Advancing Exposure Assessment of PM2.5 Using Satellite Remote Sensing: A Review

Epidemiological studies have reported the associations of adverse health outcomes with ambient particulate matter with aerodynamic diameter ≤2.5μm (PM2.5). While these studies have accumulated increasingly refined evidence on PM2.5-health associations, the needs for more advanced PM2.5 exposure models have also grown. For the last two decades, PM2.5 estimation approaches using satellite remote sensing have been developed and advanced, taking advantage of quantitative aerosol data (e.g., aerosol optical depth; AOD), the development of satellite instruments and data retrieval algorithms, and the application of statistical and machine learning techniques. Subsequently, the satellite-based PM2.5 concentrations have contributed to health effect studies by providing spatially resolved exposure estimates of ambient PM2.5 and thus reducing exposure misclassification. This article summarizes previous development and recent advancement of satellite-based PM2.5 exposure assessment in the context of satellite aerosol products and PM2.5 estimation methodologies. Furthermore, this article deals with enhanced satellite capabilities of generating the exposure estimates of PM2.5 composition and time-resolved PM2.5. Finally, the future directions of satellite-based exposure assessment are discussed based on research needs and the satellite remote sensing technology of addressing them.


INTRODUCTION
Numerous epidemiological studies have reported the associations of ambient particulate matter with aerodynamic diameter ≤2.5 μm (PM 2.5 ) with adverse health outcomes (Lepeule et al., 2012;Pope III et al., 2002;Dockery et al., 1993).The health effects of PM 2.5 include increased mortality and morbidity related to cardiovascular and respiratory functions, hospital admissions, neurological disorders, and adverse birth outcomes (Son et al., 2017;Kioumourtzoglou et al., 2016;Dominici et al., 2006).Epidemiological studies investigate PM 2.5 -health associations at two different temporal scales, short-term and long-term exposures and their associated acute and chronic health effects, respectively.In acute health effect studies, daily variation of PM 2.5 exposure is corresponded to that of health outcomes.For chronic health effect studies, on the other hand, average PM 2.5 exposures for months to years (typically annual) are related to health out comes for the same periods.
Evidence on ambient PM 2.5 and health associations has led to establish or tighten ambient air quality standards for PM 2.5 .The Harvard Six Cities study, which found an association between PM 2.5 air pollution and excess mortality (Dockery et al., 1993), contributed to ambient PM 2.5 standards (65 μg/m 3 and 15 μg/m 3 for 24-hour and annual standards, respectively) by the U.S. Environmental Protection Agency (EPA) in 1997 (U.S. EPA, 2018).Since the mandate of PM 2.5 regulations, the PM 2.5 standards have been sequentially lowered to 35 μg/m 3 (24-hour) and 12 μg/m 3 (annual).Recent studies have shown that adverse health effects still exist below the current PM 2.5 standards (i.e., 'low-level' PM 2.5 ), motivating further discussions on the standards that can better protect public health (Yu et al., 2020;Feng et al., 2016;Shi et al., 2016).
Epidemiological studies rely on the accuracy of exposure data (either measured or modeled) from PM 2.5 exposure assessment.When PM 2.5 exposure levels that are corresponded to subjects' health outcomes do not accurately represent their actual exposures (i.e., exposure misclassification), the PM 2.5 -health associations would be obscured or biased (Zeger et al., 2000).Traditionally, ground monitoring of PM 2.5 has been widely used to calculate daily or annual average PM 2.5 exposures.To alleviate the exposure misclassification due to sparsely distributed ground monitoring networks and thus generate more reliable exposure data, PM 2.5 modeling methodologies that exploit the capability of satellite remote sensing have been developed (Hoff and Christopher, 2009).These modeling approaches provide PM 2.5 exposure estimates in areas without ground monitoring in an effort to reflect actual PM 2.5 exposures more closely.Emerging evidence further suggests that the toxicity of PM 2.5 varies by its composition and sources, which may support regionally varying PM 2.5 -health associations (Bell et al., 2014;Son et al., 2012;Franklin et al., 2008).
This review article summarizes and discusses the following: (1) traditional PM 2.5 exposure assessment approaches using ground monitoring, (2) the development and advancement of satellite-based PM 2.5 exposure assessment, (3) satellite opportunities of PM 2.5 exposure assessment on PM 2.5 composition and time-resolved PM 2.5 , and (4) future satellite-based exposure assessment to address research needs.Instead of discussing previous research in an exhaustive manner, this article emphasizes more widely recognized satellite products and exposure modeling approaches around the globe.Ultimately, this article is expected to contemplate lessons learned from previous research and motivate continuing efforts to elaborate the exposure assessment of PM 2.5 .

PM 2.5 EXPOSURE ASSESSMENT USING GROUND PM 2.5 MEASUREMENTS
Traditionally, ground-based PM 2.5 measurements have been widely used as proxies of ambient PM 2.5 exposures for health effect studies in many parts of the world (Pascal et al., 2014;Son et al., 2012;Dominici et al., 2006).Ground PM 2.5 concentrations are generally reported hourly or daily (24 hours).Hourly PM 2.5 is observed by continuous monitors such as Beta Attenuation Monitor (BAM; Met One Instruments, Inc., Grants Pass, OR, USA) and Tapered Element Oscillating Microbalance (TEOM; Thermo Scientific, Waltham, MA, USA), and daily PM 2.5 concentrations are measured by 24-hour integrated filter samples (e.g., Harvard Impactor) or calculated as the averages of 24 hourly concentrations observed from the continuous monitors.In developed countries, ground PM 2.5 concentrations are routinely monitored for a regulatory purpose.Ground PM 2.5 monitors can be also deployed to address specific research objectives that require denser air monitoring by time and space.However, in developing countries without any regulatory monitors, ground PM 2.5 monitors need to be established to obtain PM 2.5 exposure data.
The most acknowledged approach of ground-based exposure assessment is to correspond ambient PM 2.5 concentrations measured at a ground monitoring site, which is located in the closest proximity to subjects' residence or within the geographic boundaries of the residence (e.g., county, census tract, and province), to the subjects' health outcomes.However, as the distance from the residence to the monitor increases, the representativeness of the measured PM 2.5 concentrations for the subjects' exposures tends to be lower, as supported by lower site-to-site correlations of PM 2.5 concentrations with larger site-to-site distance (Bell et al., 2011).The spatial heterogeneity of PM 2.5 , evaluated by the correlations and differences of concentration levels, also varies by region, causing the ground-based exposure assessment to be susceptible to regionally varying exposure misclassification (Pinto et al., 2004).Consequently, a specific constraint of distance between the locations of subjects and ground monitors is usually applied in epidemiological studies, and those subjects who do not meet the constraint are often excluded from the studies (Ebisu et al., 2014).
Taking advantage of ground PM 2.5 measurements, interpolation techniques have been employed in order to assess subjects' exposures at the locations of their residence (Neupane et al., 2010;Wu et al., 2006;Jerrett et al., 2005).A simple and common interpolation method is inverse distance weighting (IDW), and more sophisticated methods include kriging, natural neighbor, and spline among others (ESRI, 2017).The fundamental of the interpolation techniques is based on the tendency of more similar air pollution levels with closer spatial distance.With regards to PM 2.5 , these techniques calculate a PM 2.5 concentration level that is unknown at a specific location by exploiting PM 2.5 concentration levels that are known at adjacent locations.
The key advantage of using the interpolation techniques is that they only require ground PM 2.5 measurements without additional spatiotemporal information related to PM 2.5 , which makes the approaches relatively easy to apply.However, the accuracy of the PM 2.5 concentra-tion estimates depends on the distribution of ground PM 2.5 monitoring networks.As the term 'inter-polation' indicates, these techniques would be more appropriate when the location of interest with unkn own PM 2.5 concentrations is surrounded by a larger number of locations with known PM 2.5 concentrations (i.e., inbetween locations).Furthermore, the techniques may not be effective to estimate location-specific PM 2.5 concentrations that are higher than those at adjacent locations because the PM 2.5 concentration estimates are derived from spatial inter-dependency rather than direct observations from a specific location.

1 Satellite Remote Sensing for PM 2.5
Since the launch of MODerate resolution Imaging Spectroradiometer (MODIS) on Aqua (1999) andTerra (2002) satellites by the National Aeronautics and Space Administration (NASA), satellite-based aerosol data have been increasingly used to infer ground-level PM 2.5 concentrations (Hoff and Christopher, 2009).Satellite instruments and algorithms retrieve data on aerosol optical depth (AOD; unitless), which is a measure of light extinction (i.e., scattering and absorption) by There are several key properties of satellite-retrieved AOD to determine the applicability of the aerosol data for ambient PM 2.5 .First, the vertical profile of aerosols is unknown from most of satellite AOD products except for NASA's Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO).When aerosols are well-mixed within a planetary boundary layer height (i.e., near-surface level), satellite AOD tends to be closely related to ambient PM 2.5 .On the contrary, satellite AOD may not well reflect ambient PM 2.5 with a large fraction of aerosols aloft above the mixing layer.Second, AOD and PM 2.5 are not the measures of indicating exactly same aerosol properties.PM 2.5 concentration measures dry mass per unit volume of air, while AOD represents aerosols that are affected by the level of relative humidity (i.e., hygroscopic property).With the same PM 2.5 level, AOD is likely to be higher in more humid atmospheric condition.Third, AOD does not represent specific size distribution of aerosols unlike PM 2.5 characterizing aerosols or PM with the aerodynamic diameter of 2.5 μm or less.Despite the potentially non-overlapping fraction of size distributions between AOD and PM 2.5 , AOD is known to be sensitive to small-sized aerosols roughly with the diameters of 0.05-2.5 μm, which is similar to the size distribution of PM 2.5 (Liu et al., 2007;Kahn et al., 1998).Towards the end, the applicability of satellite AOD for ambient PM 2.5 concentrations depends on how well these factors, among others, are accounted for in the identification of PM 2.5 -AOD relationships.
Table 1 shows a list of AOD products from multiple satellite instruments that are currently available.One of the most popular satellite AOD products is MODIS AOD data retrieved from NASA's Aqua and Terra satellites (Levy et al., 2013).These two satellites were launched around the year 2000(i.e., Terra in 1999and Aqua in 2002), and MODIS instruments onboard Terra and Aqua have provided aerosol data for the last two decades.Under cloud-free conditions, these polar-orbiting satellites retrieve AOD data almost daily at a global scale at the spatial resolution of 3 km ('Dark Target' algorithm) and 10 km (both 'Dark Target' and 'Deep Blue' algorithms) (nominal resolution at nadir), crossing the equator approximately at 10:30 am (Terra) and 1:30 pm (Aqua) local sun time.Most recently in 2018, a new AOD data product, called Multi-Angle Implementation of Atmospheric Correction (MAIAC), was released to the public by NASA (Lyapustin et al., 2018).The MAIAC AOD retrieval algorithm also utilizes observations from MODIS and provides AOD data at 1 km resolution, which enables the investigation of local-scale aerosol distributions to be viable.In addition to AOD retrievals from MODIS, Visible Infrared Imaging Radiometer Suite (VIIRS) onboard Suomi National Polarorbiting Partnership (Suomi NPP), operated by the U.S. National Oceanic and Atmospheric Administration (NOAA), also generates spatiotemporal AOD data daily at the spatial resolution of 6 km (Environmental Data Record; EDR) and 750 m (Intermediate Product; IP) ( Jackson et al., 2013).The EDR AOD data are aggregated by 8 × 8 boxes of the IP AOD data.
Multi-angle Imaging SpectroRadiometer (MISR) onboard the Terra satellite has also retrieved AOD data by NASA's Jet Propulsion Laboratory ( JPL) (Garay et al., 2017;Diner et al., 1998).The key advantage of MISR AOD over other AOD products is that the multiangle feature of MISR enables aerosols to be monitored from 9 different cameras or angles, generating a range of quantitative aerosol properties such as fractional AOD components, representing the size, shape, and refractive index of aerosols, as well as total AOD.The spatial resolution of MISR AOD was originally 17.6 km and improved to 4.4 km in 2017.Due to MISR's rather coarse temporal resolution (every 9 days), day-to-day variability of PM 2.5 concentrations is not plausible to obtain from MISR AOD.
In addition to polar-orbiting satellites mentioned above, there are geostationary satellites that monitor aerosols at a high temporal resolution, such as Advanced Baseline Imager (ABI; Geostationary Operational Environmental Satellite (GOES-16 (East) and 17 (West); U.S. NOAA)) (Kondragunta et al., 2020), Geostationary Ocean Color Imager (GOCI; Korea Institute of Ocean Science and Technology (KIOST)) (Choi et al., 2018), and Advanced Himawari Imager (AHI; Japan Meteorological Agency ( JMA)) (Yoshida et al., 2018).The geostationary feature enables these satellites to provide AOD data hourly or sub-hourly during the daylight time and thus observe the diurnal variation of aerosols.However, the feature prevents them from covering the entire globe because they are only capable of looking down at a specific region of interest.
The selection of a specific AOD data product dep ends largely on the spatial and temporal resolutions and accuracy of AOD data to reasonably address the objectives of PM 2.5 exposure studies.In this regard, the pros and cons of the AOD products listed in Table 1 are relative and thus determined by these properties that meet specific research needs.When local or subject-specific PM 2.5 exposures are needed, AOD data at a finer spatial resolution (e.g., <10 km) would be more preferable.In case of examining PM 2.5 exposures at a global scale (e.g., global burden of disease (GBD)) (Brauer et al., 2016), AOD data at a coarser resolution (e.g., ≥10 km) are still useful to estimate PM 2.5 exposures and correspond them to health outcomes at a country level.Because AOD data are not retrieved on days with cloud and snow cover, the satellite overpass frequency in combination with the proportion of cloud-free days determines the overall frequency of AOD retrievals.Hence, the AOD products with shorter retrieval frequency lead to PM 2.5 concentration estimates at a higher temporal resolution, which is more applicable to acute health effect studies.On the other hand, the accuracy of satellite-retrieved AOD data can be evaluated by comparing the data to AOD values obtained from the AErosol RObotic NETwork (AERONET) monitoring networks as ground-truth, which is a global network of ground-based remote sensing instruments (i.e., sun photometers) to retrieve the optical, microphysical, and radiative properties of aerosols (Holben et al., 1998).A satellite AOD product that shows a better agreement with AERONET AOD (i.e., higher correlation, slope closer to 1, and intercept closer to 0 in a simple linear regression) demonstrates higher accuracy (Xiao et al., 2016).

2 Early Development of Satellite-based PM 2.5
Concentrations The simplest approach of estimating AOD-derived PM 2.5 concentrations is to employ a simple linear regression model (i.e., 1 independent variable or predictor in a model) and identify the relationship between AOD and PM 2.5 at the locations where both AOD values and measured PM 2.5 concentrations are available (Schaap et al., 2009;Engel-Cox et al., 2004;Wang and Christopher, 2003).This relationship (i.e., fitted model with an intercept and a slope estimated) is then used to estimate PM 2.5 from AOD at the locations where AOD data are available but ground PM 2.5 monitoring is not available.This initial methodology demonstrated space-dependent predictive power of AOD for PM 2.5 , which was generally higher in areas with vegetation and homogeneous land and meteorology types (e.g., eastern U.S.) than in semi-arid and barren areas or areas with diverse terrains and meteorology (e.g., western U.S.) (Li et al., 2015;Engel-Cox et al., 2004).When satellite AOD values are corresponded to ground PM 2.5 measurements, two temporal scales of PM 2.5 , i.e., PM 2.5 measured at a satellite overpass time and 24-hour average PM 2.5 , are used in the objectives of performing the synoptic evaluation of AOD-to-PM 2.5 predictability and applying AOD-derived PM 2.5 for policy assessment and health effect studies, respectively.
Multiple linear regression models are employed to account for additional spatial and/or temporal factors (e.g., local meteorology and land use information) that influence PM 2.5 concentrations and thus PM 2.5 -AOD relationships (also for PM 10 depending on ground PM data availability) (Marsha and Larkin, 2019;Seo et al., 2015;Gupta and Christopher, 2009b).The importance of adding the meteorological parameters into the model is attributed to atmospheric conditions that may affect PM 2.5 and AOD differently.For example, when planetary boundary layer height is lower (assuming the same area and amount of aerosols), the reduced air volume results in higher PM 2.5 concentrations but the column integrated AOD levels are likely to be the same.Adding data on planetary boundary layer height into the model would mitigate its disproportionate impacts on PM 2.5 and AOD to some extent.The meteorological parameters (e.g., relative humidity, temperature, wind speed, and planetary boundary layer height) can be obtained from ground weather stations (except for planetary boundary layer height) and global and regional meteorological modeling such as ERA5 (31 km resolution; European Centre for Medium-range Weather Forecasts (ECMWF)), Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2, 0.5 × 0.625 degree; NASA), and North American Regional Reanalysis (NARR, 32 km resolution; NOAA) among others.
Land use parameters are incorporated with the model in order to explain the fine-scale PM 2.5 variability around ground PM 2.5 monitors as emission proxies.The 'land use' parameters are rather loosely defined and thus inclusive and generally used to represent any temporally invari ant parameters, such as distance to roads, road density, population density, impervious areas, elevation, and distance to industrial complex.Some of the parameters vary by time in reality, but lack of frequent updates on the data leads them to be considered time-invariant in the model.On the other hand, the term 'fine-scale' is intended to represent smaller spatial scales than the spatial resolution of satellite AOD.For example, when a PM 2.5 monitor is located in close proximity to highways (i.e., traffic as a PM 2.5 emission source), measured PM 2.5 concentrations tend to be high, but satellite-based AOD may not well account for such near-source impacts depending on the relative scales of the spatial resolution of AOD data and the spatial heterogeneity of PM 2.5 around the highways.Consequently, the land use parameters are likely to be more beneficial with AOD data at a coarser spatial resolution.
In addition to the linear regression models described above, a statistical model called a generalized additive model (GAM) is utilized to allow non-linear relationships between PM 2.5 and independent variables (e.g., AOD, meteorology, and land use) (Strawa et al., 2013;Liu et al., 2009).When the non-linear or smooth function follows the actual relationships more closely than the linear function, replacing the linear function with the non-linear one would improve the PM 2.5 predictive power.However, the model predictability may also increase even when the non-linear function only follows pseudo relationships (i.e., data-driven wiggling patterns, causing model overfitting) regardless of the actual ones.To prevent the latter, those non-linear relationships need to be supported by scientific knowledge on atmospheric chemistry and physics.For instance, if the GAM displays a quadratic pattern between PM 2.5 and AOD, including both positive and inverse correlations, it is not likely to be an actual relationship that would be appropriate only with positive correlations.
Once linear or non-linear models are fitted, estimated PM 2.5 concentrations are validated by comparing them to measured PM 2.5 concentrations generally with the metrics of coefficient of determination (R 2 ), root mean squared error (RMSE), and mean absolute error (MAE).When the estimated and measured PM 2.5 concentra tions are in a better agreement (i.e., higher R 2 and lower RMSE and MAE), the model estimates are considered to be more accurate.However, in this comparison, model overfitting may be caused because the same data are used for both model fitting and validation, likely overstating the model performance.To prevent the model perfor-mance from being overestimated, cross-validation (CV) techniques have been widely implemented by randomly splitting data into the subsets of the data (e.g., 10 equal sized subsets; 10-fold CV), fitting the model with all the subsets except for one, estimating PM 2.5 for the hold-out subset using the fitted model, and repeating the same process for each of the subsets (Liu et al., 2009;Yanosky et al., 2008).The equal sized subsets can be based on observations (i.e., same number of observations in each subset) and monitoring sites (e.g., all observations from the same number of sites in each subset).The observation-based CV is more advantageous than the site-based CV because the overall model performance is not influenced by the subset-specific number of observations, while the site-based CV is more appropriate in that PM 2.5 concentrations at a given site are estimated independently from those at other sites by non-overlapping predictors (e.g., land use parameters) (Lee, 2019).All the PM 2.5 concentration estimates from the CV analysis are compared to measured PM 2.5 concentrations to evaluate the overall model performance.In addition, the CV analy sis can be performed separately by spatial and temporal portions of the model estimates (i.e., spatial and temporal CV R 2 ) (Kloog et al., 2011).Instead of the CV analysis, the model estimates have been also validated by comparing them to measured PM 2.5 concentrations obtained from a monitoring network that is independent from the network used in the modeling processes (Knibbs et al., 2018).This validation approach may be advantageous due to potential resi dual overfitting from the CV approach, but it is often not feasible to have an independent PM 2.5 monitoring net work.

3 Advanced Approaches for Estimating
Satellite-based PM 2.5 Exposures The approach of using a mixed effects model to estimate AOD-derived PM 2.5 substantially improved the predictive power of AOD for PM 2.5 (CV R 2 between daily measured and estimated PM 2.5 = 0.92) (Lee et al., 2011).Lee et al. (2011) showed that the mixed effects model outperformed the simple linear regression model (both models only with AOD), i.e., CV R 2 of 0.95 versus 0.51 (average of site-specific R 2 ), in the New England region of the U.S. (including Massachusetts, Connecticut, and Rhode Island).Since then, this modeling approach has been widely adopted by many researchers in the world, many of which incorporated additional spatiotemporal predictors into the model (Shtein et al., 2020;de Hoogh et al., 2018;Pereira et al., 2017;Ma et al., 2016;Just et al., 2015;Xie et al., 2015;Hu et al., 2014;Kloog et al., 2011).
The key feature of the mixed effects model is to allow day-to-day variability in the relationships between measured PM 2.5 concentrations and AOD values.The term 'mixed' represents a combination of 'fixed' and 'random' effects, which identifies an average relationship of PM 2.5 and AOD that is applied to the entire study period and the daily variations of the relationships (i.e., daily deviations from the average relationship; day-specific slopes and intercepts), respectively.On the contrary, the simple and multiple linear regression models and GAM only generate an average relationship between PM 2.5 and AOD for the entire period.As discussed above, the impacts of local meteorology and large-scale weather systems on the PM 2.5 -AOD relationships tend to be diverse by day in a region.The mixed effects model implicitly accounts for such temporal parameters by allowing the PM 2.5 -AOD relationships to vary on a daily basis, regardless of the feasibility of parameterizing and including each temporal parameter in the model.Moreover, the daily relationships of PM 2.5 and AOD, generated by the mixed effects model, are more robust than those derived from day-specific simple linear regression models, because the mixed effects model produces daily PM 2.5 -AOD relationships by identifying the overall average relationship and then quantifying daily deviations from this average relationship.This is an important feature of the mixed effects model particularly due to the number of PM 2.5 and AOD pairs that varies by day.Furthermore, spatial variations in PM 2.5 -AOD relationships were accounted for by adding region-specific relationships into the mixed effects model (Chudnovsky et al., 2014) or by combining the model with a geographically weighted regression (GWR) model, which generates a continuous spatial surface of each parameter and subsequently identifies the local slopes of AOD for PM 2.5 with geographical weighting (Hu et al., 2014).
The mixed effects modeling approach motivated the use of satellite-based PM 2.5 exposure estimates in acute and chronic health effect studies with low PM 2.5 modeling errors and subsequent gap-filling techniques exploiting those AOD-derived PM 2.5 (Son et al., 2017;Mc Guinn et al., 2016;Shi et al., 2016;Hyder et al., 2014).Pre vious studies developed statistical approaches to estimate PM 2.5 exposure levels for days without AOD data (largely due to cloud cover) by clustering days with simi-lar PM 2.5 spatial patterns (Lee et al., 2012) and by assuming season-specific PM 2.5 spatial patterns (Kloog et al., 2011).These statistical approaches relied on AODderived PM 2.5 concentrations (i.e., days with AOD data) as a first step of estimating all the missing PM 2.5 concentrations, leading to no spatial and temporal gaps of PM 2.5 exposure estimates.These gap-filling approaches contributed to epidemiological studies that require both short-term and long-term PM 2.5 exposures such as subject-specific trimester PM 2.5 exposures and their birth outcomes (e.g., lower birth weight) (Hyder et al., 2014) and hospital admissions (Kloog et al., 2012).Finally, such modeling approaches facilitated health effect studies associated with PM 2.5 levels below current PM 2.5 standards (i.e., 'low-level' PM 2.5 and health) (Shi et al., 2016).In the U.S., most of ground PM 2.5 monitors are located in highly populated urban or non-populated remote (i.e., forest; Interagency Monitoring of Protected Visual Environments (IMPROVE) network) areas, and satellitebased PM 2.5 exposure appro aches enable epidemiological studies to include suburban and rural populations generally with 'low-level' PM 2.5 exposures.
A combination of chemical transport models (CTM) (e.g., GEOS-Chem, global 3-D simulations) and satellite AOD data has been also employed to estimate PM 2.5 exposures.Previous research simulated relationships betw een column AOD at a satellite overpass time and ground-level 24-hour average PM 2.5 , which relied on the size, composition, and vertical and diurnal profiles of aerosols and relative humidity, and these relationships were used to calculate PM 2.5 as follows: [simulated (ground PM 2.5 /column AOD) × satellite column AOD = estima ted PM 2.5 ] (Liu et al., 2004).This appro ach is different from statistical calibrations because the simulated conversion factors are not derived from empirical relationships between the observations of ground PM 2.5 and satellite AOD.Instead, the CTM simulations tracked the atmospheric processes of emissions and meteorology and thus quantified modeldriven daily relationships of 24-hour average PM 2.5 and column AOD at a given satellite overpass time (van Donkelaar et al., 2010).With respect to GEOS-Chem, daily conversion factors were calculated originally at the spatial resolution of 2°× 2.5° (global), and nested simulations at the resolution of 0.5°× 0.67° were conducted in selected regions such as North America, Europe, and East Asia (Hammer et al., 2020).Despite the spatial resolution of estimated PM 2.5 generally following that of satellite AOD, the spatial resolution of the simulations for the conversion factors may limit the ability of capturing local PM 2.5 gradients.As a recent improvement, the GWR modeling was further adopted to calibrate annual average PM 2.5 concentrations estimated from the conversion factors and satellite AOD for measured annual average PM 2.5 concentrations, while incor porating spatial parameters such as elevation and proximity to urban land surface, leading to CV R 2 of 0.81 (van Donkelaar et al., 2016).
Due to the nature of global scale modeling, the app roach of utilizing GEOS-Chem and satellite AOD (recently also with statistical calibration) has contributed to global burden of disease (GBD) studies, providing global longterm PM 2.5 exposure estimates (Brauer et al., 2016).In addition, this modeling has been used to generate PM 2.5 exposure estimates for multi-country and nationwide health effect studies (Pinault et al., 2016;Fleischer et al., 2014).The approach of adopting GEOS-Chem and satellite AOD is useful particularly in developing countries where no ground monitoring exists and thus empirical models are not plausible.Nonetheless, ground PM 2.5 data are still necessary to further increase model predictability through statistical calibration (e.g., GWR) and vali date model estimates.

4 Machine Learning for Satellite-based PM 2.5
As artificial intelligence (AI) has infiltrated into a wide range of industries, machine learning techniques can be also applied to the estimation of satellite-based PM 2.5 exposures and environmental health research in general (National Academies of Sciences, Engineering, and Medicine, 2019).Machine learning seeks to find the most accurate relationships between PM 2.5 and a number of predictors (i.e., 'training' similar to model fitting), while considering interactions between those predictors, and thus reach the best predictive power without explicitly specifying modeling assumptions (i.e., linear or non-linear) and structures (e.g., interactions between predictors).The training process tends to be more robust as more high-quality data are added into the dataset.Though the concept to apply machine learning for PM 2.5 or PM 10 is nearly two decades old (Chaloulakou et al., 2003;Pérez et al., 2000), incorporating satellite AOD data with the machine learning frameworks is com paratively recent (Gupta and Christopher, 2009a).For the last few years, machine learning techniques have been implemented into numerous statistical and programming software packages (e.g., R packages), making the machine learning techniques more accessible to researchers.
A number of machine learning techniques such as neural network, random forest, and gradient boosting (or extreme gradient boosting) have been applied for PM 2.5 exposure estimates, taking advantage of simultaneous advancement of satellite remote sensing, machine learning tools, and computational infrastructure ( Just et al., 2020;Park et al., 2019;Hu et al., 2017;Di et al., 2016;Reid et al., 2015).These studies resulted in high predictive power for PM 2.5 in a range of CV R 2 roughly betw-Table 2. PM 2.5 estimation models developed by using satellite AOD data.The category is based on the primary focus of the models regardless of additional modeling approaches combined with.The applicability to epidemiological studies is assessed by the predictive power for PM 2.5 .

Statistical model
Simple linear regression AOD Wang and Christopher (2003) 2020) High een 0.70 and 0.90.However, it is noted that the model performance is not directly comparable to each other due to different study periods and regions and selected predictors among others, and therefore it is challenging to conclude one machine learning approach performs better than the others.An ensemble modeling was further employed in an effort to generate more refined PM 2.5 estimations collectively from multiple CTMs, statistical (i.e., mixed effects model), and machine learning approaches and enhance the overall predictive power compared to each individual one (Shtein et al., 2020).Furthermore, the machine learning approaches have imputed missing AOD or combined CTM-calibrated AOD to fill the spatial and temporal gaps of satellite-based PM 2.5 (Li et al., 2020;Di et al., 2016).Due to high accuracy of ambient PM 2.5 exposure estimates attributed to machine learning, those estimates have been employed in health effect studies (Wei et al., 2020;Di et al., 2017).A summary of the early development and recent advancement for AOD-derived PM 2.5 estimation approaches (i.e., simple linear regression through machine learning) is presented in Table 2. Despite the capability of machine learning to process complex multi-dimensional data and thus enhance the predictive power, it is worthwhile to address scientific aspects of machine learning in the context of predictive accuracy versus 'black box'-like predictions.As discussed in the National Academies of Sciences, Engineering, and Medicine (NASEM) of the U.S., data quality, transparency, and reproducibility of machine learning need to be emphasized to further promote the use of machine learning for environmental health research (National Academies of Sciences, Engineering, and Medicine, 2019).Unlike statistical models, the processes of machine learning techniques do not yield the coefficients of each model predictor and their corresponding uncertainties.Hence, it is challenging to understand and evaluate scientific mechanisms and pathways that lead to the enhanced predictive power from machine learning.In this regard, statistical approaches described above are still crucial to delve into the relationships between each predictor and PM 2.5 exposure levels and prioritize input data resources for future improvement of PM 2.5 exposure models.Moreover, during the processes of machine learning, there are a number of details to be disclosed to reproduce outputs because they may cause substantial differences in the outputs (Peng, 2020).

PM 2.5 COMPOSITION FROM SATELLITE REMOTE SENSING
PM 2.5 composition is related to the emission sources of PM 2.5 , which can lead to health effect studies associated with composition-or source-specific PM 2.5 exposures.For the last two decades, MISR onboard Terra satellite has provided aerosol optical properties including fractional AOD components, which have been used to estimate spatially resolved concentrations on PM 2.5 composition (i.e., sulfate, nitrate, organic carbon, elemental carbon, and dust) (Geng et al., 2020;Meng et al., 2018;Liu et al., 2007).MISR-derived PM 2.5 composition concentrations may be more applicable for long-term average exposures than for daily variations of exposures due to lack of day-to-day retrievals (i.e., retrieval frequency of every 9 days in cloud-free conditions).
NASA JPL plans to launch a satellite instrument called the Multi-Angle Imager for Aerosols (MAIA) in 2022 (Diner et al., 2018).As a successor of MISR, MAIA is also designed to retrieve data that can be used to estimate PM 2.5 composition.This satellite instrument specifically addresses health associations with PM 2.5 composition while collaborating with environmental epidemiologists from the planning stage of the mission.The MAIA mission selected 12 primary target areas (PTAs), 26 secondary target areas (STAs), and 3 calibration/validation target areas (CVTAs) across the world.To be eligible for the PTAs, a health cohort as well as ground PM 2.5 monitors (both total mass and composition) are required to calibrate satellite data for measured PM 2.5 composition concentrations and then apply the concentration estimates for epidemiological studies in each PTA.The MAIA instrument is designed to collect data from 3-4 revisits per week at the spatial resolution of 1 km.On the other hand, a health cohort is not required but there must be substantial data needs for PM 2.5 composition to be one of the STAs.Unlike previous satellite AOD products, the MAIA mission plans to release estimated PM 2.5 composition concentrations as well as total and fractional AOD data.A geostatistical regression model to estimate PM 2.5 composition would combine multiple data resources such as retrieved aerosol properties, land use parameters (e.g., population and urban density), and meteorology.When aerosol properties are not available from the satellite, the modeling process utilizes CTM-derived PM 2.5 composition instead of fractional AOD as a predictor.Ultimately, the MAIA mission plans to provide PM 2.5 composition data without any spatial and temporal gaps (along with PM 10 and PM 2.5 mass concentrations).Up-to-date information on the launch and retrieval plans of MAIA can be found at the NASA JPL website (https://maia.jpl.nasa.gov/).

TIME-RESOLVED PM 2.5 FROM GEOSTATIONARY SATELLITES
Geostationary satellites such as GOES-16 and GOES-17 ABI (U.S.), GOCI (South Korea), and AHI (Japan) are able to broaden the current scope of PM 2.5 exposure assessment, exploiting hourly or sub-hourly aerosol data.Thus far, PM 2.5 exposures and health effect studies focus on their associations at a daily or coarser scale (i.e., shortterm or long-term exposures and their corresponding acute or chronic health effects) largely due to limited sub-daily exposure and health data.The associations between short-term PM 2.5 exposures and acute health outcomes are generally investigated by adopting 24-hour average PM 2.5 concentrations.However, within the time window of 24 hours, PM 2.5 exposures tend to vary depending on subject-specific time-activity patterns, diurnal patterns of source emissions (e.g., higher traffic emissions during the rush hours), and meteorology (e.g., higher pla netary boundary layer height in the middle of a day).There fore, high PM 2.5 exposures at a certain time on a given day may play a critical role in triggering acute health effects that are identified in epidemiological studies exami ning daily exposure-response relationship.
Geostationary satellites are also suitable to assess PM 2.5 exposures from extreme air pollution events such as wildfires (Reid et al., 2015).These satellites are capable of tracking the transport of wildfire plumes and determining the high impact zones of the plumes.Simultaneously, hourly or sub-hourly AOD levels reflect the fastchanging behaviors of the plumes depending on local meteorology and terrains.The information on timeresolved AOD and subsequent PM 2.5 concentrations can be also used to alert the public of unhealthy PM 2.5 exposure levels.Recent studies reported the application of real-time individual health monitoring systems for air pollution and health effect studies (Pepper et al., 2020;Su et al., 2017), and the development of such health monitoring tools may facilitate access to health data at high spatial and temporal resolutions.Unlike spatially and temporally aggregated health data, these types of health data have a potential to be corresponded to time-resolved PM 2.5 exposures derived from the geostationary satellites.

FUTURE DIRECTIONS
For the last two decades, satellite remote sensing has advanced PM 2.5 exposure assessment by providing spatially resolved ambient PM 2.5 concentrations as a proxy of personal exposures to PM 2.5 (origin of outdoors).Satellite-based PM 2.5 exposure assessment is able to reasonably address local (e.g., within-urban and within-rural), regional (e.g., urban and rural areas), national (e.g., stateto-state or province-to-province), and global (e.g., country-to-country) PM 2.5 air pollution and its associated health outcomes.Despite the uncertainties inherently included in such exposure model estimates, satellite remote sensing obtains PM 2.5 exposure information directly from locations of interest, filling the spatial gaps of ground monitoring and potentially reducing PM 2.5 exposure misclassification attributed to the spatial misalignment between ground monitoring and subjects' locations.Furthermore, satellite observations have capabilities to broaden the scopes of PM 2.5 exposure assessment into PM 2.5 composition and time-resolved PM 2.5 .
Hybrid approaches across multiple satellite products, multi-tiered data resources, and modeling methodologies are expected to continue to enhance the overall capability of satellite observations for PM 2.5 exposure assessment.A number of polar-orbiting and geostationary satellites are currently in the orbit and generate aerosol products with their product-specific advantages and limitations.In the coming years, next-generation satellite instruments will be added to the orbit (e.g., NASA JPL's MAIA), motivating even more synergistic approaches.Because people bre athe air as a whole not by a specific air pollutant, a multi-pollutant approach dealing with other air pollutants as well as PM 2.5 (i.e., a mixture of air pollutants) (Dominici et al., 2010) would benefit from satellite data on nitrogen dioxide (NO 2 ), sulfur dioxide (SO 2 ), and carbon monoxide (CO), among others, obtained from Ozone Monitoring Instrument (OMI, Aura) and TROPOspheric Monitoring Instrument (TROPOMI, Sentinel-5 Precursor) and in the near future, Geostationary Environment Monitoring Spectrometer (GEMS, South Korea), Tropospheric Emissions: Monitoring Pollution (TEMPO, U.S.), and Sentinel-4 (Europe).Other ground-based monitoring technologies such as low-cost sensors and mobile monitoring are also available (Bi et al., 2020;Apte et al., 2017), which may be further combined with satellite-based approaches.In addition, the ensemble models of multiple statistical and machine learning methodologies would be more commonly used as 'big data' and more sophisticated modeling approaches and user-friendly data processing platforms are anticipated in the future.Finally, the data assimilation of current and future AOD data resources in combination with CTMs is likely to improve air quality forecasting, which is crucial for extreme air pollution events (Pang et al., 2018;Liu et al., 2011).
Size-resolved PM exposure assessment such as coarse PM (PM 10-2.5 ) would help refine the health associations of PM, leading to size-relevant PM health risks (Ebisu et al., 2016).The size, composition, and sources of PM are inter-connected (e.g., smaller PM from combustion than from mechanical processes) (Masri et al., 2015), and size-relevant PM and health associations may reflect sizespecific toxicity due to size itself and/or its relationships with composition and sources.As satellite remote sensing provides data on the size of PM (e.g., size-fractioned AOD from MISR) (Franklin et al., 2017), satellite-based research on PM 10-2.5 as well as PM 2.5 is expected to contribute to size-relevant PM epidemiology in a spatially resolved manner.The comprehensive framework of PM exposure assessment that can be enhanced by satellite remote sensing is descri bed in Fig. 1.
A global COVID-19 pandemic brings more attention to PM 2.5 air pollution due to a recent study that reported higher COVID-19 death rates in areas with higher historical long-term PM 2.5 concentrations (Wu et al., 2020).Previous studies have also reported that the exposure to ambient PM 2.5 was higher in socially more vulnerable populations (Hajat et al., 2013;Miranda et al., 2011).Hence, social vulnerability that is related to PM 2.5 exposures needs to be understood in the context of social determinants of health disparities.Socially vulnerable populations are unevenly distributed within urban and rural areas, and a larger proportion of those populations are located in closer proximity to PM 2.5 emission sources such as highways (Lee and Park, 2020).To reflect such distributions of PM 2.5 and social vulnerability, the satellite observations are critical, leading to comprehensive PM 2.5 exposure research that represents all populations and the investigation of potential PM 2.5 health risks that are linked with social vulnerability (Lee, 2019).Finally, satellite-based PM 2.5 exposure assessment in association with social vulnerability would contribute to policy interventions that mitigate the disparities of PM 2.5 exposures and adverse health outcomes.

DISCLAIMER
This review article was not written as part of statefunded research activities.Therefore, the statements and conclusions in this article are those of the author and do not represent the official views of the California Air Resources Board.

Fig. 1 .
Fig. 1.Comprehensive framework of PM exposure assessment that can be addressed by satellite remote sensing.

Table 1 .
Satellite AOD products (Level 2) that are currently used to estimate ambient PM 2.5 concentrations.