A MULTIVARIATE FREQUENCY-DOMAIN APPROACH TO LONG-LEAD CLIMATIC FORECASTING

Guided by the increasing awareness and detectability of spatiotemporal ly organized climatic variability at interannual and longer timescales , the authors motivate the paradigm of a climate system that exhibits excitations of quasi-oscillatory eigenmodes with characteristic timesc ales and large-scale spatial patterns of coherence. It is assumed that any such modes are superposed on a spatially and temporally autocorre lated stochastic noise background. Under such a paradigm, a previously described (Mann and Park) multivariate frequency-domain approach is p romoted as a particularly effective means of spatiotemporal signal ide ntification and reconstruction, and an associated forecasting methodol ogy is introduced. This combined signal detection/forecasting scheme e xhibits significantly greater skill than conventional forecasting appr oaches in the context of a synthetic example consistent with the adopt ed paradigm. The example application demonstrates statistically signif icant skill at 5-10-yr lead times. Applications to operational long-ra nge climatic forecasting are motivated and discussed.