Advances in multiangle satellite remote sensing of speciated airborne particulate matter and association with adverse health effects: from MISR to MAIA

Abstract. Inhalation of airborne particulate matter (PM) is associated with a variety of adverse health outcomes. However, the relative toxicity of specific PM types—mixtures of particles of varying sizes, shapes, and chemical compositions—is not well understood. A major impediment has been the sparse distribution of surface sensors, especially those measuring speciated PM. Aerosol remote sensing from Earth orbit offers the opportunity to improve our understanding of the health risks associated with different particle types and sources. The Multi-angle Imaging SpectroRadiometer (MISR) instrument aboard NASA’s Terra satellite has demonstrated the value of near-simultaneous observations of backscattered sunlight from multiple view angles for remote sensing of aerosol abundances and particle properties over land. The Multi-Angle Imager for Aerosols (MAIA) instrument, currently in development, improves on MISR’s sensitivity to airborne particle composition by incorporating polarimetry and expanded spectral range. Spatiotemporal regression relationships generated using collocated surface monitor and chemical transport model data will be used to convert fractional aerosol optical depths retrieved from MAIA observations to near-surface PM10, PM2.5, and speciated PM2.5. Health scientists on the MAIA team will use the resulting exposure estimates over globally distributed target areas to investigate the association of particle species with population health effects.

Advances in multiangle satellite remote sensing of speciated airborne particulate matter and association with adverse health effects: from MISR to MAIA mixtures and sources are most harmful. This information could help prioritize air quality guidelines, facilitate cost-effective monitoring and mitigation strategies, and aid research into the biological mechanisms for documented PM health effects. 12 The Weather and Air Quality panel of the 2017 Decadal Survey for Earth Science and Applications from Space 20 includes among its highest priority objectives improvement in the ability to estimate global air pollution impacts on human health along with the "establishment and maintenance of a robust, comprehensive observing strategy for the spatial distribution of PM (including speciation)." Given that the particles responsible for human health risks are situated near ground level, the Decadal Survey recognizes the need for an integrated strategy that combines space-based, aircraft, and ground-based observations, augmented by data from CTMs.
The past two decades have witnessed major advances in our ability to map aerosol abundances and particle properties from space. Aerosol retrievals over land from instruments, such as the Multi-angle Imaging SpectroRadiometer (MISR), 21 Moderate resolution Imaging Spectroradiometer (MODIS), 22 and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) 23 have been successfully used to generate global maps of near-surface fine PM concentrations and track multiyear trends. 24,25 These satellite-based maps of fine PM have been used in the GBD and many other health impact studies, including several that examined PM 2.5 exposure and lung function, kidney disease, lung cancer, breast cancer, heart attacks, and birth outcomes. 7,8,[26][27][28][29][30][31][32] These efforts have been made possible by advances in spaceborne instrumentation and associated data processing algorithms.
Current efforts in aerosol remote sensing are aimed at improving our ability to characterize particle type. Multiangle observing, implemented in satellite instruments such as MISR 33 and Polarization and Directionality of Earth's Reflectances (POLDER), 34 has been shown to provide an effective modality for achieving this objective. 21,35,36 The MISR instrument, built by the Jet Propulsion Laboratory (JPL) for flight on NASA's Terra spacecraft, has been collecting Earth science data since February 2000. In this paper, we briefly review the application of MISR to aerosol and PM remote sensing. This discussion serves as a prelude to a description of the Multi-Angle Imager for Aerosols (MAIA), 37 which builds upon MISR heritage and is currently in development at JPL. Key elements of the MAIA investigation include (1) a satellite instrument that incorporates a number of measurement advances relative to MISR, such as expanded spectral range and polarimetric imaging, (2) integration of space-based and ground-based measurements and CTM outputs to generate high-resolution maps on a 1-km spatial grid of speciated PM in a selected set of globally distributed target areas, and (3) linkage of the resulting PM exposure data to human health records to assess the impact on disease. This paper is intended to familiarize the scientific and public health communities and potential data users with the principal elements and strategies to be employed by the MAIA investigation, and to provide an overview of the current development status of the project.

Background
The MISR instrument 33  Pushbroom imagery at 275-m-to 1.1-km spatial resolution over a 400-km-wide swath is acquired in four visible/near-infrared (VNIR) spectral bands (446, 558, 672, and 866 nm) in each camera by making use of spacecraft motion and linear detector arrays. MISR was designed to improve our understanding of the Earth's climate, ecology, and environment. The suite of validated geophysical data products 38 is generated and archived for public distribution at the NASA Langley Atmospheric Science Data Center (ASDC). An extensive bibliography of peer-reviewed publications describing, applying, and validating MISR data for studies of aerosol climate, air quality, and health impacts, radiation and cloud-climate interactions, cloud-tracked winds, and surface biospheric and cryospheric science is available on the MISR website. 39

Aerosol Data Product Generation
Among the objectives of the MISR investigation is global mapping of aerosols. Direct radiative effects of aerosols, both in magnitude and sign, depend principally on the aerosol optical depth (AOD), single scattering albedo, scattering phase function, and the albedo of the underlying surface. Aerosols also have indirect climate and hydrological impacts through their effects on the albedos, lifetimes, and microphysical properties of clouds, and play a major role in human and environmental health.
Multiangle radiance observations are valuable for enhancing the aerosol signal relative to surface reflection and providing sensitivity to the aerosol scattering phase functions, which are governed by particle size, shape, and composition. [40][41][42] Radiative-transfer-based algorithms are applied to radiometrically calibrated, georectified, and cloud-screened MISR multiangle, multispectral imagery to generate the aerosol product. Over land, two main algorithms work together. The first, known as heterogeneous land, utilizes spatial contrasts to derive an empirical orthogonal function representation of the surface contribution to the measured multiangle radiances. 43 The second, known as homogeneous land, uses similarity in the angular shape of surface bidirectional reflectance factors (BRFs) among the four spectral bands as a constraint on the aerosol retrievals. 44 Both algorithms make use of the multiangular nature of the MISR observations. By employing a lookup table consisting of 74 mixtures of aerosol particles having prescribed microphysical and optical properties and using several goodness-of-fit metrics to compare modeled top-of-atmosphere radiances to the MISR observations, the retrieval algorithm provides sensitivity to both AOD and aerosol types. 35

Application to Air Quality and Human Health
Comparisons of MISR AODs with independent ground-based sunphotometer AODs from the Aerosol Robotic Network (AERONET) 45 show a high positive correlation, 46,47 including over arid land and urban areas. [48][49][50] As a result, MISR is one of several satellite instruments contributing to widely used global maps of PM 2.5 . 24,25,51 MISR's sensitivity to particle type enables separation of anthropogenic aerosols from dust, which has led to improved estimates of ground-level PM 2.5 concentrations in the arid western United States compared with singleangle approaches. 52,53 These multivariate regression models were initially developed to explore MISR's ability to quantitatively characterize ground-level concentrations of PM 2.5 components such as sulfate, nitrate, organic carbon (OC), and elemental carbon (EC). Later, a more flexible generalized additive model (GAM) using MISR fractional AOD (partitioned by particle properties) scaled by vertical profiles of aerosol loading from the GEOS-Chem transport model was able to explain 70% of the variability in sulfate concentrations measured by surface monitors. 54 Particle size and shape information from MISR retrievals has been used to associate anthropogenic pollution with significant decadal rise in AOD and ground-level PM 2.5 over urban centers and densely populated rural areas in India. 55,56 Validation of MISR aerosol retrievals using the operational 17.6-km resolution product demonstrated high accuracy over land for AOD <0.5 and systematic underestimation (though high correlation) at high aerosol loading. [46][47][48][49][50] Hierarchical Bayesian modeling and statistical analysis of this product suggested potential benefits of going to higher spatial resolution. 57,58 Given the value of finer spatial detail for studies of urban air quality, the MISR retrieval algorithm was recently adapted to operate on a 4.4-km spatial grid, and prototyping of the updated code demonstrated significant improvements in terms of accuracy, coverage, and mapping of spatial gradients. 59 Consequently, the operational aerosol product was upgraded from 17.6-km (version 22) to 4.4-km spatial resolution (version 23), and the V23 product was made publicly available in late 2017. An example of the improvement in spatial resolution and coverage is shown in Fig. 1. These data are from a Terra overpass of southeastern Texas and western Louisiana on February 14, 2013. The 4.4-km resolution product does a superior job in pinpointing elevated AODs over Houston and the Red River Valley.
Prototype versions of MISR's 4.4-km aerosol product have been used over parts of southern and central California to estimate daily-averaged PM 2.5 , PM 10 , and speciated PM 2.5 concentrations. Through leave-one-out cross-validation against the EPA's federal reference method measurements, the product was shown to capture PM 2.5 spatial variability at the grid scale and to separate PM 2.5 and PM 10 size modes in the greater Los Angeles area. 60 Another recent study applied GAMs to 15 years of the prototype 4.4-km product, and showed that the GAMs are able to explain 66%, 62%, 55%, and 58% of the variability in daily-averaged PM 2.5 sulfate, nitrate, OC, and EC concentrations. 61 3 Multi-Angle Imager for Aerosols

Background
NASA selected the MAIA investigation in 2016 as part of its Earth Venture Instrument program. The MAIA instrument builds upon MISR's legacy and adds new measurement capabilities for determining concentrations of total fine (PM 2.5 ) and coarse (PM 10 − PM 2.5 ) particles, along with the amounts of hydrated nonorganics, OC, black carbon (BC) or EC, and mineral dust in the fine particle mixtures. An integrated satellite/surface-level data and modeling strategy 62 is used to generate daily mean PM values on a 1-km grid. This approach enables further separating the nonorganics into sulfate and nitrate contributions. The main challenges that MAIA aims to address are to demonstrate that current satellite-based strategies for mapping total PM 2.5 mass can be extended to include speciation, and that the approach can be implemented on an operational basis.
MAIA's primary objective is to assess the impacts of different types of airborne PM on human health. The planned investigation consists of several elements: (1) the MAIA satellite instrument, (2) algorithms and software to generate PM maps using data from the MAIA instrument, surface monitors, and CTMs, and (3) epidemiological studies using the MAIA PM maps and geocoded health data to associate different types of PMs with adverse health outcomes. By increasing the density of spatial sampling and the coverage of PM in the targeted regions, MAIA overcomes a major impediment faced by prior studies that have examined the health impacts of specific PM types, 63-67 namely their limited ability to accurately assess exposure due to the small number of ground-based speciated PM monitors. To support other atmospheric science research, MAIA plans to collect measurements over areas that are of value for studying aerosol and cloud impacts on Earth's climate, and over extreme events such as wildfires, dust storms, and erupting volcanoes. Demonstration in Earth orbit of the new imaging technologies used in the MAIA instrument will also benefit NASA's planning for future missions.

Instrument Design
The MAIA instrument is designed to combine multispectral, polarimetric, and multiangular capabilities into a single, integrated imaging system capable of mapping total and speciated PM at the neighborhood scale. At the heart of the instrument is a pushbroom camera mounted on a two-axis gimbal.

Spectral coverage
MAIA's camera includes spectral bands in the ultraviolet (UV), VNIR, and shortwave infrared (SWIR), which improves sensitivity to aerosol particle properties compared with MISR's VNIRonly bands. UV wavelengths are useful for detecting absorption by hematite and aluminum oxide in dust particles, nitrated aromatic and polycyclic aromatic hydrocarbons in organic aerosols (e.g., brown carbon), and BC or EC (soot). 68,69 The use of VNIR bands for fine aerosols draws upon MISR, MODIS, and POLDER heritage. The SWIR is sensitive to coarse aerosols, 70 and a band located in a strong water vapor absorption feature provides enhanced cirrus screening. 71 Channels within and near the O 2 A-band are included to explore sensitivity to aerosol layer (and cloud) height. 72,73 Table 1 summarizes the MAIA spectral band set.

Polarimetry
As shown in Table 1, three of the MAIA bands are polarimetric, providing additional sensitivity to particle size and compositional proxies, such as refractive index. 74-76 By constraining these particle properties, polarization also works in conjunction with radiance to constrain aerosol absorption. 77 To capitalize on the benefits of polarimetry in future instruments, the aerosol community has established an uncertainty requirement of AE0.005 in degree of linear polarization, 78 which is more than three times stricter than POLDER performance. The MAIA camera achieves this level of accuracy at a spatial resolution of 1 km (compared to 6 km with POLDER) by using a polarization modulation technique enabled by a pair of photoelastic modulators and a pair of achromatic quarter-wave plates. 79,80 This results in a time-varying oscillation in the plane of linear polarization at a frequency near 27.5 Hz. The readout integrated circuit enables rapid sampling of the modulated signals during each pushbroom image frame. Silicon detectors are used in the UV/VNIR and mercury-cadmium-telluride detectors in the SWIR. Above the detector array is a set of spectral filters and wiregrid polarization analyzers. A similar system operating in the UV/VNIR has been implemented in JPL's Airborne Multiangle SpectroPolarimetric Imager (AirMSPI). 81 The second-generation AirMSPI-2 extends the spectral range into the SWIR. 82 MAIA makes use of heritage from both airborne instruments.

Multiangle imaging, areal coverage, and spatial resolution
The MAIA camera is a four-mirror f∕5.6 optical system with cross-track and along-track focal lengths at the center of the optical field of view of 57 and 61 mm, respectively. As the MAIA orbit is not yet known, this design accommodates any orbit altitude between 600 and 850 km. Unlike MISR, which contains multiple cameras pointed at discrete along-track view angles, MAIA's single camera is mounted on a biaxial gimbal assembly that can point the camera field of view to any along-track and cross-track position within a bidirectional field of regard. A mini dual-drive actuator (MDDA) drives each gimbal axis. The MDDA has been used on MISR and other satellite instruments, and provides each gimbal axis with 100% redundancy and resilience to single-point mechanical or electrical faults. The targeting nature of the MAIA instrument enables routine multiangle observations of a globally distributed set of study sites. The along-track (scan) gimbal has a AE58deg range of motion, while the cross-track (pan) gimbal has a AE39deg range of motion, which when added to the AE9deg cross-track field of view provides a AE48deg cross-track field of regard. The pan capability permits access to targets that are not directly situated on the subspacecraft track, making it possible to observe each target, on average, at least three times per week. Images of the same area can be observed at a set of discrete view angles in a "step-and-stare" sequence. A "sweep" mode of operation in which the scan gimbal moves continuously over its accessible range is also possible.
For most targets, images would be acquired using the step-and-stare mode (Fig. 2). In this mode, the gimbals orient the camera to view the target's leading edge, beginning at the most oblique forward view angle. Pushbroom imagery is acquired while the camera remains fixed at this angle, after which the scan gimbal moves to the next (smaller) forward view angle and imagery of the same area is reacquired. This sequence repeats until observations are acquired at all commanded angles. The pan actuator compensates for Earth rotation between views. Observing at five view angles would yield target lengths >330 km from a 600-km orbit and >420 km from an 850-km orbit. The number of view angles is selectable, with more angles resulting in a shorter along-track distance seen in common by all views. At nadir, the camera design covers a cross-track swath width of 192 km for an orbit altitude of 600 km, increasing to 272 km at 850-km altitude. Even at the lowest altitude, the target dimensions cover major metropolitan areas. Footprint sizes are on the order of 200 m at nadir and increase with view angle, particularly in the along-track direction. At the highest orbit altitude and most oblique view angle, the along-track footprint size remains below 1100 m, and is oversampled by a factor of 4.5 as a result of the pushbroom frame rate.

Instrument system
A conceptual layout of the MAIA instrument is shown in Fig. 3. A cylindrical barrel serves as a radiator to dissipate heat from the camera electronics. Another radiator, positioned to view deep space, dissipates heat from the focal plane, which is passively cooled to 225 K to limit dark current in the SWIR detectors. Other parts of the instrument include the structural supports, the biaxial gimbal assembly, the instrument electronics, an onboard calibrator (OBC), and a dark target (DT). The OBC consists of a glass diffuser and an array of wiregrid polarizers, and is illuminated by sunlight as the spacecraft traverses one of the orbital poles. The DT is a light-shielded cavity for measurement of dark levels. The biaxial gimbal enables periodically pointing the camera at these calibrators, and the data acquired are used in ground data processing to update the polarimetric and dark offset calibrations.

Science Operations
MAIA is to be launched into a low-Earth, sun-synchronous, polar orbit at an altitude in the 600to 850-km range. The orbit altitude and mean local time of equator crossing will be established once the host spacecraft has been selected. Mid-to late-morning overpass time is preferred to allow for fog burn-off and boundary layer mixing and because fewer clouds are expected in the morning than in the afternoon. 83 In addition, because the accessible area within the instrument field of regard increases with orbit altitude, target revisit frequencies generally increase as orbit altitude increases. NASA is planning to select the host spacecraft in late 2018, and launch is expected to occur no earlier than mid-2021. The baseline mission duration is 3 years. Science data would be collected, on average, over one target per orbit, resulting in about 100 acquisitions per week. Typical volume per target of the instrument data is estimated at 29 Gbit, slightly larger than the volume generated in one orbit by MISR, despite the fact that MAIA observes discrete targets, while MISR observes the illuminated side of the Earth continuously. This is a result of the larger number of spectral bands in MAIA, the collection of polarimetric data, and the use of onboard spatial averaging in MISR. Primary target areas (PTAs) are major population centers designated for conducting epidemiological investigations by the MAIA Science Team. PTAs would be observed in a step-and-stare mode and are selected to include major population centers covering a range of PM concentrations and particle types; surfacebased aerosol sunphotometers (e.g., from AERONET 45 ) for aerosol retrieval validation; PM mass, size discrimination, and chemical speciation monitors associated with various measurement networks 84,85 to enable development of statistical and machine learning regression models that relate retrieved column-integrated aerosol properties to near-surface PM; and health data geocoded by home addresses, zip codes, census block groups, or similar locations of study subjects. Secondary target areas (STAs) are regions of interest for air quality or other aerosol and cloud research (e.g., climate science) and would make use of either the step-and-stare or sweep mode, depending on the measurement objective. STAs do not have the same requirements on surface monitor availability as PTAs and the feasibility of higher-level data processing beyond generation of calibrated and georectified imagery (see § 3.4.1) would be assessed on a case-by-case basis. Calibration/validation target areas (CVTAs) would be observed routinely for instrument calibration and stability monitoring, and aerosol/PM validation. As the MAIA instrument does not contain an absolute radiometric calibrator, the prelaunch camera calibration will be routinely updated via vicarious calibrations over Railroad Valley, Nevada. The vicarious calibration technique has been widely adopted by many satellite sensor investigations and uses surface and atmospheric measurements acquired at the time of satellite overpass to compute top-of-atmosphere radiance and to update the instrument radiometric response. MAIA observations of noninstrumented but stable Earth targets, such as the Libya-4 desert site, will also be used to maintain the radiometric calibration uncertainty to within AE4% over bright targets (AE6% over dark targets). A candidate set of PTAs, STAs, and CVTAs is shown in Fig. 4. Specialized acquisitions over targets of opportunity may be acquired over episodic events, such as volcanic eruptions, major wildfires, or dust storms.
The candidate PTAs and STAs shown in Fig. 4 include historically understudied areas (e.g., Africa). The list is subject to future updates, as observability of some targets will depend on the orbit altitude of the host spacecraft and negotiations for access to the requisite surface monitors and health data are still in process.

Data Processing and Products
MAIA data products follow the NASA hierarchy from level 0 (raw instrument data) to level 1 (calibrated and georectified imagery), level 2 (geophysical products at the same location as level 1 source data), and level 4 (integration of measured and modeled results). As spatial gridding and map projection are incorporated into level 1 processing in a similar manner as is done for MISR, 86 MAIA does not identify separate level 3 products. Data processing software developed at the MAIA science computing facility at JPL (with algorithmic approaches and software partially inherited from the MISR and AirMSPI projects) will be delivered to the NASA Langley ASDC for product generation.

Level 1 calibrated and georectified imagery
Level 1 calibrated and georectified radiance and polarization image products will be map-projected to the surface terrain altitude for step-and-stare acquisitions and to the surface ellipsoid for sweep observations. For those target areas that will be subjected to higher-level aerosol and PM processing, a decision tree-based algorithm capitalizing on MISR and MODIS experience [87][88][89] will be used operationally to detect cloud-covered pixels.

Level 2 aerosol
The level 2 MAIA aerosol processing concept is envisioned to employ a nonlinear optimization algorithm to adjust the aerosol properties to match the full set of multiangular, multispectral, and polarimetric data provided by the MAIA instrument. This algorithm has been prototyped using AirMSPI data. 90 For MAIA, acceptable limits on aerosol microphysical and optical properties would be derived by configuring the CTM regionally and analyzing the aerosol climatology for each PTA. A pre-established surface BRF database based on the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm 91,92 would further constrain the retrievals. Constraints on the spatial and spectral variations of aerosol properties across neighboring pixels and temporal variations of surface reflection properties 76,90 within few days of target revisits will be imposed to stabilize the algorithm. The MAIAC surface database, which has been screened for clouds, potentially adds a supplementary layer of cloud screening. 93 This approach results in retrieval of both total AOD as well as fractional AODs associated with fine, coarse, spherical, nonspherical, absorbing, and nonabsorbing aerosols on a 1-km grid. Predicted signal-to-noise ratios (SNR) in the bands used for aerosol retrievals range from approximately 190 to 880 over dark targets (worst-case surface reflectance ∼0.02). Noise performance requirements have been specified to limit the effect of random instrument noise on the retrievals and to provide SNRs similar to those achieved with MISR.

Level 2 PM
The next step in the retrieval process transforms the retrieved total and fractional AODs to mass concentrations of PM 10 , PM 2.5 , and major PM 2.5 components including sulfates, nitrates, OC, BC or EC, and mineral dust. Reporting of BC or EC depends on the type of surface monitor available in a given PTA. Dust refers to resuspended inorganic material, such as soil, road dust, construction dust, or fly ash. There are several key differences between the level 2 AOD and PM products that must be accounted for in this transformation. First, AOD is a column-integrated quantity, whereas for studies of the impact of airborne PM on human health, the particles of greatest interest are near the surface. Second, PM concentrations are typically reported at controlled relative humidity (RH), whereas the MAIA AODs correspond to the ambient RH. Third, epidemiologists are interested in the average concentration of PM over a 24-h period, whereas the MAIA satellite flies over its targets at a specific time of day. Finally, the physical and optical characteristics of the particles that are captured in the AOD fields are only indirectly related to the chemical composition.
Transformations from total and fractional AOD at the time of satellite overpass to 24-h averaged total PM mass and PM species fractions, if derived solely based on MAIA observations alone, are likely to be fraught with systematic biases and uncertainties. However, previous studies have shown that geostatistical regression models (GRMs) derived from AOD, fractional AOD, and other environmental attributes, such as temperature, RH, wind speed, land cover type, and vertically resolved aerosol speciation from the CTM, along with collocated measurements from surface monitors, can be used to empirically calibrate the satellite data at locations where surface monitors are not present and to account for the differences in how the AOD and PM products are defined. 53,[94][95][96] An ensemble approach to GRM generation is being explored, using both a Bayesian framework as well as various machine learning methodologies, e.g., artificial neural networks, support vector machines, and random forests. 97,98 To generate level 2 maps of speciated PM 2.5 , MAIA would build upon current practice and include data from PM speciation monitors in addition to those that measure total PM 2.5 and PM 10 in generating the GRMs. Sources of such data include the Chemical Speciation Network (CSN) and Interagency Monitoring of Protected Visual Environments (IMPROVE) network, 84 Surface PARTiculate mAtter Network (SPARTAN), 85 other existing monitors within the PTAs, and additional ground monitors to be deployed by the MAIA project. Current plans are to expand the SPARTAN network with filter-based samplers in the MAIA PTAs. To deal with the severalmonth latency associated with the availability of CSN, IMPROVE, and SPARTAN data, monthly averaged species fractions from the same month in previous years, supplemented by ancillary information, such as temperature and RH, will be used to generate interim estimates of speciated PM 2.5 at the monitor locations. Once the actual data become available, MAIA level 2 products will be reprocessed.
Deployment of low-cost light-scattering-based particle sensors such as PurpleAir (PA) 99 is also under consideration to supplement existing government-sponsored PM 2.5 and PM 10 networks. Field and laboratory tests conducted by the South Coast Air Quality Management District's (SCAQMD) Air Quality Sensor Performance Evaluation Center (AQ-SPEC) indicate that while the PA tends to overestimate PM mass, a high degree of correlation with EPA's reference methods is found, 100 enabling correction for systematic biases in the PA data. JPL has deployed several PA sensors (on loan from SCAQMD) in Bakersfield, Fresno, and Visalia, California for further evaluation.

Level 4 gap-filled PM
The level 2 PM maps are populated with data only where cloud-screened aerosol retrievals using MAIA instrument data have been generated. Furthermore, level 2 maps are not generated on days for which there are no satellite overpasses. To generate the spatially and temporally gap-filled PM exposure estimates that are needed for the epidemiological investigations, the MAIA project plans to produce a daily gap-filled level 4 PM product in which spatial gaps due to cloud cover or other dropout are filled and PM estimates are generated on nonoverpass days. Three sources of data serve as input to generation of this product: the level 2 instrumentbased PM product, interpolated maps generated from surface monitor measurements, and PM mass and species fractional concentrations predicted by a CTM. Complete spatial and temporal coverage for each PTA would be obtained by fusing the satellite retrievals, ground-level concentration measurements, and CTM outputs in postretrieval processing.
The level 4 PM estimates are envisioned to be weighted averages determined by the relative predictive ability of each input source. The weights may vary across space and time, and are derived from uncertainty estimates associated with each of the inputs. Uncertainties associated with the level 2 satellite-based product would be generated as part of the retrieval algorithm. Interpolated values from surface monitors will be most accurate for locations and times closest to the monitor position and sampling period, and high uncertainties would be assigned where geographical factors, such as surface elevation changes, would make the interpolations unreliable. For the CTM, MAIA plans to use the mesoscale Weather Research and Forecasting model coupled with chemistry (WRF-Chem) model, 101,102 coupled with wildfire smoke emissions from the Fire Locating and Modeling of Burning Emissions system 103 and nested within the GEOS-Chem global model of atmospheric composition driven by meteorological observations from the Goddard Earth Observing System. 104 WRF-Chem outputs will be generated on a 4-km grid and GEOS-Chem on a 25-km grid. To account for biases that are known to plague even state-of-theart CTMs, 105 WRF-Chem outputs will be improved throughout the mission using model output statistics that are analyzed through comparison with MAIA level 2 speciated PM maps and data from surface monitors. For example, a recent study 106 calibrated GEOS-Chem outputs using speciation monitoring data combined with meteorological and land use variables using a backward propagation neural network, which allows for complex and nonlinear associations between model inputs. This model was used to predict daily PM 2.5 and constituents mass concentrations on a downscaled 1-km grid. Accuracy of the predictions was assessed using k-fold cross validation. The mean total R 2 at left out monitors was 0.85, 0.71, 0.69, 0.83, and 0.81 for PM 2.5 , EC, OC, nitrate, and sulfate, respectively.
As with MISR, archiving and distribution of MAIA data products will be the responsibility of the ASDC. To protect individual privacy, none of the publicly available geophysical data products generated by the MAIA investigation and stored at the ASDC will contain any health data. Health records accessed by epidemiologists and public health experts on the MAIA team will be handled in accordance with well-established legal and ethical requirements for confidentiality, privacy protection, and data security.

Science Investigation
Various epidemiological studies are planned for the different MAIA PTAs depending on the predominant PM species present, the type of health records available, and previous studies of the effects of air pollution in each area. Well-established epidemiological methodologies, such as time-series, case-crossover, and cohort-study designs 107-109 will be used.
Information about the candidate set of PTAs (see Fig. 4) is shown in Table 2. The MAIA science team plans to focus on health effects associated with a range of PM concentrations and different time scales of exposure. Acute exposure takes place over a period of several days and is generally associated with premature mortality and increased hospital visits due to both cardiovascular and respiratory diseases. These studies are conducted by analyzing vital statistics records (e.g., death certificates) and records of hospital admissions or emergency room visits. Subchronic exposure studies are primarily aimed at birth outcomes and pregnancy complications, such as low birth weight and preeclampsia. These outcomes are usually investigated by analyzing birth records contained in an area's vital statistics data, or by establishing a birth cohort. Chronic exposure studies usually track individual-level health effects over multiple years, and are important as they document morbidity and mortality risk increases and are often used in GBD estimates. These are generally done with an established cohort or by analyzing existing health records combined with long-term residency data. 111 As noted earlier, the baseline MAIA mission is 3 years in duration. Many epidemiological studies conducted around the world have reported associations between acute (daily) PM exposure and mortality, hospital admissions, and emergency department visits using <3 years of data in densely populated regions. [112][113][114][115] Adverse impacts on prenatal or neonatal development, e.g., restricted intrauterine growth, preterm delivery, low birth weight, congenital heart defects, and infant mortality, 5,116,117 have been associated with PM exposure during specific pregnancy trimesters. 118,119 Hence, investigations into birth outcomes targeting trimester specific effects can even utilize <1 year of data if the population of pregnant women residing in the area is large enough. [120][121][122] Long-term studies relating chronic exposure to cardiovascular disease have also benefited from only 2 to 3 years of data, and several have obtained statistically significant results using only a single year. 65,[123][124][125][126][127][128] Although this may seem surprising, PM spatial patterns and the rank order tend to be fairly stable from year to year, and results show that the inferred health impacts from shorter-term exposures are consistent with studies using longer exposure periods. 65 These epidemiological studies targeting chronic health outcomes typically make use of large cohorts (groups of people, who have been exposed to air pollutants at different levels or compositions over long periods of time).
Health studies with geocoded subject locations at high spatial resolution (address level) enable the most accurate estimation of PM exposure-related health effects. MAIA's resolution enables PM retrievals on a 1-km grid for sampling within the neighborhood scale. Although sulfate has relatively low spatial variability at urban-to-regional scales, 129 nitrate and primary OC vary over smaller spatial scales. BC aerosols are very heterogeneous due to their generation from traffic fuel combustion and biomass burning. 129,130 Recent research highlights the value of 1-km satellite-based aerosol data for health effect studies. 96,[131][132][133]

Conclusions
Building upon the success of MISR and other satellite instruments in providing aerosol observations that have contributed to numerous health studies, the MAIA investigation aims to take these efforts further by delving more deeply into assessing the contributions of different types of airborne particles to human health. Although much of the development effort is concerned with design and fabrication of the satellite instrument, the investigation also heavily relies on surface monitors and the CTM to generate PM maps needed to carry out the mission objectives. Although PM monitoring for regulatory purposes is largely concerned with absolute particle mass concentrations, epidemiological studies focus on the response associated with relative differences in exposure to ambient PM. Consequently, the MAIA data processing approach is designed at each step to eliminate systematic biases in the PM products, beginning with calibration of the instrument imagery, validation of the column AOD products, application of empirically derived GRMs to transform AOD to PM, and use of satellite and surface observations to remove biases in the CTM that provides a key element of the gap-filling strategy. The impact of random errors is mitigated by the statistical advantage of observing entire major metropolitan areas from space, and acquiring health information associated with hundreds of thousands to millions of individuals. With the inclusion of epidemiologists on the science team, MAIA is the first competitively selected NASA satellite mission with applications/societal benefits as its primary objective.

Disclosures
The authors declare that there are no conflicts of interest.