Development of ANN-Based Algorithm to Estimate Wintertime Sea Ice Temperature Profile Over the Arctic Ocean

The thermal structure of the Arctic sea ice is a critical indicator in the atmosphere-sea ice-ocean energy budget and, thus, for understanding Arctic warming and associated climate change. Therefore, understanding this thermal structure and its monitoring should be vital. However, it is challenging to obtain a 3-D view of the thermal structure of the sea ice (such as the temperature profile) through satellite measurements because of the lack of understanding of the nonlinear relationship between sea ice emission and measured radiance at the top of the atmosphere. In this study, a model was developed to estimate the temperature profile within the Arctic sea ice during winter using satellite-borne passive microwave measurements. An artificial neural network (ANN) technique based on deep learning was introduced, and the nonlinear relationship between satellite-measured brightness temperatures and buoy-measured sea ice temperature profiles was learned. The ANN model was mapped and verified using the tenfold cross-validation technique. The developed ANN model was able to restore the sea ice temperatures at all specified levels with correlation coefficients > 0.95, absolute biases < 0.1 K, and root mean square errors < 1.6 K. The retrieved temperature results well represent expected thermal structures, in addition to the snow-sea ice interface temperature similar to that in the published literature. Besides the data for validating climate model simulations, the results also promise applications for improving the sea ice growth model performance by tightly constraining the vertical thermal structure in the sea ice growth model.