Machine Learning Solutions for Big Data Analysis in the Arctic and Antarctic

Significant resources have been spent in collecting and storing large and heterogeneous datasets collected during expensive Arctic and Antarctic fieldwork. While traditional analyses provide some insight, the complexity, scale, and multidisciplinary nature of the data necessitate advanced intelligent solutions. The scientific questions that we wish to address are: What is the ice thickness of the polar ice sheets? What is the 3D ice bed topography and what are the scattering characteristics of the ice bed for this topography? What are the depth and extent of the shallow ice layers used for mapping accumulation? What are the depth and extent of the deep ice layers used for understanding past and present ice flow properties and bulk properties, such as thermal state? Intelligent answers to these questions have two-fold benefits. First, domain experts can gain useful insights about the data without unnecessary labor and generate data products automatically. Second, intelligent solutions transform the data to useful insights and knowledge. This knowledge can substantially improve future missions, benefitting the NSF. In recent years, the research community has witnessed advances in artificial intelligence (AI). Recent advances in deep neural networks (DNNs) and massive datasets have facilitated progress in AI tasks such as image classification, object detection, scene recognition, semantic segmentation, and natural language processing. Despite this progress, these algorithms are limited to electro-optical data with large labeled datasets. There is a critical need to develop more advanced machine learning and deep learning algorithms for both visual and non-visual sensors collecting data for real-world scenarios during various polar ice missions. The paper will present an overview of the machine learning tools and results from our NSF project BIGDATA: IA: Collaborative Research: Intelligent Solutions for Navigating Big Data from the Arctic and Antarctic. The Deep learning algorithms and tools include fully convolutional networks for tracking ice sheets internal layers with, regression networks for snow accumulation, Active learning for semi-automated labeling, and generative adversarial networks for data simulation.