Calibrated probabilistic forecasts of user-relevant Arctic sea ice measures on subseasonal-to-seasonal timescales

Abstract:

A rapidly declining sea-ice cover, combined with a prolonged open water season, is driving a high demand for accurate sea ice forecasts in the Arctic on user-relevant temporal and spatial scales. While sea ice forecasting on seasonal timescales has been rigorously pursued in recent years, very little attention has been directed toward sea ice prediction on subseasonal-to-seasonal (S2S) timescales. Here, we make use of two operational ensemble prediction systems, ECCC’s Ensemble GIOPS and ECMWF’s SEAS5, to investigate the S2S predictability of key sea ice metrics during the apex of the Arctic shipping season. In particular, we formulate 2-12 week lead time forecasts from daily sea ice outputs obtained from forecasts initialized monthly from the beginning of June through early November. A multidecadal re-forecast record from both systems spanning 1980-2017 is leveraged to quantify the predictive skill of local sea ice area, ice break-up timing, and freeze-up timing. The influence of systematic errors in both models is assessed in terms of lead-time and seasonal dependent bias and reliability (i.e. ensemble dispersion). Previously developed statistical postrocessing methods designed specifically for sea ice are applied to a subset of re-forecasts spanning 1999-2017. The calibration models are trained using re-forecasts from previous years on a sliding basis and include data from neighboring days. Probabilistic skill scores are then used to assess forecast performance against climatological benchmark forecasts, both before and after applying statistical postprocessing. This allows us to quantify the efficacy of the postprocessing methods, expected operational forecast skill, as well as the influence of model biases on overall forecast uncertainty.