Introduction

El \({\rm{N}}{\rm{i}}\tilde{{\rm{n}}}{\rm{o}}\)-Southern Oscillation (ENSO) is the predominant climate signal in the human scale; it exerts a great impact on weather and climate around the globe1,2,3,4. As one of the most significant predictable sources for seasonal forecasting, ENSO displays great diversity in its onset, amplitude, duration, and evolution, with great asymmetry between El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) and La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) episodes5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20.

Given its distinctive horizontal SSTA patterns, El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) has been categorized into “eastern Pacific (EP) El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)”, “central Pacific (CP) El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)” and “Mixed El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)”21,22,23,24. Some studies further distinguished El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) into “spring (SP) El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)” and “summer (SU) El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)” based on its onset season25. A recent study pointed out that “SP El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)” is associated with “EP El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)”, whereas “SU El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)” aligns with “CP El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)”26. These different types of El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) exhibit distinctive evolution patterns and climate impacts.

Some researchers attempted to classify La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) into “EP” and “CP” (or “Modoki”) types based on the SSTA spatial distributions18,22,27,28,29. However, the locations of maximum negative SSTA centers during La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) shift generally to the west of maximum positive SSTA centers during El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) 30,31. As a result, the longitudinal differences between all La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) events are relatively small compared to those of El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\). The prevailing view is that La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) is difficult to be classified based on the spatial distribution of the SSTA, unlike El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)19,23,32,33.

ENSO onset may rely on various external and internal precursory signals, such as the Pacific Meridional Mode34,35,36, interannual variability of the Pacific Decadal Oscillation37, low-frequency and high-frequency zonal winds38,39,40, warm water volume41,42, and SSTAs in Atlantic or Indian Ocean (IO)39,43,44,45,46,47. ENSO phase transition theories such as the “delayed oscillation”48,49, “recharge oscillation”50,51 and “western Pacific (WP) oscillation”52,53 emphasized the importance of ocean waves or zonal mean thermocline variation. There is a clear asymmetry between El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) and La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) onset and evolution. For instance, after reaching its peak phase, El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) often transitioned into a cold episode or a neutral state, whereas La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) frequently developed into a multi-year cold episode54. The cause of multi-year La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) events was suggested to relate to super discharge of ocean heat content driven by preceding super El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) events55,56,57,58,59. However, observations point out that there were more multi-year La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) events that did not follow a super El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\)60.

So far one of most lacking research topics is the diversity of La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) onset. This motivates the current study. The objective of this paper is to classify the La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) onset types based on an objective clustering analysis of the equatorial SSTA evolution during the past 111 years (1910–2020) and to reveal mechanisms responsible for each of the onset types identified. Understanding the different types of La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) onset and mechanisms behind is crucial for improving seasonal forecasting, reducing disaster risks, and mitigating economic losses and human impacts.

Results

Three distinguished types of La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) onset

The time series of the Oceanic \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) index (ONI) indicated that there are 33 El \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{o}}\) events and 30 La \({\rm{Ni}}\tilde{{\rm{n}}}{\rm{a}}\) events during 1910-2020 (see Methods) (Fig. 1a). A single- (multi-) year ENSO refers to an El Ni\({\rm{ {\tilde{n}} }}{\rm{o}}\) or a La Niña that persists for only 1 year (two or three years). Among the 30 La Niña events, 17 belong to either a single-year La Niña or the first year of a multi-year La Niña event.

Fig. 1: Observed La Niña onset diversity.
figure 1

a Time evolution of detrended ONI from 1910 to 2020 (units: °C). Red, blue, green and white bars denote El Niño, single-year and the first year of multi-year La Niña, the second and third year of the multi-year La Niña, and neutral years, respectively. Dashed lines are ±0.5 °C. The “N”, “C”, and “S” in the bottom represent that the La Niña onset types are “neutral year to La Niña” (N2L), “central Pacific El Niño to La Niña” (CE2L) and “super El Niño to La Niña” (SE2L), respectively. Time-zonal sections of SSTA (shaded, units: °C) and SSTA tendency (contour, units: °C mon−1) along the equator (within 5°S–5°N) for b N2L, c CE2L and d SE2L. Solid (dashed) lines represent the positive (negative) SSTA tendencies. Purple dots are the maximum negative SSTA center where SSTA exceeds −0.45 °C. Black dots represent SSTA that exceeds the 95% confidence level. Purple boxes denote La Niña onset phases. “−1” and “0” represent the preceding and developing year of La Niña.

Full size image

Based on the K-means cluster analysis (see Methods) of equatorial (5°S–5°N) SSTA evolutions from La Niña preceding summer to developing winter, all 17 first-year events can be categorized into three onset types (Fig. 1b–d; Supplementary Figs. 1–3). The first onset type is characterized by a transition from a neutral year to La Niña (N2L) (Fig. 1b). The second onset type is a transition from a CP El Niño to La Niña (CE2L) (Fig. 1c), and the third onset type is a transition from a super El Niño to La Niña (SE2L) (Fig. 1d).

The time-longitude sections of the equatorial SSTA and its tendency from preceding June to developing December illustrated that a negative SSTA starts to emerge in central-eastern Pacific (CEP, 5°S–5°N, 160°–110°W) during SO(−1) for N2L. Once being initiated, the negative SSTA persists from Oct(−1) to May(0) (Fig. 1b). Here “−1” and “0” represent the preceding year and developing year of La Niña, respectively. Then the SSTA develops rapidly with the maximum negative SSTA center shifting from 120°W in May(0) to 135°W in Nov(0).

For CE2L, a CP El Niño appears in the preceding winter, with a maximum SSTA center around 155°W (Fig. 1c). The El Niño starts to decay firstly in EP in JF(0), and then expands to CP in the following months. As a result, a negative SSTA tends to appear initially in the equatorial EP (EEP, 5°S–5°N, 120°–80°W) during JF(0), then propagates westward, reaching the threshold of La Niña by Dec(0).

For SE2L, it is evident that the preceding year is characterized by a super El Niño, with the maximum center located near 110°W and a negative SSTA in the equatorial WP (Fig. 1d). The El Niño decays firstly in CP in June, then expends eastward. As a result, this type of La Niña onsets firstly in CP (5°S–5°N, 175°–125°W) in MJ(0), subsequently expanding eastward in the following months. A primary distinction between CE2L and SE2L lies in the preceding SSTA patterns associated with CP and EP El Niños in DJF(0), whose roles will be explored in the following sections.

Cause of the N2L onset

The evolution patterns of anomalous SST, precipitation and low-level wind fields from SON(−1) to SON(0) indicate that there are obviously positive SST and precipitation anomalies in the tropical North Atlantic (TNA) in SON(−1) (Fig. 2a), which could drive easterly anomalies to its east as Kelvin wave response44,61, extending from IO to the equatorial western and central Pacific (WCP).

Fig. 2: Seasonal evolutions of circulation and diagnosed results for N2L onset.
figure 2

a–e Horizontal patterns of anomalous SST (shaded, units: °C), precipitation (crossed, units: \({mm}\) \({{day}}^{-1}\)) and wind fields at 850 hPa (vector, units: \(m\) \({s}^{-1}\)) from the preceding autumn to the developing autumn of N2L events. f Mixed-layer heat budget diagnosed results for N2L averaged over (5°S–5°N, 160°–110°W) during SO(−1) (units: °C mon−1). The x-axis from left to right represents MLTA tendency (\(\frac{\partial {T}^{{\prime} }}{\partial t}\)), the sum of advection and heat flux, the sum of advection, the sum of heat flux, the individual advection and heat flux terms: \(-\bar{u}\frac{\partial {T}^{{\prime} }}{\partial x}\), \(-{u}^{{\prime} }\frac{\partial \bar{T}}{\partial x}\), \(-{u}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial x}\), \(-\bar{v}\frac{\partial {T}^{{\prime} }}{\partial y}\), \(-{v}^{{\prime} }\frac{\partial \bar{T}}{\partial y}\), \(-{v}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial y}\), \(-\bar{w}\frac{\partial {T}^{{\prime} }}{\partial z}\), \(-{w}^{{\prime} }\frac{\partial \bar{T}}{\partial z}\), \(-{w}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial z}\), net longwave radiation (\({{lw}}^{{\prime} }\)), net solar radiation (\({{sw}}^{{\prime} }\)), surface latent heat flux (\({{lh}}^{{\prime} }\)) and surface sensible heat flux (\({{sh}}^{{\prime} }\)), respectively. The error bars represent one standard deviation. Orange solid (white hollow) dots denote statistically significant (insignificant) terms. g SSTA (shaded, units: °C), zonal wind anomaly at 850hPa (vector, units: \({{m\ s}}^{-1}\)), h zonal wind anomaly at 850hPa averaged over 160°–110°W (solid line, units: \({{m\ s}}^{-1}\)) and 150°E–100°W (dashed line, units: \({{m\ s}}^{-1}\)), i thermocline depth (shaded, units: \(m\)) and horizontal ocean currents (vector, units: \({{m\ s}}^{-1}\)), j zonal geostrophic current and Ekman current averaged over (2°S–2°N, 160°–110°W) (units: \({{m\ s}}^{-1}\)) during SO(−1) of N2L. The green (brown) crosses in a–e denote the regions of significant positive (negative) precipitation anomalies. Black dots a–e, g, i represent SSTA and thermocline depth that exceed the 95% confidence level, respectively. The blue arrow in (a) denotes the easterly anomaly along the equatorial Pacific.

Full size image

The mechanism of TNA warming generating easterly anomalies in the equatorial WCP was demonstrated by Jiang and Li (2021) using coupled GCM experiments44. Easterly anomalies are firstly generated in the tropical IO, and they oppose the climatological westerlies in the northern IO, reducing the surface latent heat flux, and inducing local SSTA warming and positive precipitation anomalies. The so-generated atmospheric heating further induces an eastward-propagating Kelvin wave, which features low-level easterly anomalies along the equator, thereby enhancing the easterly anomalies over the WP44. The easterly anomalies over the WP may initiate a cooling in the equatorial WCP, which may reinforce and maintain itself through positive air-sea feedbacks (Fig. 2a–e).

To understand the key processes triggering the onset, we conducted a mixed-layer heat budget analysis (see Eq. (1) in Methods) over CEP (5°S–5°N, 160°–110°W) in SO(−1). The result indicates that the primary contributor to the N2L onset is the zonal advective feedback \((-{u}^{{\prime} }\frac{\partial \bar{T}}{\partial x})\) (Fig. 2f).

What drives the distinct zonal advective feedback? According to Su et al.11, the anomalous zonal currents can be decomposed into thermocline-induced geostrophic currents and wind-induced Ekman currents11 (see Eqs. (2–4) in Methods). Our analysis shows that the anomalous zonal current is dominated by the geostrophic current while the Ekman current is negligible (Fig. 2j), consistent with Su et al.11. This finding complements the results of Li et al.62, who emphasized the role of meridional transport convergence in modulating thermocline variability over the full ENSO cycle, whereas our study focuses on the role of geostrophic processes in shaping zonal current anomalies during the onset phase.

Given the absence of preceding ENSO forcing in N2L, inter-basin teleconnection from TNA holds a key for this onset type. Moreover, the easterly anomaly over the WP exhibits its maximum along the equator and decreases poleward in the tropical Pacific, presenting an anticyclonic shear in both north and south of the equator, as shown by the meridional profiles of the zonal wind anomaly at 850hPa in SO(−1) (Fig. 2g, h). On one hand, the maximum easterly wind anomaly along the equator induced ocean upwelling Kelvin wave44, shoaling the thermocline depth. On the other hand, the anti-cyclonic shear of wind anomaly would generate downwelling ocean Rossby wave off the equator48, deepening the thermocline depth. The meridional distribution of thermocline depth anomaly help strengthen the westward geostrophic current anomalies (Fig. 2i, j; see Eq. (3) in Methods), leading to SSTA cooling through zonal advective feedback.

Processes controlling the CE2L and SE2L onset

Figure 3 shows the evolution patterns of anomalous SST, precipitation and lower-tropospheric wind fields from DJF(0) to SON(0) for both the CE2L and SE2L types. It is evident that a negative SSTA appears firstly in EP during MAM(0) for CE2L, while a negative SSTA appears initially in CP during JJA(0) for SE2L (Fig. 3a–d, f–i).

Fig. 3: Seasonal evolutions of circulation and diagnosis for CE2L and SE2L onsets.
figure 3

Horizontal patterns of anomalous SST (shaded, units: °C), precipitation (crossed, units: \({mm}\) day−1) and wind fields at 850hPa (vector, units: \(m\) \({s}^{-1}\)) from preceding winter to developing autumn for a–d CE2L and f–i SE2L. The green (brown) crosses denote the regions of significant positive (negative) precipitation anomalies. Black dots represent SSTA that exceeds the 95% confidence level. Wind vector with speed less than 0.2 \(m\) \({s}^{-1}\) is omitted. e, j Same with Fig. 2f but for CE2L averaged over (5°S–5°N, 120°–80°W) during developing JF(0) and for SE2L averaged over (5°S–5°N, 175°–125°W) during developing May and June MJ(0) (units: °C mon−1).

Full size image

The distinctive onset locations between these two types of La Niña are attributed to the different zonal wind anomaly induced by distinctive warming locations of two types of El Niño. For CE2L, in response to a maximum positive SSTA in CP in DJF(0), westerly anomalies appear in WP while easterly anomalies appear in EP (Fig. 3a). The EP easterly anomalies initiate a surface cooling in EP in the subsequent season [MAM(0)], and so-generated cooling then expands westward (Fig. 3b–d).

In contrast, for SE2L, a preceding maximum positive SSTA center appears in EP during DJF(0), which induced westerly anomalies across most of the equatorial Pacific (Fig. 3f). As a result, a cooling appears firstly in CP around Jun(0), and then expands eastward (Figs. 1d, 3h).

To reveal specific processes that contribute to the onset of the initial cooling for CE2L and SE2L, we conducted a mixed-layer heat budget analysis (see Eq. (1) in Methods) in their respective onset regions and months. The result revealed that the main contributors to CE2L onset are anomalous latent heat flux (\({{lh}}^{{\prime} }\)) and Ekman upwelling feedback (\(-{w}^{{\prime} }\frac{\partial \bar{T}}{\partial z}\)) (Fig. 3e). In contrast, the primary drivers for SE2L onset are the thermocline feedback (\(-\bar{w}\frac{\partial {T}^{{\prime} }}{\partial z}\)) and anomalous solar radiation forcing (\({{sw}}^{{\prime} }\)) (Fig. 3j).

As the leading contributor to the CE2L onset, the latent heat flux anomaly can be decomposed into three parts (see Eq. (8) in Methods)3,63. Our analysis result shows that the wind-induced part (−0.24 °C mon−1) dominates, while the humidity difference part is much weaker and tends to offset the wind effect (Fig. 4a). Physically, the CP El Niño induced easterly anomalies in EP strengthen the climatological easterly, leading to an enhanced latent heat flux and a surface cooling in EP.

Fig. 4: Diagnosed latent heat flux and vertical ocean currents during CE2L onset.
figure 4

a Diagnosed latent heat flux (\({{lh}}^{{\prime} }\)) and its linearized components averaged over (5°S–5°N, 120°–80°W) during JF(0) for CE2L (units: °C mon−1). “wind_anom”, “q_anom” and “wind_q_anom” represent the components \(-{\rho }_{a}{C}_{E}{L}_{v}{\left|U\right|}^{{\prime} }\overline{\left({q}_{s}-{q}_{a}\right)}\), \(-{\rho }_{a}{C}_{E}{L}_{v}\bar{U}{\left({q}_{s}-{q}_{a}\right)}^{{\prime} }\), and \(-{\rho }_{a}{C}_{E}{L}_{v}{\left|U\right|}^{{\prime} }{\left({q}_{s}-{q}_{a}\right)}^{{\prime} }\), respectively. b Anomalous vertical ocean current (\({w}^{{\prime} }\)) at the bottom of the mixed layer (units: \({3\times 10}^{-5}\ m\ {s}^{-1}\)), mixed-layer averaged divergence (\({{DIV}}^{{\prime} }\)), divergence calculated by geostrophic currents (\({{{DIV}}_{g}}^{{\prime} }\)), divergence calculated by Ekman currents (\({{{DIV}}_{e}}^{{\prime} }\)), zonal component of \({{{DIV}}_{e}}^{{\prime} }\), and meridional component of \({{{DIV}}_{e}}^{{\prime} }\) (units: \({10}^{-6}\ {s}^{-1}\)) averaged over (5°S–5°N, 120°–80°W). The error bars in a, b represent one standard deviation. Orange solid dots denote statistically significant terms. Horizontal distribution of anomalous c vertical ocean current at the bottom of the mixed layer (units: \({3\times 10}^{-5}\ m\ {s}^{-1}\)) and d mixed-layer averaged horizontal divergence (shaded, units: \({10}^{-6}\ {s}^{-1}\)) and ocean currents (vector, units: \(m\ {s}^{-1}\)). Green dots in c, d represent the anomalous vertical ocean current and horizontal divergence that exceed the 95% confidence level.

Full size image

A question related to the second leading contributor (\(-{w}^{{\prime} }\frac{\partial \bar{T}}{\partial z}\)) for CE2L is what drives the anomalous vertical upwelling. Our diagnosis shows that the anomalous vertical velocity pattern is consistent with the divergence of mixed-layer horizontal currents (Fig. 4c, d). A further diagnosis shows that the divergence of the horizontal ocean currents is predominately driven by the divergence of the Ekman currents (Fig. 4b) while the divergence of the geostrophic currents is negligibly small. A further analysis demonstrates that the meridional component of the Ekman divergence contributes 60%, while the zonal component contributes 40%.

To sum up, the marked easterly anomalies in the EEP excited by the CP El Niño play an important role in triggering the CE2L onset. The easterly anomalies, on one hand, strengthen the climatological mean wind to increase the surface evaporation and initiate a cooling in EP, and, on the other hand, induce divergent Ekman current and upwelling anomalies, initiating a cold SSTA in EP during JF(0).

For SE2L, because it is evolved from a preceding super El Niño, initial cooling happens relatively late [MJ(0)] compared to the other two types (Fig. 1d; Fig. 3f–i). According to our budget analysis, the two leading contributors to SE2L onset at that period are the thermocline feedback and anomalous downward solar radiation (Fig. 3j). Physically, in response to a super El Niño forcing in DJF(0), strong upwelling Rossby waves are generated off the equator, which, after westward propagation and reflection at the western boundary, become upwelling Kelvin waves propagating eastward along the equator48, resulting in a strong discharge of ocean thermocline depth at the equator in the subsequent seasons (Supplementary Fig. 4). The colder subsurface water is transported upward by climatological mean upwelling, promoting a surface cooling firstly in CP and later in EP (Fig. 1d; Fig. 5a). Therefore, the exceptionally strong thermocline discharge associated with the preceding super El Niño is fundamental in generating this type onset. The contrast in the strength of the discharged thermocline depth anomaly between SE2L and CE2L is clearly evident, as shown in Supplementary Fig. 4.

Fig. 5: Anomalous oceanic and atmospheric circulations during SE2L onset.
figure 5

a Anomalous oceanic temperature (shaded, units: °C) and climatological subtropical cell (vector, units: \(m\ {s}^{-1}\)), b anomalous vertical overturning circulation (vector, units: \(m\) \({s}^{-1}\) for \(v\) and 10−2 Pa s−1 for ω), vertical wind (shaded, units: 10−2 Pa s−1), SSTA (red curve, units: °C) and precipitation (green curve, units: mm day-1) averaged over 175°–125°W during SE2L onset phase (MJ[0]). Green dots in a, b represent the anomalous ocean temperature and vertical wind velocity that exceed the 95% confidence level.

Full size image

An additional contributor to SE2L onset is anomalous solar radiation. As seen from Fig. 1d, the average SSTA during MJ(0) is still positive. This promotes a local vertical overturning circulation over 175°W–125°W, with anomalous ascending motion near the equator (10°S–5°N) and anomalous descending around 10°N (Fig. 5b). The meridional distribution of the anomalous vertical motion is consistent with the distribution of the SST and precipitation anomalies. The ascending motion at the equator leads to the increase of cloud cover and thus less downward solar radiation into the surface, which further cools the ocean surface.

Discussion

In this work, the diversified onset characteristics of La Niña were investigated using K-means cluster analysis. Based on the SSTA evolution patterns at the equator, the La Niña onset is classified into three types, N2L (neutral year to La Niña), CE2L (CP El Niño to La Niña) and SE2L (super El Niño to La Niña). The N2L onset can be traced back to the preceding autumn, whereas the CE2L and SE2L onsets start in the concurrent winter and spring.

The most pronounced precursory signal prior to the N2L onset is positive SST and precipitation anomalies in TNA during SO(−1). Physically, the positive SSTA in TNA may induce easterly anomalies in WCP through an atmospheric Kelvin wave response44. The easterly anomalies, on one hand, induce upwelling oceanic Kelvin waves, shoaling the thermocline at the equator, and on the other hand, induce downwelling oceanic Rossby waves off the equator due to the anticyclonic wind stress curl, deepening the thermocline off the equator. Such a meridional thermocline distribution generates a westward geostrophic current, leading to a negative SSTA in CEP through anomalous zonal advection.

For CE2L onset, the preceding CP El Niño, with a maximum SSTA center in CP, holds a key in generating low-level easterly anomalies in EP. The easterly anomalies in EP, on one hand, enhance the background mean easterly and thus the surface latent heat flux, and on the other hand, induce the divergence of the surface ocean currents and thus anomalous upwelling. Both processes lead to the initiation of a cooling in EP.

During CE2L onset phase, the cooling tendency is evident across the entire equatorial Pacific during JF(0) despite the stronger magnitude in the EP (Fig. 1c). To understand the cooling in the CP (~160°E–150°W) where westerly wind anomalies dominate (Fig. 3a), we conducted a mixed-layer heat budget analysis. The results show that the primary contributor is the reduced net surface solar radiation (Supplementary Fig. 5). The positive SSTA and precipitation in the equatorial CP in JF(0) induced ascending motion, which leads to more cloud cover, reducing the net surface solar radiation, thus cooling the SST.

For SE2L, the preceding super El Niño and associated strong discharge of the ocean heat content play a key role. Anomalous cyclonic shear of the wind stress anomaly associated with a preceding super El Niño excites strong upwelling Rossby waves off the equator, and after their westward propagation and reflection in the western boundary, the Rossby waves transform into upwelling equatorial Kelvin waves and propagate eastward along the equator, leading to a strong discharge of the ocean heat content at the equator. The strong discharge induced cooling first appears in CP, and then expands to EP. Such a feature is in a great contrast to CE2L in which the initial cooling starts in the EP.

The dominant processes mentioned above for each onset type were confirmed by the mixed-layer heat budget analysis. It is worth mentioning that in the case selection we intentionally excluded the 1910 and 2017 La Niña events, simply because it is not regarded as a first-year La Niña. In the preceding year 1909, a La Niña had already occurred in the equatorial Pacific. Although the year 2016 has not reached the threshold of La Niña, it was very close to −0.5 °C. The SSTA evolution patterns in 2017 were quite similar to the second year of multi-year La Niña events.

Given the large uncertainties and known biases in ocean reanalyses and surface heat flux, the diagnosed heat budget should be interpreted with appropriate caution64,65. These issues are particularly evident in products such as Climate Forecast System Reanalysis (CFSR)64, and some variables, especially surface heat flux and precipitation, remain unreliable65. Although such data limitations may affect the quantitative aspects of the heat budget analysis, the spatial and temporal anomaly patterns are physically consistent and provide reliable insights into the dominant processes contributing to La Niña onset.

During the onset periods of three types of La Nina events, the residual term (R) exhibits substantial variability: −0.06 ± 0.53 °C \({{mon}}^{-1}\) for N2L, −0.11 ± 0.63 °C \({{mon}}^{-1}\) for CE2L and −0.33 ± 0.29 °C mon−1 for SE2L. These residuals reflect not only the undiagnosed processes—such as sub-grid-scale ocean mixing, vertical diffusion, and high-frequency advective rectification9,66,67—but also the uncertainties inherent in the diagnosed terms themselves, which may be amplified by the aforementioned data quality issues. Their magnitude highlights the limited accuracy of the heat budget closure and emphasizes the need for caution when interpreting the individual contributions. Future research would benefit from multi-product ensemble approaches and improved observational constraints to better quantify these uncertainties.

One important question related to the N2L onset is what caused the positive SSTA over TNA in the preceding autumn. One may hypothesize that the TNA warming is the lag response to the El Niño-like warming prior to Aug(−1) (Fig. 1b). However, our analysis showed no clear evidence of preceding El Niño–like signals (Supplementary Fig. 6), suggesting that the TNA warming in SON(–1) is unlikely to be a response to a prior ENSO event. In the absence of evident ENSO precursors, the role of SST variability in other regions, such as the subtropical Pacific and high-latitude regions, warrants further consideration in N2L onset. While our results emphasize the role of TNA-induced remote forcing (i.e., Kelvin wave responses) on N2L onset, we do not exclude the potential effect of Rossby wave responses and the stochastic atmospheric variability in the subtropical eastern North Pacific, which may also contribute to equatorial wind anomalies. The future work is needed to quantify their relative importance.

An additional relevant issue is the origin of the westerly wind anomaly over the west IO (Fig. 2a–e). The easterly anomalies induced by the Kelvin wave response could weaken the climatological westerlies and reduce the surface latent heat flux over the east IO, enhancing the local SST and precipitation. The associated atmospheric heating may in turn induce westerly anomalies over the west IO. As the anomalous SST, surface latent heat flux and precipitation centers shift gradually eastward towards the eastern Maritime Continent, the atmospheric heating may further stimulate the westerly anomalies over the west IO. Furthermore, the orographic lifting effect over the Maritime Continent may directly enhance precipitation, contributing to the westerly anomalies over the west IO44.

A further open question is why N2L events have not occurred since 1970s. We speculate that it may be attributed to the increased frequency of CP and super eastern Pacific (SEP) El Niño events after 1970s. According to Pan et al.68, CP and SEP El Niño tend to initiate in WCP. The higher mean SST and increased mean moisture over the WCP after 1970s have led to a more unstable air-sea coupled system, facilitating the development of local SST perturbation. The climatological mean state after 1970s favors the development of CP and SEP El Niño, which in turn promote more CE2L and SE2L events, potentially reducing the likelihood of N2L onset.

For CE2L and SE2L, besides the preceding tropical Pacific SSTA patterns, there are significant SSTA signals in other basins such as tropical Atlantic and IO during DJF(0) ~ MAM(0) (Fig. 3). To what extent do the cross-basin interactions play in contributing to the La Niña onset types? All these questions require further observational, modeling and theoretical investigations.

Methods

Observational datasets

The monthly mean SST field used for this study is derived from the ensemble mean of the Extended Reconstructed SST analyses version 5 (ERSSTv5)69 and the Hadley Centre Sea Ice and Sea Surface Temperature reanalysis version 1.1 (HadISSTv1.1)70 data products, covering the period from 1910 to 2020. ERSSTv5 has a 2° × 2° horizontal resolution, while HadISSTv1.1 has a 1° × 1° horizontal resolution.

Monthly atmospheric wind fields are derived from the following reanalysis products: NOAA Twentieth Century Reanalysis version 2c (20CRv2c)71 (1910–1978), the National Centers for Environmental Prediction/Department of Energy (NCEP/DOE) Reanalysis II (NCEP2)72 (1979–2020), ECMWF reanalysis of the 20th Century (ERA20C)73 (1910-1957), ECMWF Reanalysis ERA40 (1958–2001)74, ECMWF reanalysis Interim (ERA-Interim)75 (2002–2018), and ECMWF reanalysis 5 (ERA5)76 (2002–2020).

To ensure temporal consistency, the mean state of ERA-Interim and ERA5 datasets is calibrated to ERA40 based on their overlapping period from 1979 to 2001. In addition, to minimize potential discontinuities and uncertainties associated with individual reanalysis products, the ensemble mean of the NOAA and ECMWF datasets is applied. These two centers were chosen because they provide century-long reanalysis products (i.e., 20CRv2c and ERA20C), allowing for a more coherent and consistent long-term reconstruction.

The precipitation datasets are the ensemble mean of the following datasets: 20CRv2c (1910–2014), ERA20C (1910–2010), ERA-Interim (1979–2018), NCEP2 (1979–2020), NCEP/National Center for Atmospheric Research (NCAR) Reanalysis I (NCEP1)77 (1948–2020), and the Global Precipitation Climatology Project Version 2.3 Combined Precipitation Dataset (GPCPv2.3)78 (1979–2020). Among them, GPCP is an observationally based dataset after 1979. For earlier periods lacking reliable observations, the long-term reanalysis products were adopted to ensure temporal continuity. The ensemble mean was applied to reduce uncertainties associated with individual datasets and to provide a more robust estimate of large-scale precipitation variability.

Monthly mixed-layer ocean temperature and three-dimensional ocean currents are derived from the ensemble mean of the following oceanic reanalysis datasets: the Simple Ocean Data Assimilation version 2.2.4 (SODAv2.2.4)79 at a 0.5° × 0.5° horizontal resolution and 40 vertical levels, and the NCEP Global Ocean Data Assimilation System (GODAS)80 at a 1° zonal resolution, \(\frac{1}{3}\)° meridional resolution, and 40 vertical levels (with a 10-m resolution in the upper 105 m). The continuity equation was applied to calculate the vertical ocean currents. SODAv2.2.4 covers the period from 1910 to 2008, while GODAS spans from 2009 to 2020. Oceanic thermocline depth, which is defined as the depth of 20 °C isotherm, is obtained from both SODAv2.2.4 and GODAS. The climatological mean of GODAS is calibrated to SODAv2.2.4 based on their overlapping period from 1980 to 2008.

The surface heat flux fields are also taken from the ensemble mean of the following six datasets: 20CRv2c (1910–2014), NCEP2 (1979–2020), ERA20C (1910–2010), ERA40 (1958–2001), ERA-Interim (1979–2018) and Woods Hole Oceanographic Institute (WHOI) Objectively Analyzed Air-Sea Fluxes (OAFlux)81 (1984–2009). OAFlux is included as an objectively analyzed observational dataset, while the other reanalysis datasets are used to extend the coverage over the full period from 1910 to 2020. Although reanalysis-derived heat fluxes are not directly constrained by observations, they provide internally consistent long-term estimates that are widely used in large-scale climate studies82. All the above datasets are interpolated into 1° × 1° horizontal resolution, and the merged datasets cover the period from 1910 to 2020.

Definition of El Niño and La Niña

In this study, La Niña (El Niño) events are defined when the ONI [defined as the monthly mean SSTA averaged over the Niño3.4 region (5°S–5°N, 170°–120°W) averaged for October-November-December-January-February (ONDJF)] exceeds −0.5 °C (0.5 °C)82. As shown in Fig. 1a, there are 33 El Niño events and 30 La Niña events over the 110-year period.

K-means cluster analysis

In this study, we applied a K-means cluster analysis82,83 to examine the evolution of equatorial (5°S–5°N) SSTAs for 17 first-year La Niña cases (i.e., blue bars in Fig. 1a). The K-means cluster analysis employs squared Euclidean distance to measure the “similarity” between each cluster member and its corresponding cluster centroid. The silhouette score was used to evaluate the performance of cluster analysis. It measures how similar a point is to its own cluster compared to other clusters, with values ranging from −1 to +1. A high score indicates a better-defined cluster. Three clusters are selected to represent different evolution patterns based on statistical significance and stability.

Ocean mixed-layer heat budget analysis

A mixed-layer heat budget analysis8,9,11,84,85,86 was applied in this work:

$$\frac{\partial {T}^{{\prime} }}{\partial t}=-\left(\bar{u}\frac{\partial {T}^{{\prime} }}{\partial x}{+u}^{{\prime} }\frac{\partial \bar{T}}{\partial x}{+u}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial x}\right)-\left(\bar{v}\frac{\partial {T}^{{\prime} }}{\partial y}{+v}^{{\prime} }\frac{\partial \bar{T}}{\partial y}{+v}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial y}\right)-\left(\bar{w}\frac{\partial {T}^{{\prime} }}{\partial z}{+w}^{{\prime} }\frac{\partial \bar{T}}{\partial z}{+w}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial z}\right)+\frac{{Q}_{n{\rm{e}}t}^{{\prime} }}{\rho {CpH}}+R$$
(1)

where \(T\) denotes the mixed-layer temperature; \(u\), \(v\), and \(w\) represent three-dimensional mixed-layer current velocity. A prime denotes the anomaly field and a bar represents the climatological mean annual cycle field. \({Q}_{{net}}^{{\prime} }\) denotes the net surface heat flux anomaly that includes anomalous net longwave radiation, net shortwave radiation, latent heat flux and sensible heat flux; \(\rho =1015\ {{kg\ m}}^{-3}\) denotes the density of water; \({Cp}=4000\ J\ {{kg}}^{-1}\ {K}^{-1}\) is the specific heat of water. R is a residual term representing processes other than heat fluxes and temperature advection84, which may be related to sub-grid ocean mixing process, diffusion, high-frequency advective rectification processes9,66,67. As seen from Figs. 2f, 3e, j, the mixed-layer heat budget analysis is not perfectly closed, but the cooling tendencies during the La Niña onset phases are well captured, which can be seen from the summation of all dynamic and thermodynamic terms. Therefore, the budget analyses identify key physical processes relevant to the initial cooling during the La Niña onset phases. The actual value of the residual term R may be inferred from the difference of the first two terms in the figures (i.e., the tendency term and the sum term). In this study, the climatological mixed-layer depth (\(H\)), defined as the depth at which the temperature is 0.5 °C lower than surface temperature, is used. Although the mixed-layer depth varies seasonally and spatially87,88, its climatological mean shows limited variation along the equatorial Pacific. Therefore, for simplicity, constant values of 50 m and 30 m are adopted for the CP and EP, respectively.

Ocean geostrophic and Ekman currents

The anomalous ocean currents can be separated into thermocline-induced geostrophic currents and wind-induced Ekman currents11. Their formula are as follows:

$${u}^{{\prime} }={{u}_{g}}^{{\prime} }+{{u}_{e}}^{{\prime} }$$
(2)
$${{u}_{g}}^{{\prime} }=-\frac{{g}^{{\prime} }}{\beta }\frac{{\partial }^{2}{h}^{{\prime} }}{{\partial y}^{2}}$$
(3)
$${{u}_{e}}^{{\prime} }=\frac{1}{\rho H}\frac{{r}_{s}{{\tau }_{x}}^{{\prime} }+\beta y{{\tau }_{y}}^{{\prime} }}{{{r}_{s}}^{2}+{(\beta y)}^{2}}$$
(4)
$${v}^{{\prime} }={{v}_{g}}^{{\prime} }+{{v}_{e}}^{{\prime} }$$
(5)
$${{v}_{g}}^{{\prime} }=\frac{{g}^{{\prime} }}{\beta }\frac{{\partial }^{2}{h}^{{\prime} }}{\partial x\partial y}$$
(6)
$${{v}_{e}}^{{\prime} }=\frac{1}{\rho H}\frac{{r}_{s}{{\tau }_{y}}^{{\prime} }-\beta y{{\tau }_{x}}^{{\prime} }}{{{r}_{s}}^{2}+{(\beta y)}^{2}}$$
(7)

where \({u}^{{\prime} }\), \({{u}_{g}}^{{\prime} }\), \({{u}_{e}}^{{\prime} }\), \({v}^{{\prime} }\), \({{v}_{g}}^{{\prime} }\), \({{v}_{e}}^{{\prime} }\) represent the anomalous zonal current, zonal geostrophic current, zonal Ekman current, meridional current, meridional geostrophic current and meridional Ekman current. \({g}^{{\prime} }=0.026\ {{m\ s}}^{-2}\) is the reduced gravity. \(\beta\) is the planetary vorticity gradient. \({h}^{{\prime} }\) is the thermocline depth anomaly. \({r}_{s}=0.5\ {{day}}^{-1}\) is a dissipation coefficient. \({{\tau }_{x}}^{{\prime} }\) and \({{\tau }_{y}}^{{\prime} }\) are anomalous zonal and meridional wind stress, respectively.

Linearization of Latent heat flux

The latent heat flux can be decomposed into three components, i.e., the latent heat flux determined by the anomalous wind speed, by the anomalous specific humidity and by the combined effect of both wind speed and specific humidity anomalies3,63.

$${{lh}}^{{\prime} }=-{\rho }_{a}{C}_{E}{L}_{v}{{|U|}}^{{\prime} }\overline{\left({q}_{s}-{q}_{a}\right)}-{\rho }_{a}{C}_{E}{L}_{v}\bar{U}{\left({q}_{s}-{q}_{a}\right)}^{{\prime} }-{\rho }_{a}{C}_{E}{L}_{v}{{|U|}}^{{\prime} }{\left({q}_{s}-{q}_{a}\right)}^{{\prime} }$$
(8)

where |U| is the total wind speed at 1000 hPa. \({q}_{s}\) and \({q}_{a}\) represent the saturated specific humidity at the sea surface and air specific humidity at 1000 hPa. \({\rho }_{a}=1.225\,{kg}\ {m}^{-3}\) is the air density, and \({c}_{E}=1.5\times {10}^{-3}\) denotes the moisture exchange coefficient. \({L}_{v}=2.5\times {10}^{6}\,J\ {{kg}}^{-1}\) is the coefficient of latent heat of evaporation. \({q}_{s}=\frac{0.622{e}_{s}}{{slp}-0.378{e}_{s}}\), where \({e}_{s}=6.112{e}^{\frac{17.67{sst}}{{sst}+243.5}}\), \({sst}\) and \({slp}\) are the sea surface temperature and sea level pressure, respectively.