An EOF methodology is applied to a hemispheric two-layer, baroclinic, primitive equation model to better understand the advantages and drawbacks of EOF modeling in numerical prediction. A spectral transform version of the model (truncated at T31) is integrated to generate a 100 year "nature" run. EOFs, from that run, are analyzed and used as basis functions in a simplified version of the model. By employing the EOFs as basis functions in the dynamical model, we obtain a system of quadratic nonlinear equations involving the EOF coefficients. / A statistical analysis is based on the spectral domain EOF expansion of the streamfunction, velocity potential and thickness of each layer. A comparison with spatially and temporally uncorrelated noise suggests that the first 65 PCs (6.4% of the total number of degrees of freedom) are significant and provide for a relatively good resolution of the individual modes; These account for about 96% of variance. / The predictability of the EOF model is examined when the number of EOF modes is truncated at 50, 100 and 150. A set of 1800 experiments of EOF model forecasts were run with different initial conditions for each EOF truncation. The root mean square error and anomaly correlation are used as measures of predictability. It was found that, at short-range, the forecast skill improves as the number of EOF modes retained in the model increases. For example, an EOF model truncated at 150 improves the average anomaly correlation of 3 day forecasts from 0.63 to 0.73 when compared with the model truncated at 50. However, there is a cross-over point (at about 6 days) after which the low resolution EOF model out-performs the high resolution EOF model in predicting the dominant modes. A similar result holds when the high resolution model is the nature run and initial errors are sufficiently large. This points to a potentially important application of truncated EOF models to extended-range forecasting. It remains to be seem, however, whether in applications to the atmosphere, the advantage of neglecting unresolved/uncertain scales in predicting the dominant modes occurs at a point when forecast skill is still useful. / Source: Dissertation Abstracts International, Volume: 55-04, Section: B, page: 1349. / Major Professor: Albert Barcilon. / Thesis (Ph.D.)--The Florida State University, 1994.
| Identifer | oai:union.ndltd.org:fsu.edu/oai:fsu.digital.flvc.org:fsu_77158 |
| Contributors | Chang, Yehui., Florida State University |
| Source Sets | Florida State University |
| Language | English |
| Detected Language | English |
| Type | Text |
| Format | 155 p. |
| Rights | On campus use only. |
| Relation | Dissertation Abstracts International |
Page generated in 0.0021 seconds