Aggregate-area radiative flux bias corrections over sea ice

The spatial and temporal variability of surface, cloud, and radiative properties of sea ice are examined using new satellite-derived products during the Surface Heat Budget of the Arctic (SHEBA) field experiment. Downwelling shortwave and longwave fluxes exhibit temporal correlation over about 180 days, but cloud optical depth and cloud fraction show almost no correlation over time. The spatial variance of surface properties is shown to increase much less rapidly with distance than that of cloud properties. Most sea ice and climate models treat surface and atmospheric properties as being horizontally homogeneous at scales of 100-250 km, and compute surface radiative fluxes with average grid cell properties. Given the spatial variability of surface and cloud properties and, in some cases, their nonlinear relationship with radiative fluxes, using mean surface and cloud properties in a climate or ice model grid cell to compute radiative fluxes could result in substantial errors.

This study shows that large biases can occur if subgrid cell variability is ignored, where bias is defined as the difference between the average of fluxes computed at high resolution within a model cell and the flux computed with the average surface and cloud properties within the cell. The dependence of the downwelling shortwave and longwave radiation flux bias on cloud fraction, cloud optical depth and solar zenith angle (shortwave only) is described, and a method for correcting the bias that implicitly accounts for small-scale variability is presented. Results indicate that for low cloud amounts the shortwave flux is overestimated and the longwave flux is underestimated; high cloud amounts have the opposite effect. Cloud optical depth is also important in estimating the radiative fluxes in that the larger the cloud optical depth, the larger the flux biases. The biases are nearly scale-invariant, especially for the downwelling longwave flux. A simple regression approach to correcting the fluxes for errors that result from horizontal variability was found to reduce the average bias by up to 80%. The correction can be easily implemented in numerical models.