AAAR 37th Annual Conference October 14 - October 18, 2019 Oregon Convention Center Portland, Oregon, USA
Abstract View
The Role of Size Distribution Representation in Aerosol Data Inversion
RICHARD FLAGAN, Amanda Grantz, Yuanlong Huang, California Institute of Technology
Abstract Number: 396 Working Group: Instrumentation and Methods
Abstract Recovery of the particle size distribution from aerosol measurements involves: (i) the transfer function that describes the signals produced by particles of different sizes over the size range of the measurement; (ii) an algorithm for extracting a set of parameters that characterize the particle size distribution; and (iii) a mathematical representation of the particle size distribution. The transfer function is generally based upon empirical characterization of the instrument, or a theoretical model of the instrument that may be combined with empirical corrections that account for deviations of the real instrument from an idealized model. The best-studied measurement system is the differential mobility particle sizer (DMPS). The scanning electrical mobility spectrometer (SEMS or SMPS) has recently been advanced to similar fidelity through detailed numerical models of the real instrument. Algorithms for inverting the signals range from ad hoc successive approximation methods such as the Twomey algorithm, direct matrix inversion, least-squares and non-negative least-squares methods, and constrained regularization, among others. The representation of the particle size distribution has received less critical attention than the first two components of the data inversion problem. They include multi-modal log-normal representions, nodal (delta-function) representations, histograms or linear splines. Many papers fail to even mention which representation is being employed. Different size distribution representations impact the fidelity of data inversion. We simulate measurements to quantify biases and sources of uncertainty introduced during data inversion. The results reveal increased uncertainty when the size distribution extends beyond the measurement range of a single instrument, and that the magnitude of that uncertainty depends strongly upon the representation of the size distribution that is employed. We further demonstrate ways to reduce that uncertainty without increasing the number of parameters required to represent the size distribution, using codes that will be released open-source.