The GEWEX Radiative Flux Assessment (GEWEX-RFA) is an international scientific project commissioned by the GEWEX Radiation Panel to examine the current state of both the top of atmosphere (TOA) and surface Earth Radiation Budget (ERB) data sets and to give recommendations for future improvements of these climate data records. Earth radiation budget data producers from various institutions around the globe have responded to the GEWEX-RFA data request and submitted their scientific data sets into the GEWEX-RFA archive. Here we present results of an evaluation of the radiative fluxes at the Earth's surface provided by satellite data or models (GCMs and reanalyses) using ground measurements as a reference. These results have important implications regarding the current accuracy of the radiative budget parameters as an essential climate variable.

Ground measurements are drawn from the Baseline Surface Radiation Network (BSRN), Global Energy Balance Archive (GEBA), the University of Oregon Solar Radiation Monitoring Laboratory's archive, and the Alpine Surface Radiation Budget data set. Taken together, these data sets make it possible to evaluate both incoming shortwave and longwave fluxes at a variety of spatial scales.

Monthly mean satellite data at a spatial resolution of 2.5°x2.5° are evaluated over their entire record length (in most cases 1983-2004) and for the CERES (2000-2003) period. The satellite fluxes compare well to the BSRN observations with monthly averaged root-mean-square (RMS) differences ranging from 15-24 W/m2 for SW and 13-20 W/m2 for the LW. However, the agreement between the surface measurements and satellite products depends on the geographic location and surface type at the measurement sites. Ensemble SW (LW) anomalies using a continuous set of BSRN measurements give a 1-sigma difference standard deviation of about 5.5 W/m2 (2.2 W/m2) for most of the satellite algorithms. This is taken as a measure of the precision of these data sets. The time series also shows a SW ensemble bias that is decreasing negatively in time and the reasons for this are explored. Such a change is not seen in the longwave.

The GCM and reanalysis data sets are evaluated over their entire common period of record (1992-1999) and for the CERES (2000-2003) period. The RMS differences for these data sets relative to the BSRN values are somewhat higher than those for the satellite products, at 20-25 W/m2 for SW and 15-23 W/m2 for the LW. However, the NCEP reanalysis is an outlier for the SW. The geographic areas where the models perform poorly are notably different from those for the satellite products. Significant long-term trends computed for 12-site ensembles are not found in the model, satellite, or surface data.