Dynamical seasonal predictions with the COLA atmospheric model

Abstract Seasonal hindcasts were made using the COLA (Center for Ocean‐Land‐Atmosphere Studies) model for 16 winter seasons (mid‐December through March for 1981/82 to 1996/97). For each season, a nine‐member ensemble was generated using observed initial conditions in mid‐December and observed global sea surface temperatures (SSTs). It is found that in the presence of large tropical SST anomalies, the model is quite successful in simulating seasonal‐mean height anomalies over the Pacific‐North America (PacNA) region. A local spatial partem correlation field is computed for the ensemble seasonal mean of 500 hPa height from the General Circulation Model (GCM) and the seasonal means from the re‐analyses of the National Centers for Environmental Prediction, hereafter ‘observations’. This field exceeds 0.6 over the eastern tropical Pacific and western North America; the maximum values are greatly enhanced during El Niño Southern Oscillation (ENSO) years. Similar results are obtained for 200 hPa winds. The observed enhancement of intraseasonal low‐pass (10–90 day) variability of 500 hPa height during cold events is simulated, as is the shift in the storm tracks (2–10 day variability of 850 hPa meridional heat flux) during warm events. Empirical orthogonal function (EOF) analysis is applied in the PacNA region to the ensemble seasonal means of GCM 500 hPa height and the corresponding observed seasonal means; the global combined Arctic/North Atlantic Oscillation is removed from both GCM and observations. The leading EOF mode explains about 50% of the variance for both GCM and observations; the two patterns are nearly identical. Singular‐value decomposition (SVD) analysis between the tropical Pacific SST and 500 hPa height in the PacNA region shows that the nature of the coupling between SST and height is nearly the same in the GCM and observations. The great similarity between the height patterns in the first SVD mode and the corresponding leading mode EOF pattern indicates that the leading height variations are forced by the SST. The SST‐forced variance of height was also estimated by regression analysis of (ensemble) seasonal means for the GCM and observations for the 16 years onto an index of tropical Pacific SSTs derived from SVD analysis of a long (30‐year) record of observed heights and SSTs. The pattern of percentage variance explained in the GCM and in the observations are very similar to each other (and to the EOF described above). The higher absolute values in the GCM case reflect the effectiveness of the ensemble in filtering out variability unrelated to SST forcing. SVD analysis was applied to 100 GCM ‘samples’ coupled with the observations; a sample is defined as a single 16‐year record obtained by picking one ensemble member randomly for each of the 16 years. Probability distribution functions (p.d.f.s) of the pattern correlation for the leading SVD patterns, the percentage explained squared covariance, and the time series correlations all indicated sharp peaks at values of 0.87, 87%, and 0.82 respectively. The p.d.f. of the projection of individual 5‐day means onto the leading EOF described above is quite dramatically shifted during strong warm and cold tropical SST events; the warm (cold) event p.d.f. has almost all its weight in the negative (positive) EOF region. GCM and observations agree well. The intra‐ensemble spread was estimated by computing the PacNA anomaly correlation coefficient (ACC) for each of 36 possible intra‐ensemble pairs. For eight years in which histograms of the ACC indicate predominantly positive values, the ACC of the ensemble mean with the observed seasonal mean is also relatively high. Brier skill scores and reliability diagrams were computed for the ‘event’ of the 500 hPa height being one standard deviation above (or below) the normal, with all such events pooled over the entire northern hemisphere or North America only. All skill scores are positive and statistically significant at the 99% level. The North American scores are higher than the whole hemisphere scores; the ENSO year scores are higher than those for all years.