Evaluating the impact of anthropogenic drivers and meteorological factors on air pollutants by explainable machine learning in Shandong Province, China

Based on the daily monitoring data of urban air quality in Shandong province from 2016 to 2020, combined with socio-economic data such as population density and urbanization rate, as well as meteorological data such as wind speed, temperature, and relative humidity, the methods of geographic weighted regression (GWR), multiscale geographically weighted regression (MGWR), and wavelet analysis were comprehensively applied to explore the temporal and spatial distribution characteristics of air pollutants and their relationship with socio-economic and meteorological elements. The results showed that:① In the past five years, the air quality in Shandong province has shown an overall improvement trend. Except for ozone, the concentrations of SO2, NO2, PM2.5, and PM10 decreased annually. Additionally, their distribution had obvious spatial differences, which was reflected in the lower concentration of air pollutants in coastal areas. ② PM2.5 in Shandong province had an extremely significant positive correlation with population density and the proportion of secondary industry (P<0.01) but had a very negative correlation with urbanization rate (P<0.01). Moreover, there were scale differences in the spatial relationship. The spatial relationship between population density, civil vehicle volume, industrial power consumption, and PM2.5 was relatively stable, whereas the spatial heterogeneity of the impact of urbanization rate and the proportion of secondary industry on PM2.5 concentration was high. ③ Meteorological factors had different effects on PM2.5 in Heze and Weihai. PM2.5 in Heze had a stronger correlation with air temperature, relative humidity, and sunshine hours, whereas sea land breeze prevailed in Weihai, resulting in a higher correlation between PM2.5 and wind speed. ④ Wavelet analysis showed that the frequency of air pollution in Heze was higher than that in Weihai, approximately one-two weeks/time in winter. In the annual cycle, the PM2.5 in Heze lagged behind the wind speed, whereas the PM2.5 and wind speed in Weihai were in the same phase. To summarize, there were obvious temporal and spatial differences in air quality in Shandong province, which was comprehensively affected by socio-economic and meteorological factors.

Meteorological factors and air pollutants are associated with the spread of pulmonary tuberculosis (PTB), but few studies have examined the effects of their interactions on PTB. Therefore, this study investigated the impact of meteorological factors and air pollutants and their interactions on the risk of PTB in Urumqi, a city with a high prevalence of PTB and a high level of air pollution. The number of new PTB cases in eight districts of Urumqi from 2014 to 2019 was collected, along with data on meteorological factors and air pollutants for the same period. A generalized additive model was applied to explore the effects of meteorological factors and air pollutants and their interactions on the risk of PTB incidence. Segmented linear regression was used to estimate the nonlinear characteristics of the impact of meteorological factors on PTB. During 2014-2019, a total of 14,402 new cases of PTB were reported in eight districts, with March to May being the months of high PTB incidence. The exposure-response curves for temperature (Temp), relative humidity (RH), wind speed (WS), air pressure (AP), and diurnal temperature difference (DTR) were generally inverted "U" shaped, with the corresponding threshold values of - 5.411 °C, 52.118%, 3.513m/s, 1021.625hPa, and 8.161 °C, respectively. The effects of air pollutants on PTB were linear and lagged. All air pollutants were positively associated with PTB, except for O3, which was not associated with PTB, and the ER values for the effects on PTB were as follows: 0.931 (0.255, 1.612) for PM2.5, 1.028 (0.301, 1.760) for PM10, 5.061 (0.387, 9.952) for SO2, 2.830 (0.512, 5.200) for NO2, and 5.789 (1.508, 10.251) for CO. Meteorological factors and air pollutants have an interactive effect on PTB. The risk of PTB incidence was higher when in high Temp-high air pollutant, high RH-high air pollutant, high WS-high air pollutant, lowAP-high air pollutant, and high DTR-high air pollutant. In conclusion, both meteorological and pollutant factors had an influence on PTB, and the influence on PTB may have an interaction.

In observational studies, a possible correlation between atmospheric environmental factors and the number of daily outpatient visits by Otitis media with effusion(OME) patients has been observed. However, the causal relationship is not clear.To study the relationship between the incidence of OME and meteorological factors and air pollutants in the main urban areas of Lanzhou, it is helpful to further understand the health effects of meteorological and environmental factors on OME and to prevent and treat the disease, it is of great academic and practical significance to the prevention, treatment and prognosis of diseases. The levels of AQI、PM2.5、PM10、NO2、O3、SO2、CO、AP、RH、W and T were obtained from local monitor stations. Data of patients with OME were collected from two Grade A Level hospitals in Lanzhou from January 1, 2014 to December 31, 2016. Descriptive analysis of data was carried out for the study subjects. Spearman correlation coefficients between atmospheric environmental factors and daily visits of patients with OME were calculated by SPSS statistical software. Lag effects, relative risks(RR) and exposure-response curves were calculated by generalized additive model (GAM) with R software. (1) The incidence of OME in winter and spring was more than that in summer and autumn, which was consistent with the seasonal variation of meteorological environmental factors of Lanzhou. That was, the meteorological conditions and air quality in winter and spring were poor, while in summer and autumn they were relatively good. (2) The number of male outpatients were 1.05, 1.08 and 1.09 times of female outpatients during the period 2014-2016, respectively. And aged 0-10 years old outpatients accounted for 31% of the total OME outpatients. (3) Exposure-response curve showed that PM2.5, PM10, NO2 and SO2 were positively correlated with OME, T was negatively correlated with OME. When concentration < 1mg/m3, CO was positively correlated with OME. When concentration>1mg/m3, CO was negatively correlated with OME. When concentration<30 ug/m3, O3 was positively correlated with OME. When concentration>30ug/m3, O3 was negatively correlated with OME. (4) The factors we studied could significantly affect the number of OME outpatients within 2-3 days of single lag effects, and 3-4 days of cumulative lag effects.(5) The influential factors on OME were as follows: PM2.5、NO2、SO2、O3、 CO and T. The daily average number of OME patients in different seasons was different in major region of Lanzhou city, with more in winter、spring and fewer in summer、autumn. Age and sex were the important factors affecting the daily average number of OME patients. Males were more susceptible to OME than females and children awere moresusceptible to OME than adults. The change of OME patients was related to air quality, air pressure and temperature. The worse the air quality, the higher the air pressure, the lower the temperature, the more the average daily number of OME patients. Meteorological environmental factors affected the visits of OME, and the lagging effect time of different factors were different. Most of the research factors within 3-4 days had a significant impact on the number of patients of OME. 1.The number of OME visits in the Lanzhou was more seasonal in winter and spring than in summer and fall. 2.Age and sex were the most important factors affecting the number of patients with OME. According to the prevalence of OME in Lanzhou, children were more likely to have OME than adults and men were more likely to have OME than women. 3.The number of OME patients was related to air quality, air pressure and temperature. 4.The meteorological factors have a delayed effect on the onset of OME, and the time of delayed effect is different for different factors. The single delayed effect of 2-3 days and the cumulative delayed effect of 3-4 days have a significant effect on the change of the number of patients with OME. To study the relationship between the incidence of OME and meteorological factors and air pollutants in the main urban areas of Lanzhou, it is helpful to further understand the health effects of meteorological and environmental factors on OME and to prevent and treat the disease, it is of great academic and practical significance to the prevention, treatment and prognosis of diseases.

The health risks of PM2.5-bound metals highlight the need for burden assessment, metal prioritization, and key factor analysis to support effective air quality management, yet relevant studies remain limited. Shandong Province is one of the most polluted regions in northern China, providing an ideal setting for this investigation. We monitored 17 PM2.5-bound metals for three years across Shandong, China and performed disease burden assessment based on disability-adjusted life years (DALYs). Furthermore, key influencing factors contributing to high-hazard metals were identified through explainable machine learning. The results showed that PM2.5-bound metal concentrations were generally higher in inland areas than in coastal regions, with Ni concentrations elevated in coastal areas. K, Ca, Zn, and Mn exhibited the highest three-year average concentrations among the metals, while Cr averaged 6.12 ng/m3, significantly exceeding the recommended annual limit of 0.025 ng/m3 set by Chinese Ambient Air Quality Standards. Jinan carried the greatest burden at 4.67 DALYs per 1000 people, followed by Zibo (3.78), Weifang (2.98), and Rizhao (2.80). CKD, interstitial pneumonia, and chronic respiratory diseases account for the highest DALYs from PM2.5-bound metals in Shandong Province. Industrial emissions are the largest contributors to the disease burden (>34%), with Cr, Cd, and Pb as the primary contributing metals requiring priority control. Fractional vegetation cover was identified as the key factor contributing to the reduction in their concentrations. These results underscore that prioritizing the regulation of industrial combustion, particularly concerning Cr, Cd, and Pb, and enhancing fractional vegetation cover could reduce disease burden and provide public health benefits.

Association of viruses causing acute respiratory infections with meteorological factors and air pollutants in hospitalised children in Macao: a retrospective analysis

In this work, the relationships between air quality and pollutant emissions were investigated during the COVID-19 pandemic in Shandong Province. During the quarantine period (from January 24 to February 7, 2020), the concentrations of atmospheric pollutants decreased significantly relative to the period before controls were imposed (from January 15 to 23, 2020). Specifically, except for an increase in the concentration of O3, concentrations of PM10, PM2.5, NO2, SO2, and CO decreased for 72.6 μg·m-3 (45.86%), 47.4 μg·m-3(41.24%), 25.6 μg·m-3 (58.00%), 3.0 μg·m-3 (17.71%), and 0.5 mg·m-3 (31.40%), respectively. RAMS-CMAQ simulation showed that meteorological diffusion had an essential role in improving air quality. Influenced by meteorological factors, emissions of PM10, PM2.5, NO2, SO2, and CO were reduced 26.04%, 33.03%, 28.35%, 43.27%, and 23.29%, respectively. Furthermore, the concentrations of PM10, PM2.5, NO2, SO2, and CO were reduced by 19.82%, 8.21%, 29.65%, -25.56%, and 8.12%, respectively, due to pollution emissions reductions during the quarantine period. O3 concentrations increased by 20.51% during quarantine, caused by both meteorological factors (10.47%) and human activities (10.04%). These results indicate that primary pollutants were more sensitive to emissions reductions; however, secondary pollutants demonstrated a lagged response the emissions reduction and were significantly affected by meteorological factors. The linear relationship between ozone and the emissions reduction was not significant, and was inverse overall. Further investigation are now required on the impact of emissions reduction on ozone pollution control.

Although studies have reported the modification effect of air pollutants on heat-related health risk, little is known on the modification effect among various particulate matter with different particle size on mortality. We aimed to investigate whether the associations of hot temperatures with daily mortality were modified by different air pollutant levels in Shandong Province, China. Daily data of air pollutants, meteorological factors, and mortality of 1,822 subdistricts in Shandong province from 2013 to 2018 were collected. We used a time-stratified case-crossover model with an interaction term between the cross-basis term for ambient temperature and the linear function of particulate matter ≤1 µm (PM1), PM2.5, nitrogen dioxide (NO2), and ozone to obtain heat-mortality associations during the hot season. Results showed that the cumulative odds ratio of extreme heat on mortality over 0 to 10 days was 3.66 (95% CI: 3.10-4.31). The mortality risk during hot seasons was stronger at high air pollutant levels. The modification effect of particulate matters on heat-related mortality decreased by its aerodynamic diameter. Females and older adults over 75 years were more vulnerable to the modification effect of air pollutants, and significant differences were detected in the association between temperatures and mortality stratified by PM1 and PM2.5. Higher heat-related mortality risks were observed at high NO2 levels, especially for cardiorespiratory disease. The findings suggest that more consideration should be given to the combined effect of very fine particles and NO2 with ambient heat when developing healthcare strategies, and women and older adults should be given priority in health-related settings.

Effects of meteorological factors on daily outpatient visits for skin diseases: a time series study in a Chinese population

To analyze the relationship between levels of air pollution and number of children hospitalizations because of respiratory tract infection in Shenmu County, the data regarding meteorological factors, environmental pollutants, that is SO2 and NO2, Particulate Matter 10 (PM10), and hospitalizations of children less than 16 years of age was collected during the time duration of November 2009 to October 2012. Using SAS 9.3, descriptive data analysis for meteorological and environmental factors and hospital admissions were performed along with main air pollutants determination. Using the statistical software R 3.0.1, a generalized additive Poisson regression model was established, the linear fitting models of the air pollutant concentrations and meteorological factors were introduced considering the lag effect, and the relative risk of the main atmospheric pollutants on children hospitalization was evaluated. The results showed that the primary air pollutant in Shenmu County is PM10 and its Pearson correlation coefficient with Air Pollution Index (API) is 0.917. After control of long term climate trend, “week day effect,” meteorological factors, and impact of other contaminants, it was found that, on the same day and during the lag of 1 to 10 days, PM10 concentrations had no significant effect on children hospitalization rate.

Objective: The number of patients requiring prolonged mechanical ventilation (PMV) is increasing worldwide, but the weaning outcome prediction model in these patients is still lacking. We hence aimed to develop an explainable machine learning (ML) model to predict successful weaning in patients requiring PMV using a real-world dataset.Methods: This retrospective study used the electronic medical records of patients admitted to a 12-bed respiratory care center in central Taiwan between 2013 and 2018. We used three ML models, namely, extreme gradient boosting (XGBoost), random forest (RF), and logistic regression (LR), to establish the prediction model. We further illustrated the feature importance categorized by clinical domains and provided visualized interpretation by using SHapley Additive exPlanations (SHAP) as well as local interpretable model-agnostic explanations (LIME).Results: The dataset contained data of 963 patients requiring PMV, and 56.0% (539/963) of them were successfully weaned from mechanical ventilation. The XGBoost model (area under the curve [AUC]: 0.908; 95% confidence interval [CI] 0.864–0.943) and RF model (AUC: 0.888; 95% CI 0.844–0.934) outperformed the LR model (AUC: 0.762; 95% CI 0.687–0.830) in predicting successful weaning in patients requiring PMV. To give the physician an intuitive understanding of the model, we stratified the feature importance by clinical domains. The cumulative feature importance in the ventilation domain, fluid domain, physiology domain, and laboratory data domain was 0.310, 0.201, 0.265, and 0.182, respectively. We further used the SHAP plot and partial dependence plot to illustrate associations between features and the weaning outcome at the feature level. Moreover, we used LIME plots to illustrate the prediction model at the individual level. Additionally, we addressed the weekly performance of the three ML models and found that the accuracy of XGBoost/RF was ~0.7 between weeks 4 and week 7 and slightly declined to 0.6 on weeks 8 and 9.Conclusion: We used an ML approach, mainly XGBoost, SHAP plot, and LIME plot to establish an explainable weaning prediction ML model in patients requiring PMV. We believe these approaches should largely mitigate the concern of the black-box issue of artificial intelligence, and future studies are warranted for the landing of the proposed model.

BackgroundWith the rapid development of China’s economy, air pollution has attracted public concern because of its harmful effects on health.MethodsThe source apportioning of air pollution, the spatial distribution characteristics, and the relationship between atmospheric contamination, and the risk of exposure were explored. The in situ daily concentrations of the principal air pollutants (PM2.5, PM10, SO2, NO2, CO and O3) were obtained from 188 main cities with many continuous air-monitoring stations across China (2014 and 2015).ResultsThe results indicate positive correlations between PM2.5 and SO2 (R2 = 0.395/0.404, P < 0.0001), CO (R2 = 0.187/0.365, P < 0.0001), and NO2 (R2 = 0.447/0.533, P < 0.0001), but weak correlations with O3 (P > 0.05) for both 2014 and 2015. Additionally, a significant relationship between SO2, NO2, and CO was discovered using regression analysis (P < 0.0001), indicating that the origin of air pollutants is likely to be vehicle exhaust, coal consumption, and biomass open-burning. For the spatial pattern of air pollutants, we found that the highest concentration of SO2, NO2, and CO were mainly distributed in north China (Beijing-Tianjin-Hebei regions), Shandong, Shanxi and Henan provinces, part of Xinjiang and central Inner Mongolia (2014 and 2015).ConclusionsThe highest concentration and risk of PM2.5 was observed in the Beijing–Tianjin–Hebei economic belts, and Shandong, Henan, Shanxi, Hubei and Anhui provinces. Nevertheless, the highest concentration of O3 was irregularly distributed in most areas of China. A high-risk distribution of PM10, SO2 and NO2 was also observed in these regions, with the high risk of PM10 and NO2 observed in the Hebei and Shandong province, and high-risk of PM10 in Urumchi. The high-risk of NO2 distributed in Beijing-Yangtze River Delta region-Pearl River Delta region-central. Although atmospheric contamination slightly improved in 2015 compared to 2014, humanity faces the challenge of reducing the environmental and public health effects of air pollution by altering the present mode of growth to achieve sustainable social and economic development.

Climate change, environmental factors, and COVID-19: Current evidence and urgent actions

Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model.

Spatial-temporal heterogeneity of air pollution and its relationship with meteorological factors in the Pearl River Delta, China

In this paper, the heavily polluted cities of Jinan, Zibo, Dongying and Linyi in Shandong Province were taken as the study area. Based on the daily air quality and pollutant concentration data from 2015 to 2019, the change of air quality and pollutant concentration in the study area was analyzed from four time scales of year, month, working day and day by time series analysis, combined with meteorological, industrial and other influencing factors. The analysis showed that: (1) With the implementation of the blue Sky Defense plan in 2018, the air quality in the study area was significantly improved, and the decrease of SO2, PM2.5 and CO was the largest, the percentage reduction was 41.5%, 62.36%, 39.35%; (2): Monthly variation: it shows a "W" -shaped variation, with severe pollution in winter and light pollution in early autumn; (3) Working days and rest days show an "S" -shaped change, with Tuesday as the pollution trough and Saturday as the pollution peak; (4) A "convex" type of fluctuation was observed, with the heaviest pollution in the middle of the month and lighter pollution in the beginning and end of the month.