Deep Learning of Systematic Sea Ice Model Errors From Data Assimilation Increments

Data assimilation is often viewed as a framework for correcting short-term error growth in dynamical climate model forecasts. When viewed on the time scales of climate however, these short-term corrections, or analysis increments, can closely mirror the systematic bias patterns of the dynamical model. In this study, we use convolutional neural networks (CNNs) to learn a mapping from model state variables to analysis increments, in order to showcase the feasibility of a data-driven model parameterization which can predict state-dependent model errors. We undertake this problem using an ice-ocean data assimilation system within the Seamless system for Prediction and EArth system Research (SPEAR) model, developed at the Geophysical Fluid Dynamics Laboratory, which assimilates satellite observations of sea ice concentration every 5 days between 1982 and 2017. The CNN then takes inputs of data assimilation forecast states and tendencies, and makes predictions of the corresponding sea ice concentration increments. Specifically, the inputs are states and tendencies of sea ice concentration, sea-surface temperature, ice velocities, ice thickness, net shortwave radiation, ice-surface skin temperature, sea-surface salinity, as well as a land-sea mask. We find the CNN is able to make skillful predictions of the increments in both the Arctic and Antarctic and across all seasons, with skill that consistently exceeds that of a climatological increment prediction. This suggests that the CNN could be used to reduce sea ice biases in free-running SPEAR simulations, either as a sea ice parameterization or an online bias correction tool for numerical sea ice forecasts. To make predictions of the Earth's climate system we use expensive computer simulations, called climate models. These models are not perfect however, as we often need to approximate certain physical laws in order to save on compute time. On the other hand we have observational climate data, however these data have limited space and time coverage and also contain errors because of noise and assumptions about how our measurements relate to the quantity we are interested in. Therefore we often use a process called data assimilation to combine our climate model predictions together with observations, to produce our "best guess" of the climate system. The difference between our best-guess-model and our original climate model prediction then gives us clues as to how wrong our original climate model is. In this work we use some fancy statistics, called machine learning, where we show a computer algorithm lots of examples of sea ice, atmosphere and ocean climate model predictions, and see if it can learn its own inherent sea ice errors. We find that it can do this well, which means that we can hopefully incorporate the machine learning algorithm into the original climate model to improve its future climate predictions. We show that sea ice data assimilation increments closely reflect the systematic bias patterns of a global ice-ocean modelConvolutional neural networks can make skillful predictions of sea ice data assimilation increments, using only model state variablesThe skillful predictions suggest the network could be used as a parameterization to reduce sea ice biases in free-running model simulations