Aerosol Radiative Effect

Introduction Conclusions References Tables Figures ◭ ◮ ◭ ◮ Back Close Full Screen / Esc Abstract Introduction Conclusions References Tables Figures ◭ ◮ ◭ ◮ Back Close Full Screen / Esc Abstract The statistical uncertainty of globally averaged MODIS aerosol optical thickness at 0.55 µm (AOT) and top of atmosphere CERES cloud-free shortwave radiative effect (SWRE) is presented. All analysis is presented at the CERES footprint level which we call " raw data ". Statistical uncertainty may result from the raw data not being nor-5 mally distributed. Both the AOT and SWRE data derived from clear-sky CERES-SSF products show significant deviations from a normal distribution as evidenced by high skewness values. The spatial and temporal distribution of the data is also not uniform, with a greater concentration of data being in aerosol heavy regions. As a result, globally averaged AOT and SWRE are overestimated when derived from raw data. Raw 10 data are gridded into 2×2 degree grid-cells (called " gridded " data) to reduce the effect of spatial non-uniformity. However, the underlying non-normal distribution remains and manifests itself by increasing the uncertainty of grid-cell values. Globally averaged AOT and SWRE derived from a gridded dataset are substantially lower than those derived from raw data alone. The range of globally averaged AOT and SWRE val-15 ues suggests that up to a 50% statistical uncertainty exists, much of which is directly tied to how the data are manipulated prior to averaging. This uncertainty increases when analyzing aerosol components (e.g. anthropogenic) since component AOT (and SWRE) may not exist at all locations were AOT is present. As a result, regions where a particular component AOT does not exist must either not be included in the global 20 average or have data within these regions set to null values. However, each method produces significantly different results. The results of this work indicate that placing raw observations on to a uniform grid is a necessary step before calculating global statistics. However, this by no means eliminates statistical uncertainty, while adding its own set of assumptions. When reporting any globally averaged statistic, it is important 25 to report corresponding distribution and coverage information, in the form of skewness values, probability density functions, and spatial distribution plots, to help quantify its usefulness and robustness. Abstract Introduction Conclusions References Tables Figures ◭ ◮ ◭ ◮