Research Themes: statistical methods, Arctic, sea ice, model bias

A recent paper in the Journal of Climate proposes a new technique for forecasting the sea ice edge contour, defined as the boundary between regions with sea ice cover less than or greater than 15%. This study, with first author Hannah M. Director, develops a statistical model that identifies and corrects errors in forecasts in a physical sea ice model. This correction reduces the amount of area incorrectly forecast by 21%.

The need for short-term sea ice forecasts is driven by changes in the climate of the Arctic. Reductions in sea ice cover impact wildlife and coastal communities and have increased the amount of shipping in the region.

Current sea ice forecasts come from numerical prediction systems, which use equations to represent the physical behavior of the ocean, atmosphere, and sea ice. These models are able to accurately represent many of the processes affecting where sea ice will be located; however, they do make some forecast errors repeatedly. In this research, a statistical technique that anticipates the spatial and temporal patterns of these errors is introduced, allowing for the errors to be corrected before a forecast is issued.

While both the fields of statistics and climate science focus on modeling, the ways in which models are developed, interpreted, and used are quite different. Communicating about and combining these different frameworks for modeling was a challenge, but doing so enabled the development of more accurate forecasting methods than could be obtained using the tools from either discipline alone.

This work was in collaboration with Adrian E. Raftery, Professor of Statistics and Sociology, and Cecilia M. Bitz, Professor of Atmospheric Sciences and Director of the Program on Climate Change.