Ozone Prediction Research Articles

Machine Learning‐Driven Spatiotemporal Analysis of Ozone Exposure and Health Risks in China

AbstractAccurate and fine‐scaled prediction of ozone concentrations across space and time, as well as the assessment of associated human risks, is crucial for protecting public health and promoting environmental conservation. This paper introduces NetGBM, an innovative machine‐learning model designed to comprehensively model ozone levels across China's diverse topography and analyze the spatiotemporal distribution of ozone and exposure. Our model focuses on daily, weekly, and monthly predictions, achieving commendable coefficients of 0.83, 0.77, and 0.79, respectively. By constructing a gridded map of ozone and incorporating both land use and meteorological features into each grid, we achieved ozone prediction at a high spatiotemporal resolution, outperforming previous research in terms of performance and scale, particularly in regions with limited monitoring stations. The results can be further improved when applied to regional research using meteorological and ozone data from regional stations. Additionally, our research revealed that temperature is the most significant factor affecting ozone concentrations across China. In health risk assessment, we retrieved a high‐resolution spatial distribution of ozone‐attributed mortality for 5‐COD and daily ozone inhalation distributions during our study period. We concluded that ozone‐attributed mortality is predominantly caused by stroke and IHD, accounting for more than 70% of the total deaths in 2021, with the highest mortality rates in developed urban areas such as the NCP and the YRD. Our experiment demonstrated the potential of NetGBM in robustly modeling ozone across China with high spatiotemporal resolution and its applicability in measuring associated health risks.

A deep learning model integrating a wind direction-based dynamic graph network for ozone prediction

Ozone pollution is an important environmental issue in many countries. Accurate forecasting of ozone concentration enables relevant authorities to enact timely policies to mitigate adverse impacts. This study develops a novel hybrid deep learning model, named wind direction-based dynamic spatio-temporal graph network (WDDSTG-Net), for hourly ozone concentration prediction. The model uses a dynamic directed graph structure based on hourly changing wind direction data to capture evolving spatial relationships between air quality monitoring stations. It applied the graph attention mechanism to compute dynamic weights between connected stations, thereby aggregating neighborhood information adaptively. For temporal modeling, it utilized a sequence-to-sequence model with attention mechanism to extract long-range temporal dependencies. Additionally, it integrated meteorological predictions to guide the ozone forecasting. The model achieves a mean absolute error of 6.69 μg/m3 and 18.63 μg/m3 for 1-h prediction and 24-h prediction, outperforming several classic models. The model's IAQI accuracy predictions at all stations are above 75 %, with a maximum of 81.74 %. It also exhibits strong capabilities in predicting severe ozone pollution events, with a 24-h true positive rate of 0.77. Compared to traditional static graph models, WDDSTG-Net demonstrates the importance of incorporating short-term wind fluctuations and transport dynamics for data-driven air quality modeling. In principle, it may serve as an effective data-driven approach for the concentration prediction of other airborne pollutants.

Innovative approaches for accurate ozone prediction and health risk analysis in South Korea: The combined effectiveness of deep learning and AirQ+

Short-term exposure to ground-level ozone (O3) poses significant health risks, particularly respiratory and cardiovascular diseases, and mortality. This study addresses the pressing need for accurate O3 forecasting to mitigate these risks, focusing on South Korea. We introduce Deep Bias Correction (Deep-BC), a novel framework leveraging Convolutional Neural Networks (CNNs), to refine hourly O3 forecasts from the Community Multiscale Air Quality (CMAQ) model. Our approach involves training Deep-BC using data from 2016 to 2019, including CMAQ's 72-hour O3 forecasts, 31 meteorological variables from the Weather Research and Forecasting (WRF) model, and previous days' station measurements of 6 air pollutants. Deep-BC significantly outperforms CMAQ in 2021, reducing biases in O3 forecasts. Furthermore, we utilize Deep-BC's daily maximum 8-hour average O3 (MDA8 O3) forecasts as input for the AirQ+ model to assess O3's potential impact on mortality across seven major provinces of South Korea: Seoul, Busan, Daegu, Incheon, Daejeon, Ulsan, and Sejong. Short-term O3 exposure is associated with 0.40 % to 0.48 % of natural cause and respiratory deaths and 0.67 % to 0.81 % of cardiovascular deaths. Gender-specific analysis reveals higher mortality rates among men, particularly from respiratory causes. Our findings underscore the critical need for region-specific interventions to address air pollution's detrimental effects on public health in South Korea. By providing improved O3 predictions and quantifying its impact on mortality, this research offers valuable insights for formulating targeted strategies to mitigate air pollution's adverse effects. Moreover, we highlight the urgency of proactive measures in health policies, emphasizing the significance of accurate forecasting and effective interventions to safeguard public health from the deleterious effects of air pollution.

Outdoor air pollution and risk of incident adult haematologic cancer subtypes in a large US prospective cohort

BackgroundOutdoor air pollution and particulate matter (PM) are classified as Group 1 human carcinogens for lung cancer. Pollutant associations with haematologic cancers are suggestive, but these cancers are aetiologically heterogeneous and sub-type examinations are lacking.MethodsThe American Cancer Society Cancer Prevention Study-II Nutrition Cohort was used to examine associations of outdoor air pollutants with adult haematologic cancers. Census block group level annual predictions of particulate matter (PM2.5, PM10, PM10-2.5), nitrogen dioxide (NO2), ozone (O3), sulfur dioxide (SO2), and carbon monoxide (CO) were assigned with residential addresses. Hazard ratios (HR) and 95% confidence intervals (CI) between time-varying pollutants and haematologic subtypes were estimated.ResultsAmong 108,002 participants, 2659 incident haematologic cancers were identified from 1992–2017. Higher PM10-2.5 concentrations were associated with mantle cell lymphoma (HR per 4.1 μg/m3 = 1.43, 95% CI 1.08–1.90). NO2 was associated with Hodgkin lymphoma (HR per 7.2 ppb = 1.39; 95% CI 1.01–1.92) and marginal zone lymphoma (HR per 7.2 ppb = 1.30; 95% CI 1.01–1.67). CO was associated with marginal zone (HR per 0.21 ppm = 1.30; 95% CI 1.04–1.62) and T-cell (HR per 0.21 ppm = 1.27; 95% CI 1.00–1.61) lymphomas.ConclusionsThe role of air pollutants on haematologic cancers may have been underestimated previously because of sub-type heterogeneity.

Elucidating contributions of volatile organic compounds to ozone formation using random forest during COVID-19 pandemic: A case study in China

Ozone has been reported to increase despite nitrogen oxides reductions during the COVID-19 pandemic, and ozone formation needs to be revisited using volatile organic compounds (VOCs), which are rarely measured during the pandemic. Here, a total of 98 VOCs species were monitored in an economy-active city in China from January 2021 to August 2022 to assess contributions to ozone formation during the pandemic. Total VOCs concentrations were 35.55 ± 21.47 ppb during the entire period, among which alkanes account for the largest fraction (13.78 ppb, 38.0%), followed by aromatics (6.16 ppb, 16.8%) and oxygenated VOCs (OVOCs, 5.69 ppb, 15.7%). Most VOCs groups (e.g., alkenes, OVOCs) and individual species (e.g., isoprene, methyl vinyl ketone) display obvious seasonal and diurnal variations, which are related to their sources and reactivities. No weekend effects of VOCs suggest limited influences from traffic emissions during pandemic. Aromatics and alkenes are the major contributors (39% and 33%) to ozone formation potential, largely driven by o/m/p-xylene (21%), ethylene (15%), toluene (9%). Secondary organic aerosol formation potential is dominated by toluene (>50%) despite its low proportion (5%). Further inclusion of VOCs and meteorology in the Random Forest model shows good ozone prediction performance (R2 = 0.77–0.86, RMSE = 11.95–19.91 μg/m3, MAE = 8.89–14.58 μg/m3). VOCs and NO2 contribute >50% of total importance with the largest difference in importance ratio of VOCs/NO2 in the summer and winter, implying ozone formation regime may vary. No seasonal variations in importance of meteorology are observed, while importance of other variables (e.g., PM2.5) is highest in the summer. This work identifies critical VOCs groups and species for ozone formation during the pandemic, and demonstrates the feasibility of machine learning algorithms in elucidation of ozone formation mechanisms.

Current status of model predictions of volatile organic compounds and impacts on surface ozone predictions during summer in China

Abstract. Volatile organic compounds (VOCs) play a crucial role in the formation of tropospheric ozone (O3) and secondary organic aerosols. VOC emissions are generally considered to have larger uncertainties compared to other pollutants, such as sulfur dioxide and fine particulate matter (PM2.5). Although predictions of O3 and PM2.5 have been extensively evaluated in air quality modeling studies, there has been limited reporting on the evaluation of VOCs, mainly due to a lack of routine VOC measurements at multiple sites. In this study, we utilized VOC measurements from the “Towards an Air Toxic Management System in China” (ATMSYC) project at 28 sites across China and assessed the predicted VOC concentrations using the Community Multiscale Air Quality (CMAQ) model with the widely used Multi-resolution Emission Inventory for China (MEIC). The ratio of predicted to observed total VOCs was found to be 0.74 ± 0.40, with underpredictions ranging from 2.05 to 50.61 ppbv (5.77 % to 85.40 %) at 24 sites. A greater bias in VOC predictions was observed in industrial cities in the north and southwest, such as Jinan, Shijiazhuang, Lanzhou, Chengdu, and Guiyang. In terms of different VOC components, alkanes, alkenes, non-naphthalene aromatics (ARO2MN), alkynes, and formaldehyde (HCHO) had prediction-to-observation ratios of 0.53 ± 0.38, 0.51 ± 0.48, 0.31 ± 0.38, 0.41 ± 0.47, and 1.21 ± 1.61, respectively. Sensitivity experiments were conducted to assess the impact of the VOC prediction bias on O3 predictions. While emission adjustments improved the model performance for VOCs, resulting in a change in the ratio of total VOCs to 0.86 ± 0.47, they also exacerbated O3 overprediction relative to the base case by 0.62 % to 6.27 % across the sites. This study demonstrates that current modeling setups and emission inventories are likely to underpredict VOC concentrations, and this underprediction of VOCs contributes to lower O3 predictions in China.

The Capability of Deep Learning Model to Predict Ozone Across Continents in China, the United States and Europe

AbstractData‐driven methods have been extensively applied to predict atmospheric compositions. Here, we explore the capability of a deep learning (DL) model to make ozone (O3) predictions across continents in China, the United States (US) and Europe. The DL model was trained and validated with surface O3 observations in China and the US in 2015–2018. The DL model was applied to predict hourly surface O3 over three continents in 2015–2022. Compared to baseline simulations using GEOS‐Chem (GC) model, our analysis exhibits mean biases of 2.6 and 4.8 μg/m3 with correlation coefficients of 0.94 and 0.93 (DL); and mean biases of 3.7 and 5.4 μg/m3 with correlation coefficients of 0.95 and 0.92 (GC) in Europe in 2015–2018 and 2019–2022, respectively. The comparable performances between DL and GC indicate the potential of DL to make reliable predictions over spatial and temporal domains where a wealth of local observations for training is not available.

A comparative analysis of linear regression, neural networks and random forest regression for predicting air ozone employing soft sensor models

The proposed methodology presents a comprehensive analysis of soft sensor modeling techniques for air ozone prediction. We compare the performance of three different modeling techniques: LR (linear regression), NN (neural networks), and RFR (random forest regression). Additionally, we evaluate the impact of different variable sets on prediction performance. Our findings indicate that neural network models, particularly the RNN (recurrent neural networks), outperform the other modeling techniques in terms of prediction accuracy. The proposed methodology evaluates the impact of different variable sets on prediction performance, finding that variable set E demonstrates exceptional performance and achieves the highest average prediction accuracy among various software sensor models. In comparing variable set E and A, B, C, D, it is observed that the inclusion of an additional input feature, PM10, in the latter sets does not improve overall performance, potentially due to multicollinearity between PM10 and PM2.5 variables. The proposed methodology provides valuable insights into soft sensor modeling for air ozone prediction.Among the 72 sensors, sensor NNR[Y]C outperforms all other evaluated sensors, demonstrating exceptional predictive performance with an impressive R2 of 0.8902, low RMSE of 24.91, and remarkable MAE of 19.16. With a prediction accuracy of 81.44%, sensor NNR[Y]C is reliable and suitable for various technological applications.

Improved O3 predictions in China by combining chemical transport model and multi-source data with machining learning techniques

Accurate ozone (O3) predictions is crucial for assessing its impact on public health and developing effective prevention and control measures. While ground-based observations are considered highly accurate, the limited and uneven spatial coverage poses challenges. Air quality models can provide complete spatiotemporal coverage but are subject to biases due to simplifications in physicochemical mechanisms and uncertainties associated with emission and meteorology inputs. In this study, we aimed to improve the accuracy of the predicted O3 concentrations from the Community Multiscale Air Quality (CMAQ) model by using the machine learning techniques. First, we compared three machine learning algorithms, namely Light Gradient Boosting Machine (LightGBM), Random Forest, and eXtreme Gradient Boosting. Results showed that LightGBM exhibited the highest correlation coefficient (R) of 0.84, making it the preferred algorithm for further analysis. Subsequently, two multi-source data prediction models based on LightGBM were constructed to improve the accuracy of predicted daily maximum 8-h O3 (O3-Max8h). The first model, referred to as LGBR, utilized the predicted air pollutant concentrations and meteorological fields as input variables, while the second model, named LGBR-CHAP, incorporated ChinaHighAirPollutants (CHAP) O3-Max8h as an additional input variable. Validation results demonstrated significant improvements in the LGBR and LGBR-CHAP models compared to the original CMAQ model. On the daily scale, the root mean square error and mean bias of the predicted O3-Max8h decreased by 3.15 and 2.07 μg/m3, respectively, for the LGBR model, while they decreased by 5.61 and 4.18 μg/m3, respectively, for the LGBR-CHAP model. On the monthly scale, the R of the original CMAQ model varied from 0.2 in the March to 0.91 in June, and LGBR improved all months (0.4–0.92) as did LGBR-CHAP (0.5–0.94). Spatially, the O3-Max8h simulated by the original CMAQ model is better in eastern China but less skillful in western China. After optimization by the LGBR and LGBR-CHAP models, the national average R improved from 0.77 to 0.83 and 0.88, respectively. The LGBR-CHAP model exhibited better predictive capacity than the LGBR model and was subsequently employed to generate high-resolution (10 km × 10 km) and full-coverage (100%) O3-Max8h data. This dataset will prove valuable for future air pollution and health studies.

Nighttime ozone in the lower boundary layer: insights from 3-year tower-based measurements in South China and regional air quality modeling

Abstract. Nighttime ozone in the lower boundary layer regulates atmospheric chemistry and surface ozone air quality, but our understanding of its vertical structure and impact is largely limited by the extreme sparsity of direct measurements. Here we present 3-year (2017–2019) measurements of ozone in the lower boundary layer (up to 500 m) from the Canton Tower in Guangzhou, the core megacity in South China, and interpret the measurements with a 1-month high-resolution chemical simulation from the Community Multiscale Air Quality (CMAQ) model. Measurements are available at 10, 118, 168, and 488 m, with the highest (488 m) measurement platform higher than the typical height of the nighttime stable boundary layer that allows direct measurements of ozone in the nighttime residual layer (RL). We find that ozone increases with altitude in the lower boundary layer throughout the day, with a vertical ozone gradient between the 10 and 488 m heights (ΔO3/ΔH10–488 m) of 3.6–6.4 ppbv hm−1 in nighttime and 4.4–5.8 ppbv hm−1 in daytime. We identify a high ozone residual ratio, defined as the ratio of ozone concentration averaged over nighttime to that in the afternoon (14:00–17:00 LT), of 69 %–90 % in January, April, and October, remarkably higher than that in the other three layers (29 %–51 %). Ozone in the afternoon convective mixing layer provides the source of ozone in the RL, and strong temperature inversion facilitates the ability of RL to store ozone from the daytime convective mixing layer. The tower-based measurement also indicates that the nighttime surface Ox (Ox= O3+NO2) level can be an effective indicator of RL ozone if direct measurement is not available. We further find significant influences of nocturnal RL ozone on both the nighttime and the following day's daytime surface ozone air quality. During the surface nighttime ozone enhancement (NOE) event, we observe a significant decrease in ozone and an increase in NO2 and CO at the 488 m height, in contrast to their changes at the surface, a typical feature of enhanced vertical mixing. The enhanced vertical mixing leads to an NOE event by introducing ozone-rich and NOx-poor air into the RL to enter the nighttime stable boundary layer. The CMAQ model simulations also demonstrate an enhanced positive contribution of vertical diffusion (ΔVDIF) to ozone at the 10 and 118 m heights and a negative contribution at the 168 and 488 m heights during the NOE event. We also observe a strong correlation between nighttime RL ozone and the following day's surface maximum daily 8 h average (MDA8) ozone. This is tied to enhanced vertical mixing with the collapse of nighttime RL and the development of a convective mixing layer, which is supported by the CMAQ diagnosis of the ozone budget, suggesting that the mixing of ozone-rich air from nighttime RL downward to the surface via the entrainment is an important mechanism for aggravating ozone pollution the following day. We find that the bias in CMAQ-simulated surface MDA8 ozone the following day shows a strong correlation coefficient (r= 0.74) with the bias in nighttime ozone in the RL, highlighting the necessity to correct air quality model bias in the nighttime RL ozone for accurate prediction of daytime ozone. Our study thus highlights the value of long-term tower-based measurements for understanding the coupling between nighttime ozone in the RL, surface ozone air quality, and boundary layer dynamics.

Hybrid machine learning model for hourly ozone concentrations prediction and exposure risk assessment

With the development and improvement of optimization algorithms, machine learning models have become more important in air quality prediction. These models can be used to address problems associated with atmospheric environmental pollutants like ozone and facilitate the development of appropriate control policies. Firstly, the data sets of meteorological and pollution in Nanjing and four nearby cities from 2018 to 2021 were divided into four sections to build mixed models. The extreme gradient boosting (XGBoost) algorithm was improved by the permutation importance method (PIM) and Pearson correlation coefficient to acquire the most important features of the seasonal model and analysis of the effect on ozone prediction. The results reflect that the NO2, PM2.5, and PM10 parameters have an important influence, accounting for an average of 23% of prediction, and the average influence of the meteorological parameters of the solar radiation exceeds 11%. Secondly, the particle swarm optimization (PSO), grey wolf optimizer (GWO), and ant lion optimizer (ALO) algorithms were used to optimize the SVR model parameters to predict hourly ozone concentration in January, April, July, and September 2022. The R2 values are 0.95, 0.92, 0.85, and 0.84, respectively, and the RMSE values are 5.2, 12.4, 18.0 and 16.6 μg/m3, respectively. Finally, the impact of hourly ozone concentration on population health was studied by a risk assessment of the population density-weighted pollution exposure. The result shows that the influence on humans gradually decreased yearly from 2018 to 2021, the average variation from 2.6% to −1.1%, and health risks during the evening rush hours are still prevalent and require further attention.

The Influence of Meteorology Initialization on Ozone Forecasting in the Great Lakes Region during MOOSE Study

This study investigates the influence of meteorology initialization on surface ozone prediction in the Great Lakes region using Canada’s operational air quality model (GEM-MACH) at a 2.5 km horizontal resolution. Two different initialization techniques are compared, and it is found that the four-dimensional incremental analysis updating (IAU) method yields improved model performance for surface ozone prediction. The IAU run shows better ozone regression line statistics (y = 0.7x + 14.9, R2 = 0.2) compared to the non-IAU run (y = 0.6x + 23.1, R2 = 0.1), with improved MB and NMB values (3.9 ppb and 8.9%, respectively) compared to the non-IAU run (4.1 ppb and 9.3%). Furthermore, analyzing ozone prediction sensitivity to model initialization time reveals that the 18z initialization leads to enhanced performance, particularly during high ozone exceedance days, with an improved regression slope of 0.9 compared to 0.7 for the 00z and 12z runs. The MB also improves to −0.2 ppb in the 18z run compared to −2.8 ppb and −3.9 ppb for the 00z and 12z runs, respectively. The analysis of meteorological fields reveals that the improved ozone predictions at 18z are linked to a more accurate representation of afternoon wind speed. This improvement enhances the transport of ozone, contributing to the overall improvement in ozone predictions.

Sensitivity of northeastern US surface ozone predictions to the representation of atmospheric chemistry in the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMMv1.0).

Chemical mechanisms describe how emissions of gases and particles evolve in the atmosphere and are used within chemical transport models to evaluate past, current, and future air quality. Thus, a chemical mechanism must provide robust and accurate predictions of air pollutants if it is to be considered for use by regulatory bodies. In this work, we provide an initial evaluation of the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMMv1.0) by assessing CRACMMv1.0 predictions of surface ozone (O3) across the northeastern US during the summer of 2018 within the Community Multiscale Air Quality (CMAQ) modeling system. CRACMMv1.0 O3 predictions of hourly and maximum daily 8 h average (MDA8) ozone were lower than those estimated by the Regional Atmospheric Chemistry Mechanism with aerosol module 6 (RACM2_ae6), which better matched surface network observations in the northeastern US (RACM2_ae6 mean bias of +4.2 ppb for all hours and +4.3 ppb for MDA8; CRACMMv1.0 mean bias of +2.1 ppb for all hours and +2.7 ppb for MDA8). Box model calculations combined with results from CMAQ emission reduction simulations indicated a high sensitivity of O3 to compounds with biogenic sources. In addition, these calculations indicated the differences between CRACMMv1.0 and RACM2_ae6 O3 predictions were largely explained by updates to the inorganic rate constants (reflecting the latest assessment values) and by updates to the representation of monoterpene chemistry. Updates to other reactive organic carbon systems between RACM2_ae6 and CRACMMv1.0 also affected ozone predictions and their sensitivity to emissions. Specifically, CRACMMv1.0 benzene, toluene, and xylene chemistry led to efficient NO x cycling such that CRACMMv1.0 predicted controlling aromatics reduces ozone without rural O3 disbenefits. In contrast, semivolatile and intermediate-volatility alkanes introduced in CRACMMv1.0 acted to suppress O3 formation across the regional background through the sequestration of nitrogen oxides (NO x ) in organic nitrates. Overall, these analyses showed that the CRACMMv1.0 mechanism within the CMAQ model was able to reasonably simulate ozone concentrations in the northeastern US during the summer of 2018 with similar magnitude and diurnal variation as the current operational Carbon Bond (CB6r3_ae7) mechanism and good model performance compared to recent modeling studies in the literature.

Prediction of daytime and nighttime ground-level ozone using the hybrid regression models

Ozone is one of the major challenges for the air quality community due to its adverse impact on the environment and human health. This study seeks to improve the understanding of underlying mechanisms for several developed models for ozone prediction. We aim to establish a robust prediction model for ozone concentration up to the next four hours. Three years dataset including ozone (O3), nitrogen oxide (NOx), nitric oxide (NO), sulphur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), particulate matter (PM10, PM2.5), wind speed, solar radiation, temperature, and relative humidity (RH) were used in this study. The data were analyzed by using Multiple Linear Regression (MLR), Principal Component Regression (PCR), and Cluster-Multiple Linear Regression (CMLR) in predicting the next hours of O3 concentration. Results show that the MLR models executed high accuracy for O3t+1 (R2= 0.313), O3,t+2 (R2= 0.265), O3,t+3 (R2= 0.227) and O3,t+4 (R2= 0.217) as the best fitted-model. In conclusion, the MLR model is suitable for the next hour's O3 concentration prediction.

Development of a recurrent spatiotemporal deep-learning method coupled with data fusion for correction of hourly ozone forecasts

Ambient ozone (O3) predictions can be very challenging mainly due to the highly nonlinear photochemistry among its precursors, and meteorological conditions and regional transport can further complicate the O3 formation processes. The emission-based chemical transport models (CTM) are broadly used to predict O3 formation, but they may deviate from observations due to input uncertainties such as emissions and meteorological data, in addition to the treatment of O3 nonlinear chemistry. In this study, an innovative recurrent spatiotemporal deep-learning (RSDL) method with model-monitor coupled convolutional recurrent neural networks (ConvRNN) has been developed to improve O3 predictions of CTM. The RSDL method was first used to build the ConvRNN within a 24-h scale to characterize the spatiotemporal relationships between the monitored O3 data and CTM simulations, and then incorporated the recurrent pattern to achieve 72-h multi-site forecasts based on a pilot study over the Pearl River Delta (PRD) region of China. The results showed that the RSDL method predicted O3 with high accuracy over this case study, with an increase of 27.54% in the correlation coefficient (R) average for all sites as well as an increase in R of 0.14–0.21 for all cities compared to CTM. Moreover, the regional distribution of CTM was further improved by the RSDL predictions with the data fusion technique, which greatly reduced the underpredictions of O3 concentrations, particularly in high O3-level areas (concentrations >160 μg/m3), with a 33.55% reduction in the mean absolute error (MAE).

O3ResNet: A Deep Learning–Based Forecast System to Predict Local Ground-Level Daily Maximum 8-Hour Average Ozone in Rural and Suburban Environments

Abstract With the impact of tropospheric ozone pollution on humankind, there is a compelling need for robust air quality forecasts. Here, we introduce a novel deep learning (DL) forecasting system called O3ResNet that produces a 4-day forecast for ground-level ozone. O3ResNet is based on a convolutional neural network with residual blocks. The model has been trained on 22 yr of ozone and nitrogen oxides in situ measurements and ERA5 reanalysis data from 2000 to 2021 at 328 stations in central Europe located in rural and suburban environments. Our model outperforms the state-of-the-art Copernicus Atmosphere Monitoring Service regional forecast model ensemble for ground-level ozone with respect to the mean-square error and mean absolute error of the daily maximum 8-h running-average ozone, thus marking a major milestone for DL-based ozone prediction. O3ResNet has a very small bias without requiring additional postprocessing, and it generalizes well so that new stations can be added with no need to retrain the neural network. Because the model works on hourly data, it can be easily adapted to output other air quality metrics. We conclude that O3ResNet is sufficiently advanced and robust to become a test application for operational air quality forecasting with DL. Significance Statement In this paper, we introduce a novel deep learning approach to forecast ground-level ozone for rural and suburban environments on a local scale. Our model is able to outperform the state-of-the-art Copernicus Atmosphere Monitoring Service regional ensemble forecast and is a major milestone toward a more reliable ozone prediction. This is important because local-scale ozone forecasts using conventional methods show significant bias or require site-dependent postprocessing. The findings suggest that the model presented in this article can become an important tool for air quality prediction.

A New Hybrid Forecasting Model Based on Dual Series Decomposition with Long-Term Short-Term Memory

In recent years, ozone (O3) has gradually become the primary pollutant plaguing urban air quality. Accurate and efficient ozone prediction is of great significance to the prevention and control of ozone pollution. The air quality monitoring network provides multisource pollutant concentration monitoring data for ozone prediction, but ozone prediction based on multisource monitoring data still faces the challenges of each station’s series of data. Aiming at the problems of low prediction accuracy and low computational efficiency in traditional atmospheric ozone concentration prediction, ozone concentration prediction using dual series decomposition was proposed by variational mode decomposition (VMD), ensemble empirical mode decomposition (EEMD), and long short-term memory (LSTM). First, the historical data series of Nanjing air quality monitoring stations is decomposed by VMD, and then the EEMD algorithm is applied to the residual of VMD to obtain several characteristic intrinsic mode function (IMF) components; each characteristic IMF component is trained by LSTM to obtain the prediction result of each component, and then the final result can be obtained by linear superposition. The proposed method achieved the best results with R2 = 99%, MSE = 5.38, MAE = 4.54, and MAPE = 3.12. Because LSTM has strong adaptive learning ability and good memory function, it has the learning advantage of long-term memory for long-term data, and the prediction results are more accurate. According to the data, the proposed method is superior to the baseline models in terms of statistical metrics. As a result, the proposed hybrid method can serve as a reliable model for ozone forecasting.

Assessing uncertainty and heterogeneity in machine learning-based spatiotemporal ozone prediction in Beijing-Tianjin- Hebei region in China

Accurate prediction of spatiotemporal ozone concentration is of great significance to effectively establish advanced early warning systems and regulate air pollution control. However, the comprehensive assessment of uncertainty and heterogeneity in spatiotemporal ozone prediction remains unknown. Here, we systematically analyze the hourly and daily spatiotemporal predictive performances using convolutional long short term memory (ConvLSTM) and deep convolutional generative adversarial network (DCGAN) models over the Beijing-Tianjin-Hebei region in China from 2013 to 2018. In extensive scenarios, our results show that the machine learning-based (ML-based) models achieve better spatiotemporal ozone concentration prediction performance with multiple meteorological conditions. A further comparison to the air pollution model-Nested Air Quality Prediction Modelling System (NAQPMS) and monitoring observations, the ConvLSTM model demonstrates the practical feasibility of identifying high ozone concentration distribution and capturing spatiotemporal ozone variation patterns at a high spatial resolution (here 15 km × 15 km).

Machine learning based quantification of VOC contribution in surface ozone prediction

The prediction of surface ozone is essential attributing to its impact on human and environmental health. Volatile organic compounds (VOCs) are crucial in driving ozone concentration; particularly in urban areas where VOC limited regimes are prominent. The limited measurements of VOCs, however, hinder assessing the VOC-ozone relationship. This work applies machine learning (ML) algorithms for temporal forecasting of surface ozone over a metropolitan city in India. The availability of continuous VOCs measurement data along with meteorology and other pollutants during 2014–2016 makes it possible to deduce the influence of various input parameters on surface ozone prediction. After evaluating the best ML model for ozone prediction, simulations were carried out using varied input combinations. The combination with isoprene, meteorology, NOx, and CO (Isop + MNC) was the best with RMSE 4.41 ppbv and MAPE 6.77%. A season-wise comparison of simulations having all data, only meteorological data and Isop + MNC as input showed that Isop + MNC simulation gives the best results during the summer season (RMSE: 5.86 ppbv, MAPE: 7.05%). This shows the increased ability of the model to capture ozone peaks (high ozone during summer) relatively better when isoprene data is used. The overall results highlight that using all available data doesn't necessarily give best prediction results; also critical thinking is essential when evaluating the model results.

Improving the accuracy of O3 prediction from a chemical transport model with a random forest model in the Yangtze River Delta region, China

Due to inherent errors in the chemical transport models, inaccuracies in the input data, and simplified chemical mechanisms, ozone (O3) predictions are often biased from observations. Accurate O3 predictions can better help assess its impacts on public health and facilitate the development of effective prevention and control measures. In this study, we used a random forest (RF) model to construct a bias-correction model to correct the bias in the predictions of hourly O3 (O3-1h), daily maximum 8-h O3 (O3-Max8h), and daily maximum 1-h O3 (O3-Max1h) concentrations from the Community Multi-Scale Air Quality (CMAQ) model in the Yangtze River Delta region. The results show that the RF model successfully captures the nonlinear response relationship between O3 and its influence factors, and has an outstanding performance in correcting the bias of O3 predictions. The normalized mean biases (NMBs) of O3-1h, O3-Max8h, and O3-Max1h decrease from 15.8%, 20.0%, and 17.0.% to 0.5%, −0.8%, and 0.1%, respectively; correlation coefficients increase from 0.78, 0.90, and 0.89 to 0.94, 0.95, and 0.94, respectively. For O3-1h and O3-Max8h, the original CMAQ model shows an obvious bias in the central and southern Zhejiang region, while the RF model decreases the NMB values from 54% to −1% and 34% to −4%, respectively. The O3-1h bias is mainly caused by the bias of nitrogen dioxide (NO2). Relative humidity and temperature are also important factors that lead to the bias of O3. For high O3 concentrations, the temperature bias and O3 observations are the major reasons for the discrepancy between the model and the observations.