Application of an ensemble-trained source apportionment approach at a site impacted by multiple point sources

High-spatiotemporal-resolution mapping of PM2.5 traffic source impacts integrating machine learning and source-specific multipollutant indicator

Traffic sources are a major contributor to fine particulate matter (PM2.5) pollution, with their emissions and diffusion exhibiting complex spatiotemporal patterns. Receptor models have limitations in estimating high-resolution source contributions due to insufficient observation networks of PM2.5 compositions. This study developed a source apportionment method that integrates machine learning and emission-based integrated mobile source indicator (IMSI) to rapidly and accurately estimate PM2.5 traffic source impacts with high spatiotemporal resolution in the Beijing-Tianjin-Hebei region. Firstly, we utilized multisource data and developed various machine learning models to optimize the traffic-related pollutant concentration fields simulated by a chemical transport model. Results demonstrated that the Extreme Gradient Boosting (XGBoost) model exhibited excellent prediction accuracy of nitrogen oxide (NO2), carbon oxide (CO), and elemental carbon (EC), with the cross-validated R values increasing to 0.87-0.92 and error indices decreasing by 50-67%. Furthermore, we estimated and predicted daily mappings of PM2.5 traffic source impacts using the IMSI method based on optimized concentration fields, which improved spatially resolved source contributions to PM2.5. Our findings reveal that PM2.5 traffic source impacts display significant spatial heterogeneity, and these hotspots can be precisely identified during the pollution processes with sharp changes. The evaluation results indicated that there is a good correlation (R of 0.79) between PM2.5 traffic source impacts by IMSI method and traffic source contributions apportioned by a receptor model at Beijing site. Our study provides deeper insights of estimating the spatiotemporal distribution of PM2.5 source-specific impacts especially in regions without PM2.5 compositions, which can provide more complete and timely guidance to implement precise air pollution management strategies.

An explainable integrated optimization methodology for source apportionment of ambient particulate matter components

Source apportionment of fine particulate matter (PM_2.5) components is crucial for air pollution control. Prediction accuracies by the chemical transport model (CTM) significantly affect source apportionment results. Many efforts have been made to improve source apportionment results based on the CTM using mathematical algorithms, but the reasons for uncertainties in source apportionment results are less concerned. Here, an integrated optimization methodology is developed to quantify deviations from emission inventory and chemical mechanism in the model for improving prediction and source apportionment accuracies. Emission deviations of primary aerosols and gaseous pollutants are firstly calculated by an optimization algorithm with observation and receptor model constraints. Emission inventory is then adjusted for a new CTM simulation. Deviations from chemical mechanism for secondary conversions are evaluated by biases between observations and new predictions. Source apportionment results are adjusted according to both emission and chemical mechanism deviations. A winter month in 2016 at the Qingpu supersite in eastern China is selected as a case study. Results show that our integrated optimization methodology can successfully adjust emissions to pull original predictions towards observations. Total deviations of emissions for elemental carbon, organic carbon, primary sulfate, primary nitrate, primary ammonium, sulfur dioxide (SO₂), nitrogen oxides (NO_x) and ammonia (NH₃) are estimated +59.6%, +95.9%, +72.9%, +82.2%, +75.9%, -6.4%, +67.6% and -17.6%, respectively. Also, major directions of deviations from chemical mechanisms can be captured. Deviations from SO₂ to secondary sulfate, nitrogen dioxide (NO₂) to secondary nitrate and NH₃ to secondary ammonium conversions are estimated -77.3%, +27.1% and -38.8%, respectively. Consequently, source apportionment results are significantly improved. This developed methodology provides an efficient way to quantify deviations from emissions and chemical mechanisms to improve source apportionment for air pollution management.

Identification of Soil Heavy Metal Sources in a Large-Scale Area Affected by Industry

Heavy metals (HMs) in soil are some of the most serious pollutants due to their toxicity and nonbiodegradability. Especially across large-scale areas affected by industry, the complexity of pollution sources has attracted extensive attention. In this study, an approach based on zoning to analyze the sources of heavy metals in soil was proposed. Qualitative identification of pollution sources and quantification of their contributions to heavy metals in soil are key approaches in the prevention and control of heavy metal pollution. The concentrations of five HMs (Cd, Hg, As, Pb and Cr) in the surface soil of the Chenzhou industrial impact area were the research objects. Multiple methods were used for source identification, including positive matrix factorization (PMF) analysis combined with multiple other analyses, including random forest modeling, the geo-accumulation index method and hot spot analysis. The results showed that the average concentrations of the five heavy metals were 9.46, 2.36, 2.22, 3.27 and 1.05 times the background values in Hunan soil, respectively. Cd was associated with moderately to strongly polluted conditions, Hg, As and Pb were associated with unpolluted to moderately polluted conditions and Cr was associated with practically unpolluted conditions. The mining industry was the most significant anthropogenic factor affecting the content of Cd, Pb and As in the whole area, with contribution rates of 87.7%, 88.5% and 62.5%, respectively, and the main influence area was within 5 km from the mining site. In addition, we conducted hot spot analysis on key polluting enterprises and identified hot spots, cold spots, and areas insignificantly affected by enterprises, used this information as the basis for zoning treatment and discussed the sources of heavy metals in the three subregions. The results showed that Cd originated mainly from agricultural activities, with a contribution rate of 63.6%, in zone 3. As originated mainly from sewage irrigation, with a contribution rate of 65.0%, in zone 2, and the main influence area was within 800 m from the river. This element originated mainly from soil parent materials, with a contribution rate of 69.7%, in zone 3. Pb mainly originated from traffic emissions, with a contribution rate of 72.8%, in zone 3, and the main influence area was within 500 m from the traffic trunk line. Hg was mainly derived from soil parent materials with a contribution rate of 92.1% in zone 1, from agricultural activities with a contribution rate of 77.5% in zone 2, and from a mixture of natural and agricultural sources with a contribution rate of 74.2% in zone 3. Cr was mainly derived from the soil parent materials in the whole area, with a contribution rate of 90.7%. The study showed that in large-scale industrial influence areas, the results of heavy metal source analysis can become more accurate and detailed by incorporating regional treatment, and more reasonable suggestions can be provided for regional enterprise management and soil pollution control decision making.

A partition computing-based positive matrix factorization (PC-PMF) approach for the source apportionment of agricultural soil heavy metal contents and associated health risks

Apportion soil heavy metal sources across large-scale regions is a challenging task. The present study developed a modified receptor model to estimate the contributions of various sources to soil heavy metals and the associated health risks at a large scale. A positive matrix factorization model based on a partition computing approach was employed; the entire study area was divided into several zones for the source apportionment and then calculated together, termed partition computing-PMF (PC-PMF). The agricultural soil in Tianjin, China, was chosen for the case study. The PC-PMF results showed that irrigation, atmospheric deposition and sludge application were the main anthropogenic sources, with contributions of 26.60 %, 19.56 % and 2.86 %, respectively. We subsequently combined PC-PMF with a human health risk assessment model (HHRA) to obtain the human health risk of every source category. The natural background was regarded as a major factor influencing human health in the study area, with contributions of 38.03 % for the noncarcinogenic risk and 28.68 % for the carcinogenic risk. The results indicated that PC-PMF performed better at the source apportionment of soil heavy metals than PMF. This study provides a good example of how the spatial variability can be utilized to reduce the uncertainty in source apportionment.

The mechanisms of PM2.5 and its main components penetrate into HUVEC cells and effects on cell organelles

Atmospheric particulate matter (PM_2.5) is associated with the morbidity and mortality of cardiovascular diseases. However, whether PM_2.5 penetrates into the cells and the potential mechanisms are unknown. Hence, the study firstly indicated that PM_2.5 could penetrate into the HUVEC cells, and phagocytosis, micropinocytosis, caveolin as well as clathrin mediated the internalization of PM_2.5 into HUVEC cells. Particularly, the components of PM_2.5-Metal, PAHs and WSC could enter into HUVEC cells mainly via the micropinocytosis, clathrin and caveolin mediated endocytosis, respectively. The current data of environmental assessments indicated that PM_2.5-Metal were extremely harmful to the ecological environment and human health. Moreover, accompanying with mitochondrial fusion gene Mfn1 was increased and fission genes Opa1 and Drp1 were decreased, and the lysosome related genes LAMP2 and LAMP3 were decreased, the phenomenon that the morphology of mitochondrial and lysosome injured was observed in HUVEC cells treated with PM_2.5 and/or PM_2.5-Metal. These data suggest that PM_2.5 and its main components depend on different endocytosis penetrate into HUVEC cells and cause the mitochondrial and lysosomal damages. Thereby, our study provides the potential mechanism of haze particles penetration into HUVEC cells and damage to organelles.

Development of PM2.5 Source Profiles Using a Hybrid Chemical Transport-Receptor Modeling Approach

Laboratory-based or in situ PM_2.5 source profiles may not represent the pollutant composition for the sources in a different study location due to spatially and temporally varying characteristics, such as fuel or crustal element composition, or due to differences in emissions behavior under ambient versus laboratory conditions. In this work, PM_2.5 source profiles were estimated for 20 sources using a novel optimization approach that incorporates observed concentrations with source impacts from a chemical transport model (CTM) to capture local pollutant characteristics. Nonlinear optimization was used to minimize the error between source profiles, CTM source impacts, and observations. In a 2006 U.S. application, spatial and seasonal variability was seen for coal combustion, dust, fires, metals processing, and other source profiles when compared to the reference profiles, with variability in species fractions over 400% (calcium in dust) compared to mean contributions of the same species. Revised profiles improved the spatial and temporal bias in modeled concentrations of several trace metal species, including Na, Al, Ca, Mn, Cu, As, Se, Br, and Pb. In an application of the CMB-iteration model for two U.S. cities, revised profiles estimated higher biomass burning and dust impacts for summer compared with previous studies. Source profile optimization can be useful for source apportionment studies that have limited availability of source profile data for the location of interest.

Review of receptor-based source apportionment research of fine particulate matter and its challenges in China

As the key for haze control, atmospheric fine particulate matter with aerodynamic diameter <2.5μm (or PM_2.5) is of great concern lately in China. It is closely linked to fast pace of urbanization, industrialization and economic development, especially in eastern China. A good understanding of its sources is required for effective pollution abatement. Receptor models are one of the major methods for source apportionment used in China. The major objective of this study is to understand sources that contribute to fine particulate matter in China and key challenges in this area. Spatial distribution of fine particulate matter concentration, chemical composition and dominant sources in North and South China are summarized. Based on chemical speciation results from 31 cities and source apportionment results from 21 cities, it is found that secondary sources and traffic emission have higher contribution in South China while the percentage of coal combustion, dust and biomass burning to total PM_2.5 are higher in North China. Source profiles established in China from 44 cities and areas are also summarized as references for future source apportionment studies. Suggestions for future research are also provided including methods for evaluating source apportionment results, ways for integrating multiple source apportionment methods, the need for standardizing protocols and developing tools for high-time resolution source apportionment.

pH of Aerosols in a Polluted Atmosphere: Source Contributions to Highly Acidic Aerosol

Acidity (pH) plays a key role in the physical and chemical behavior of PM_2.5. However, understanding of how specific PM sources impact aerosol pH is rarely considered. Performing source apportionment of PM_2.5 allows a unique link of sources pH of aerosol from the polluted city. Hourly water-soluble (WS) ions of PM_2.5 were measured online from December 25th, 2014 to June 19th, 2015 in a northern city in China. Five sources were resolved including secondary nitrate (41%), secondary sulfate (26%), coal combustion (14%), mineral dust (11%), and vehicle exhaust (9%). The influence of source contributions to pH was estimated by ISORROPIA-II. The lowest aerosol pH levels were found at low WS-ion levels and then increased with increasing total ion levels, until high ion levels occur, at which point the aerosol becomes more acidic as both sulfate and nitrate increase. Ammonium levels increased nearly linearly with sulfate and nitrate until approximately 20 μg m^-3, supporting that the ammonium in the aerosol was more limited by thermodynamics than source limitations, and aerosol pH responded more to the contributions of sources such as dust than levels of sulfate. Commonly used pH indicator ratios were not indicative of the pH estimated using the thermodynamic model.

Associations between Source-Specific Fine Particulate Matter and Emergency Department Visits for Respiratory Disease in Four U.S. Cities

BACKGROUND

Short-term exposure to ambient fine particulate matter (PM2.5) concentrations has been associated with increased mortality and morbidity. Determining which sources of PM2.5 are most toxic can help guide targeted reduction of PM2.5. However, conducting multicity epidemiologic studies of sources is difficult because source-specific PM2.5 is not directly measured, and source chemical compositions can vary between cities.

OBJECTIVES

We determined how the chemical composition of primary ambient PM2.5 sources varies across cities. We estimated associations between source-specific PM2.5 and respiratory disease emergency department (ED) visits and examined between-city heterogeneity in estimated associations.

METHODS

We used source apportionment to estimate daily concentrations of primary source-specific PM2.5 for four U.S. cities. For sources with similar chemical compositions between cities, we applied Poisson time-series regression models to estimate associations between source-specific PM2.5 and respiratory disease ED visits.

RESULTS

We found that PM2.5 from biomass burning, diesel vehicle, gasoline vehicle, and dust sources was similar in chemical composition between cities, but PM2.5 from coal combustion and metal sources varied across cities. We found some evidence of positive associations of respiratory disease ED visits with biomass burning PM2.5; associations with diesel and gasoline PM2.5 were frequently imprecise or consistent with the null. We found little evidence of associations with dust PM2.5.

CONCLUSIONS

We introduced an approach for comparing the chemical compositions of PM2.5 sources across cities and conducted one of the first multicity studies of source-specific PM2.5 and ED visits. Across four U.S. cities, among the primary PM2.5 sources assessed, biomass burning PM2.5 was most strongly associated with respiratory health. Citation: Krall JR, Mulholland JA, Russell AG, Balachandran S, Winquist A, Tolbert PE, Waller LA, Sarnat SE. 2017. Associations between source-specific fine particulate matter and emergency department visits for respiratory disease in four U.S. cities. Environ Health Perspect 125:97-103; http://dx.doi.org/10.1289/EHP271.

Review of receptor modeling methods for source apportionment

Efforts have been made to relate measured concentrations of airborne constituents to their origins for more than 50 years. During this time interval, there have been developments in the measurement technology to gather highly time-resolved, detailed chemical compositional data. Similarly, the improvements in computers have permitted a parallel development of data analysis tools that permit the extraction of information from these data. There is now a substantial capability to provide useful insights into the sources of pollutants and their atmospheric processing that can help inform air quality management options. Efforts have been made to combine receptor and chemical transport models to provide improved apportionments. Tools are available to utilize limited numbers of known profiles with the ambient data to obtain more accurate apportionments for targeted sources. In addition, tools are in place to allow more advanced models to be fitted to the data based on conceptual models of the nature of the sources and the sampling/analytical approach. Each of the approaches has its strengths and weaknesses. However, the field as a whole suffers from a lack of measurements of source emission compositions. There has not been an active effort to develop source profiles for stationary sources for a long time, and with many significant sources built in developing countries, the lack of local profiles is a serious problem in effective source apportionment. The field is now relatively mature in terms of its methods and its ability to adapt to new measurement technologies, so that we can be assured of a high likelihood of extracting the maximal information from the collected data.

IMPLICATIONS

Efforts have been made over the past 50 years to use air quality data to estimate the influence of air pollution sources. These methods are now relatively mature and many are readily accessible through publically available software. This review examines the development of receptor models and the current state of the art in extracting source identification and apportionments from ambient air quality data.

Fine particulate matter components and emergency department visits for cardiovascular and respiratory diseases in the St. Louis, Missouri-Illinois, metropolitan area

BACKGROUND

Given that fine particulate matter (≤ 2.5 μm; PM2.5) is a mixture of multiple components, it has been of high interest to identify its specific health-relevant physical and/or chemical features.

OBJECTIVES

We conducted a time-series study of PM2.5 and cardiorespiratory emergency department (ED) visits in the St. Louis, Missouri-Illinois metropolitan area, using 2 years of daily PM2.5 and PM2.5 component measurements (including ions, carbon, particle-phase organic compounds, and elements) made at the St. Louis-Midwest Supersite, a monitoring site of the U.S. Environmental Protection Agency Supersites ambient air monitoring research program.

METHODS

Using Poisson generalized linear models, we assessed short-term associations between daily cardiorespiratory ED visit counts and daily levels of 24 selected pollutants. Associations were estimated for interquartile range changes in each pollutant. To allow comparison of relationships among multiple pollutants and outcomes with potentially different lag structures, we used 3-day unconstrained distributed lag models controlling for time trends and meteorology.

RESULTS

Considering results of our primary models, as well as sensitivity analyses and models assessing co-pollutant confounding, we observed robust associations of cardiovascular disease visits with 17α(H),21β(H)-hopane and congestive heart failure visits with elemental carbon. We also observed a robust association of respiratory disease visits with ozone. For asthma/wheeze, associations were strongest with ozone and nitrogen dioxide; observed associations of asthma/wheeze with PM2.5 and its components were attenuated in two-pollutant models with these gases. Differential measurement error due to differential patterns of spatiotemporal variability may have influenced patterns of observed associations across pollutants.

CONCLUSIONS

Our findings add to the growing field examining the health effects of PM2.5 components. Combustion-related components of the pollutant mix showed particularly strong associations with cardiorespiratory ED visit outcomes.

Coupling chemical transport model source attributions with positive matrix factorization: application to two IMPROVE sites impacted by wildfires

Source contributions to total fine particle carbon predicted by a chemical transport model (CTM) were incorporated into the positive matrix factorization (PMF) receptor model to form a receptor-oriented hybrid model. The level of influence of the CTM versus traditional PMF was varied using a weighting parameter applied to an object function as implemented in the Multilinear Engine (ME-2). The methodology provides the ability to separate features that would not be identified using PMF alone, without sacrificing fit to observations. The hybrid model was applied to IMPROVE data taken from 2006 through 2008 at Monture and Sula Peak, Montana. It was able to separately identify major contributions of total carbon (TC) from wildfires and minor contributions from biogenic sources. The predictions of TC had a lower cross-validated RMSE than those from either PMF or CTM alone. Two unconstrained, minor features were identified at each site, a soil derived feature with elevated summer impacts and a feature enriched in sulfate and nitrate with significant, but sporadic contributions across the sampling period. The respective mean TC contributions from wildfires, biogenic emissions, and other sources were 1.18, 0.12, and 0.12 ugC/m(3) at Monture and 1.60, 0.44, and 0.06 ugC/m(3) at Sula Peak.

The impact of source contribution uncertainty on the effects of source-specific PM2.5 on hospital admissions: a case study in Boston, MA

Epidemiologic studies of particulate sources and adverse health do not account for the uncertainty in the source contribution estimates. Our goal was to assess the impact of uncertainty on the effect estimates of particulate sources on emergency cardiovascular (CVD) admissions. We examined the effects of PM2.5 sources, identified by positive matrix factorization (PMF) and absolute principle component analysis (APCA), on emergency CVD hospital admissions among Medicare enrollees in Boston, MA, during 2003-2010, given stronger associations for this period. We propagated uncertainty in source contributions using a block bootstrap procedure. We further estimated average across-methods source-specific effect estimates using bootstrap samples. We estimated contributions for regional, mobile, crustal, residual oil combustion, road dust, and sea salt sources. Accounting for uncertainty, same-day exposures to regional pollution were associated with an across-methods average effect of 2.00% (0.18, 3.78%) increase in the rate of CVD admissions. Weekly residual oil exposures resulted in an average 2.12% (0.19, 4.22%) increase. Same-day and 2-day exposures to mobile-related PM2.5 were also associated with increased admissions. Confidence intervals when accounting for the uncertainty were wider than otherwise. Agreement in PMF and APCA results was stronger when uncertainty was considered in health models. Accounting for uncertainty in source contributions leads to more stable effect estimates across methods and potentially to fewer spurious significant associations.

Bayesian-based ensemble source apportionment of PM2.5

A Bayesian source apportionment (SA) method is developed to provide source impact estimates and associated uncertainties. Bayesian-based ensemble averaging of multiple models provides new source profiles for use in a chemical mass balance (CMB) SA of fine particulate matter (PM2.5). The approach estimates source impacts and their uncertainties by using a short-term application of four individual SA methods: three receptor-based models and one chemical transport model. The method is used to estimate two seasonal distributions of source profiles that are used in SA for a long-term PM2.5 data set. For each day in a long-term PM2.5 data set, 10 source profiles are sampled from these distributions and used in a CMB application, resulting in 10 SA results for each day. This formulation results in a distribution of daily source impacts rather than a single value. The average and standard deviation of the distribution are used as the final estimate of source impact and a measure of uncertainty, respectively. The Bayesian-based source impacts for biomass burning correlate better with observed levoglucosan (R(2) = 0.66) and water-soluble potassium (R(2) = 0.63) than source impacts estimated using more traditional methods and more closely agrees with observed total mass. The Bayesian approach also captures the expected seasonal variation of biomass burning and secondary impacts and results in fewer days with sources having zero impact. Sensitivity analysis found that using non-informative prior weighting performed better than using weighting based on method-derived uncertainties. This approach can be applied to long-term data sets from speciation network sites of the United States Environmental Protection Agency (U.S. EPA). In addition to providing results that are more consistent with independent observations and known emission sources being present, the distributions of source impacts can be used in epidemiologic analyses to estimate uncertainties associated with the SA results.