A Pilot Study on the Challenges in Ice Layer Annotations

Abstract

Radargrams or radiostratigraphy images of polar ice sheets reveal layers under the ice sheets. The study of ice sheets through these radargrams that capture layers of ice accumulation plays a crucial role in understanding climate, past snowfall trends, the impact of climate change, and sea level rise. Current methodologies for ice layer annotation in the radargrams, crucial for understanding ice dynamics and predicting sea level rise, face numerous challenges. These include the variability in annotation accuracy due to differing interpretation methodologies, the labor-intensive nature of manual annotations, and the need for improved automated techniques that can consistently match or surpass human-level precision. This study aims to address these challenges by evaluating current annotation methods, identifying limitations, and suggesting areas for improvement. By focusing on the current challenges and status of these techniques, we aim to contribute to the development of more robustand efficient tools for glaciological studies. Current ice layer-tracing algorithms encounter multifaceted challenges. First, manual annotation of ice layers by experts (i.e., glaciologists), although of high quality, requires considerable time and effort and may be incomplete. Second, the absence of a standardized automated approach leads to significant variability in annotation accuracy. Third, there is a lack of standardized evaluation metrics for assessing the reliability of annotations, which are crucial for validating the accuracy and scalability of automated annotation techniques. Figure 1 illustrates an example of expert annotation (e.g., incomplete) on a radargram image. In this work, we study the challenges of ice layer annotation methods and their evaluation metrics.