainfall and temperature have a negative impact on PM_2.5concentrations.The intensity of the negative impact of rainfall on PM_2.5concentrations is high in the north and low in the south,while the negative impact of temperature on PM_2.5gradually increases from north to south.The analysis results of geographical detectors show that altitude has the greatest impact on the spatial pattern of PM_2.5concentrations,followed by land use and road density.Among the interactions among various factors,altitude and land use have the greatest impact on the spatial pattern of PM_2.5concentrations.There are differences in the impact of various factors on PM_2.5concentrations among different subregions,and the formation of high pollution areas of PM_2.5in northern Henan Province is the result of the interaction of multiple factors in high-risk areas.（4）Estimation,validation,and result analysis of spatiotemporal PM_2.5concentrationsUsing the INLA-SPDE modeling method,integrating land use,altitude,road density,and meteorological covariates,and incorporating spatial correlation and first-order autoregressive structures,a spatiotemporal model was constructed to estimate PM_2.5concentrations,and daily estimates of PM_2.5concentrations were obtained for 196×203 3 km grids in the study area.The validation results indicate that the predicted values are in good agreement with the observed values（cross validation with ten fold R²=0.9407;root mean square error RMSE=10.7135μg/m³）,the actual coverage probability of the 95%confidence interval is 96.04%.Compared with the validation results of the STK model（R²==0.9003,RMSE=15.7972μg/m³）,INLA-SPDE results showed better accuracy and effectiveness.Especially,INLA-SPDE is more sensitive to values with higher pollution concentrations and can more accurately reflect the distribution of severely polluted areas.In practical applications,the INLA-SPDE model has more advantages in reflecting the spatial differences of PM_2.5concentrations in heavily polluted areas,thus having higher prediction accuracy and strong practical value and practical significance.The INLA-SPDE model accurately estimated the daily concentrations of PM_2.5in Henan Province,and the estimation results reflected the detailed spatial and spatiotemporal changes of PM_2.5concentrations,improving our understanding of the spatiotemporal changes of regional PM_2.5concentrations.This can provide reference for effectively developing measures to control and reduce PM_2.5pollution,and the data results can also be used to evaluate the health effects related to air pollution.