Mapping real-time air pollution health risk for environmental management: Combining mobile and stationary air pollution monitoring with neural network models.

Air pollution poses health concerns at the global scale. The challenge of managing air pollution is significant because of the many air pollutants, insufficient funds for monitoring and abatement programs, and political and social challenges in defining policy to limit emissions. Some governments provide citizens with air pollution health risk information to allow them to limit their exposure. However, many regions still have insufficient air pollution monitoring networks to provide real-time mapping. Where available, these risk mapping systems either provide absolute concentration data or the concentrations are used to derive an Air Quality Index, which provides the air pollution risk for a mix of air pollutants with a single value. When risk information is presented as a single value for an entire region it does not inform on the spatial variation within the region. Without an understanding of the local variation residents can only make a partially informed decision when choosing daily activities. The single value is typically provided because of a limited number of active monitoring units in the area. In our work, we overcome this issue by leveraging mobile air pollution monitoring techniques, meteorological information and land use information to map real-time air pollution health risks. We propose an approach that can provide improved health risk information to the public by applying neural network models within a framework that is inspired by land use regression. Mobile air pollution monitoring campaigns were conducted across Hamilton from 2005 to 2013. These mobile air pollution data were modelled with a number of predictor variables that included information on the surrounding land use characteristics, the meteorological conditions, air pollution concentrations from fixed location monitors, and traffic information during the time of collection. Fine particulate matter and nitrogen dioxide were both modelled. During the model fitting process we reserved twenty percent of the data to validate the predictions. The models' performances were measured with a coefficient of determination at 0.78 and 0.34 for PM2.5 and NO2, respectively. We apply a relative importance measure to identify the importance of each variable in the neural network to partially overcome the black box issues of neural network models.

Spatial patterns of fetal loss and infant death in an arsenic-affected area in Bangladesh.

BACKGROUND: Arsenic exposure in pregnancy is associated with adverse pregnancy outcome and infant mortality. Knowledge of the spatial characteristics of the outcomes and their possible link to arsenic exposure are important for planning effective mitigation activities. The aim of this study was to identify spatial and spatiotemporal clustering of fetal loss and infant death, and spatial relationships between high and low clusters of fetal loss and infant death rates and high and low clusters of arsenic concentrations in tube-well water used for drinking. METHODS: Pregnant women from Matlab, Bangladesh, who used tube-well water for drinking while pregnant between 1991 and 2000, were included in this study. In total 29,134 pregnancies were identified. A spatial scan test was used to identify unique non-random spatial and spatiotemporal clusters of fetal loss and infant death using a retrospective spatial and spatiotemporal permutation and Poisson probability models. RESULTS: Two significant clusters of fetal loss and infant death were identified and these clusters remained stable after adjustment for covariates. One cluster of higher rates of fetal loss and infant death was in the vicinity of the Meghna River, and the other cluster of lower rates was in the center of Matlab. The average concentration of arsenic in the water differed between these clusters (319 µg/L for the high cluster and 174 µg/L for the low cluster). The spatial patterns of arsenic concentrations in tube-well water were found to be linked with the adverse pregnancy outcome clusters. In the spatiotemporal analysis, only one high fetal loss and infant death cluster was identified in the same high cluster area obtained from purely spatial analysis. However, the cluster was no longer significant after adjustment for the covariates. CONCLUSION: The finding of this study suggests that given the geographical variation in tube-well water contamination, higher fetal loss and infant deaths were observed in the areas of higher arsenic concentrations in groundwater. This illustrates a possible link between arsenic contamination in tube-well water and adverse pregnancy outcome. Thus, these areas should be considered a priority in arsenic mitigation programs.

Spatial modelling of individual arsenic exposure via well water: evaluation of arsenic in urine, main water source and influence of neighbourhood water sources in rural Bangladesh.

Arsenic concentrations in well water often vary even within limited geographic areas. This makes it difficult to obtain valid estimates of the actual exposure, as people may take their drinking water from different wells. We evaluated a spatial model for estimation of the influence of multiple neighbourhood water sources on the actual exposure, as assessed by concentrations in urine in a population in rural Bangladesh. In total 1307 individuals (one per bari, group of families) were randomly selected. Arsenic concentrations of urine and water were analysed. Simple average and inverse distance weighted average of arsenic concentrations in the five nearest water sources were calculated for each individual. Spatial autocorrelation was evaluated using Moran's I statistics, and spatial regression models were employed to account for spatial autocorrelation. The average distance from a household to the nearest tube-well was 32 metres (Inter-Quartile Range 1-49 metres). Water arsenic concentrations of the reported main water sources were significantly correlated with concentrations in urine (R(2) = 0.41, rho < 0.0001, R(2) for women = 0.45 and for men = 0.36). General model fit improved only slightly after spatial adjustment for neighbouring water sources (pseudo-R(2) = 0.53, spatial lag model), compared to covariate adjusted regression coefficient (R(2) = 0.46). Arsenic concentration in urine was higher than arsenic in main water source with an intercept of 57 microg L(-1), indicating exposure from food. A suitable way of estimating an individual's past exposure to arsenic in this rural setting, where influence of neighbouring water sources was minimal, was to consider the reported main source of drinking water.

Code Red: Explaining Average Age of Death in the City of Hamilton.

The aim of this study is to identify the underlying factors that explain the average age of death in the City of Hamilton, Ontario, Canada, as identified in the Code Red Series of articles that were published in the city's local newspaper in 2010. Using a combination of data from the Canadian Census, the Government of Ontario and the Canadian Institute for Health Information, factor analysis was performed yielding three factors relating to poverty, working class, and health and aging. In a regression analysis these factors account for 42% of the total variability in the average ages of death observed at the census tract level of geography within the city.

Assessing the health impacts of air pollution: a re-analysis of the Hamilton children's cohort data using a spatial analytic approach.

The objective of this paper was to reassess children's exposure to air pollution as well as investigate the importance of other covariates of respiratory health. We re-examined the Hamilton Children's Cohort (HCC) dataset with enhanced spatial analysis methods, refined in the approximately two decades since the original study was undertaken. Children's exposure to air pollution was first re-estimated using kriging and land-use regression. The land-use regression model performed better, compared to kriging, in capturing local variation of air pollution. Multivariate linear and logistic regression analysis was then applied for the study of potential risk factors for respiratory health. Findings agree with the HCC study-results, confirming that children's respiratory health was associated with maternal smoking, hospitalization in infancy and air pollution. However, results from this study reveal a stronger association between children's respiratory health and air pollution. Additionally, this study demonstrated associations with low-income, household crowding and chest illness in siblings.