A combined multi-source data and deep learning approach for retrieving snow depth on Antarctic Sea ice during the melting season

Abstract

Snow on the Antarctic sea ice is a crucial component of the cryosphere. In response to the dynamic and highly heterogeneous Antarctic snow during the sea ice melting season, this study employed a combined multi-source data and deep learning method to accurately retrieve snow depth on Antarctic sea ice. Initially, we integrate multiple datasets, including satellite remote sensing, geospatial information, and meteorological data. Subsequently, a Convolutional Neural Network (CNN) is utilized to construct a snow depth retrieval model (PSDCNN-5_7 model). Compared to snow depth measurements from Alfred Wegener Institute (AWI) snow buoys, the PSDCNN-5_7 model outperforms existing algorithms, exhibiting a deviation of only −3.38 cm. The uncertainty of the snow depth caused by the model input is only 1.64 cm. In West Antarctica, snow depth is more affected by snowfall (SF), 2-m air temperature (T2m), and sea ice velocity (SIV). Conversely, in East Antarctica, snow depth is primarily influenced by SIV. The proposed approach accurately retrieves snow depth on Antarctic sea ice and facilitates the derivation of long-term variations and trends in snow depth, contributing to a better understanding of the relationship between sea ice, snow, and climate change.

Keywords

• Snow depth
• Antarctic sea ice
• deep learning
• multi-source data