Preceding winter La Niña reduces Indian summer monsoon rainfall

Leaving out the strong El Niño Southern Oscillation (ENSO) years, our understanding in the interannual variation of the Indian summer monsoon rainfall (ISMR) stands poor for the rest. This study quantifies the role of ENSO in the preceding winter on ISMR with a particular emphasis on ENSO-neutral summer and La Niña winter. Results show that, unlike the simultaneous ENSO-ISMR relationship, La Niña of previous winter reduces mean rainfall over the country by about 4% even during ENSO neutral summer. Moreover, when ENSO changes phase from La Niña in winter to El Niño in summer, ISMR is anomalously lower than during persisting El Niño years (−14.5% and −5.3%, respectively), increasing the probability of severe drought. This suppression effect of La Niña of the preceding winter on summer monsoon precipitation over India is mostly experienced in its western and southern parts. Principal component analysis of the zonal propagation of surface pressure anomalies from winter to summer along Northern Hemisphere subtropics decomposes interannual variations of seasonally persisting anomalies from zonal propagations. The dominant modes are associated with the seasonal transition of the ENSO phase, and are well correlated with date of onset and seasonal mean rainfall of monsoon over India. These results improve our understanding of the interannual variations of ISMR and could be used for diagnostics of general circulation models.


Introduction
The interannual variation of Indian summer monsoon rainfall (ISMR) bears crucial socio-economic values to the country of teeming millions [1,2]. ISMR exhibits quasi-periodicity at time periods from biennial to multi-decadal [3]. The identification of the physical mechanism driving this interannual variation would not only provide a pathway to improve the seasonal prediction of ISMR, but is also central to understanding organized tropical convection and associated interaction with the non-linear, transient global climate [4][5][6][7].
The gigantic circulation of the Asian summer monsoon [8,9], of which the Indian monsoon is a part, shows notable east-west asymmetry at lead zonal wave-number. It is known that the El Niño Southern Oscillation (ENSO) impacts ISMR by controlling the strength and position of this circulation [10,11].
ENSO could also impact ISMR through modulation of the stationary Rossby wave of the midlatitude [12,13]. On the other hand, the strength of Indian monsoon in summer (weak or strong) skews the probability of development of El Niño or La Niña in the following winter [14,15]. Such interactions have associated time scales that control the lagged response between components of the system. This time-lag in response between the ENSO and Asian monsoon, combined with the seasonality of their phases, is possibly the key to explaining the observed asymmetry in the East Asian summer monsoon between the developing and decaying phases of ENSO [16][17][18], and its dependence on the mean state of climate [19]. It has been shown that the decaying (developing) mean sea level pressure of Darwin from winter to spring increases (decreases) the following ISMR [20]. This was used, in conjunction with other parameters, to develop a statistical prediction model for ISMR [21]. Tendencies of eastern Pacific sea surface temperature (SST) are also used for the real-time prediction of ISMR [22] using the statistical model. However, the skill of such models could vary on decadal time scales depending on the mean state [23,24].
Although there are several studies addressing the importance of the change of phase of ENSO on the East Asian monsoon [16,25,19], such quantifications are limited for Indian summer monsoon. This is especially important for the ENSO neutral summer for which identification of an exact reason for the spatial distribution of precipitation anomalies over India has been elusive. This study quantifies the changing footprint of ENSO from winter to summer on ISMR with special emphasis on the ENSO neutral summer. It also puts forward a mechanism that can explain the observed delayed response of ISMR to winter ENSO conditions.

Methods
We use precipitation data sets from two different sources. The first one is a 144 year long record of raingauge-based precipitation over India [26], available from 1871 through 2014. We also use raingauge-based precipitation at every 1 • × 1 • grid over India, obtained from the India Meteorological Department (IMD), available as daily averages from 1901 through 2014 [27].
We use Hadley Centre Interpolated SST observations [28], available at global regular grids, monthly averages from 1870 through present. Monthly mean atmospheric parameters such as surface pressure (p ), winds and humidity were taken from the National Center for Environmental Prediction /National Center for Atmospheric Research (NCEP/NCAR) Reanalysis 1 (R1). This data is available from 1948 through present [29].
In this study ISMR is defined as June-September mean area averaged precipitation over India, as in [26]. Prior to any calculation, anomalies were created removing the monthly mean annual cycle from all data sets. We average SST of Nino 3.4 region (190 • -240 • E, 5 • S-5 • N; termed as N34SST henceforth) during Northern Hemispheric summer (June-September: JJAS) and the preceding winter (December-February; DJF) to categorize them into phases of ENSO. If N34SST is less than −0.5 • C (greater than 0.5 • C), we identify that season as La Niña (El Niño), the rest being ENSO neutral. Owing to the slowly varying nature of N34SST, such definitions of ENSO using four (June-September) and three (December-February) months averages is close to the definition of ENSO that uses three month running average (supplementary table S1 available at stacks.iop.org/ERL/13/054030/mmedia). In this study, we identify years into nine possible categories: from DJF La Niña, neutral or El Niño to JJAS La Niña, neutral or El Niño. List of these years from 1871 through 2014 are provided in supplementary table S2. The correspond-ing composites of N34SST from December through September are shown in figure S1. While using a shorter length data set, we use the corresponding subset from this list.
Vertically integrated moisture flux is calculated at every grid of R1 using standard procedure.
To understand interannual modes of timelongitude evolution of p , principal component (PC) analysis is performed on the 20 • -30 • N mean p anomalies from January through September.
where p (X, M, T) is surface pressure anomaly averaged along 20 • -30 • N. and are the nth empirical orthogonal function (EOF) pattern and PC time series, respectively. X is number of longitude grids (144 for R1). M is number of months (January-September: nine). T is number of years. N is number of PCs. In other words, we replace the traditional 'space' (longitudelatitude) of standard PC analysis with Hovmoller (longitude-month) patterns. The time T is interannual variations of this pattern (1948 through 2014: 67 years).
The date of the onset of monsoon over Kerala (MoK) was calculated using daily rainfall averaged between 74 • -78 • E, 8 • -12 • N, and vertically integrated (surface to 600 hPa) zonal wind averaged between 55 • -75 • E, 5 • -12.5 • N. The MoK is defined as the first day of the monsoon season (after 15 May) when the rainfall exceeds 4 mm d −1 and remains above this value for at least three consecutive days, and the zonal wind is above 3 m s −1 for at least ten consecutive days. This definition is same as that in [30].
We do not remove any long-term trend from any data set. Results, especially those involve computation of spatial pattern, are dependent on the method used for trend removal [31]. This is because it is possible to modify some real modes/oscillations by removing a trend from a short time series. Moreover, the observed trend of N34SST, which is the focus of this study, is weak. Despite this, please note that the identified years in all nine categories (supplementary table S2) are evenly distributed along the 144 year timeline.

Results
A scatter plot between N34SST in JJAS and ISMR reconfirms their general out-of-phase relationship ( figure 1(a)). The linear correlation coefficient is −0.57, suggesting that N34SST in JJAS is associated with 32.5% of the total interannual variance of ISMR. While most of the La Niña summer experience +ve ISMR anomaly (mean and standard deviation of ISMR for those years are 0.59 and 0.35 mm d −1 , respectively), El Niño summers show larger scatter (mean = −0.62 mm d −1 , standard deviation = 0.66 mm d −1 ). The scatters also suggest stronger droughts than floods, not necessarily relating to summer El Niño. Besides, most of the variance in ISMR is noticed when JJAS N34SST is neutral. About 28% of the summer ENSO neutral years experience drought or flood (ISMR < −10% or > +10% of its long-term climatology).
The colors and size of the markers on figure 1(a) indicate N34SST in the preceding winter (DJF); the larger the marker, the stronger the anomaly. Note the skewed distribution of DJF N34SST along the ISMR axis. Many of the warm (cold) winters seem to be associated with increased (decreased) ISMR for a given value of N34SST in summer. This is particularly true for years when JJAS ENSO is in a state of neutral or El Niño. Now, we identify all JJAS ENSO neutral years and calculate average N34SST in the preceding winter for different ranges of ISMR ( figure 1(b)). This quantifies the relationship between N34SST in winter and ISMR for JJAS ENSO neutral years. It is clear that warm SST in DJF results in increased ISMR even when JJAS ENSO is neutral. This is especially true for ISMR anomalies < −0.75 mm d −1 and > 0.75 mm d −1 . We explore this further computing conditional composites based on N34SST.
ISMR composited separately for years when JJAS ENSO is neutral, El Niño or La Niña are shown in figure 1(c). Even when summer ENSO is neutral, ISMR appears to be dependent of the phase of ENSO in the preceding winter. If preceding winter was La Niña, the mean of ISMR is about 4% below normal, and about 70% of those years experience −ve ISMR anomaly. Although we do not include the 2017 summer season in this study, from observations we know that there were several such years in the past, including 2017 when the ISMR anomaly was close to −5% with an ENSO neutral summer that followed winter La Niña. When we compare this with winter El Niño years, summer neutral conditions of ENSO produce mostly close to normal ISMR. The hypothesis that mean ISMR anomaly during JJAS ENSO neutral for DJF La Niña years is zero is rejected at 97.9% significant level.
Summer El Niño conditions result in below normal ISMR. These anomalies are stronger compared to JJAS ENSO neutral years preceded by DJF La Niña, suggesting that the simultaneous impact of ENSO on ISMR is stronger than the delayed impact, and is consistent with previous findings. However, when summer El Niño was preceded by winter La Niña, the negative anomaly is more severe (mean ∼−14.5%). In contrast, summer El Niño years preceded by El Niño do not experience such large deficit in ISMR (mean ∼−5.3%). However, when El Niño is preceded by winter ENSO neutral conditions (white bars of figure 1(c)), the ISMR is significantly below normal. Although there are not too many years of summer El Niño preceded by winter El Niño, consistent with previous studies [32], we note that the composite N34SST in JJAS is similar for winter El Niño, La Niña and neutral states (figure S1). However, it must be kept in mind that ENSO-ISMR relationship could be non-linear and the differences seen could partly be associated with different amplitudes of the sampled events.
Summer La Niña conditions produce positive ISMR anomaly (3rd set of bars in figure 1(c)), again consistent with previous findings. But we do not find any significant impact of the previous winter's ENSO on ISMR during La Niña summer. However, there are large spatial differences, which are discussed later. Figures 1(d)-(h) show that summer rainfall over southern and north-western parts of India are mostly sensitive to preceding winter ENSO condition. For example, over peninsular India, winter El Niño and La Niña years produce opposite signs of rainfall anomaly that is significant at least at the 95% level. Similarly, winter El Niño with summer La Niña produces large positive anomaly in rainfall ( > 15%) that is significant at the 99.2% level, and larger than for winter La Niña and neutral conditions.
The spatial patterns of JJAS mean precipitation composites based on IMD 1 • × 1 • observations for six different combinations of winter and summer ENSO states (of figure 1(c)) are shown in figure 2. Summer ENSO neutral years with preceding winter El Niño (figure 2(a)) are marked by negative precipitation anomaly over western and southern India. The anomalies are close to zero for summer ENSO neutral years with preceding winter El Niño ( figure 2(d)). Similarly, summer El Niño years those come from winter La Niña ( figure 2(b)) shows stronger negative anomalies all over the Indian region as compared to years of persisting El Niño ( figure 2(e)). In figures 2(c) and (f) we see that previous winter El Niño or La Niña redistributes rainfall spatially during La Niña summer. While Northeast India receives more precipitation for persisting La Niña conditions ( figure 2(c)), the transformation of El Niño to La Niña produces more precipitation over western and southern India. Differences between the composites for winter La Niña and El Niño are statistically significant over several parts of Indian land (figures S2 and S3). Panels of figure 2(g) show the probability distribution of JJAS mean precipitation over five homogeneous regions of Indian summer monsoon (area weighted average of these values is the 144 year ISMR data from [26]), separately constructed for winter La Niña and El Niño, irrespective of the summer ENSO state. Over peninsular India (PEN), winter La Niña increase probability of strong and moderate drought in summer. At the same time, winter El Niño increases the probability of rainfall > 1 mm d −1 over this region. Similarly, the probability curves for west central India (WC) and north-west India (NW) are skewed toward negative (positive) side of the distribution for La Niña (El Niño) winter, signifying the increased probability of drought (flood). On the contrary, the probability curve for precipitation over central Northeast India (CNE) is positively skewed for winter La Niña (consistent with the spatial map). The difference in summer precipitation over Northeast India (NE) between winter La Niña and El Niño is insignificant.

Mechanism
It is known that large-scale circulations associated with p are a dominant signature in the subtropics and midlatitudes. The subtropical high of the western North Pacific Ocean induces large anticyclonic circulation that has substantial interannual variation [33]. On the other hand, to the western side of India (over West Asia), climatological surface pressure is lower than that at the equator along the same longitude. This pressure gradient drives the strength and position of the cross-equatorial low-level winds over the Arabian Sea. Interannual variations of this gradient affect the date of the onset of the monsoon over central India [30]. Figures 3(a)-(d) show spatial variation of regression coefficients between N34SST in DJF with surface pressure (shaded) and vertically integrated moisture flux (vector) in four different seasons. In January-February ( figure 3(a)), p of tropical western Pacific increases with an increase in N34SST. This is a typical response of the atmosphere during ENSO in winter [25]. Associated with this increase in p is The crossed grids are regions where this anomaly is different than zero at least at 90% level. Panels of (g) show probability distribution of precipitation over five homogeneous regions of India for preceding winter La Niña (blue lines) and El Niño (red lines) conditions, irrespective of the summer ENSO state. The 90% confidence band, calculated using a binomial distribution, is shown as shades (supplementary text S1). anomalous anticyclonic moisture flux over western Pacific and anomalous westward moisture flux over equatorial Indian Ocean. We also notice a decrease (increase) in p northwest of India with El Niño (La Niña) condition in winter.
In March-April ( figure 3(b)), the equatorial western Pacific p anomaly weakens and is mostly centered at around 15 • -20 • N. The negative p anomaly of western Asia also weakens but its center moves southeast. In May-June (figure 3(c)) the high of western Pacific weakens further but extends westward and merges with the high over the Indian region. At the same time, p over the northern Arabian Sea and northwestern India increases. This results in the weakening of the low-level westerlies and a decrease in moisture flux to the Indian region from the Arabian Sea, consistent with [30]. This can have an impact on the onset date of the monsoon, which is discussed later. However, note that the low p region over 20-40 E, 30-50 N still persists, although weak. In July-August-September (figure 3(d)) this low p region moves southeast close to the Indian monsoon, strengthening the low level moisture flux through the Arabian Sea toward India. We also notice the persistence of the western North Pacific high pressure, transporting anomalous moisture toward Indian region from west. This increase (decrease) in moisture flux toward India in summer following warm (cold) N34SST of the preceding winter increases (decreases) precipitation over India, as seen in figures 1(b) and 2.
Composites of moisture flux are calculated for summer ENSO neutral years. In figure 3(f), we find that the westward influx of moisture to the Indian region at 85 • E in JJAS is higher for DJF El Niño years as compared to La Niña years. This is due to the presence of anomalous subtropical high (low) of western north Pacific, persistent from preceding winter El Niño (La Niña)  [25].
How do these p anomalies propagate in space from winter to summer? Figures 4(a) and (b) show the longitude-time evolution of monthly p anomaly along 20 • -30 • N during ENSO neutral summer, composited separately for La Niña and El Niño winters. The La Niña (El Niño) winters are associated with increased (decreased) p in northern subtropics. These anomalies are seasonally persistent from December to April. It is interesting to note here that during northern winter and spring, p anomalies are of opposite signs between 0-180 • E and 0-180 • W. In May, p anomalies become +ve (−ve) between 0-60 • E for winter La Niña (El Niño) condition. The is more prominent for winter La Niña to summer ENSO neutral years (figure 4(a)) for which +ve p anomalies develop over the north and west of the Indian region and persist up to September. This could be responsible for reduced eastward moisture flux over the Arabian Sea that can affect ISMR [30].
We now analyze the time evolution and propagation of p anomalies for all years, averaged between 20 • -30 • N, from winter to summer using principal component analysis. The first mode of time-longitude evolution shows a standing wave with prominent high from 10 W to 120 E, covering the European and Asian midlatitude. The time evolution of this pattern does not show any prominent propagation in longitude direction. The associated PC shows interannual variation amid increase since 1960s and decrease after mid-1990s. We found that this PC is closely associated with the date of onset of the monsoon over peninsular India (r = 0.44). Note that early onset before the 1960s and delayed onset afterward is associated with this pattern of p . Reference [30] has shown that an increase in p over western Asia can delay the onset of the monsoon because of a decrease in westerly moisture flux toward India. The current finding, in addition, sheds light on the observed long-term variation of the onset over peninsular India.
The second mode (figures 5(c) and (d)) is closely associated with the winter ENSO condition (r = 0.6). El Niño (La Niña) decreases (increases) p of subtropical eastern Pacific Ocean and increases (decreases) p over western Pacific Ocean. The anomalies of the eastern Pacific fizzle out after winter. However, the anomalies of western Pacific associated with this principal component strengthen in northern summer. This is consistent with figure 3. The associated p anomalies move westward from winter to summer at a speed of about 7 degrees of longitude per month from winter to summer.
The third and fourth modes are associated with winter and summer N34SST, and their difference, although somewhat weakly. The fifth mode shows prominent eastward propagation (figures 5(i) and (j)), with speed of about 32 degrees of longitude per month. p anomalies of eastern Pacific Ocean in winter propagate eastward and become more prominent between 0 • -120 • E after April. At the same time, p anomalies of central Pacific Ocean (180 • -240 • E) propagate eastward and appear over western Pacific Ocean by JJAS. A positive (negative) value of the associated PC time series results in a decrease (increase) in p over west Asia (west Pacific Ocean) that increase (decrease) moisture flux toward the Indian region. This increases precipitation over India, as seen by the correlation between this PC time series with ISMR (r = 0.48). Note that this value of correlation coefficient between 5 and ISMR is close to that between ISMR and JJAS N34SST (−0.57).

Discussions
In this study, we quantify the relative impacts of ENSO in summer and in the preceding winter on summer monsoon rainfall over India. We focus on years of ENSO neutral summer since interannual variations of ISMR during those years are least understood. The primary findings are: a. Winter La Niña reduces following summer monsoon rainfall over India by about 4% even during ENSO neutral summer. The partial correlation between DJF N34SST and ISMR, after regressing out the effect of JJAS N34SST on ISMR is 0.27 (significant at 95% level). These results are consistent with previous findings [21] that use the tendency of Darwin surface pressure. However, the present study uses SST that has higher auto-correlation in time than surface pressure.
b. This impact of preceding winter La Niña is strongest during El Niño summer (ISMR anomaly about −14.5%, as opposed to −5.3% for a persistent El Niño), increasing the probability of severe drought. This is reported for the first time and would have high socio-economic importance.
c. The western and southern parts of Indian land are most prone to reduced precipitation when the preceding winter was La Niña. Such spatial variations of the impact of winter ENSO is also reported for the first time. d. As ENSO weakens after the strongest amplitude in winter, surface pressure anomalies propagate along northern subtropical longitude, both toward east (faster) and west (slower). This modulates surface pressure over west Asia and west Pacific Ocean during May-September, impacting moisture transport through the Arabian Sea and the Bay of Bengal. The anomalous moisture transport modifies monsoon precipitation over India.
The delayed impact of El Niño on the Indian Ocean region is known to be through the thermal capacitor effect [34]. The continued warming signal of the tropical Indian Ocean after a strong El Niño event during winter could change the climate around this region [35,36], especially weakening monsoon circulation. Our results show that the effect of preceding winter La Niña on ISMR is more as compared to that of preceding winter El Niño. Composite plots of seasonal evolution of global SST, done separately for winter La Niña and El Niño (supplementary figures 2 and 3) show that while the Indian Ocean is warmer during JJAS for preceding winter El Niño, SST is close to normal for winter La Niña years. In other words, the thermal capacitor effect (in this case cooling) is not observed for La Niña to neutral years. The increased SST over Indian Ocean for summers preceded by El Niño could reduce the impact of mid-latitude p anomalies on monsoon rainfall over India, as was seen in figure 1(c).

Conclusion
This study, for the first time, quantifies the precipitation anomaly over different regions of India resulting from combined states of ENSO in winter and summer. The results not only highlight this spatial variation of precipitation, but also underlines the asymmetry of the impact of El Niño and La Niña of winter and summer on the summer monsoon. The method of identifying modes of time-longitude evolution of surface pressure decomposed standing anomalies from propagations without pre-assuming periodicity. This method and other findings could be used for the evaluation of numerical models employed for seasonal prediction of the monsoon, which normally show poor skill in capturing the impact of ENSO on ISMR and its spatial variations.