Lorenzo Zampieri, National Center for Atmospheric Research

Atmospheric Reanalyses are widely used to estimate the past atmospheric near-surface state over the sea ice, providing crucial boundary conditions for uncoupled sea ice and ocean simulations. These products are widely used because physically consistent, available over the last 40 to 70 years, and with uniform spatial coverage. Nevertheless, previous research revealed systematic near-surface temperature biases over sea ice for most atmospheric reanalyzes, a fact that has been linked to a poor representation of the snow over the sea ice in the forecast models used to produce the reanalysis. Developed and continuously updated by the ECMWF, ERA5 is arguably one of the most mature and detailed reanalyses currently available, reaching a spatial resolution of 30km and a time resolution of 1h. As with other products, also ERA5 shows positive temperature biases over the Arctic sea ice when compared to in-situ and satellite observations (up to +10K in winter under clear-sky conditions), compromising the employment of this product in support of sea ice research. While ECMWF and other numerical weather prediction centers will likely correct this issue in future products, this study explores the possibility of improving the existing generation of reanalyzes, starting with the ERA5 2m and skin temperatures, by training a machine-learning algorithm that learns from in-situ and remote sensing observations. The impact of the correction on uncoupled sea ice and ocean simulations will be quantified and compared to the effect of tuning key thermodynamical parameters of the sea ice model: the snow conductivity and surface albedo.

свернуть