Interdecadal changes in potential predictability of the summer monsoon in East Asia and South Asia

Interdecadal changes in the potential predictability of the East Asian summer monsoon (EASM) and South Asian summer monsoon (SASM) were investigated using the signal‐to‐noise ratio of daily reanalysis data from 1948 to 2017. Results reveal that variations in the SASM predictability are out of phase with those in EASM predictability, and that SASM predictability is higher (lower) than EASM predictability before (after) the 1980s. The out‐of‐phase relationship of the predictability of the two monsoon systems is attributed mainly to temporal changes in their relationships with El Niño‐Southern Oscillation (ENSO). The relationship between the EASM and ENSO is related to the position of the ENSO heating center, which is located in the central Pacific before the 1980s but shifted to the eastern Pacific after the 1980s, accompanied by a strengthening (weakening) in the relationship between ENSO and the EASM (SASM). The shift in the position of the ENSO heating center was found to have contributions from the modulation of the Pacific decadal oscillation.


| INTRODUCTION
The East Asian summer monsoon (EASM) and the South Asian summer monsoon (SASM) are two major components of the Asian monsoon system, which itself is part of the global monsoon system, and they play an important role in modulating precipitation and wind fields, which have important effects on climate change in Asia (Lau and Weng, 2001;Ding and Chan, 2005;Chang et al., 2011). Thus, accurate predictions of the summer monsoon are vital to obtaining reliable forecasts of intense precipitation and associated flood events across East Asia (Webster et al., 1998;Wu et al., 2006Wu et al., , 2009. As predictability is an inherent nature of the monsoon system (Li and Chou, 1997), a comparison of the predictability of these two monsoon systems will benefit efforts to improve monsoon and related precipitation forecasts.
To explore the monsoon predictability at different timescales will likely lead to improved forecasts. Several studies have reported the multitimescale changes in predictability of the summer monsoon, from synoptic to interannual timescales. Besides, the relationship between the EASM or the SASM and El Niño-Southern Oscillation (ENSO) has also been explored, but with a focus on seasonal and interannual timescales. For example, Ai et al. (2017) investigated the predictability limit of EASM indices on a synoptic timescale using the nonlinear local Lyapunov exponent method proposed by Li (2007, 2009) and demonstrated that EASM indices defined by zonal wind shear tend to be more predictive than those defined by sea-level pressure and meridional wind shear. Consequently, the monsoon index used in this study is based on zonal wind. The predictability of the monthly and seasonal-scale circulation and rainfall of the EASM and the SASM have also been examined using numerical simulations. Wang et al. (2004) found that nearly all atmospheric general circulation models perform poorly in simulating circulation and precipitation anomalies over the monsoon area at the monthly scale. Lee et al. (2010) investigated the potential ability of two climate forecast models to predict EASM precipitation and circulation one or two seasons in advance. Variability in the seasonal predictability of the summer monsoon has also been investigated (e.g., Wang et al., 2005;Zhou and Zou, 2010;Yang et al., 2012;Seo et al., 2015;Wu and Yu, 2016), revealing that ENSO is the primary indicator of the seasonal predictability of the summer monsoon. The relationship between the EASM (SASM) and ENSO at interannual timescales has been studied extensively, demonstrating that the sea surface temperature (SST) heating center and phases of ENSO have varying effects on the monsoon system (Parthasarathy and Pant, 1985;Huang and Wu, 1989;Chen, 2002;Wu et al., 2003). Whether the interdecadal relationship between the EASM (SASM) and ENSO is an important indicator of monsoon predictability remains an open question. The effects of ENSO events on the relationship between the EASM (SASM) and ENSO at interdecadal timescales also remain poorly known. These uncertainties are the primary motivation of the present study.
Here, we compare interdecadal changes in the predictability of two monsoon systems and investigate the potential effects of the ENSO-Monsoon relationship on predictability at interdecadal timescale. The reminder of this article is organized as follows. The data, indices and methods used are described in Section 2. Section 3 compares the interdecadal changes in predictability of EASM and SASM. A discussion and conclusion are provided in Section 4.

| Reanalysis data
The SST dataset used in this study is the Extended Reconstructed SST version 3b (ERSSTv3b) dataset from the National Oceanic and Atmospheric Administration with a 2 × 2 horizontal grid resolution and covering the period 1948-2017 (Smith et al., 2008). Daily mean atmospheric fields were taken from the National Centers for Environmental Prediction (NCEP1) reanalysis data, with a horizontal resolution of 2.5 × 2.5 for the period 1948-2017 (Kalnay et al., 1996;Kistler et al., 2001). Monthly precipitation data were obtained from the Climate Prediction Center Merged Analysis of Precipitation dataset  and used to examine the mechanism of the ENSO-Monsoon relationship in this paper. The climatology is derived from the whole period of each dataset.

| Indices
The Niño 3.4 index is defined as the area-average of monthly sea surface temperature anomaly (SSTA) in the region (5 S-5 N, 150 -90 W) and is often used to explore the relationship between the summer monsoon and ENSO. The Niño 3 index and the El Niño Modoki index (EMI) are similarly used to investigate two types of ENSO events with heating centers located in the eastern and central Pacific, respectively. The Niño 3 index is the area-average of monthly SSTA in the region (5 S-5 N, 170 -120 W). The EMI (Ashok et al., 2007) is defined as where [] represents the area-averaged SSTA over the central The SASM index using the area-averaged 850-hPa zonal wind (U850) minus the area-averaged 200-hPa zonal wind (U200) over the monsoon region (0 -20 N, 40 -110 E) from June to September (Webster and Yang, 1992) is defined as 2.3 | Signal-to-noise ratio method The signal-to-noise ratio (SNR) method has been widely used to investigate atmospheric predictability (Trenberth, 1984(Trenberth, , 1985Goswami, 2004). This method estimates atmospheric predictability by quantifying the relative contributions of the predictable climate signal and the unpredictable climate noise: where Var(signal) represents the variance of interannual variability, and Var(noise) is the variance of the intra-annual (seasonal) variability.
One limitation of the SNR method is that only one value can be calculated from a time series to represent the potential predictability over a given period. We chose an 11-year sliding window to investigate interdecadal changes in the EASM and the SASM. In addition, the signal and noise of monsoon predictability can be separated using the SNR method to analyze their respective effects on the predictability.

| RESULTS
The SNR method was used to calculate potential predictability from 1948 to 2017, using an 11-year window. The interdecadal changes in EASM and SASM predictability are shown in Figure 1a. The EASM and SASM predictabilities are inversely correlated. The SASM predictability is higher than that of EASM before the 1980s, with EASM predictabilities below 8 and SASM predictabilities in the range 10-16. After this time, the SASM predictability decreases to around 8 and the EASM predictability rises to between 16 and 22 during 1977-2000. The SASM predictability then increases to above that of EASM during the 2000s.
The interdecadal monsoon signals (Figure 1b) follow the trends in predictability shown in Figure 1a. However, the EASM and SASM noise patterns differ from the signal and predictability patterns (Figure 1c), with the EASM noise (~0.4) being slightly larger than the SASM noise (0.1-0.2), which may contribute to the complexity of predicting EASM.
To examine whether the signal or the noise is the primary contributor to predictability, we calculated the correlation coefficient between the predictability in EASM and SASM, as well as their respective signals and noise. The correlation coefficient between predictability and signal for the EASM is slightly lower than that for the SASM (0.996 and 0.786, respectively; significant at the 99% confidence level). However, neither the EASM nor SASM noise is significantly correlated with predictability. Thus, the interdecadal change in monsoon predictability is correlated primarily with monsoon signal, whereas there is no significant correlation with noise.
Previous studies have revealed that ENSO is a primary contributor to the predictability of the summer monsoon. The physical mechanisms through which ENSO is related to the summer monsoon have been investigated (Nigam, 1994;Goswami, 1998;Wu et al., 2012;Li et al., 2018), showing that ENSO affects the SASM in developing summer and the EASM in the decayed year. The Niño 3.4 index was chosen to represent ENSO to investigate its correlation with the two summer monsoons (Figure 2). The Niño 3.4 and the SASM are significantly negatively correlated before the 1980s, whereas the positive relationship between the Niño 3.4 and the EASM becomes significant after the 1980s. Overall, EASM and SASM show opposite relationships with ENSO; that is, ENSO may contribute to SASM (EASM) before (after) the 1980s. Combined with the monsoon predictability (see Figure 1a), interdecadal changes in the relationship between ENSO and the summer monsoons are well correllated with predictability ( Figure 2). During the period in which ENSO is significantly correlated with the EASM, predictability is high, as is the case with the SASM. Thus, interdecadal changes in the relationships between ENSO and the summer monsoon systems lead to interdecadal changes in EASM and SASM predictability. Whether these interdecadal relationships are caused by interdecadal variations in ENSO type warrants further investigation. Thus, due to the turning points of predictability and correlation coefficient occuring in Figures 1a and 2, SST anomalies related to El Niño events are divided into two preiods, 1948-1976 and 1977-2000, and shown in Figures 3a  and b. The identification of El Niño events was based on the winter (December-February) Niño 3.4 index. Following the method of Yoon and Yeh (2010), events were identified for the composite analysis when the normalized index was greater than 0.5. El Niño events were identified in 1951, 1953, 1957, 1958, 1963, 1965, 1968, 1969, 1972, and 1976 during the period 1948-1976 and in 1997, 1979, 1982, 1986, 1987, 1990, 1991, 1994, and 1997  To examine whether the location of the ENSO heating center affects the summer monsoon systems, the Niño 3 index (EMI) is used to represent the ENSO heating center, which locates in the eastern (central) Pacific. The 850-hPa horizontal winds and precipitation anomalies were regressed onto the Niño 3 index (EMI) for June-August during ENSO developing and decaying years. As shown in Figures 4a and  b, different ENSO heating centers lead to different wind fields and precipitation in the tropics during the summer of ENSO developing years. Compared to ENSO events with its heating center in the eastern Pacific, the ENSO events with heating centers in the central Pacific will cause more precipitation in the northwestern Pacific and less precipitation in the maritime continent and the Indian subcontinent. Precipitation anomalies cause unequal changes in latent heat release, leading to changes in the wind field. With this kind of heating, cross-scale air flow in the southern Indian Ocean is enhanced and exhibits anomalous cyclonic changes between 60 and 120 E, further affects the onset and development of the SASM. Weak precipitation anomalies caused by ENSO events with heating centers in the eastern Pacific are unable to stimulate such changes in the wind field, leading to negligible impacts on the SASM. The Indian subcontinent is more arid during El Niño events than during normal years because of the downward branch of the Walker circulation. Thus, ENSO events with heating centers in the central Pacific, compared with those in the eastern Pacific, have more noticeable impacts on the SASM and lead to less precipitation over the Indian subcontinent. At the same time, warm SST anomalies coupled with heating centers in the central Pacific may enhance convection and trigger an anomalous cyclone in the lower troposphere closer to the Indian Ocean via the Gill-Matsuno mechanism, compared with the case of ENSO events with heating centers in the eastern Pacific (Matsuno, 1966;Gill, 1980). Wu et al. (2017aWu et al. ( , 2017b revealed that by maintaining the western North Pacific anomalous anticyclone (WNPAC), ENSO affects EASM in subsequent years. Therefore, the location of the WNPAC plays a key role in the EASM. In Figure 4c, the location of the WNPAC center caused by ENSO events with a heating center in the central Pacific is more easterly and northerly (140 E, 28 N), whereas the WNPAC center caused by ENSO events with a heating center in the eastern Pacific is located around (135 E, 20 N), near East Asia, and has a greater influence on the EASM (Figure 4d). Figure 5 shows the average correlation coefficients between the wind fields (area defined by the monsoon index) and the Niño 3 index (EMI). During ENSO developing years, ENSO events with central-Pacific heating centers, as indicated by the EMI, show a stronger relationship with the SASM than for ENSO events with heating centers in the eastern Pacific (average correlation coefficients of 0.15 and 0.03, respectively). A weaker relationship with EASM is found for ENSO events with heating centers in the central Pacific during ENSO decaying years (an average correlation coefficient of 0.31) than for ENSO events with heating centers in the eastern Pacific (an average correlation coefficient of 0.93). These results are consistent with those shown in Figure 4. Thus, ENSO events with heating centers in the central Pacific have a significant effect on the SASM, whereas those with heating centers in the eastern Pacific have similar effects on the EASM. The location of the ENSO heating center modulates the interdecadal changes in the interannual relationship between ENSO and these monsoon systems.

| CONCLUSION AND DISCUSSION
The SNR method was used to analyze interdecadal changes in the predictability of the EASM and the SASM from 1948 to 2017 using reanalysis data. Results suggest that variations in SASM predictability are out of phase with those in EASM predictability. The SASM predictability is higher than EASM predictability before the 1980s, and lower during 1977-2000. Here we reveal that monsoon predictability is affected by ENSO and that the location of the ENSO heating center varies on an interdecadal timescale, which causes an interdecadal change in the ENSO-monsoon relationship. Before 1976, SASM predictability was higher than EASM predictability because the heating centers of ENSO events were in the central Pacific. Precipitation fields generated by  . All shown coefficients are statistically significant at a 90% confidence level this type of ENSO excite wind anomalies that facilitate SASM onset and maintenance. From 1977 to 2000, the ENSO heating center was located primarily in the eastern Pacific and showed a stronger relationship with EASM in subsequent years due to the location of the WNPAC. The Pacific decadal oscillation (PDO) modulates the interannual relationships between monsoon systems, and ENSO has been revealed in several studies (e.g., Zhu and Yang, 2003;Yoon and Yeh, 2010). Krishnamurthy and Krishnamurthy (2014) reported that the phase of the PDO may lead to better long-term predictions of seasonal monsoon rainfall and the impacts of ENSO on monsoons. Several studies have found that the PDO influences the type of ENSO. For example, Lin et al. (2018) pointed out that positive SST anomalies during El Niño events in the equatorial eastern Pacific are much stronger during positive PDO phases than during negative phases. Feng et al. (2014) found that El Niño decays slowly (rapidly) during positive (negative) PDO phases. As shown in Figure 3a and b, warm SST anomalies were located in the central Pacific for the time period from 1948 to 1976, which is correspond with the PDO negative phase, as warm SST anomalies in eastern Pacific when PDO in positive phase for 1977-2000. In addition, the PDO transfers to negative phase in recent years, and warm SST anomalies located in the central Pacific even with increased ENSO amplitudes during 2001-2016 (see Figure 3c). Thus, the PDO may be the primary contributor to interdecadal changes in the location of the ENSO heating center. However, the underlying physical processes should be explored in future work, which is also helpful for us to deepen the independence and connection between EASM and SASM.
To aid the development of seamless multiscale weather forecasts and our understanding of monsoon predictability, we investigated interdecadal changes in the EASM and the SASM. Such a comparison of the EASM and the SASM implies us that EASM (SASM) is more likely be predicted with the ENSO heating center in eastern (central) Pacific. Thus, this research enhances the understanding of predictability changes in the monsoon season, but also providing a complete framework for the monsoon prediction. Nonetheless, future work should eventually incorporate these findings into monsoon predictability over broad timescales.