Modelled and measured effects of clouds on UV Aerosol Indices on a local, regional, and global scale
Abstract. The UV Aerosol Indices (UVAI) form one of very few available tools in satellite remote sensing that provide information on aerosol absorption. The UVAI are also quite insensitive to surface type and are determined in the presence of clouds – situations where most aerosol retrieval algorithms do not work. The UVAI are most sensitive to elevated layers of absorbing aerosols, such as mineral dust and smoke, but they can also be used to study non-absorbing aerosols, such as sulphate and secondary organic aerosols. Although UVAI are determined for cloud-contaminated pixels, clouds do affect the value of UVAI in several ways: (1) they shield the underlying scene (potentially containing aerosols) from view, (2) they enhance the apparent surface albedo of an elevated aerosol layer, and (3) clouds unpolluted by aerosols also yield non-zero UVAI, here referred to as "cloudUVAI".
The main purpose of this paper is to demonstrate that clouds can cause significant UVAI and that this cloudUVAI can be well modelled using simple assumptions on cloud properties. To this aim, we modelled cloudUVAI by using measured cloud optical parameters – either with low spatial resolution from SCIAMACHY, or high resolution from MERIS – as input. The modelled cloudUVAI were compared with UVAI determined from SCIAMACHY reflectances on different spatial (local, regional and global) and temporal scales (single measurement, daily means and seasonal means). The general dependencies of UVAI on cloud parameters were quite well reproduced, but several issues remain unclear: compared to the modelled cloudUVAI, measured UVAI show a bias, in particular for large cloud fractions. Also, the spread in measured UVAI is larger than in modelled cloudUVAI.
In addition to the original, Lambert Equivalent Reflector (LER)-based UVAI algorithm, we have also investigated the effects of clouds on UVAI determined using the so-called Modified LER (MLER) algorithm (currently applied to TOMS and OMI data). For medium-sized clouds the MLER algorithm performs better (UVAI are closer to 0), but like for LER UVAI, MLER UVAI can become as large as −1.2 for small clouds and deviate significantly from zero for cloud fractions near 1. The effects of clouds should therefore also be taken into account when MLER UVAI data are used.
Because the effects of clouds and aerosols on UVAI are not independent, a simple subtraction of modelled cloudUVAI from measured UVAI does not yield a UVAI representative of a cloud-free scene when aerosols are present. We here propose a first, simple approach for the correction of cloud effects on UVAI. The method is shown to work reasonably well for small to medium-sized clouds located above aerosols.