Genetic Algorithm Retrieval of Real Refractive Index from Aerosol Distributions that are not Lognormal

Abstract

Size distributions of real-world aerosols typically deviate substantially from log-normal or other simple mathematical descriptions. The effect of assuming a lognormal shape for distributions that are not well described by this function on the genetic algorithm (GA) retrieval of the real refractive index (mr) from scattering properties is investigated. Tests using several laboratory-generated ammonium sulfate aerosols with relatively broad bimodal distributions and with synthetic distributions generated by combining two log-normal distributions are examined. Scattering of the ammonium sulfate aerosols was measured with a dual polarization polar nephelometer. Analysis of the GA retrieval with the assumption that the distribution is lognormal for both the ammonium sulfate aerosol and synthetic distorted distributions show that the retrievedmr is within 0.015 of the expected value in all cases tested. The retrieved values for both the mean and the standard deviation of the aerosol size distribution become more inaccurate as the level of distortion increases. In our experiments, the GA determined mean aerosol diameter is always larger than the measured mean because in all of the distorted distributions there are significantly more large particles than the lognormal assumption. The intensity of particle scattering is generally a function of the square of the particle diameter hence as the distortion increases, the larger particles contribute more to the overall scattering on which the GA retrieval is based. We show that assuming a distribution is lognormal could potentially introduce errors in calculating radiative forcing by at least 12%. The total scattering coefficient measured by integrating nephelometers can be off by a factor of 2 for distortion levels similar to those seen in some of our experiments.