%0 PDF
%T Optimizing the Wind Variable Formulation in a Multiple Linear Regression Model of Particle Number Concentration
%A Collins, Caitlin Laurel
%8 2005-06-20
%I Digital Collections and Archives, Tufts University
%R http://localhost/files/vq280048s
%X Airborne ultrafine particles (UFP; diameter < 100 nm) are typically present at high levels near highways. Due to their small size, UFP may have more adverse health impacts than larger particles. The concentration of UFP can vary widely over a small area, especially in an urban setting, so many monitoring locations are necessary to accurately estimate UFP concentrations on an urban scale. Because of this, a model of particle number concentration (PNC; a proxy for UFP) based on mobile monitoring data can present a less costly alternative for estimating UFP levels than deploying a monitoring network. Wind is known to have an effect on PNC, so it is important that any model of PNC with high temporal resolution include an effective wind variable. Because wind is a vector quantity (has both direction and magnitude), it can be difficult to represent in a model. Many air pollution models entirely omit the effects of wind or only include wind speed as a variable. The objectives of this thesis were (1) investigate a suite of different wind variables used in air pollution modeling by including them in multiple linear regressions of PNC, (2) compare the regressions using goodness-of-fit statistics such as adjusted R2, and (3) validate the regressions by using regression diagnostics to determine whether modeling assumptions were met. To investigate the relative merits of different wind variables, a multiple linear regression of the logarithm of hourly PNC (Adj. R2 = 0.22) was augmented with one of four types of wind variables: wind speed, wind sectors, a vector-based variable, or a Fourier-based variable. Wind sectors based on wind direction relative to the highway performed best (Adj. R2 > 0.30), while the vector-based variables performed poorly (Adj. R2 < 0.25; less than wind speed alone). The wind sectors may have performed better than vector-based variables because only one weather station was used, which meant that local turbulence was not accounted for, thus leading to incorrect wind direction input to the vector variables. The highway-relative wind sectors may have performed better than the conventional sectors because they took the effect of the highway, which is considered to be a major source of UFP within the study area, into account. The findings from this research will inform a regression model of PNC that will aid in assessing exposure to UFP for the study participants of CAFEH.
%[ 2019-10-09
%~ Tufts Digital Library
%W Institution