Dynamical seasonal predictions with the COLA atmospheric model

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