Retrieval of aerosol optical thickness by means of the least-median-squares robust algorithm

Atmospheric aerosols play an active role in the atmospheric radiative energy transfer by interacting with the solar energy through light scattering and/or absorption. Such an interaction has the potential to counter balance or enhance the warming caused by the greenhouse gases (e.g, carbon dioxide, water vapor, and methane). The interactions would depend on a variety of factors such as the microphysics and chemistry of aerosol particles, and the environmental conditions. To model the energy transfer processes accurately, several aerosol parameters such as aerosol optical thickness (AOT), single scattering albedo, scattering phase function, refractive index, and chemical composition of aerosols have to be known. In this report, a robust method for aerosol optical thickness is presented. The method is based on the least median squares (LMS) regression technique. The LMS-based technique is fundamentally different From the traditional least-squares (LS) technique commonly used today to retrieve AOT. The LMS technique was found to resist influential outliers and sustain the impacts of outliers much better than the LS in our application. The outlier-resistance property is an important design consideration for an automatic AOT retrieval algorithm. We demonstrated the strength of the new technique by using the shortwave irradiance measurements taken by one of the Multi-filter Rotating Shadow-band Radiometers installed at the Southern Great Plains in Oklahoma, U.S.A., one of the three sites operated by the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) Program. The data cover a time period from September 1995 to May 1997. The LMS-retrieved mean top-of-atmosphere solar irradiance values (i.e. I-o) were stable based on the one-tailed Student's t tests. It was also found that the difference between the 75th and 25th percentiles (a non-parametric robust estimate of variation) of I-o over the 2-year period was 0.2 for the 499-nm channel, and 0.05 for the 860-nm channel. The monthly averaged AOT values over the 2-year period were from 0.05 to 0.30 for the 499-nm channel and from 0.03 to 0.15 for the 860-nm channels. AOT values peaked between May and September and reached minima in between November and January. The synoptic flow from the North might have contributed to the low level of AOT values during the wintertime, while the southerly flow, traversing through more industrial areas before reaching the site, could have contributed to the influx of particles during the summertime. (C) 1999 Elsevier Science Ltd. All rights reserved.