A hybrid model for spatially and temporally resolved ozone exposures in the continental United States

Abstract

Ground-level ozone is an important atmospheric oxidant, which exhibits considerable spatial and temporal variability in its concentration level. Existing modeling approaches for ground-level ozone include chemical transport models, land-use regression, Kriging, and data fusion of chemical transport models with monitoring data. Each of these methods has both strengths and weaknesses. Combining those complementary approaches could improve model performance. Meanwhile, satellite-based total column ozone, combined with ozone vertical profile, is another potential input. We propose a hybrid model that integrates the above variables to achieve spatially and temporally resolved exposure assessments for ground-level ozone. We used a neural network for its capacity to model interactions and nonlinearity. Convolutional layers, which use convolution kernels to aggregate nearby information, were added to the neural network to account for spatial and temporal autocorrelation. We trained the model with AQS 8-hour daily maximum ozone in the continental United States from 2000 to 2012 and tested it with left out monitoring sites. Cross-validated R2 on the left out monitoring sites ranged from 0.74 to 0.80 (mean 0.76) for predictions on 1 km×1 km grid cells, which indicates good model performance. Model performance remains good even at low ozone concentrations. The prediction results facilitate epidemiological studies to assess the health effect of ozone in the long term and the short term.

Association of Short-term Exposure to Air Pollution With Mortality in Older Adults

“…Details have been published elsewhere. 12,13 Warm season is defined to be from April 1 to September 30, which is the specific time window to examine the association between ozone and mortality. Meteorological variables including air and dew point temperatures were retrieved from North American Regional Reanalysis data and estimated daily mean values were determined for each 32 km × 32 km grid in the continental US.…”

A Random Forest Approach to Estimate Daily Particulate Matter, Nitrogen Dioxide, and Ozone at Fine Spatial Resolution in Sweden

“…In contrast with PM 2.5 and PM 10 , there are very few studies applying machine-learning methods to predict coarse PM [20], NO 2 [50], or O 3 [19] at fine spatiotemporal resolution over large geographical areas, and none of them conducted in Sweden. In a previous study conducted in Italy, we applied the same methodology proposed here to predict coarse PM, and we were able to predict 77% of PM 2.5-10 variability in OOB samples and 62% in held-out monitors [20].…”

“…Ozone levels are much higher today than in the pre-industrial era, and there are concerns of future increases related to global warming [18]. However, predicting ozone concentrations at fine spatial and temporal concentrations is extremely difficult because many parameters related to local sources, land-use characteristics, and meteorological conditions are involved in ozone formation and removal, resulting in high spatial and temporal variability [19].…”

Informatics and Data Analytics to Support Exposome-Based Discovery for Public Health

“…Methods such as canonical correlation analysis or graphical least angle regression (graphical LASSO) techniques can enable investigators to map one large data set (e.g., external exposures) to another (e.g., internal exposures) (13), but missing are methods to consider longitudinal data. Computational methods enjoying a resurgence in the data science community, such as neural networks, may also be harnessed to assimilate data over different dimensions (11). …”