Aerosol Retrievals from Individual AVHRR Channels. Part II: Quality Control, Probability Distribution Functions, Information Content, and Consistency Checks of Retrievals

This second part of a two-part study evaluates retrievals of aerosol optical depths, 1 and 2 , in Advanced Very High Resolution Radiometer (AVHRR) channels 1 and 2 centered at 1 0.63 and 2 0.83 m, and an effective A ngstro m exponent, , derived therefrom as ln( 1 / 2 )/ln( 1 / 2 ). The retrievals are made with the Second Simulation of the Satellite Signal in the Solar Spectrum (6S) radiative transfer model from four NOAA-14 AVHRR datasets, collected between February 1998 and May 1999 in the latitudinal belt of 5-25S. A series of quality control (QC) checks applied to the retrievals to identify outliers are described. These remove a total of 1% of points, which presumably originate from channel misregistration, residual cloud in AVHRR cloud-screened pixels, and substantial deviations from the assumptions used in the retrieval model (e.g., bright coastal and high altitude inland waters). First, from examining histograms of the derived parameters it is found that and are accurately fit by lognormal and normal probability distribution functions (PDFs), respectively. Second, the scattergrams 1 versus 2 are analyzed to see if they form a coherent pattern. They do indeed converge at the origin, as expected, but frequently are outside of the expected domain in 1 -2 space, defined by two straight lines corresponding to 0 and 2. This results in a low bias in , which tends to fill in an interval of [1, 1] rather than [0, 2]. Third, scattergrams of versus are used to empirically confirm a previously drawn theoretical conclusion that errors in are inversely proportional to . More in-depth quantitative analyses suggest that the AVHRR-derived A ngstro m exponent becomes progressively more meaningful when 0.2. Geographical trends are studied to demonstrate that the selected ocean area is reasonably uniform to justify application of consistency checks to reveal angular trends in the retrievals. These checks show that in most cases, the artifacts in the retrieved and are statistically insignificant. On average, the analyses suggest that the retrieved 1 , 2 , and show a high degree of self-and interconsistency, with the exception of a troublesome May 1999 dataset. The most prominent problem noticed so far is the inconsistency between 1 and 2 , persistent from one dataset to another, which calls for fine-tuning some (not aerosol-model related) elements of the retrieval algorithm. These adjustments will be discussed elsewhere.