Uncertainty assessment of surface net radiation derived from Landsat images

The net radiation flux available at the Earth's surface drives evapotranspiration, photosynthesis and other physical and biological processes. The only cost-effective way to capture its spatial and temporal variability at regional and global scales is remote sensing. However, the accuracy of net radiation derived from remote sensing data has been evaluated up to now over a limited number of in situ measurements and ecosystems. This study aims at evaluating estimates and uncertainties on net radiation derived from Landsat-7 images depending on reliability of the input surface variables albedo, emissivity and surface temperature. The later includes the reliability of remote sensing information (spectral reflectances and top of canopy brightness temperature) and shortwave and longwave incoming radiations. Primary information describing the surface is derived from remote sensing observations. Surface albedo is estimated from spectral reflectances using a narrow-to-broadband conversion method. Land surface temperature is retrieved from top of canopy brightness temperature by accounting for land surface emissivity and reflection of atmospheric radiation; and emissivity is estimated using a relationship with a vegetation index and a spectral database of soil and plant canopy properties in the study area. The net radiation uncertainty is assessed using comparison with ground measurements over the Crau-Camargue and lower Rhone valley regions in France. We found Root Mean Square Errors between retrievals and field measurements of 0.25-033 (14-19%) for albedo, similar to 1.7 K for surface temperature and similar to 20 W.m(-2) (5%) for net radiation. Results show a substantial underestimation of Landsat-7 albedo (up to 0.024), particularly for estimates retrieved using the middle infrared, which could be due to different sources: the calibration of field sensors, the correction of radiometric signals from Landsat-7 or the differences in spectral bands with the sensors for which the models where originally derived, or the atmospheric corrections. We report a global uncertainty in net radiation of 40-100 W.m(-2) equally distributed over the shortwave and longwave radiation, which varies spatially and temporally depending on the land use and the time of year. In situ measurements of incoming shortwave and longwave radiation contribute the most to uncertainty in net radiation (10-40 W.m(-2) and 20-30 W.m(-2), respectively), followed by uncertainties in albedo (<25 W.m(-2)) and surface temperature (similar to 8 W.m(-2)). For the latter, the main factors were the uncertainties in top of canopy reflectances (