Step or Trend? Detecting and Analyzing Structural Breaks in the Evolution of Arctic Open Water

The area and timing of seasonally navigable open water is critical to Arctic industrialization activities: freight, tourism, mining, and commercial fishing critically rely on shipping in the Arctic, which in turn requires reduced or clear ice area over Arctic ocean sectors for transit navigability. Due to the present low predictability of the navigable season, economic benefits have not accrued at the expected rate. As shipping across the Northern Sea Route and other economic activity increases in the Arctic Ocean, understanding the evolution of sea ice and open water become increasingly important. However, while many researchers take basic sea ice data and apply a linear detrending model before performing other analyses, it is not immediately clear that a linear detrend is the best description of the data. If not, applying a linear detrend to the data could add noise instead of removing it, making predictions more difficult.

In these analyses, we mix breakpoint analyses along with econometric time series techniques generally used in economics or finance to see if sea ice data is better described by a series of structural breaks than by a simple linear trend model. We apply these techniques to the September open water fraction in the Arctic using the satellite era record of ice concentration (1979-2014). We find evidence that three breakpoints (shifts in the mean) occurred in the Pacific sector, with higher amounts of open water starting in 1989, 2002, and 2007. Breakpoints in the Atlantic sector record of open water are evident in 1971 in longer records, and around 2000 and 2011. Multiple breakpoints are also evident in the Canadian and Russian halves. Notably, models that use detected breakpoints of the Pacific and Atlantic sectors, as well as models with breakpoints in the Canadian and Russian halves and the Arctic as a whole, outperform linear trend models in fitting the data. From a physical system standpoint, the results support the thesis that Arctic sea ice may have critical points beyond which a return to the previous state is less likely. From an analysis standpoint, the findings imply that de-meaning the data using the breakpoint means is less likely to cause spurious signals than employing a linear detrend. Applying econometric time series to climate data can help improve predictions, and possibly help Arctic shipping.