Forecasting PM2.5 Concentrations with Machine Learning: Accuracy, Efficiency, and Public Health Implications

Many countries are concerned about high particulate matter (PM) concentrations caused by rapid industrial development, which can harm both human health and the environment. To manage PM, the prediction of PM concentrations based on historical data is actively being conducted. Existing technologies for predicting PM mostly assess the model performance for the prediction of existing PM concentrations; however, PM must be forecast in advance, before it becomes highly concentrated and causes damage to the citizens living in the affected regions. Thus, it is necessary to conduct research on an index that can illustrate whether the PM concentration will increase or decrease. We developed a model that can predict whether the PM concentration might increase or decrease after a certain time, specifically for PM (fine PM) generated by anthropogenic volatile organic compounds. An algorithm that can select a model on an hourly basis, based on the long short-term memory (LSTM) and artificial neural network (ANN) models, was developed. The proposed algorithm exhibited a higher F1-score than the LSTM, ANN, or random forest models alone. The model developed in this study could be used to predict future regional PM concentration levels more effectively.

The objectives of the present study were to investigate particulate matter (PM) and polycyclic aromatic hydrocarbon (PAH) concentrations in ambient air during rice straw open burning and non-open burning periods. In the ambient air of a rice field, the mean PM concentration during and after an open burning event were 1828 and 102 μg m−3, respectively, which demonstrates that during a rice field open burning event, the PM concentration in the ambient air of rice field is over 17 times higher than that of the non-open burning period. During an open burning event, the mean total PAH and total toxic equivalence (BaP eq ) concentrations in the ambient air of a rice field were 7206 ng m−3 and 10.3 ng m−3, respectively, whereas after the open burning event, they were 376 ng m−3 and 1.50 ng m−3, respectively. Open burning thus increases total PAH and total BaP eq concentrations by 19-fold and 6.8-fold, respectively. During a rice straw open burning event, in the ambient air of a rice field, the mean dry deposition fluxes of total PAHs and total BaP eq were 1222 μg m−2 day−1 and 4.80 μg m−2 day−1, respectively, which are approximately 60- and 3-fold higher than those during the non-open burning period, respectively. During the non-open burning period, particle-bound PAHs contributed 79.2–84.2% of total dry deposition fluxes (gas + particle) of total PAHs. However, an open burning event increases the contribution to total PAH dry deposition by particle-bound PAHs by up to 85.9–95.5%. The results show that due to the increased amount of PM in the ambient air resulting from rice straw open burning, particle-bound PAHs contributed more to dry deposition fluxes of total PAHs than they do during non-open burning periods. The results show that biomass (rice straw) open burning is an important PAH emission source that significantly increases both PM and PAH concentration levels and PAH dry deposition in ambient air.

Air pollution has become a major environmental and public health concern, particularly in urban areas where rapid industrialization, traffic emissions, and population density contribute to elevated levels of airborne contaminants. In educational settings, such as secondary schools, poor air quality can adversely affect the health and cognitive performance of students, making the assessment of indoor and outdoor pollutants essential for risk management. This study comprehensively assessed the levels of particulate matter concentrations in indoor and outdoor air at some secondary schools in Port Harcourt Local Government Area (LGA), Rivers State. The study focused on measuring levels of particulate matter (PM1, PM2.5, PM4, PM7, PM10) and evaluating seasonal variations. Sample collection across nine secondary schools covered a period of twelve months (twice per term during academic session), covering the dry and wet seasons. Particulate matter concentrations were obtained using a calibrated handheld Aerocet Particle Mass Counter 531S. The results demonstrated higher indoor particulate matter (PM) concentrations compared to outdoors across schools. Indoor PM1 ranged from 3.9–8.1 µg/m³ versus 2.90–7.45 µg/m³ outdoors, while PM2.5 was 18.5–24.5 µg/m³ indoors compared to 13.3–20.07 µg/m³ outdoors. The coarser particles (PM4, PM7, and PM10) also exhibited higher concentrations indoors compared to outdoors with location PCSN having the highest average concentration indoor for PM10 at 153.9±118.5 µg/m³. Overall, particulate levels tend to be higher indoors than outdoors, especially for larger particles, although some outdoor values remain comparably high in certain locations. Overall, particulate matter concentrations were consistently higher during the dry season than wet season across all schools and particulate sizes, with larger particles exhibiting more pronounced increase. Generally, the concentrations of particulate matter for all nine schools were within permissible limit from PM1-PM10 (both indoor and outdoor) except location PCSN school which exceeded the permissible limit at PM10 for indoor. The study highlighted significant indoor particulate accumulation, across schools within Port Harcourt L.G A. It was recommended that schools authorities should implement improved ventilation systems, limit indoor combustion sources, and establish regular cleaning and dust control practices to reduce indoor particulate matter.

The smart buildings’ load forecasting is necessary for efficient energy management, and it is easily possible because of the data availability based on widespread use of Internet of Things (IoT) devices and automation systems. The information of buildings’ occupancy is directly associated with energy consumption. Therefore, we present a hybrid model consisting of a Long Short-Term Memory (LSTM) network, Extreme Gradient Boosting (XgBoost), Random Forest (RF) and Linear Regression (LR) for commercial and academic buildings’ load forecasting. The correlation between occupants’ count and total load of the building is calculated using Pearson Correlation Coefficient (PCC). The comparative analysis of the proposed approach with LSTM, XgBoost, RF and Gated Recurrent Unit (GRU) is also performed. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE) and Normalized Root Mean Square Error (NRMSE) are used as performance indicators for evaluating performance. Findings indicate that the proposed hybrid approach outperforms other models. The RMSE and MAE of 2.99 and 2.18, respectively, are recorded by the proposed model for commercial building dataset while for academic building the RMSE and MAE are 4.48 and 2.85, respectively. Occupancy and load consumption have a positive correlation as evident from PCC analysis. Therefore, we have scheduled the forecasted load based on occupancy patterns for two different cases. Cost is reduced by 17.42% and 33.40% in case 1 and case 2, respectively. Moreover, the performance of the proposed hybrid approach is compared with different techniques presented in literature for buildings load forecasting.

Travel planning is a process that allows users to obtain maximum benefit from their time, cost and energy. When planning a route from one place to another, it is an important option to present alternative travel areas on the route. This study proposes a travel route planning (TRP) architecture using a Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) model to improve both travel efficiency and environmental sustainability in route selection. This model incorporates carbon emissions directly into the route planning process by unifying user preferences, location recommendations, route optimization, and multimodal vehicle selection within a comprehensive framework. By merging environmental sustainability with user-focused travel planning, it generates personalized, practical, and low-carbon travel routes. The carbon emissions observed with TRP’s artificial intelligence (AI) recommendation route are presented comparatively with those of the user-determined route. XGBoost, Random Forest (RF), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), (Extra Trees Regressor) ETR, and Multi-Layer Perception (MLP) models are applied to the TRP model. LSTM is compared with Recurrent Neural Networks (RNNs) and Gated Recurrent Unit (GRU) models. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Normalized Root Mean Square Error (NRMSE) error measurements of these models are carried out, and the best result is obtained using XGBoost and LSTM. TRP enhances environmental responsibility awareness within travel planning by integrating sustainability-oriented parameters into the decision-making process. Unlike conventional reservation systems, this model encourages individuals and organizations to prioritize eco-friendly options by considering not only financial factors but also environmental and socio-cultural impacts. By promoting responsible travel behaviors and supporting the adoption of sustainable tourism practices, the proposed approach contributes significantly to the broader dissemination of environmentally conscious travel choices.

The prediction of wind energy generation is important to enhance the performance and dependability of renewable energy systems due to the rising demand for wind-generated electricity and advancements in wind energy technology competitiveness. This study leverages advanced machine learning (ML) and some other statistical and deep learning based time series forecasting models to enhance the accuracy of wind energy predictions. This comprehensive analysis includes nine ML models—Linear Regression, Random Forests (RF), Gradient Boosting Machines (GBM), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), AdaBoost, XGBoost, Support Vector Regression (SVR), and Neural Networks—as well as Four time-series forecasting models—ARIMA, Temporal Convolutional Networks (TCNs), Long Short-Term Memory (LSTM) networks and GRU. Each ML model underwent rigorous cross-validation to ensure optimal performance. The assessment criteria utilized here comprised the Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the R² Score. It was found that among the nine ML models, Random Forests, GBM and KNN consistently provided superior accuracy and robustness, making them the top choices for wind energy prediction whereas the performance of linear regression, SVM and SVR were very poor for the considered dataset. From the experiment, Random Forest, GBM, and KNN showed the best performance with low MSE values of 0.77, 1.95, and 1.51 respectively, while other models had MSEs above 7.5, with AdaBoost reaching 30. Their RMSEs (0.88, 1.40, 1.23) and MAEs (0.093, 0.73, 0.10) also indicate strong predictive accuracy compared to the rest.In this paper, time series forecasting, TCNs, LSTM and GRU networks showed strong capabilities in capturing temporal dependencies and trends within the wind energy data. Visualization techniques were employed to compare model performances comprehensively, providing clear insights into their predictive power. Therefore, this present study offers a robust framework for researchers and practitioners aiming to leverage machine learning and time series forecasting in the realm of renewable energy prediction.

Investigating the impact of relative humidity and air tightness on PM sedimentation and concentration reduction

HighlightsParticulate matter (PM) data were analyzed to identify PM emission characteristics among different animal types.The PM concentrations were higher in broiler chicken and swine farrowing houses and were higher in winter.The PM emissions were also higher in broiler chicken houses and swine farrowing rooms.The PM in the layer chicken house in Indiana had narrower distributions with a greater percentage of smaller particles.Abstract. Understanding the characteristics of particulate matter (PM) emissions from animal feeding operations (AFOs) is essential to address the associated health and environmental impacts and to develop control strategies to mitigate such impacts. This article reports a study of PM concentrations and emission characteristics from 26 poultry and swine production houses to investigate the similarities and differences in PM emission characteristics, e.g., concentrations, emission rates, and particle size distribution (PSD), among different animal and housing types. Concentration and emission data for PM2.5, PM10, and total suspended particulates (TSP) collected by the National Air Emission Monitoring Study (NAEMS) were used to compare the differences among different production practices and animal types. The PSDs of the PM were examined based on the PM2.5/PM10 and PM10/TSP emission rate ratios. It was discovered that the concentrations of PM varied among animal types. For poultry, the concentrations of PM were higher in broiler houses than in other poultry houses. For swine, the average concentrations of PM were higher in farrowing rooms than in swine barns. Moreover, the PM concentrations in poultry and swine houses exhibited significant seasonal trends, with higher concentrations in winter and lower concentrations in summer, which were in a reverse relationship with ventilation rates. The PM emissions also varied among animal types. For poultry, the PM emissions were significantly higher for poultry production houses in California. For swine, the PM emissions were significantly higher for farrowing rooms than other swine houses. The PSD of PM varied among animal types, with mass median diameters (MMD) in the ranges of 6.51 to 13.62 µm for poultry houses and 7.94 to 17.19 µm for swine houses. The geometric standard deviations (GSD) were in the ranges of 1.66 to 2.71 and 1.65 to 2.9 for poultry and swine PM, respectively. The PM in the layer house in Indiana had a narrower distribution (smaller GSD) with a greater percentage of smaller particles than the other poultry houses, while the PM in the broiler house in California had a broader distribution (larger GSD) than the other poultry houses. For swine, the PM in the sow barn in North Carolina had a narrower distribution (smaller GSD) than the other swine houses, while the PM in the farrowing houses in Oklahoma had a broader distribution (larger GSD) than the other swine houses. The knowledge gained from this research may provide insights for addressing the PM emissions from various animal production systems. Keywords: Concentration and emission, Particulate matter, PM2.5, PM10, Poultry, swine, PSD, TSP.

Abstract. As air pollution is regarded as the single largest environmental health risk in Europe it is important that communication to the public is up to date and accurate and provides means to avoid exposure to high air pollution levels. Long- and short-term exposure to outdoor air pollution is associated with increased risks of mortality and morbidity. Up-to-date information on present and coming days' air quality helps people avoid exposure during episodes with high levels of air pollution. Air quality forecasts can be based on deterministic dispersion modelling, but to be accurate this requires detailed information on future emissions, meteorological conditions and process-oriented dispersion modelling. In this paper, we apply different machine learning (ML) algorithms – random forest (RF), extreme gradient boosting (XGB), and long short-term memory (LSTM) – to improve 1, 2, and 3 d deterministic forecasts of PM10, NOx, and O3 at different sites in Greater Stockholm, Sweden. It is shown that the deterministic forecasts can be significantly improved using the ML models but that the degree of improvement of the deterministic forecasts depends more on pollutant and site than on what ML algorithm is applied. Also, four feature importance methods, namely the mean decrease in impurity (MDI) method, permutation method, gradient-based method, and Shapley additive explanations (SHAP) method, are utilized to identify significant features that are common and robust across all models and methods for a pollutant. Deterministic forecasts of PM10 are improved by the ML models through the input of lagged measurements and Julian day partly reflecting seasonal variations not properly parameterized in the deterministic forecasts. A systematic discrepancy by the deterministic forecasts in the diurnal cycle of NOx is removed by the ML models considering lagged measurements and calendar data like hour and weekday, reflecting the influence of local traffic emissions. For O3 at the urban background site, the local photochemistry is not properly accounted for by the relatively coarse Copernicus Atmosphere Monitoring Service ensemble model (CAMS) used here for forecasting O3 but is compensated for using the ML models by taking lagged measurements into account. Through multiple repetitions of the training process, the resulting ML models achieved improvements for all sites and pollutants. For NOx at street canyon sites, mean squared error (MSE) decreased by up to 60 %, and seven metrics, such as R2 and mean absolute percentage error (MAPE), exhibited consistent results. The prediction of PM10 is improved significantly at the urban background site, whereas the ML models at street sites have difficulty capturing more information. The prediction accuracy of O3 also modestly increased, with differences between metrics. Further work is needed to reduce deviations between model results and measurements for short periods with relatively high concentrations (peaks) at the street canyon sites. Such peaks can be due to a combination of non-typical emissions and unfavourable meteorological conditions, which are rather difficult to forecast. Furthermore, we show that general models trained using data from selected street sites can improve the deterministic forecasts of NOx at the station not involved in model training. For PM10 this was only possible using more complex LSTM models. An important aspect to consider when choosing ML algorithms is the computational requirements for training the models in the deployment of the system. Tree-based models (RF and XGB) require fewer computational resources and yield comparable performance in comparison to LSTM. Therefore, tree-based models are now implemented operationally in the forecasts of air pollution and health risks in Stockholm. Nevertheless, there is big potential to develop generic models using advanced ML to take into account not only local temporal variation but also spatial variation at different stations.

Identification of technical problems affecting performance of DustTrak DRX aerosol monitors.

The PM accumulation process by plants is quite energetic, and even after one day, fluctuations in PM load on foliage can be significant. Rain and, to a lesser extent, wind influenced PM deposition on leaves, with the latter being more species-specific. This research explored the temporal and spatial variations in the concentrations of Particulate Matter (PM) collected on two evergreen plant species (Nerium oleander and Ligustrum lucidum) leaves commonly grown along with the roadside Quetta city Balochistan, Pakistan. The impact of wind and rain on the quantity of PM collected on vegetation was investigated. The PM (g⋅m−2) concentrations held by N. oleander and L. lucidum leaves considerably varied among the places (from 7.70 – 10.7 & 6.24 – 9.53) with significant variation and over time (from 5.94 – 18.0 & 5.32 – 16.5). The highest PM concentrations on the foliage of N. oleander and L. lucidum growing at the most contaminated site, Saryab road, were determined. The largest and lowest levels of accumulation PM followed in August and January, respectively, throughout the year. Rainfall events eliminated a significant percentage of the accumulated PM on leaves (30%, 42% and 55% of PM from leaves of N. oleander and 40, 62 and 95% from L. lucidum leaves) and strong winds (20%. 35% and 47% of PM N. oleander and 25%, 45% and 71% from L. lucidum), It’s also possible that heavier precipitation or a higher maximum wind speed will help to eliminate more PM from the leaves. Rainfall primarily cleared coarse and large particles, but small fragments clung to the foliage more tenaciously. These findings suggested that when assessing total PM accumulation on leaves, the influence of regional weather circumstances (such as strong wind or rainfall), altered seasons, and levels of pollution should be judged.

Hybrid intelligent system for air quality forecasting using phase adjustment

In Transit-Oriented Development (TOD), the close integration of residential structures with community activities and traffic heightens residents' exposure to traffic-related pollutants. Despite traffic being a primary source of particulate matter (PM), the compact design of TODs, together with the impact of urban heat island (UHI), increases the likelihood of trapping emitted PM from traffic, leading to heightened exposure of TOD residents to PM. Although PM originates from two distinct sources in road traffic, exhaust and non-exhaust emissions (NEE), current legislation addressing traffic-related PM from non-exhaust emissions sources remains limited. This paper focuses on two TOD typologies in Manchester City—Manchester Piccadilly and East Didsbury—to understand the roles of TOD traffic as a PM generator and TOD place design as a PM container and trapper. The investigation aims to establish correlations between street design canyon ratios, vehicular Speed, and PM10/PM2.5, providing design guidance and effective traffic management strategies to control PM emissions within TODs. Through mapping the canyon ratio and utilising the Breezometer API for PM monitoring, the paper revealed elevated PM levels in both TOD areas, exceeding World Health Organization (WHO) recommendations, particularly for PM2.5. Correlation analysis between canyon configuration and PM2.5/PM10 highlighted the importance of considering building heights and avoiding the creation of deep canyons in TOD design to minimise the limited dispersion of PM. Leveraging UK road statistics and the PTV Group API for vehicle speed calculations, the paper studied the average speeds on the TOD roads concerning PM. Contrary to conventional assumption, the correlation analyses have revealed a noteworthy association shift between vehicular speed and PM concentrations. A positive correlation existed between speed increase and PM increases on arterial roads. However, a negative correlation emerged on main, collector, and local streets, indicating that PM levels rise for both PM10 and PM2.5 as Speed decreases. These findings challenge the traditional assumption that higher Speed leads to increased emissions, highlighting the potential impact of NEE on PM concentrations. This paper calls for thorough design considerations and traffic management strategies in TOD, especially in dense areas, considering building height, optimising traffic flow, and enhancing compromised air quality associated with vehicular emissions.

Indoor air quality of 5,000 households and its determinants. Part A: Particulate matter (PM2.5 and PM10–2.5) concentrations in the Japan Environment and Children's Study

Regional air quality over East Asia, including South Korea, has recently become a center of public attention because of a few episodes of very high particulate matter (PM) concentrations. Predicting PM variation with lead time of a few hours up to days is one of the key areas that the governments are working on because it can benefit from an early warning system to short‐term mitigation effort. In this study, the influence of synoptic weather conditions on regional air quality was investigated with the occurrence frequencies of PM episodes as a function of various synoptic weather patterns during winter and spring. During winter, dry moderate (DM) types occur frequently alongside high PM cases (24 hr mean PM10 concentration >). The composite weather map showed a weak northwesterly wind field as a potential cause. On the contrary, it is interesting to note that dry polar types can be associated with low PM cases (24 hr mean PM10 concentration <) as well as high PM depending on near‐surface wind speed. Furthermore, DM and dry tropical types were found to be highly correlated with high (higher) PM concentrations during spring season, likely because of the enhanced static stability in the lower troposphere. It should be noted that PM concentration depends on the lower atmospheric stability. The close relationship between synoptic weather patterns and PM concentration suggests that synoptic weather can play an important role in regional air quality.