Land-use regression modeling of wood smoke in Rochester, NY

During the last decade, a substantial rise in the use of wood for space and water heating followed the increased cost of fossil fuels in Northern America. The use of this renewable source of energy has many advantages related to the carbon cycling, greenhouse gas emissions and climate. However, emissions from wood combustion sources may seriously impact air quality. The relationships between wood smoke exposure and health effects have been studied: epidemiology suggests that the toxicity of wood combustion particles is similar to urban particulate matter. During the 2015/16 heating season, ambient particulate matter (PM) concentrations were measured at 23 sites across the Monroe County using low-cost monitors. Data were corrected based on the results of previous calibration studies. The corrected data were assessed to have good reproducibility, 10% precision and 10 μg/m3 limit of detection. Most of the sites showed clear diurnal patterns with higher concentration in the late afternoon and evening hours. Weekly pattern were more variable, but a general decrease of concentrations was found during the weekends. The hourly and weekly patterns of PM were spatially interpolated. The results show that ambient PM concentrations are generally higher over the more densely urbanized areas of the county. Data were then used as input for a land use regression model. Raster surfaces (spatial resolution 10 x 10 m) were developed for a series of predictors which are potentially related to the concentration of PM, including number of bedroom, fireplaces, kitchens, property value, property year built, road type and road traffic densities, elevation and density of various land cover data features. Circular buffers were calculated around the 23 sampling sites with radius ranging from 50 m to 5000 m (50 m steps) and buffer statistics were computed. Pearson correlation coefficients were calculated between hourly and weekday averages of PM concentrations and the statistics of predictors for all the buffers. In general, the relationships between PM concentrations and the predictors are moderate and comparable with the results of previous studies. A Deletion/Substitution/Addition (D/S/A) algorithm was then used to maximize the prediction accuracy for locations without measurements.