Repetitive genetic inversion of optical extinction data

We describe a genetic method of deriving aerosol size distributions from multiwavelength extinction measurements. The genetic inversion searches for log-normal size distribution parameters whose calculated extinctions best fit the data. By repetitively applying the genetic inversion using different random number seeds, we are able to generate multiple solutions that fit the data equally well. When these solutions are similar, they lend confidence to an interpretation, whereas when they vary widely, they demonstrate nonuniqueness. In this way we show that, even in the case of a single log-normal distribution, many different distributions can fit the same set of extinction data unless the misfit is reduced below typical measurement error levels. In the case of a bimodal distribution, we find many dissimilar size distributions that fit the data to within 1% at six wavelengths. To recover the original bimodal distribution satisfactorily, we found that extinctions at ten wavelengths must be fitted to within 0.5%. Our results imply that many size distributions recovered from existing extinction measurements can be highly nonunique and should be treated with caution.