Peak-summer East Asian rainfall predictability and prediction part I: Southeast Asia

摘要

The interannual variation of East Asia summer monsoon (EASM) rainfall exhibits considerable differences between early summer [May-June (MJ)] and peak summer [July-August (JA)]. The present study focuses on peak summer. During JA, the mean ridge line of the western Pacific subtropical High (WPSH) divides EASM domain into two sub-domains: the tropical EA (5 degrees N-26.5 degrees N) and subtropical-extratropical EA (26.5 degrees N-50 degrees N). Since the major variability patterns in the two sub-domains and their origins are substantially different, the Part I of this study concentrates on the tropical EA or Southeast Asia (SEA). We apply the predictable mode analysis approach to explore the predictability and prediction of the SEA peak summer rainfall. Four principal modes of interannual rainfall variability during 1979-2013 are identified by EOF analysis: (1) the WPSH-dipole sea surface temperature (SST) feedback mode in the Northern Indo-western Pacific warm pool associated with the decay of eastern Pacific El Nio/Southern Oscillation (ENSO), (2) the central Pacific-ENSO mode, (3) the Maritime continent SST-Australian High coupled mode, which is sustained by a positive feedback between anomalous Australian high and sea surface temperature anomalies (SSTA) over Indian Ocean, and (4) the ENSO developing mode. Based on understanding of the sources of the predictability for each mode, a set of physics-based empirical (P-E) models is established for prediction of the first four leading principal components (PCs). All predictors are selected from either persistent atmospheric lower boundary anomalies from March to June or the tendency from spring to early summer. We show that these four modes can be predicted reasonably well by the P-E models, thus they are identified as the predictable modes. Using the predicted PCs and the corresponding observed spatial patterns, we have made a 35-year cross-validated hindcast, setting up a bench mark for dynamic models' predictions. The P-E hindcast prediction skill represented by domain-averaged temporal correlation coefficient is 0.44, which is twice higher than the skill of the current dynamical hindcast, suggesting that the dynamical models have large rooms to improve. The maximum potential attainable prediction skills for the peak summer SEA rainfall is also estimated and discussed by using the PMA. High predictability regions are found over several climatological rainfall centers like Indo-China peninsula, southern coast of China, southeastern SCS, and Philippine Sea.