Seasonal predictable source of the East Asian summer monsoon rainfall in addition to the ENSO–AO

Abstract

Improvement in the seasonal forecasting of East Asian summer monsoon rainfall (EASMR) remains a great challenge, as it is influenced by varied and complex impacts from (1) external forcings and slowly varying internal variabilities, which are potentially predictable, and (2) internal dynamics on intraseasonal time scales, which is basically unpredictable beyond a season. In this work, a (co-)variance decomposition method is applied to identify the leading potentially predictable (slow) patterns of the EASMR [the seasonal mean rainfall in the region (5°–50° N, 100°–140° E) in June–July–August] during 1979–2019 by separating the unpredictable noise (intraseasonal). We focus on the most critical predictable sources that are additional to the decaying (DC) El Niño–Southern Oscillation (ENSO), developing (DV) ENSO, and spring Arctic Oscillation (AO)—the three most important and well-recognized predictors for EASMR. We find that (1) the indices that represent the EASMR predictability related to the DC ENSO, spring AO and DV ENSO are the preceding November to March Niño1 + 2 sea surface temperature (SST), the April–May AO, and the May Niño4 SST, respectively; (2) the dominant additional predictable EASMR signals that are linearly independent of the DC ENSO, spring AO and DV ENSO have apparent relationships with the interannual variability of the SST in the western North Pacific, tropical and southern Atlantic, southern Indian, and Arctic oceans during boreal springtime, as well as the linear trend; and (3) by applying a principal component regression scheme to evaluate the EASMR predictability arising from DC/DV ENSO–AO and these additional predictors, the cross-validated fraction variance skill of the total seasonal mean EASMR is 11% (8%—land; 13%—ocean) for the former, and 15% (15%—land; 15%—ocean) for the latter, with a total of 26% that comprises more than 80% of the potential predictability of the EASMR. The considerable skill stemming from the predictors additional to DC/DV ENSO–AO indicates that they are worthy of attention in the seasonal forecasting of EASMR, especially for terrestrial areas.

Appendix

The steps in deriving the slow or predictable EOF patterns are:

• 1. Derive the eigenvalues and eigenmodes of the total covariance matrix of the seasonal mean rainfall using an EOF analysis.

• 2. Project the monthly rainfall anomalies onto the obtained leading Y − 1 (Y is the total number of years) eigenmodes (EOF modes) to form the Y − 1 truncated monthly PC time series.

• 3. Estimate the (Y − 1) × (Y − 1) covariance matrix of the intraseasonal component for each pair of the Y − 1 truncated monthly PC time series using Eq. (3.1).

• 4. The symmetric matrix from (3), above, is then pre-multiplied by the matrix of the leading Y − 1 EOF modes and post-multiplied by its transpose to form the full, or raw, intraseasonal covariance matrix.

• 5. If the rth diagonal element of the raw intraseasonal covariance matrix is greater than its corresponding total variance, then the rth row and the rth column of the raw intraseasonal covariance matrix are replaced by rth row and the rth column of the total covariance matrix. This adjusted covariance matrix is the estimated intraseasonal covariance matrix.

• 6. The slow covariance matrix is the difference between the total covariance matrix and the estimated intraseasonal covariance matrix.

• 7. An EOF analysis is then applied to the slow covariance matrix to obtain the corresponding patterns.