Search

• What is the primary question addressed by this study? The present study sought to understand whether the presence of the APOE -ε4 allele results in differential cognitive outcomes following the interacting effects of decreasing in PM 2.5 levels over time and aging. • What is the main finding of this study? The results showed high levels of PM 2.5 across all visits were significantly associated with worsening of scores on the overall MMSE. Participants demonstrated cognitive decline with an average MMSE score decline of 1.11 per year. Carriers of ε4/ε4 alleles in APOE had significantly 3.68-fold risks of MMSE decline. • What is the meaning of the finding? Although the annual PM 2.5 levels decreased over time, the results still indicated that long-term exposure to PM 2.5 , was associated with an increased risk of MMSE decline. Limited research has explored the long-term effect of reduced PM 2.5 exposure on cognitive function. This study aimed to investigate the effects of time-dependent PM 2.5 exposure and the interactions of PM 2.5 and aging on declines in Mini-Mental State Examination (MMSE) scores, in carriers and non-carriers of the APOE -ε4 allele. Participants aged over 60 were recruited for this cohort study, undergoing MMSE tests twice from the Taiwan Biobank Program from 2008 to 2020. Participants with dementia or baseline MMSE scores <24 were excluded. Annual PM 2.5 levels were estimated using a hybrid kriging/land use regression model with extreme gradient boosting, treated as a time-dependent variable. Generalized estimating equations were used to assess the impacts of repeated PM 2.5 on MMSE decline, further stratified by the presence of APOE -ε4 alleles. After follow-up, 290 participants out of the overall 7,000 community residents in the Biobank dataset demonstrated incidences of MMSE declines (<24), with an average MMSE score decline of 1.11 per year. Participants with ε4/ε4 alleles in the APOE gene had significantly 3.68-fold risks of MMSE decline. High levels of PM 2.5 across all visits were significantly associated with worsening of scores on the overall MMSE. As annual levels of PM 2.5 decreased over time, the impact of PM 2.5 on MMSE decline also slowly diminished. Long-term PM 2.5 exposure may be associated with increased risk of MMSE decline, despite improvements in ambient PM 2.5 levels over time. Validation of these results necessitates a large-scale prospective cohort study with more concise cognitive screening tools. [ABSTRACT FROM AUTHOR]

PM 2.5 concentrations are higher during rush hours at background stations compared to the average concentration across these stations. Few studies have investigated PM 2.5 concentration and its spatial distribution during rush hours using machine learning models. This study employs a geospatial-artificial intelligence (Geo-AI) prediction model to estimate the spatial and temporal variations of PM 2.5 concentrations during morning and dusk rush hours in Taiwan. Mean hourly PM 2.5 measurements were collected from 2006 to 2020, and aggregated into morning (7 a.m.–9 a.m.) and dusk (4 p.m.–6 p.m.) rush-hour mean concentrations. The Geo-AI prediction model was generated by integrating kriging interpolation, land-use regression, machine learning, and a stacking ensemble approach. A forward stepwise variable selection method based on the SHapley Additive exPlanations (SHAP) index was used to identify the most influential variables. The performance of the Geo-AI models for morning and dusk rush hours had accuracy scores of 0.95 and 0.93, respectively and these results were validated, indicating robust model performance. Spatially, PM 2.5 concentrations were higher in southwestern Taiwan for morning rush hours, and suburban areas for dusk rush hours. Key predictors included kriged PM 2.5 values, SO 2 concentrations, forest density, and the distance to incinerators for both morning and dusk rush hours. These PM 2.5 estimates for morning and dusk rush hours can support the development of alternative commuting routes with lower concentrations. [Display omitted] • Rush-hour exposure to PM 2.5 is high, but the spatiality is less understood. • PM 2.5 variations during morning and dusk rush hours are captured by Geo-AI models. • The Geo-AI model performance for morning/dusk is 0.95 and 0.93, respectively. • PM 2.5 is higher in southern Taiwan and urban areas in the morning. • Low-exposure commuting routes could be identified by the rush-hour Geo-AI models. [ABSTRACT FROM AUTHOR]

For indoor air modelling, difficulties in collecting indoor parameters including life activity patterns and building characteristics are dilemmas when conducting a large-area study. Land-use/land cover information which is easier to obtain could represent as surrogates of emission sources for assessing indoor air quality. Moreover, low-cost sensors and machine learning provide a better way to enhance model accuracy. This study proposed an alternative estimation approach to assess daily PM 2.5 concentration for indoor environments of schools in a large area by integrating low-cost sensors, land-use/land cover predictors, and machine learning-based modelling approaches. Indoor PM 2.5 data was collected from 145 indoor AirBox sensors in Kaohsiung and Pingtung Counties of Taiwan. Geospatial predictors were extracted from the circular buffers surrounding each AirBox sensor. Spearman correlation analysis and stepwise variable selection procedures were performed to select variables for land-use regression (LUR) and integrated with XGBoost, Random Forest (RF), and LGBM machine learning models. The results revealed that outdoor PM 2.5 and distance to the nearest thermal power plant were the main determinants of indoor estimation variations, when there were no indoor sources. When incorporating machine learning, the R2 increased from 0.59 for LUR to 0.85 for LUR-XGBoost while the RMSE decreased from 8.63 to 5.27 μg/m3, which performed better than both LUR-RF and LUR-LGBM. This study demonstrates the value of the proposed alternative approach by incorporating data from a low-cost sensor with LUR model and machine learning algorithm in estimating the spatiotemporal variability of indoor PM 2.5 for a large area. [Display omitted] • An alternative approach was proposed to estimate indoor PM 2.5 for school campuses. • Low-cost sensor and land-use/land cover were incorporated using machine learning. • Important land-use/land cover variables affecting indoor PM 2.5 were identified. • The R2 of indoor LUR-XGBoost model was 0.85 with RMSE 5.27 μg/m3. [ABSTRACT FROM AUTHOR]

It is likely that exposure surrogates from monitoring stations with various limitations are not sufficient for epidemiological studies covering large areas. Moreover, the spatiotemporal resolution of air pollution modelling approaches must be improved in order to achieve more accurate estimates. If not, the exposure assessments will not be applicable in future health risk assessments. To deal with this challenge, this study featured Land-Use Regression (LUR) models that use machine learning to assess the spatial-temporal variability of Nitrogen Dioxide (NO 2). Daily average NO 2 data was collected from 70 fixed air quality monitoring stations, belonging to the Taiwanese EPA, on the main island of Taiwan. Around 0.41 million observations from 2000 to 2016 were used for the analysis. Several datasets were employed to determine spatial predictor variables, including the EPA environmental resources dataset, the meteorological dataset, the land-use inventory, the landmark dataset, the digital road network map, the digital terrain model, MODIS Normalized Difference Vegetation Index database, and the power plant distribution dataset. Regarding analyses, conventional LUR and Hybrid Kriging-LUR were performed first to identify important predictor variables. A Deep Neural Network, Random Forest, and XGBoost algorithms were then used to fit the prediction model based on the variables selected by the LUR models. Lastly, data splitting, 10-fold cross validation, external data verification, and seasonal-based and county-based validation methods were applied to verify the robustness of the developed models. The results demonstrated that the proposed conventional LUR and Hybrid Kriging-LUR models captured 65% and 78%, respectively, of NO 2 variation. When the XGBoost algorithm was further incorporated in LUR and hybrid-LUR, the explanatory power increased to 84% and 91%, respectively. The Hybrid Kriging-LUR with XGBoost algorithm outperformed all other integrated methods. This study demonstrates the value of combining Hybrid Kriging-LUR model and an XGBoost algorithm to estimate the spatial-temporal variability of NO 2 exposure. For practical application, the associations of specific land-use/land cover types selected in the final model can be applied in land-use management and in planning emission reduction strategies. [Display omitted] • Estimating long-term daily NO 2 concentration with machine learning models. • Land-use patterns were included in machine learning models by using land-use regression. • The most contributed predictors were identified by stepwise variable selection. • Explanatory power of daily NO 2 concentration was increased from 0.65 to 0.91. • XGboost outperformed RF and DNN machine learning algorithms. Capsule: The explanatory power of Hybrid Kriging-LUR coupled with XGBoost algorithm on daily NO 2 variations reached 91% and outperformed all other integrated methods. [ABSTRACT FROM AUTHOR]

Ambient fine particulate matter (PM 2.5) has been ranked as the sixth leading risk factor globally for death and disability. Modelling methods based on having access to a limited number of monitor stations are required for capturing PM 2.5 spatial and temporal continuous variations with a sufficient resolution. This study utilized a land use regression (LUR) model with machine learning to assess the spatial-temporal variability of PM 2.5. Daily average PM 2.5 data was collected from 73 fixed air quality monitoring stations that belonged to the Taiwan EPA on the main island of Taiwan. Nearly 280,000 observations from 2006 to 2016 were used for the analysis. Several datasets were collected to determine spatial predictor variables, including the EPA environmental resources dataset, a meteorological dataset, a land-use inventory, a landmark dataset, a digital road network map, a digital terrain model, MODIS Normalized Difference Vegetation Index (NDVI) database, and a power plant distribution dataset. First, conventional LUR and Hybrid Kriging-LUR were utilized to identify the important predictor variables. Then, deep neural network, random forest, and XGBoost algorithms were used to fit the prediction model based on the variables selected by the LUR models. Data splitting, 10-fold cross validation, external data verification, and seasonal-based and county-based validation methods were used to verify the robustness of the developed models. The results demonstrated that the proposed conventional LUR and Hybrid Kriging-LUR models captured 58% and 89% of PM 2.5 variations, respectively. When XGBoost algorithm was incorporated, the explanatory power of the models increased to 73% and 94%, respectively. The Hybrid Kriging-LUR with XGBoost algorithm outperformed the other integrated methods. This study demonstrates the value of combining Hybrid Kriging-LUR model and an XGBoost algorithm for estimating the spatial-temporal variability of PM 2.5 exposures. [Display omitted] • Estimating long-term daily PM 2.5 concentration with machine learning models. • Land-use patterns were included in machine learning models by using land-use regression. • Explanatory power of daily PM 2.5 concentration was increased from 0.58 to 0.94. • XGboost outperformed random forest and deep neural network algorithms. [ABSTRACT FROM AUTHOR]

This paper is the first of its kind to use machine learning algorithms in conjunction with a Land-use Regression (LUR) model for predicting the spatiotemporal variation of CO concentrations in Taiwan. We used daily CO concentration from 2000 to 2016 to develop model and data from 2017 to 2018 as external data to verify the model reliability. Location of temples was used as a predictor to account for Asian culturally specific sources. With the ability to capture nonlinear relationship between observations and predictions, three LUR-based machine learning algorithms were used to estimate CO concentrations, including deep neural network (DNN), random forest (RF), and extreme gradient boosting (XGBoost). The results showed that LUR-based machine-learning model (LUR-XGBoost) has the best computation efficiency and improved adjusted R2 from 0.69 to 0.85. Our studies demonstrate the ability of the LUR-based machine learning algorithms to estimate long-term spatiotemporal CO concentration variations in fine resolution. [Display omitted] • Long-term daily CO concentrations were estimated with LUR-machine learning models. • Land-use patterns were included in machine learning models by using land-use regression. • The most contributed predictors were identified by stepwise variable selection. • Explanatory power of daily CO concentration was increased from 0.69 to 0.85. • XGboost outperformed RF and DNN machine learning algorithms. [ABSTRACT FROM AUTHOR]

It is well known benzene negatively impacts human health. This study is the first to predict spatial-temporal variations in benzene concentrations for the entirety of Taiwan by using a mixed spatial prediction model integrating multiple machine learning algorithms and predictor variables selected by Land-use Regression (LUR). Monthly benzene concentrations from 2003 to 2019 were utilized for model development, and monthly benzene concentration data from 2020, as well as mobile monitoring vehicle data from 2009 to 2019, served as external data for verifying model reliability. Benzene concentrations were estimated by running six LUR-based machine learning algorithms; these algorithms, which include random forest (RF), deep neural network (DNN), gradient boosting (GBoost), light gradient boosting (LightGBM), CatBoost, extreme gradient boosting (XGBoost), and ensemble algorithms (a combination of the three best performing models), can capture how nonlinear observations and predictions are related. The results indicated conventional LUR captured 79% of the variability in benzene concentrations. Notably, the LUR with ensemble algorithm (GBoost, CatBoost, and XGBoost) surpassed all other integrated methods, increasing the explanatory power to 92%. This study establishes the value of the proposed ensemble-based model for estimating spatiotemporal variation in benzene exposure. [Display omitted] • Estimating long-term ambient benzene concentrations using a mixed spatial prediction model. • Land-use patterns in the developed model can be used for targeted control strategies. • Explanatory power of monthly benzene concentration was improved from 0.79 to 0.92. • Ensemble-based model outperformed all other selected machine learning algorithms. [ABSTRACT FROM AUTHOR]

Estimating short-term exposure to PM 2.5 has been achieved for population health studies using the land use regression with machine learning (LUR_ML) and microenvironmental exposure (ME) models. However, there is a lack of clarity regarding the performance of these models in predicting PM 2.5 exposure for individuals residing in diverse environments, and the factors influencing the variations in accuracy between these models. This study performed the LUR_ML and ME models to estimate daily exposure concentrations of PM 2.5 for elders residing in urban, suburban, rural, and industrial regions in Taiwan. The accuracy of the model predictions was assessed by comparing them with personal PM 2.5 monitoring for both overall and regional assessments. The LUR_ML model demonstrated reasonably moderate agreement (R2 = 0.516) overall with personal exposure to PM 2.5 , while the ME models exhibited relatively higher predictions (R2 = 0.535–0.575) and lower biases. The agreement of PM 2.5 predictions varies across regions, particularly in areas with higher exposure contrast. The ME model 1, utilizing region-specific microenvironmental measurements rather than generic data, highlights the potential for accurate prediction of personal PM 2.5 exposure. This study contributed to the understanding of variations in prediction accuracy across different regions and support the need for improved exposure models of air pollution. [Display omitted] • The land use regression with machine learning moderately agreed (R2 = 0.516) with PM 2.5 exposure. • Microenvironmental exposure models had a higher prediction (R2 = 0.535–0.575) and lower bias. • Exposure predictions of daily PM 2.5 concentrations vary by regions. • Indoor exposure is a stronger predictor of personal PM 2.5 exposure than outdoor exposure. • Higher PM 2.5 exposure agreement is likely due to greater exposure contrast in the population. [ABSTRACT FROM AUTHOR]