Icy Waters: Developing a Test-Suite to Benchmark Sea Ice Concentration Forecasting (Papers Track)
Kiernan McGuigan (University of Waterloo); Sirisha Rambhatla (University of Waterloo); K Andrea Scott (University of Waterloo)
Abstract
Artificial intelligence (AI) for Climate Change efforts have made significant progress in forecasting atmospheric weather patterns and events. Despite this, translating these gains in the context of phenomenon on earth surface, e.g. sea-ice concentration, has been limited because of differences in how these physical processes evolve. Sea ice concentration is one of the key indicators of climate change and is also critical for a number of different applications and indigenous peoples. Consequently, there is an acute need to develop a baseline of a diverse set of modern machine learning techniques within the Arctic. Our work aims to fill this gap, with the goal of both informing current research, as well as pointing out limitations with certain architectures. We achieve this by providing baselines for a number of different convolutional LSTMs, transformer based, and neural operator based machine learning methods.