The application of machine learning to air pollution research: A bibliometric analysis

Exploration of transfer learning techniques for the prediction of PM10

Modelling of pollutants provides valuable insights into air quality dynamics, aiding exposure assessment where direct measurements are not viable. Machine learning (ML) models can be employed to explore such dynamics, including the prediction of air pollution concentrations, yet demanding extensive training data. To address this, techniques like transfer learning (TL) leverage knowledge from a model trained on a rich dataset to enhance one trained on a sparse dataset, provided there are similarities in data distribution. In our experimental setup, we utilize meteorological and pollutant data from multiple governmental air quality measurement stations in Graz, Austria, supplemented by data from one station in Zagreb, Croatia to simulate data scarcity. Common ML models such as Random Forests, Multilayer Perceptrons, Long-Short-Term Memory, and Convolutional Neural Networks are explored to predict particulate matter in both cities. Our detailed analysis of PM10 suggests that similarities between the cities and the meteorological features exist and can be further exploited. Hence, TL appears to offer a viable approach to enhance PM10 predictions for the Zagreb station, despite the challenges posed by data scarcity. Our results demonstrate the feasibility of different TL techniques to improve particulate matter prediction on transferring a ML model trained from all stations of Graz and transferred to Zagreb. Through our investigation, we discovered that selectively choosing time spans based on seasonal patterns not only aids in reducing the amount of data needed for successful TL but also significantly improves prediction performance. Specifically, training a Random Forest model using data from all measurement stations in Graz and transferring it with only 20% of the labelled data from Zagreb resulted in a 22% enhancement compared to directly testing the trained model on Zagreb.

Can " Zero waste city" policy promote green technology? Evidence from econometrics and machine learning

The promotion of green technology innovation (GTI) is regarded as an effective way to protect the environment and achieve sustainable development. The "Zero waste city" construction pilot policy (ZWCP), is an important policy for the promotion of waste management, and the achievement of sustainable development and circular economy. The paper explores the effect of ZWCP on GTI using the difference-in-differences model and machine learning. The result shows that ZWCP significantly promotes GTI. In addition, causal forest modeling, a type of machine learning method, similarly validates this result. The mechanism effect suggests that ZWCP promotes GTI by increasing research expenditure and enhancing informational development. The heterogeneity effect suggests that ZWCP is more powerful in promoting GTI when it is implemented in western, low administrative level, and resource-based cities. The paper provides a reference for waste management policies and development of GTI in other countries.

Bioavailability and ecological risk assessment of metal pollutants in ambient PM2.5 in Beijing

Metal pollutants in fine particulate matter (PM2.5) are physiologically toxic, threatening ecosystems through atmospheric deposition. Biotoxicity and bioavailability are mainly determined by the active speciation of metal pollutants in PM2.5. As a megacity in China, Beijing has suffered severe particulate pollution over the past two decades, and the health effects of metal pollutants in PM2.5 have received significant attention. However, there is a limited understanding of the active forms of metals in PM2.5 and their ecological risks to plants, soil or water in Beijing. It is essential that the ecological risks of metal pollutants in PM2.5 are accurately evaluated based on their bioavailability, identifying the key pollutants and revealing historic trends to future risks control. A two-year project measured the chemical speciation of pollution elements (As, Cd, Cu, Cr, Ni, Mn, Pb, Sb, Sr, Ti, and Zn) in PM2.5 in Beijing, in particular their bioavailability, assessing ecological risks and identifying key pollutants. The mass concentrations of total and active species of pollution elements were 199.12 ng/m3 and 114.97 ng/m3, respectively. Active fractions accounted for 57.7 % of the total. Cd had the highest active proportion. Based on the risk assessment code (RAC), most pollution elements except Ti had moderate or high ecological risk, with RAC exceeding 30 %. Cd, with an RAC of 70 %, presented the strongest ecological risk. Comparing our data with previous research shows that concentrations of pollution elements in PM2.5 in Beijing have decreased over the past decade. However, although the total concentrations of Cd in PM2.5 have decreased by >50 % over the past decade, based on machine model simulation, its ecological risk has reduced by only 10 %. Our research shows that the ecological risks of pollution elements remain high despite their decreasing concentrations. Controlling the active species of metal pollutants in PM2.5 in Beijing in the future is vital.

Predicting tropospheric nitrogen dioxide column density in South African municipalities using socio-environmental variables and Multiscale Geographically Weighted Regression

Atmospheric nitrogen dioxide (NO2) pollution is a major health and social challenge in South African induced mainly by fossil fuel combustions for power generation, transportation and domestic biomass burning for indoor activities. The pollution level is moderated by various environmental and social factors, yet previous studies made use of limited factors or focussed on only industrialised regions ignoring the contributions in large parts of the country. There is a need to assess how socio-environmenral factors, which inherently exhibit variations across space, influence the pollution levels in South Africa. This study therefore aimed to predict annual tropospheric NO2 column density using socio-environmental variables that are widely proven in the literature as sources and sinks of pollution. The environmental variables used to predict NO2 included remotely sensed Enhanced Vegetation Index (EVI), Land Surface Temperature and Aerosol Optical Depth (AOD) while the social data, which were obtained from national household surveys, included energy sources data, settlement patterns, gender and age statistics aggregated at municipality scale. The prediction was accomplished by applying the Multiscale Geographically Weighted Regression that fine-tunes the spatial scale of each variable when building geographically localised relationships. The model returned an overall R2 of 0.92, indicating good predicting performance and the significance of the socio-environmental variables in estimating NO2 in South Africa. From the environmental variables, AOD had the most influence in increasing NO2 pollution while vegetation represented by EVI had the opposite effect of reducing the pollution level. Among the social variables, household electricity and wood usage had the most significant contributions to pollution. Communal residential arrangements significantly reduced NO2, while informal settlements showed the opposite effect. The female proportion was the most important demographic variable in reducing NO2. Age groups had mixed effects on NO2 pollution, with the mid-age group (20-29) being the most important contributor to NO2 emission. The findings of the current study provide evidence that NO2 pollution is explained by socio-economic variables that vary widely across space. This can be achieved reliably using the MGWR approach that produces strong models suited to each locality.

Key toxic components and sources affecting oxidative potential of atmospheric particulate matter using interpretable machine learning: Insights from fog episodes

Fog significantly affects the air quality and human health. To investigate the health effects and mechanisms of atmospheric fine particulate matter (PM2.5) during fog episodes, PM2.5 samples were collected from the coastal suburb of Qingdao during different seasons from 2021 to 2022, with the major chemical composition in PM2.5 analyzed. The oxidative potential (OP) of PM2.5 was determined using the dithiothreitol (DTT) method. A positive matrix factorization model was adopted for PM2.5. Interpretable machine learning (IML) was used to reveal and quantify the key components and sources affecting OP. PM2.5 exhibited higher oxidative toxicity during fog episodes. Water-soluble organic carbon (WSOC), NH4+, K+, and water-soluble Fe positively affected the enhancement of DTTV (volume-based DTT activity) during fog episodes. The IML analysis demonstrated that WSOC and K+ contributed significantly to DTTV, with values of 0.31 ± 0.34 and 0.27 ± 0.22 nmol min-1 m-3, respectively. Regarding the sources, coal combustion and biomass burning contributed significantly to DTTV (0.40 ± 0.38 and 0.39 ± 0.36 nmol min-1 m-3, respectively), indicating the significant influence of combustion-related sources on OP. This study provides new insights into the effects of PM2.5 compositions and sources on OP by applying IML models.