David A Rutan — Science Systems and Applications. Inc./NASA - LRC
Seiji Kato — NASA - Langley Research Center
Fred G Rose — Science Systems and Applications. Inc./NASA - LRC
David Robert Doelling — Science Systems and Applications, Inc.
Category
Radiation
Description
Surface sites used in study. White diamonds are land sites (including islands) blue diamonds indicate ocean buoy locations.
Using surface observations to validate surface downward irradiance from radiative transfer models, particularly those embedded in re-analyses and Global Climate Models (GCMs), is commonplace thanks to projects such as the DOE ARM and WCRP BSRN. Establishing the quality of the statistics retrieved is more problematic. GCMs and re-analyses typically operate on scales significantly larger than that observed by upward facing instruments and the number of surface sites is limited, in particular to the Northern Hemisphere.
We have collected surface observations at 85 sites across 8 years, thirty-seven sites located on land (including islands) and 48 buoy sites. All have shortwave (SW) observations while only 56 have longwave (LW) observations. Few include the entire temporal period in their archive. In this poster we first give statistics for the sites compared to calculated fluxes from 6 well-known data products. Secondly we present a sampling study of the sites used, focusing on the comparison of monthly means between observations and the NASA CERES Synoptic (SYN-1Degree) data product.
To see the impact of different sample sizes on the mean statistics, a simple ‘bootstrap’ analysis is done. The original sites are re-sampled randomly, removing more sites with each sample resulting in a set of distributions describing the effect of fewer and fewer sites on the comparison statistics between observed and modeled fluxes. Though individual sites are removed randomly, the original bias distribution is not re-sampled randomly as most sites do not have a complete 8 year record and bias at many sites is often skewed positive or negative due to other factors such as spatial representativeness of the. Thus the standard error of the mean bias is derived empirically from each new sampling distribution.
As expected, as more sites are removed from the comparisons, the distributions of mean and standard deviation become broader and the probability of retrieving the original mean and sigma, in any given sample, decreases. For both LW and SW, the standard error of the mean increases by a factor of 3. For LW the standard error goes up from ~0.35Wm-2 (52 sites) to 1.0Wm-2 (32 sites). For SW, the standard error increases from ~0.15Wm-2 (81 sites) to 0.45Wm-2 (61 sites). The conclusion though obvious is important. More surface sites, consistently measuring over long periods of time are crucial to establish definitive conclusions regarding the veracity of our modeling efforts.
