Correcting surface solar radiation of two data assimilation systems against FLUXNET observations in North America

Solar radiation at the Earth's surface is an important driver of meteorological and ecological processes. The objective of this study is to evaluate the accuracy of the reanalysis solar radiation produced by NARR (North American Regional Reanalysis) and MERRA (Modern-Era Retrospective Analysis for Research and Applications) against the FLUXNET measurements in North America. We found that both assimilation systems systematically overestimated the surface solar radiation flux on the monthly and annual scale, with an average bias error of +37.2 Wm^-2 for NARR and of +20.2 Wm^-2 for MERRA. The bias errors were larger under cloudy skies than under clear skies. A postreanalysis algorithm consisting of empirical relationships between model bias, a clearness index, and site elevation was proposed to correct the model errors. Results show that the algorithm can remove the systematic bias errors for both FLUXNET calibration sites (sites used to establish the algorithm) and independent validation sites. After correction, the average annual mean bias errors were reduced to +1.3 Wm^-2 for NARR and +2.7 Wm^-2 for MERRA. Applying the correction algorithm to the global domain of MERRA brought the global mean surface incoming shortwave radiation down by 17.3 W m^-2 to 175.5 W m^-2. Under the constraint of the energy balance, other radiation and energy balance terms at the Earth's surface, estimated from independent global data products, also support the need for a downward adjustment of the MERRA surface solar radiation. Key Points NARR and MERRA systematically overestimated the surface incoming solar radiationA simple post-reanalysis correction algorithm was proposedGlobal surface energy balance was revised according to the correction