Application of Self-Organizing Maps to characterize subglacial bedrock properties in East Antarctica based on gravity, magnetic and radar data

Abstract. Subglacial bedrock properties are a key to understand and predict the dynamics and future evolution of the Antarctic Ice Sheet. However, the ice sheet bed is largely inaccessible for direct sampling. Therefore, it is crucial to efficiently combine various attributes derived from satellite and airborne geophysical surveys to characterize subglacial properties. To reduce subjective choices in the joint analysis of data and related biases, we evaluate a Self-Organizing Map (SOM), an unsupervised machine learning technique. The concept of SOMs, an unsupervised machine learning approach, is briefly discussed, but we focus on data selection and their associated attributes for the case at hand. For this, we analysis the correlation between attributes in order to provide a validation of an appropriate choice. The SOM is trained on attributes derived from gravity, magnetics and ice-penetrating radar data for the Wilkes Land area in East Antarctica, a region where basal conditions may be of high importance to ice sheet flow and corresponding sea level rise, and where also suitable data sets for the application of the SOM exist. In contrast to the earlier studies, our approach uses original line data as far as possible, which have much higher resolution/sampling than in smooth gridded products, which were used for previous analyses. Previous analysis indicated the presence of both crystalline basement and sedimentary basins in the area, and our SOM shows a remarkable agreement, but suggests some points of difference, as for example, some highlands appear similar on previous interpretations, but have quite dissimilar physical settings, which is also expressed in our results. These variations can potentially be exploited further in describing subglacial properties and the coupling between bed and overlying ice-sheets.