Exploring Atmospheric Variability Due to Arctic Sea Ice Loss with Machine Learning

Abstract:

Arctic sea ice is declining rapidly as greenhouse gas levels and global temperatures continue to rise due to human activity. Although a robust response of the mean atmospheric circulation to sea ice loss is well established, the impact of future Arctic sea ice loss on atmospheric variability remains an open question. In this study, we analyze results from two fully coupled Whole Atmosphere Community Climate Model (WACCM4) simulations, one with 1980-1999 seasonally varying sea ice conditions and the other with sea ice nudged to projected RCP 8.5 values over the period of 2080-2099. This model setup allows us to observe changes in atmospheric conditions directly caused by sea ice loss. To characterize atmospheric variability, we use an artificial neural network method called self organizing maps (SOMs) applied to daily DJF data over North America. The SOM method identifies the dominant patterns of atmospheric variability and how often they occur in the two simulations, allowing us to quantify the impact of sea ice loss on atmospheric circulation. Furthermore, we composite days associated with each SOM node for additional variables in order to provide an understanding of the physical processes responsible for changes in variability between these simulations. This novel application of self-organizing maps allows us to gain new insight into how weather patterns over North America may change as a result of future Arctic sea ice loss.