Subseasonal variations of Eurasian wintertime surface air temperature: two distinct leading modes

To better understand subseasonal evolutions of wintertime surface air temperature (SAT) over Eurasia, this study investigates the first two leading modes of SAT month-to-month variations based on Season-reliant Empirical Orthogonal Function (S-EOF) analysis. The first S-EOF mode is characterized by a continuous in-phase SAT anomaly throughout the whole winter and labelled as the Consistent Mode (C-Mode). The second S-EOF mode is labeled as the Reversal Mode (R-Mode), featuring a warmer (or colder) early winter (November–December, ND) with a colder (or warmer) late winter (January–February, JF). These two distinct modes are linked to different circulation patterns and physical processes. The cold C-Mode is associated with an eastward shift of negative North Atlantic Oscillation (NAO) pattern, and highly correlated with Arctic Oscillation (AO) index. The C-Mode can largely be attributed to atmospheric internal variability, for simultaneously AO (NAO) index explains 57.8% (46.2%) of the total variance of the corresponding principal component. The R-Mode, on the other hand, is accompanied by a nearly out-of-phase change of large-scale circulation. The R-mode is not only closely related to the preceding AO, but also low boundary forcing anomalies which exhibit a North Pacific sea surface temperature (SST) dipole pattern (NPDP) and an Arctic sea ice-SST coupled pattern (ACP). Further analysis shows that an AO-like circulation anomaly in September–October induces the NPDP, which in turn generates feedback to the atmosphere and modulates Eurasian SAT anomaly in ND. The ACP significantly contributes to an out-of-phase change of SAT anomaly in JF due to a delayed yet pronounced atmospheric response to surface heat flux anomalies in November. Potential vorticity dynamics is also applied to illustrate how the NPDP and the ACP affect the R-Mode. A physically-based empirical model is established using the AO and the ACP indices, which shows a promising prediction skill. The results indicate that the AO and ACP provide crucial predictability sources for the R-Mode.