Presentation PDF (424.7 kB)
The method of principal components was used to reduce dimensionality in both the SLP and proxy data sets. Reconstructions retaining 3-4 SLP components, explaining approximately 40-45% of the observed variance, and 9-11 proxy components, explaining approximately 38-45% of the observed variance, provided the most reasonable patterns. Variance in the reconstructed patterns coincided with observed variance. The zone of greatest residual standard deviation, 5 mb, was in the Canada Basin, in the Arctic Ocean, with a small secondary maximum just west of the British Isles, where residual standard deviation approached 4 mb. Beyond these specific regions, overall predictive capacity was generally good for the training period, with RMSE values for individual years ranging between 0.9 and 2.5 mb and averaging around 1.3 mb over the 1977 element grid. These were considered reasonable considering the size and diversity of the region of study.