| Summary: | Thesis (Master's)--University of Washington, 2016-06 In this work we improve the skill of climate field reconstructions (CFRs) through the use of an online paleoclimate data assimilation (PDA) method within the Last Millennium Reanalysis framework (LMR). A computationally cheap forecast model, known as a linear inverse model (LIM), is employed here to provide 1-year forecasts between analysis times of the reconstruction. CFRs of annual mean 2m air temperature are compared between the previous offline and new online method. We test LIMs calibrated on surface temperatures from the Berkeley Earth observational dataset, the 20th Century Reanalysis (20CR), and coupled general circulation model last-millennium simulations (Community Climate System Model v4, CCSM4; Max Planck Institute Earth System Model, MPI). In all cases skill metrics are assessed for both spatial fields and global averages. Generally, we find that the usage of online PDA can increase reconstruction agreement with the instrumental record for both spatial fields and the global mean temperature (GMT). Spatial field skill increases tend to occur over Northern Hemisphere land areas and in the high-latitude North Atlantic - Barents Sea corridor. These regions of increased skill are associated with better agreement in temperature anomaly amplitude or trend, and not associated with changes in anomaly timing. Overall, the CCSM4 LIM provides the best performance when considering both spatial fields and GMT. A comparison with a persistence forecast experiment suggests that the skill of LIM forecasts are associated with the usage of the LIM, rather than simply the persistence of existing temperature anomalies. Results from this study are directly applicable in providing more dynamically consistent CFRs over the past two millennia.
|
|---|