Evaluation and Improvement of a SVD-Based Empirical Atmospheric Model

An empirical atmospheric model (EAM) based on the singularvalue decomposition (SVD) method is evaluated using the composite ElNino/Southern Oscillation (ENSO) patterns of sea surfacetemperature (SST) and wind anomalies as the target scenario. Twoversions of the SVD-based EAM were presented for comparisons. Thefirst version estimates the wind anomalies in response to SSTvariations based on modes that were calculated from a pair of globalwind and SST fields (i.e., conventional EAM or CEAM). The secondversion utilizes the same model design but is based on modes thatwere calculated in a region-wise manner by separating the tropicaldomain from the remaining extratropical regions (i.e., region-wiseEAM or REAM).Our study shows that, while CEAM has shown successful modelperformance over some tropical areas, such as the equatorial easternPacific (EEP), the western North Pacific (WNP), and the tropicalIndian Ocean (TIO), its performance over the North Pacific (NP)seems poor. When REAM is used to estimate the wind anomalies insteadof CEAM, a marked improvement over NP readily emerges. Analyses ofcoupled modes indicate that such an improvement can be attributed toa much stronger coupled variability captured by the firstregion-wise SVD mode at higher latitudes compared with that capturedby the conventional one. The newly proposed way of constructing theEAM (i.e., REAM) can be very useful in the coupled studies becauseit gives the model a wider application beyond the commonly acceptedtropical domain.