Weakening of the biennial relationship between Central American and equatorial South American rainfall in recent decades

Abstract There is a rainfall variability biennial relationship between Central America (CA) and equatorial South America (ESA) over the tropical western hemisphere, which is known to have arisen due to the combined effects of ENSO and tropical North Atlantic (TNA) SST. Here, the authors report that this biennial rainfall relationship between CA and ESA has weakened remarkably since 2000, with weakening in both in-phase and out-of-phase rainfall transitions. The observed decadal changes in the biennial relationship between CA and ESA rainfall can be attributed to changes in the effects of ENSO and TNA SST since 2000, which may be associated with more frequent occurrences of the central Pacific or ‘Modoki’ type El Niño. The weakening of the association with ENSO for CA rainfall since 2000 might have given rise to the weakening of the in-phase rain transition from CA rainfall to the following ESA rainfall. The weakened linkage between boreal-winter ESA rainfall and the subsequent boreal-summer TNA SST since 2000 may have resulted in the weakening of the out-of-phase rainfall transition from boreal-winter ESA rainfall to the subsequent boreal-summer CA rainfall.


Introduction
In the tropical western hemisphere, a biennial relationship exists between the interannual rainfall anomalies over Central America (CA) and equatorial South America (ESA) (Wu and Zhang 2010), which is analogous to the well-documented biennial rainfall relationship in the Indian-Australian monsoons involved in the tropospheric biennial oscillation (TBO) over the tropical eastern hemisphere (Meehl 1987(Meehl , 1993(Meehl , 1997Meehl and Arblaster 2002). Understanding the behavior and mechanisms of such a biennial rainfall relationship has important applications, due to the importance of rainfall variability to agriculture, economies and society in these regions.
According to the results reported by Wu and Zhang (2010), this biennial rainfall relationship manifests as an in-phase transition from CA rainfall to ESA rainfall (i.e. strong rainfall over CA in boreal summer followed by strong rainfall over ESA in the following austral summer, and vice versa), and an out-of-phase transition from ESA rainfall back to CA rainfall in the following year (i.e. strong rainfall over ESA in austral summer followed by weak rainfall over CA in the following boreal summer, and vice versa). Thus, these in-phase and out-of-phase transitions between CA and ESA rainfall are two key features of this biennial rainfall relationship, and are referred to as the 'biennial rainfall transition' in this study.
The biennial transitions between CA and ESA rainfall may form because of the combined effects of ENSO and tropical North Atlantic (TNA) SST (Wu and Zhang 2010), which involves interactions between the seasonal migration of convection in the northwest-southeast direction (Horel, Hahnmann, and Geisler 1989) and the underlying These positive sliding correlations are statistically significant at the 95% confidence level during nearly the entire analysis period, except for the year 2003. The values of the positive correlations between JJA CA rainfall and DJF ESA rainfall decrease after 2000 (Figure 1(a)), with a mean 11-year sliding correlation of 0.79 during the pre-2000 period , and 0.66 during the post-2000 period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016). A similar sliding analysis between the DJF ESA rainfall index and the subsequent JJA CA rainfall index shows negative correlation throughout the analysis period (Figure 1(b)), which indicates the expected out-of-phase transition from ESA rainfall to the following boreal-summer CA rainfall. The negative correlations are statistically significant at the 90 or 95% confidence level during ~1990-2000, but become non-significant after 2000. It is interesting to note that the minimum value of the positive sliding correlations (0.47) occurs in 2003 (dip in Figure 1(a)) and the maximum value of the negative sliding correlations (−0.13) in 2004 (peak in Figure 1(b)). This analysis indicates that the biennial rainfall transition between CA and ESA rainfall may have weakened since 2000.
To quantitatively assess the strength of the biennial rainfall relationship between CA and ESA rainfall, we define a biennial relationship index (BRI) as follows: where Cor1 is the 11-year sliding correlation coefficient between the JJA CA rainfall index and the subsequent DJF ESA rainfall index, and Cor2 is the 11-year sliding correlation coefficient between the DJF ESA rainfall index and the subsequent JJA CA rainfall index. The factor −1 is included in Equation (1) considering the fact that the biennial rainfall relationship is a combination of an in-phase transition (i.e. a positive correlation coefficient) and an out-of-phase transition (i.e. a negative correlation coefficient) between CA and ESA rainfall. Thus, the larger the BRI value, the stronger the biennial relationship between CA and ESA rainfall. The BRI values calculated from the 11-year sliding correlations between CA and ESA rainfall (Figure 1(c)) show prominent decadal variation. After 2000, the BRI values tend to be smaller than the long-term mean value (0.33) of the BRI during the whole analysis period. On the contrary, the BRI values before 2000 are generally greater than the long-term mean value. These results suggest that the biennial relationship between CA and ESA rainfall has weakened remarkably since 2000. A regime shift index, calculated following Rodionov (2004), further confirms the occurrence of a regime shift around 2000 for the BRI between JJA CA rainfall and DJF ESA rainfall. This weakening of the biennial rainfall relationship since 2000 may have been contributed by both the weakening of the in-phase rain transition (i.e. the value of positive correlation coefficients decreasing after 2000) and the weakening of the out-phase rain transition (i.e. the negative correlation coefficients becoming non-significant after 2000). (1) Pacific-Atlantic Ocean. It is very similar to the interactions between the seasonal migration of convection and the underlying Pacific-Indian Ocean in the biennial rainfall transitions of the TBO over the tropical eastern hemisphere (e.g. Li et al. 2006;Meehl 1987Meehl , 1993Meehl, Arblaster, and Loschingg 2003;Wu 2008Wu , 2009Wu and Kirtman 2007;Yu, Weng, and Farrara 2003).
Prominent decadal variation has been observed in the biennial rainfall relationship over the tropical eastern hemisphere (e.g. Ashok et al. 2014;Meehl and Arblaster 2011;Pillai and Mohankumar 2010;Wu 2016). However, it remains necessary to determine if there are significant decal changes in the biennial rainfall relationship between CA and ESA over the tropical western hemisphere. In the present study, we report that the biennial relationship between CA and ESA has weakened remarkably since 2000.

Data
The rainfall data, spanning the period 1979-2016, are from the CMAP product (Xie and Arkin 1997); the SST data are from the ERSST analysis (Smith et al. 2008); and the wind fields are from the NCEP-NCAR reanalysis (Kalnay et al. 1996).
Anomalies are calculated by first removing the longterm trend, and then the mean seasonal cycle, for the period 1979-2016. The level of statistical significance is determined based on the two-tailed P-value using the Student's t-test.

Results
To elucidate possible decadal changes in the biennial rainfall transition between CA and ESA, we perform an 11-year sliding correlation analysis between the CA rainfall index during boreal summer (June-July-August; JJA) and the ESA rainfall index during the austral summer (December-January-February; DJF) for the period 1979-2016 (Figure 1(a) and (b)). The correlation coefficients between the JJA CA rainfall index and the following DJF ESA rainfall index are positive throughout the analysis period (Figure 1(a)), which indicates that the expected in-phase transition from strong (weak) CA rainfall to strong (weak) ESA rainfall dominates the time series.  for CA rainfall since 2000 may be closely associated with more the frequent occurrences of CP or 'Modoki' El Niño during recent decades (e.g. Lübbecke and McPhaden 2014;McPhaden 2012). For DJF ESA rainfall, significant negative correlation is still observed, with simultaneous SST in the central and eastern equatorial Pacific east of 174°E during the post-2000 period (Figure 2(b)). The different distribution patterns in the correlation with equatorial Pacific SST for JJA CA rainfall and DJF ESA rainfall (Figure 2(b)) weakens the in-phase rainfall transition between CA and ESA.
But why did the out-phase rain transition between DJF ESA rainfall and the subsequent JJA CA rainfall weaken around 2000? The TNA SST anomalies in response to ENSOgenerated ESA rainfall/convection anomalies were likely the main contributor to this out-of-phase rainfall transition (Wu and Zhang 2010). Warm (cold) TNA SST anomalies can give rise to above-normal (below-normal) rainfall over CA through a Gill-type response (e.g. Wang, Lee, and Enfield 2008;Wu and Zhang 2010). Remarkable weakening is observed in the linkage between DJF ESA rainfall and the subsequent JJA TNA SST since 2000 (Figure 3). For the correlation between DJF ESA rainfall and the subsequent JJA TNA SST, the mean value of the 11-year sliding correlation is −0.77 for the pre-2000 period, which is significant at the 95% confidence level; however, the correlation for the post-2000 period is −0.42, which is not significant at the 90% confidence level. Regression analysis (Figure 4) also indicates that strong ESA rainfall during boreal winter may have caused cold TNA SST anomalies during the subsequent JJA in the pre-2000 period; however, such associations weakened greatly during the post-2000 period. During boreal spring, cold SST anomalies are apparent in the TNA region during both the pre-2000 (Figure 4(b)) and post-2000 (Figure 4(e)) periods. These cold TNA SST anomalies persist throughout the following boreal summer during the pre-2000 period (Figure 4(b) and (c)), which may suppress the JJA CA rainfall anomalies through a Gilltype response and thus contribute to the out-phase rainfall transition from DJF ESA rainfall to JJA CA rainfall. In contrast, the cold TNA SST anomalies decay in JJA during the post-2000 period (Figure 4(f )). Therefore, the significant weakening of the effects from the TNA SST anomalies may have weakened the out-phase rainfall transition during the post-2000 period, as shown in Figure 1(b). These changes in the TNA SST anomalies should be closely associated with different characteristics of the Pacific SST anomalies associated with the evolution of ENSO. Compared with the pre-2000 period, the maximum SST anomalies in the tropical Pacific during the post-2000 period move westward (Figure 4(d) and (e)), making them more likely to be the CP or 'Modoki' type of ENSO. Moreover, the warm SST anomalies in the eastern equatorial Pacific occur in boreal summer during the post-2000 period (Figure 4(f )), suggesting But why did the in-phase rain transition between JJA CA rainfall and the following ESA rainfall weaken around 2000? The in-phase rainfall transition may have been induced by the direct effects of ENSO (Wu and Zhang 2010), in which anomalous convection over the eastern equatorial Pacific associated with warm (cold) ENSO events generate anomalous descent (ascent) and suppress (enhance) rainfall over both CA and ESA. The 11-year sliding correlation between the JJA CA rainfall index and the following DJF Niño3.4 index shows prominent weakening from 2000, although no such changes are observed for the correlation between the DJF ESA rainfall index and the simultaneous Niño3.4 index (not shown). For the correlation between CA rainfall and the Niño3.4 index, the mean value of the 11-year sliding correlation is −0.82 for the pre-2000 period and −0.65 for the post-2000 period. In contrast, the corresponding value for the correlation between ESA rainfall and the Niño3.4 index is −0.92 for the pre-2000 period and −0.91 for the post-2000 period. Therefore, the weakening of the association with ENSO for CA rainfall might have given rise to the weakening of the in-phase rain transition from JJA CA rainfall to the following DJF ESA rainfall.
Since 2000, changes in the characteristics and variability of ENSO have been observed (e.g. Hu et al. 2013;Lübbecke and McPhaden 2014;McPhaden 2012). In particular, central Pacific (CP) or 'Modoki' El Niño events, with the center of SST anomalies shifted westward (e.g. Lübbecke and McPhaden 2014;McPhaden 2012), have occurred more frequently. The weakening of the relationship between CA rainfall and ENSO since 2000 may be closely associated with this twenty-first century shift in ENSO characteristics. During the pre-2000 period , very similar distribution patterns are observed in the correlations with equatorial Pacific SST for both JJA CA rainfall and DJF ESA rainfall (Figure 2(a)), which supports the prominent in-phase rainfall transition between JJA CA rainfall and the following DJF ESA rainfall under the same influence from the equatorial Pacific SST anomalies. Significant negative correlations are observed with DJF SST in both the central and eastern equatorial Pacific from 174°E to 80°W for both CA rainfall and ESA rainfall during the pre-2000 period (Figure 2(a)). In contrast, remarkable changes can be found for CA rainfall in the correlation distribution with equatorial Pacific SST during the post-2000 period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016), with significant negative correlation located in the central equatorial Pacific from 170°E to 144°W (Figure 2(b)). The correlation between JJA CA rainfall and the following DJF SST in the eastern equatorial Pacific (east of 144°W) becomes non-significant during the post-2000 period (Figure 2(b)). These results suggest that CA rainfall during the post-2000 period might have been influenced more by CP or 'Modoki' El Niño events, with the center of SST anomalies in the central equatorial Pacific. This change  Figure 2. the zonal distributions of correlation coefficients between rainfall indices (JJA cA rainfall and dJF esA rainfall) and dJF equatorial pacific sst averaged between 5°s and 5°n for (a) the pre-2000 period   note: the red dots represent correlations that are significant at the 95% confidence level.
persist from boreal spring to boreal summer during the pre-2000 period (Figure 4(b) and (c)), but not during the post-2000 period (Figure 4(e) and (f )), which is similar to the evolution of the cold TNA SST anomalies. These different evolutionary characteristics in TNA and Indian Ocean SST anomalies are closely associated with differing ENSO evolution during the pre-and post-2000 periods.
a faster biennial phase transition in ENSO. Wang and Ren (2017) also suggested that the dominant period of ENSO has tended to be more biennial since 2000. This fast ENSO phase transition from the cold to warm phase may be the root cause for the observed weaker persistence in the cold TNA SST anomalies from boreal spring to boreal summer during the post-2000 period. It is interesting to note that the cold SST anomalies in the tropical Indian Ocean also Figure 4. regressions of sst anomalies (color shading) and 850-hpa winds (vectors) with respect to dJF esA rainfall from dJF to the following JJA during (a-c) the pre-2000 period  and (d-f) the post-2000 period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016).
notes: only the values at the 90% confidence level or higher are shown. the red box in (c) and (f) represents the domain of the tnA sst index. ORCID WANG Lei http://orcid.org/0000-0002-9015-5422