Search

Ground-level PM 2.5 data derived from satellites with machine learning are crucial for health and climate assessments. However, uncertainties persist due to the absence of spatially covered observations. To address this, we propose a novel testbed using nontraditional numerical simulations to evaluate PM 2.5 estimation across the entire spatial domain. The testbed emulates the general machine-learning approach by training the model with grids corresponding to ground monitoring sites and subsequently testing its predictive accuracy for other locations. Our approach enables comprehensive evaluation of various machine-learning methods' performance in estimating PM 2.5 across the spatial domain for the first time. Unexpected results are shown in the application in China, with larger absolute PM 2.5 biases found in densely populated regions with abundant ground observations across all benchmark models due to the higher baseline concentration, though the relative error (approximately 20 %) is smaller compared to that in rural areas (over 50 %). The imbalance in training samples, mostly from urban areas with high emissions, is the main reason, leading to significant overestimation due to the lack of monitors in downwind areas where PM 2.5 is transported from urban areas with varying vertical profiles. Our proposed testbed also provides an efficient strategy for optimizing model structure or training samples to enhance satellite-retrieval model performance. Integration of spatiotemporal features, especially with conventional neural network (CNN)-based deep-learning approaches like the residual neural network (ResNet) model, has successfully mitigated PM 2.5 overestimation (by 5–30 µ g m -3) and the corresponding exposure (by 3 million people ⋅ µ g m -3) in the downwind area over 9 years (2013–2021) compared to the traditional approach. Furthermore, the incorporation of 600 strategically positioned ground monitoring sites identified through the testbed is essential for achieving a more balanced distribution of training samples, thereby ensuring precise PM 2.5 estimation and facilitating the assessment of the associated impacts in China. In addition to presenting the retrieved surface PM 2.5 concentrations in China from 2013 to 2021, this study provides a testbed dataset derived from physical modeling simulations which can serve to evaluate the performance of data-driven methodologies, such as machine learning, in estimating spatial PM 2.5 concentrations for the community (Li et al., 2024a; https://doi.org/10.5281/zenodo.11122294). [ABSTRACT FROM AUTHOR]

Ground-level PM2.5 data derived from satellites with machine learning are crucial for health and climate assessments, however, uncertainties persist due to the absence of spatially covered observations. To address this, we propose a novel testbed using untraditional numerical simulations to evaluate PM2.5 estimation across the entire spatial domain. The testbed emulates the general machine-learning approach, by training the model with grids corresponding to ground monitor sites and subsequently testing its predictive accuracy for other locations. Our approach enables comprehensive evaluation of various machine-learning methods' performance in estimating PM2.5 across the spatial domain for the first time. Unexpected results are shown in the application in China, with larger PM2.5 biases found in densely populated regions with abundant ground observations across all benchmark models, challenging conventional expectations and are not explored in the recent literature. The imbalance in training samples, mostly from urban areas with high emissions, is the main reason, leading to significant overestimation due to the lack of monitors in downwind areas where PM2.5 is transported from urban areas with varying vertical profiles. Our proposed testbed also provides an efficient strategy for optimizing model structure or training samples to enhance satellite-retrieval model performance. Integration of spatiotemporal features, especially with CNN-based deep-learning approaches like the ResNet model, successfully mitigates PM2.5 overestimation (by 5–30 µg m-3) and corresponding exposure (by 3 million people • µg m-3) in the downwind area over the past nine years (2013–2021) compared to the traditional approach. Furthermore, the incorporation of 600 strategically positioned ground-measurement sites identified through the testbed is essential to achieve a more balanced distribution of training samples, thereby ensuring precise PM2.5 estimation and facilitating the assessment of associated impacts in China. In addition to presenting the retrieved surface PM2.5 concentrations in China from 2013 to 2021, this study provides a testbed dataset derived from physical modeling simulations which can serve to evaluate the performance of data-driven methodologies, such as machine learning, in estimating spatial PM2.5 concentrations for the community. [ABSTRACT FROM AUTHOR]

Accurate estimation of atmospheric chemical concentrations from multiple observations is crucial for assessing the health effects of air pollution. However, existing methods are limited by imbalanced samples from observations. Here, we introduce a novel deep-learning model-measurement fusion method (DeepMMF) constrained by physical laws inferred from a chemical transport model (CTM) to estimate NO 2 concentrations over the Continental United States (CONUS). By pretraining with spatiotemporally complete CTM simulations, fine-tuning with satellite and ground measurements, and employing a novel optimization strategy for selecting proper prior emission, DeepMMF delivers improved NO 2 estimates, showing greater consistency and daily variation alignment with observations (with NMB reduced from -0.3 to -0.1 compared to original CTM simulations). More importantly, DeepMMF effectively addressed the sample imbalance issue that causes overestimation (by over 100%) of downwind or rural concentrations in other methods. It achieves a higher R 2 of 0.98 and a lower RMSE of 1.45 ppb compared to surface NO 2 observations, overperforming other approaches, which show R 2 values of 0.4-0.7 and RMSEs of 3-6 ppb. The method also offers a synergistic advantage by adjusting corresponding emissions, in agreement with changes (-10% to -20%) reported in the NEI between 2019 and 2020. Our results demonstrate the great potential of DeepMMF in data fusion to better support air pollution exposure estimation and forecasting.

Despite emerging evidence on the health impacts of fine particulate matter (PM 2.5 ) from wildland fire smoke, the specific effects of PM 2.5 composition on health outcomes remain uncertain. We developed a three-level, chemical transport model-based framework to estimate daily full-coverage concentrations of smoke-derived carbonaceous PM 2.5 , specifically Organic Carbon (OC) and Elemental Carbon (EC), at a 1 km 2 spatial resolution from 2002 to 2019 across the contiguous U.S. (CONUS) and Southern Canada (SC). Cross-validation demonstrated that the framework performed well at both the daily and monthly levels. Modeling results indicated that increases in wildland fire smoke have offset approximately one-third of the improvements in background air quality. In recent years, wildland fire smoke has become more frequent and carbonaceous PM 2.5 concentrations have intensified, especially in the Western CONUS and Southwestern Canada. Smoke exposure is also occurring earlier throughout the year, leading to more population being exposed. We estimated that long-term exposure to fire smoke carbonaceous PM 2.5 is responsible for 7,462 and 259 non-accidental deaths annually in the CONUS and SC, respectively, with associated annual monetized damage of 68.4 billion USD for the CONUS and 1.97 billion CAD for SC. The Southeastern CONUS, where prescribed fires are prevalent, contributed most to these health impacts and monetized damages. Given the challenges posed by climate change for managing prescribed and wildland fires, our findings offer critical insights to inform policy development and assess future health burdens associated with fire smoke exposure., Competing Interests: Competing Interest Statement: The authors declare no conflict of interests.