- SLP Index
Because the signature of the NAO is strongly regional, a simple NAO index was defined as the difference between the normalized mean winter SLP anomalies at locations representative of the relative strengths of the AH and IL. The first NAO index was defined by Walker and Bliss (1932) and simplified by Rogers (1984), who constructed an NAO index starting in 1894, using SLP anomalies from Ponta Delgados, Azores and Akuyreyri, Iceland. Hurrell (1995) selected Lisbon, Portugal and Stykkisholmur, Iceland in order to extend the record another 30 years.
- SST Index
The NAO SLP Index was spatially correlated with the analysis of global 5šx 5š standardized SST anomalies from 1856-1991 (Kaplan et al., 1996). Spatial correlation was performed in order to determine areas of highest sensitivity to the NAO Index. Five "centers of action" (CSST) were selected from the North Atlantic sector, each corresponding to an area of 5šx 5š. Once the 5 centers (CSST1 - CSST5) were selected and standardized, the time series (I(t)) was then constructed.
- Tree Ring Reconstruction
Investigation of the power spectrum of the NAO index for the 130 winters (1864-1995) reveals dominant periods of NAO-related North Atlantic sector variability, centered on 6-10 years and 2-3 years (Hurrell, 1996a) . Also significant is the red portion of the NAO power spectrum, which suggests NAO-related interdecadal-centennial scale variability may play a principle role in Holocene climate variability. The tree-ring based NAO index, extending from 1701-1980, confirms this low frequency power and suggests it is on the order of 70 years (Cook et al., in press) .
- Proxy Tree Ring Reconstruction
Multi-proxy `consensus' reconstructions of North Atlantic Oscillation (NAO) indices are presented spanning the last three centuries. These consensus time series draw on four nearly independent reconstructions derived from different combinations of North Atlantic sector tree rings, ice cores and long instrumental records. One consensus reconstruction based on all four series (R4) extends from 1750-1979 while another based on only three series (R3) extends from 1701-1979. Both consensus reconstructions outperform the individual series based on correlation analysis and verification statistics. R4, containing 2 records made up of a blend of proxy and long instrumental records, outperforms R3, which contains only 1 such record. These results suggest each individual series reflects different aspects of the NAO, such that when combined they yield more robust reconstructions. As to be expected, these consensus reconstructions, based on both proxy and long instrumental data, yield more optimal results than individual or consensusreconstructions based on proxy data alone.The reconstructions presented here provide an improved means of addressing features evident in the instrumental record with respect to those of prior centuries. In conclusion, this analysis further elucidates the need for proxy-based extensions of the instrumental record to better understand variability in the climate system.