Improving surface PM2.5 forecasts in the United States using an ensemble of chemical transport model outputs: 2. bias correction with satellite data for rural areas

This work serves as the second of a two-part study to improve surface PM2.5 forecasts in the continental U.S. through the integrated use of multi-satellite aerosol optical depth (AOD) products (MODIS Terra/Aqua and VIIRS DT/DB), multi chemical transport model (CTM) (GEOS-Chem, WRF-Chem and CMAQ) outputs and ground observations. In part I of the study, a multi-model ensemble Kalman filter (KF) technique using three CTM outputs and ground observations was developed to correct forecast bias and generate a single best forecast of PM2.5 for next day over non-rural areas that have surface PM2.5 measurements in the proximity of 125 km. Here, with AOD data, we extended the bias correction into rural areas where the closest air quality monitoring station is at least 125 - 300 km away. First, we ensembled all of satellite AOD products to yield the single best AOD. Second, we corrected daily PM2.5 in rural areas from multiple models through the AOD spatial pattern between these areas and non-rural areas, referred to as "extended ground truth" or EGT, for today. Lastly, we applied the KF technique to update the bias in the forecast for next day using the EGT. Our results find that the ensemble of bias-corrected daily PM2.5 from three models for both today and next day show the best performance. Together, the two-part study develops a multi-model and multi-AOD bias correction technique that has the potential to improve PM2.5 forecasts in both rural and non-rural areas in near real time, and be readily implemented at state levels.

Comparison of Aerosol Optical Depth from MODIS Product Collection 6.1 and AERONET in the Western United States

Recent observations reveal that dust storms are increasing in the western USA, posing imminent risks to public health, safety, and the economy. Much of the observational evidence has been obtained from ground-based platforms and the visual interpretation of satellite imagery from limited regions. Comprehensive satellite-based observations of long-term aerosol records are still lacking. In an effort to develop such a satellite aerosol dataset, we compared and evaluated the Aerosol Optical Depth (AOD) from Deep Blue (DB) and Dark Target (DT) product collection 6.1 with the Aerosol Robotic Network (AERONET) program in the western USA. We examined the seasonal and monthly average number of Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua DB AOD retrievals per 0.1∘×0.1∘ from January 2003 to December 2017 across the region’s different topographic, climatic, and land cover conditions. The number of retrievals in the southwest United States was on average greater than 37 days per 90 days for all seasons except summer. Springtime saw the highest number of AOD retrievals across the southwest, consistent with the peak season for synoptic-scale dust events. The majority of Arizona, New Mexico, and western Texas showed the lowest number of retrievals during the monsoon season. The majority of collocating domains of AOD from the Aqua sensor showed a better correlation with AERONET AOD than AOD from Terra, and the correlation coefficients exhibited large regional variability across the study area. The correlation coefficient between the couplings Aqua DB AOD-AERONET AOD and Terra DB AOD-AERONET AOD ranges from 0.1 to 0.94 and 0.001 to 0.94, respectively. In the majority of the sites that exhibited less than a 0.6 correlation coefficient and few matched data points at the nearest single pixel, the correlations gradually improved when the spatial domain increased to a 50 km × 50 km box averaging domain. In general, the majority of the stations revealed significant correlation between MODIS and AERONET AOD at all spatiotemporal aggregating domains, although MODIS generally overestimated AOD compared to AERONET. However, the correlation coefficient in the southwest United States was the lowest and in stations from a higher latitude was the highest. The difference in the brightness of the land surface and the latitudinal differences in the aerosol inputs from the forest fires and solar zenith angles are some of the factors that manifested the latitudinal correlation differences.

An ensemble-based model of PM2.5 concentration across the contiguous United States with high spatiotemporal resolution

Various approaches have been proposed to model PM2.5 in the recent decade, with satellite-derived aerosol optical depth, land-use variables, chemical transport model predictions, and several meteorological variables as major predictor variables. Our study used an ensemble model that integrated multiple machine learning algorithms and predictor variables to estimate daily PM2.5 at a resolution of 1 km × 1 km across the contiguous United States. We used a generalized additive model that accounted for geographic difference to combine PM2.5 estimates from neural network, random forest, and gradient boosting. The three machine learning algorithms were based on multiple predictor variables, including satellite data, meteorological variables, land-use variables, elevation, chemical transport model predictions, several reanalysis datasets, and others. The model training results from 2000 to 2015 indicated good model performance with a 10-fold cross-validated R2 of 0.86 for daily PM2.5 predictions. For annual PM2.5 estimates, the cross-validated R2 was 0.89. Our model demonstrated good performance up to 60 μg/m3. Using trained PM2.5 model and predictor variables, we predicted daily PM2.5 from 2000 to 2015 at every 1 km × 1 km grid cell in the contiguous United States. We also used localized land-use variables within 1 km × 1 km grids to downscale PM2.5 predictions to 100 m × 100 m grid cells. To characterize uncertainty, we used meteorological variables, land-use variables, and elevation to model the monthly standard deviation of the difference between daily monitored and predicted PM2.5 for every 1 km × 1 km grid cell. This PM2.5 prediction dataset, including the downscaled and uncertainty predictions, allows epidemiologists to accurately estimate the adverse health effect of PM2.5. Compared with model performance of individual base learners, an ensemble model would achieve a better overall estimation. It is worth exploring other ensemble model formats to synthesize estimations from different models or from different groups to improve overall performance.

Regional Estimates of Chemical Composition of Fine Particulate Matter Using a Combined Geoscience-Statistical Method with Information from Satellites, Models, and Monitors

An accurate fine-resolution surface of the chemical composition of fine particulate matter (PM_2.5) would offer valuable information for epidemiological studies and health impact assessments. We develop geoscience-derived estimates of PM_2.5 composition from a chemical transport model (GEOS-Chem) and satellite observations of aerosol optical depth, and statistically fuse these estimates with ground-based observations using a geographically weighted regression over North America to produce a spatially complete representation of sulfate, nitrate, ammonium, black carbon, organic matter, mineral dust, and sea-salt over 2000-2016. Significant long-term agreement is found with cross-validation sites over North America (R² = 0.57-0.96), with the strongest agreement for sulfate (R² = 0.96), nitrate (R² = 0.90), and ammonium (R² = 0.86). We find that North American decreases in population-weighted fine particulate matter (PM_2.5) concentrations since 2000 have been most heavily influenced by regional changes in sulfate and organic matter. Regionally, the relative importance of several chemical components are found to change with PM_2.5 concentration, such as higher PM_2.5 concentrations having a larger proportion of nitrate and a smaller proportion of sulfate. This data set offers information for research into the health effects of PM_2.5 chemical components.

Assessing PM2.5 Exposures with High Spatiotemporal Resolution across the Continental United States

A number of models have been developed to estimate PM2.5 exposure, including satellite-based aerosol optical depth (AOD) models, land-use regression, or chemical transport model simulation, all with both strengths and weaknesses. Variables like normalized difference vegetation index (NDVI), surface reflectance, absorbing aerosol index, and meteoroidal fields are also informative about PM2.5 concentrations. Our objective is to establish a hybrid model which incorporates multiple approaches and input variables to improve model performance. To account for complex atmospheric mechanisms, we used a neural network for its capacity to model nonlinearity and interactions. We used convolutional layers, which aggregate neighboring information, into a neural network to account for spatial and temporal autocorrelation. We trained the neural network for the continental United States from 2000 to 2012 and tested it with left out monitors. Ten-fold cross-validation revealed a good model performance with a total R(2) of 0.84 on the left out monitors. Regional R(2) could be even higher for the Eastern and Central United States. Model performance was still good at low PM2.5 concentrations. Then, we used the trained neural network to make daily predictions of PM2.5 at 1 km × 1 km grid cells. This model allows epidemiologists to access PM2.5 exposure in both the short-term and the long-term.

Global Estimates of Fine Particulate Matter using a Combined Geophysical-Statistical Method with Information from Satellites, Models, and Monitors

We estimated global fine particulate matter (PM2.5) concentrations using information from satellite-, simulation- and monitor-based sources by applying a Geographically Weighted Regression (GWR) to global geophysically based satellite-derived PM2.5 estimates. Aerosol optical depth from multiple satellite products (MISR, MODIS Dark Target, MODIS and SeaWiFS Deep Blue, and MODIS MAIAC) was combined with simulation (GEOS-Chem) based upon their relative uncertainties as determined using ground-based sun photometer (AERONET) observations for 1998-2014. The GWR predictors included simulated aerosol composition and land use information. The resultant PM2.5 estimates were highly consistent (R(2) = 0.81) with out-of-sample cross-validated PM2.5 concentrations from monitors. The global population-weighted annual average PM2.5 concentrations were 3-fold higher than the 10 μg/m(3) WHO guideline, driven by exposures in Asian and African regions. Estimates in regions with high contributions from mineral dust were associated with higher uncertainty, resulting from both sparse ground-based monitoring, and challenging conditions for retrieval and simulation. This approach demonstrates that the addition of even sparse ground-based measurements to more globally continuous PM2.5 data sources can yield valuable improvements to PM2.5 characterization on a global scale.

High-resolution satellite-based analysis of ground-level PM2.5 for the city of Montreal

Satellite remote sensing offers the opportunity to determine the spatial distribution of aerosol properties and could fill the gap of ground-level observations. Various algorithms have recently been developed in order to retrieve the aerosol optical depth (AOD) at continental scales. However, they are, to some extent, subject to coarse spatial resolutions which are not appropriate for intraurban scales as usually needed in health studies. This paper presents an improved AOD retrieval algorithm for satellite instrument MODIS at 1-km resolution for intraurban scales. The MODIS-retrieved AODs are used to derive the ground-level PM2.5 concentrations using the aerosol vertical profiles and local scale factors obtained from the GEOS-Chem model simulation. The developed method has been applied to retrieve the AODs and to evaluate the ground-level PM2.5 over the city of Montreal, Canada for 2009 on daily, monthly and annual scales. The daily and monthly results are compared with the monitoring values with correlations R(2) ranging from 0.86 to 0.93. Especially, the annual mean PM2.5 concentrations are in good agreement with the measurement values at all monitoring stations (r=0.96, slope=1.0132 ± 0.0025, intercept=0.5739 ± 0.0013). This illustrates that the developed AOD retrieval algorithm can be used to retrieve AODs at a higher spatial resolution than previous studies to further derive the regional full coverage PM2.5 results at finer spatial and temporal scales. The study results are useful in health risk assessment across this region.

High-Resolution Satellite-Derived PM2.5 from Optimal Estimation and Geographically Weighted Regression over North America

We used a geographically weighted regression (GWR) statistical model to represent bias of fine particulate matter concentrations (PM2.5) derived from a 1 km optimal estimate (OE) aerosol optical depth (AOD) satellite retrieval that used AOD-to-PM2.5 relationships from a chemical transport model (CTM) for 2004-2008 over North America. This hybrid approach combined the geophysical understanding and global applicability intrinsic to the CTM relationships with the knowledge provided by observational constraints. Adjusting the OE PM2.5 estimates according to the GWR-predicted bias yielded significant improvement compared with unadjusted long-term mean values (R(2) = 0.82 versus R(2) = 0.62), even when a large fraction (70%) of sites were withheld for cross-validation (R(2) = 0.78) and developed seasonal skill (R(2) = 0.62-0.89). The effect of individual GWR predictors on OE PM2.5 estimates additionally provided insight into the sources of uncertainty for global satellite-derived PM2.5 estimates. These predictor-driven effects imply that local variability in surface elevation and urban emissions are important sources of uncertainty in geophysical calculations of the AOD-to-PM2.5 relationship used in satellite-derived PM2.5 estimates over North America, and potentially worldwide.

High-resolution MODIS aerosol retrieval during wildfire events in California for use in exposure assessment

Retrieval of aerosol optical depth (AOD) from the Moderate Resolution Imaging Spectroradiometer (MODIS) using the Collection 5 (C005) algorithm provides large-scale (10 × 10 km) estimates that can be used to predict surface layer concentrations of particulate matter with aerodynamic diameter smaller than 2.5 μm (PM_2.5). However, these large-scale estimates are not suitable for identifying intraurban variability of surface PM_2.5 concentrations during wildfire events when individual plumes impact populated areas. We demonstrate a method for providing high-resolution (2.5 km) kernel-smoothed estimates of AOD over California during the 2008 northern California fires. The method uses high-resolution surface reflectance ratios of the 0.66 and 2.12 μm channels, a locally derived aerosol optical model characteristic of fresh wildfire plumes, and a relaxed cloud filter. Results show that the AOD derived for the 2008 northern California fires outperformed the standard product in matching observed aerosol optical thickness at three coastal Aerosol Robotic Network sites and routinely explained more than 50% of the variance in hourly surface PM_2.5 concentrations observed during the wildfires.

Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application

BACKGROUND

Epidemiologic and health impact studies of fine particulate matter with diameter < 2.5 microm (PM2.5) are limited by the lack of monitoring data, especially in developing countries. Satellite observations offer valuable global information about PM2.5 concentrations.

OBJECTIVE

In this study, we developed a technique for estimating surface PM2.5 concentrations from satellite observations.

METHODS

We mapped global ground-level PM2.5 concentrations using total column aerosol optical depth (AOD) from the MODIS (Moderate Resolution Imaging Spectroradiometer) and MISR (Multiangle Imaging Spectroradiometer) satellite instruments and coincident aerosol vertical profiles from the GEOS-Chem global chemical transport model.

RESULTS

We determined that global estimates of long-term average (1 January 2001 to 31 December 2006) PM2.5 concentrations at approximately 10 km x 10 km resolution indicate a global population-weighted geometric mean PM2.5 concentration of 20 microg/m3. The World Health Organization Air Quality PM2.5 Interim Target-1 (35 microg/m3 annual average) is exceeded over central and eastern Asia for 38% and for 50% of the population, respectively. Annual mean PM2.5 concentrations exceed 80 microg/m3 over eastern China. Our evaluation of the satellite-derived estimate with ground-based in situ measurements indicates significant spatial agreement with North American measurements (r = 0.77; slope = 1.07; n = 1057) and with noncoincident measurements elsewhere (r = 0.83; slope = 0.86; n = 244). The 1 SD of uncertainty in the satellite-derived PM2.5 is 25%, which is inferred from the AOD retrieval and from aerosol vertical profile errors and sampling. The global population-weighted mean uncertainty is 6.7 microg/m3.

CONCLUSIONS

Satellite-derived total-column AOD, when combined with a chemical transport model, provides estimates of global long-term average PM2.5 concentrations.