Limited research has assessed the spatio-temporal distribution and chronic health effects of NO2 exposure, especially in developing countries, due to the lack of historical NO2 data. A gap-filling model was first adopted to impute the missing NO2 column densities from satellite, then an ensemble machine learning model incorporating three base learners was developed to estimate the spatiotemporal pattern of monthly mean NO2 concentrations at 0.05° spatial resolution from 2005 to 2020 in China. Further, we applied the exposure dataset with epidemiologically derived exposure response relations to estimate the annual NO2 associated mortality burdens in China. The coverage of satellite NO2 column densities increased from 46.9% to 100% after gap-filling. The ensemble model predictions had good agreement with observations, and the overall, temporal and spatial cross-validation (CV) R2 were 0.88, 0.82 and 0.73, respectively. In addition, our model can provide accurate historical NO2 concentrations, with both by-year CV R2 and external separate year validation R2 achieving 0.80. The estimated national NO2 levels showed a increasing trend during 2005-2011, then decreased gradually until 2020, especially in 2012-2015. The estimated annual mortality burden attributable to long-term NO2 exposure ranged from 305 thousand to 416 thousand, and varied considerably across provinces in China. This satellite-based ensemble model could provide reliable long-term NO2 predictions at a high spatial resolution with complete coverage for environmental and epidemiological studies in China. Our results also highlighted the heavy disease burden by NO2 and call for more targeted policies to reduce the emission of nitrogen oxides in China.

Estimating surface-level fine particulate matter from satellite remote sensing data can expand the spatial coverage of ground-based monitors. This approach is particularly effective in assessing rapidly changing air pollution events such as wildland fires that often start far away from centralized ground monitors. We developed Deep Neural Network algorithm to bias correct hourly PM2.5 levels informed by GOES-R satellites, NOAA meteorology forecasts, and real-time PM2.5 observations from the Environmental Protection Agency (EPA) via AirNow. The surface-satellite-model collocated datasets for the period of 2020-2021 was used to assess the biases in GOES-GWR PM2.5 against AirNow measurements at hourly and daily scales. Then a deep neural network (DNN) based bias correction algorithm is used to improve the accuracies of GOES-GWR PM2.5. The DNN uses GOES-GWR PM2.5, GOES-R aerosol parameters, and HRRR meteorological fields as input and AirNow PM2.5 is used as target variable. The application of DNN reduced the PM2.5 biases as compared to GOES-GWR estimates. RMSE was also reduced to 6.55 µg/m3 from 8.72 µg/m3 in GOES-GWR estimates. The DNN model was also evaluated on two sets of independent datasets for its robustness. In the first independent dataset for the first half of 2020, ~89% of stations show an increase in correlation (r) and, ~76% and ~62% of stations show a reduction in bias. The IOA and r for the independent data were 0.77 and 0.61 (GWR: 0.68 and 0.53) and RMSE was 4.48 µg/m3 (GWR=6.13 µg/m3) for the same period.

Introduction Pathophysiologic pathways of sickle cell disease (SCD) and air pollution involve inflammation, oxidative stress, and endothelial damage. It is therefore plausible that children with SCD are especially prone to air pollution’s harmful effects. Methods Patient data were collected from a single center, urban/peri-urban cohort of children with confirmed SCD. Daily ambient concentrations of particulate matter (PM 2.5) were collected via satellite-derived remote-sensing technology, and carbon monoxide (CO), nitrogen dioxide (NO 2), and ozone from local monitoring stations. We used multivariable regression to quantify associations of pollutant levels and daily counts of emergency department (ED) visits, accounting for weather and time trends. For comparison, we quantified the associations of pollutant levels with daily all-patient (non-SCD) ED visits to our center. Results From 2010-2018, there were 17 731 ED visits by 1740 children with SCD (64.8% HbSS/HbSβ 0). Vaso-occlusive events (57.8%), respiratory illness (17.1%), and fever (16.1%) were the most common visit diagnoses. Three-day (lags 0-2) rolling mean PM 2.5 and CO levels were associated with daily ED visits among those with SCD (PM 2.5 incident rate ratio (IRR) 1.051 (95% CI 1.010-1.094) per 9.4 µg/m 3 increase; CO 1.088 (1.045-1.132) per 0.5 ppm). NO 2 showed positive associations in secondary analyses; ozone levels were not associated with ED visits. The comparison, all-patient ED visit analyses showed lower IRR for all pollutants. Conclusions Our results suggest short-term air pollution levels as triggers for SCD events and that children with SCD may be more vulnerable to air pollution than those without SCD. Targeted pollution-avoidance strategies could have significant clinical benefits in this population.

Low-cost air quality monitors (LCAQMs) are promising supplements to regulatory monitors for PM2.5 exposure assessment. However, the application of LCAQM in spatially extensive exposure modeling is hindered by the difficulty in performing calibration at large spatial scales and the adverse influence of LCAQM residual uncertainty after calibration. We aimed to develop an efficient spatially scalable calibration method for LCAQM and design a residual uncertainty-derived down-weighting strategy to optimize the use of LCAQM data with regulatory monitoring data in PM2.5 modeling. In California, for each monitor from PurpleAir, a global LCAQM network, we identified a station within a 500-m radius from the Air Quality System (AQS), a U.S. regulatory monitoring network. Regional calibration of PurpleAir to AQS was performed at the hourly level with Geographically Weighted Regression (GWR). The calibrated PurpleAir measurements were down-weighted according to their residual uncertainty and then incorporated into a Random Forest (RF) prediction model as a dependent variable to generate 1-km daily PM2.5 exposure estimates. The state-level PurpleAir calibration reduced the systematic bias to ~0 ug/m3 and decreased the random error by 38%. The considerably large samples also enabled quantitative analyses regarding potential factors related to the PurpleAir bias. The RF-based model with both AQS and down-weighted PurpleAir data outperformed the RF model based solely on AQS with an improved CV R2 of 0.86, an improved spatial CV R2 of 0.81, and a lower prediction error of 5.40 ug/m3. The down-weighting allowed the prediction model to show more spatial details of PM2.5 and to better detect pollution hot-spots. Our spatially scalable calibration and down-weighting strategies, for the first time, allowed an effective application of a state-level LCAQM network in high-resolution PM2.5 exposure modeling. The proposed framework can be generalized to regions worldwide for advancing the evaluation of heavy PM2.5 episodes and health-related applications.