Circulation Regimes: Chaotic Variability versus SST-Forced Predictability

Abstract The circulation regimes in the Pacific–North American region are studied using the NCEP–NCAR reanalyses for the 18-winter period (1981/82–1998/99; NCEP18) and for the 54-winter period (1948/49–2001/02; NCEP54). The sampling properties of the regimes are estimated using very large ensembles (of size 55) of winter simulations made for the NCEP18 period with the atmospheric general circulation model of the Center for Ocean–Land–Atmosphere Studies, forced by observed SST and sea ice. The regimes are identified using a modified version of the k-means method. From the NCEP54 dataset a set of four clusters was found i.e., the Alaskan ridge (AR), Arctic low (AL), Pacific trough (PT), and the Arctic high (AH), which are significant (vis-à-vis a multinormal background), and more reproducible (within randomly chosen half-length samples) than would be expected from a multinormal process. The frequency of occurrence of the PT (AH) has increased (decreased) significantly during the past two decades. The PT cluster obtained from NCEP18 dataset more closely resembles the El Niño–forced seasonal mean pattern of recent decades than it does the traditional PNA. The GCM simulates the AR, AL, and PT clusters (but not the AH). The simulated AR and PT patterns have errors (cf. the NCEP18 results), which are outside the range of internal variability. The simulated frequency of occurrence agrees with the NCEP18 results within sampling variability. The differences in cluster properties of the PT and AR regimes between the NCEP18 and NCEP54 datasets are due to changes in SST forcing, not sampling error. Year-to-year changes in the frequency of occurrence of the PT, AL, and AR clusters in the simulations and the NCEP18 dataset are generally consistent with each other.