Satellite-derived data and ground-based measurements relationships for assessing local and regional distributions of PM2.5: model development and applications
Export citation
Abstract
Estimating concentrations of ground-level PM2.5 (particulate matter with an aerodynamic diameter less than or equal to 2.5 microns) from satellite-derived Aerosol Optical Depth (AOD) through statistical models is a promising method to evaluate the spatial and temporal distribution of PM2.5. Although PM concentrations are most accurately measured using ground-based instruments, the spatial coverage is often too sparse to determine local and regional variations in PM2.5. AOD satellite data offers the opportunity to overcome the spatial limitation of ground-based measurements. Combining ground-based measurements with satellite and reanalysis data can be a robust tool to assess air pollution models. However, estimating PM2.5 surface concentrations from AOD satellite data is challenging, since multiple factors can affect the relationship between the total-column of AOD and the surface-concentration of PM2.5.
This study aims to establish a relationship between AOD satellite data and ground-based data that will allow designing relational models for the study of local PM2.5 pollution and the regional PM2.5 distributions in northeastern Mexico (NEM). First, an Ensemble Multiple Linear Regression Model (MLR) and a Neural Network Model (NN) were developed to estimate the relationship between the AOD and ground-concentrations of PM2.5 within the MMA. The best performance of the models was obtained using a daily scheme, an AOD at 550 µm from the MYD04_3k product in combination with Temperature, Relative Humidity, Wind Speed and Wind Direction ground-based data. For the MLR developed, a correlation coefficient of R ~ 0.57 and mean percentage error ~ -6% were obtained. The NN showed a better performance than the MLR, with a correlation coefficient of R ~ 0.73 and mean percentage error ~ –3%. The results obtained confirmed that satellite-derived AOD in combination with meteorological fields may allow to estimate PM2.5 local distributions.
Then, both the observed and model-estimated daily PM2.5 concentrations were classified according to the categories of the Mexican Air Quality Index for PM2.5
(PM2.5 MX-AQI). The observed PM2.5 MX-AQI revealed a distinct seasonal variation: a decreasing trend was observed from spring to summer, but then concentrations increased from fall to winter, indicating that air quality across the region is worse in winter than in summer. The developed NN showed 90.4% overall accuracy, 2.5% overestimation and
7.2% underestimation. The results obtained confirmed that the developed model can be used to estimate PM2.5 MX-AQI with high accuracy.
Finally, a method that uses neural networks (NN) was developed to combine five years of PM2.5 data recorded at local monitoring sites with MERRA-2 reanalysis data. Based on these data, monthly pollution maps were generated to analyze the spatial and temporal patterns of PM2.5 concentration fields. AOD, temperature, specific humidity, dust PM2.5, sea salt PM2.5, black carbon (BC), organic carbon (OC) and sulfate (SO42–) reanalysis data were identified as factors that significantly influence the regional distribution of aerosol. The NN estimated a PM2.5 monthly mean of 29.4 µg m–3, with the following performance: correlation coefficient R ~ 0.82, root mean square error = 8.4 µg m–3 and mean percentage
error = –2.7%. Significant spatial clustering of high PM2.5 was observed, with the highest PM2.5 levels located in the MMA, which is the major source to air pollution in this entire area. The estimated data indicates that PM2.5 are not distributed uniformly throughout the region, varying spatially as well as temporally, as the mixed mountainous-and-valley topography complicates dispersion of local emissions.
In conclusion, the magnitude of air pollution varies with seasons and regions, and correlates with meteorological factors. The methodology developed here can be used to identify new monitoring sites and can help face information gaps. These findings can also help epidemiologists to better understand the adverse health effects related to PM2.5, and serve as a stepping stone towards designing effective regional environmental planning and emissions control strategies.