Calibrated Seasonal Ensemble Forecasts of Arctic Sea Ice Advance and Retreat Dates
Abstract
Arctic sea ice is well known to have undergone considerable reductions in recent decades. This has led to significant effort being directed toward developing and improving sea ice prediction capabilities on seasonal timescales. Through initial engagements with potential stakeholders, it has been established that forecasts of the date of local sea ice retreat (ice-free date; IFD) and advance (freeze-up date; FUD) may be particularly useful for strategic decision making. While these forecasts have been, and are currently being produced with the Canadian Seasonal to Interannual Prediction System (CanSIPS) in an experimental mode, at present only the gridded ensemble-mean date (and date anomaly) is considered.
Here, we present on a newly-developed approach for producing probabilistic IFD/FUD forecasts in the Arctic based on ensemble model output statistics (EMOS). This approach, termed heteroscedastic censored gaussian regression (NCGR), explicitly handles the pre-occurrence and non-occurrence of IFD/FUD events using a doubly-censored normal predictive distribution, and also accounts for non-stationarity in IFDs/FUDs associated with the lengthening of the open water season over the satellite record. We show that NCGR leads to overall more idealized forecast behavior than a quantile-mapping based bias correction approach (lower bias, greater reliability, higher skill), and that probabilistic skill is qualitatively consistent with the corresponding deterministic-style forecasts. This work emphasizes the care required to avoid parameter overfitting when post-processing seasonal forecasts with limited training samples, including in the processes of predictor selection and regression formulation. Averaged over the Arctic, sea ice retreat (advance) dates can be forecast skillfully up to ~2.5 months (~3.5 months) prior to the climatological day of retreat (advance) after post-processing. In some regions however, skillful forecasts can be obtained from much earlier.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.A21O2768D
- Keywords:
-
- 3309 Climatology;
- ATMOSPHERIC PROCESSES;
- 0799 General or miscellaneous;
- CRYOSPHERE;
- 1899 General or miscellaneous;
- HYDROLOGY;
- 3245 Probabilistic forecasting;
- MATHEMATICAL GEOPHYSICS