Informal seminar: A machine learning correction model for the temperature bias over Arctic sea ice in atmospheric reanalyses
Atmospheric reanalyses are widely used as an estimate for the past atmospheric near-surface state over the sea ice, providing crucial boundary conditions for sea ice and ocean numerical simulations. Previous research revealed the existence of large near-surface (mostly warm) temperature biases over the Arctic sea ice in the current generation of atmospheric reanalyses that are linked to a poor representation of the snow over the sea ice in the forecast models used to produce the reanalyses. These errors can compromise the employment of these products in support of sea ice research. Here, we train a fully connected neural network that learns from remote sensing infrared observations to correct the existing generation of uncoupled atmospheric reanalyses (ERA5, JRA-55) based on a set of sea ice and atmospheric predictors, which are themselves reanalysis products. The ML correction model proved to be skillful in reducing the model bias in the Arctic regions experiencing clear sky conditions, and therefore a strong radiative cooling of the ice (or snow) surface. Furthermore, we investigate the seasonality and interannual variability of the corrected temperature fields, and we observe that the correction has only a minor impact on the warming trends in the Arctic as described by the reanalyses. The advantages of the proposed correction scheme are its consistency with the physical mechanism responsible for the bias, and a self-emerging seasonality and multi-decadal trend compatible with the mutating sea ice state in the Arctic. These are compared to previous calibration attempts and discussed particularly in the context of forced sea ice and ocean simulations, which rely on reanalysis surface fields as boundary conditions. Finally, we will also describe the limitations of our correction approach and discuss our results in relation to expected developments of fully coupled reanalysis systems.