Interdecadal changes in interannual variability of the global monsoon precipitation and interrelationships among its subcomponents

Abstract

The interdecadal and the interannual variability of the global monsoon (GM) precipitation over the area which is chosen by the definition of Wang and Ding (Geophys Res Lett 33: L06711, 2006) are investigated. The recent increase of the GM precipitation shown in previous studies is in fact dominant during local summer. It is evident that the GM monsoon precipitation has been increasing associated with the positive phase of the interdecadal Pacific oscillation in recent decades. Against the increasing trend of the GM summer precipitation in the Northern Hemisphere, its interannual variability has been weakened. The significant change-point for the weakening is detected around 1993. The recent weakening of the interannual variability is related to the interdecadal changes in interrelationship among the GM subcomponents around 1993. During P1 (1979–1993) there is no significant interrelationship among GM subcomponents. On the other hand, there are significant interrelationships among the Asian, North American, and North African summer monsoon precipitations during P2 (1994–2009). It is noted that the action center of the interdecadal changes is the Asian summer (AS) monsoon system. It is found that during P2 the Western North Pacific summer monsoon (WNPSM)-related variability is dominant but during P1 the ENSO-related variability is dominant over the AS monsoon region. The WNPSM-related variability is rather related to central-Pacific (CP) type ENSO rather than the eastern-Pacific (EP) type ENSO. Model experiments confirm that the CP type ENSO forcing is related to the dominant WNPSM-related variability and can be responsible for the significant interrelationship among GM subcomponents.

Appendix: Change-point detection methods

In this study, two methods were used for change-point detection. The Lepage test (Lepage 1971; Yonetani and Gregory 1994) is a non-parametric test that investigates significant differences between two samples, even if the distributions of the parent populations are unknown. The Lepage statistic, HK (symbol used by Yonetani and Gregory 1994), is combination of standardized statistics of Wilcoxon and Ansari-Bradley (Lepage 1971). If HK is greater than 5.99 or 9.21, the mean change between the two samples is significant at the 95 or 99 % confidence level, respectively.

The second methods is the Pettitt test (1979, 1980a). Pettitt derived the test statistic on the basis of changes in the median of the observation sequence by examining the rank of the observations. The Pettitt test uses a remarkably stable distribution and provides a stout test of the change-point resistant to outliers (Pettitt 1980b). The Pettitt test procedure is as follows:

First, each of the observations X ₁, X ₂, …, X _N is ranked from 1 to N. Let r _i be the rank associated with the observations X _i . Then, at each place j in the series, we calculate

$$ W_{j} = \sum\limits_{i = 1}^{j} {r_{i} } ,\quad j = 1,2, \ldots ,N - 1, $$

(1)

which is the sum of the ranks of the variables at or before point j. Next, for each point in the sequence, we calculate 2W _j  − j(N + 1) to set

$$ {\text{K}}_{\text{m,n}} = \hbox{max} \left| {2{\text{W}}_{\text{j}} - {\text{j}}({\text{N}} + 1)} \right|,\quad j = 1, 2, \ldots ,N - 1. $$

(2)

The value of j where the maximum in K _m,n occurs is the estimated change-point in the sequence and is denoted by m. The number of observations after the change-point is n = N − m. The sampling distribution of W _m is normally distributed with mean m(N + 1)/2 and variance mn(N + 1)/12 when N is large. We then perform a statistical test to determine whether the estimated change-point m is significant by using the sampling distribution of W _m .