A Regional Model for Seasonal Sea Ice Prediction in the Pacific Sector of the Arctic

Abstract:

The recent results from a linear Markov model for seasonal prediction of pan-Arctic sea ice concentration (SIC) show that sea ice in the Pacific sector has the lowest predictability compared to other regions. One reason could be that the climate variability in the Atlantic sector is so dominant that other signals in the Arctic climate system do not appear in the leading modes used for model construction. This study develops a regional Markov model to improve seasonal forecasting of SIC in the Pacific sector. The model climate system consists of various combinations of the monthly mean series of SIC, sea surface temperature (SST), surface air temperature (SAT), pressure/geopotential height fields and winds at pressure levels. Multivariate empirical orthogonal functions (MEOF) and rotated MEOF are applied to each set of data to reduce the model dimensions. After a series of experiments, the final model configuration selects 23 rotated MEOF modes from a data matrix of three variables (SIC, SST and SAT). This regional model shows considerable improvement in the prediction skill in the Pacific sector in all seasons. The anomaly correlation skill increases by 0.2 at 1- to 4-month leads in the Bering Sea, and by 0.1 at 1- to 10-month leads in the Sea of Okhotsk. In general, the model performs better in summer and fall than in winter and spring. On average, the correlation skill can reach 0.6 at a 2-month (4-month) lead in the Bering Sea (the Sea of Okhotsk).