Triple‐Dip La Niña in 2020–23: North Pacific Atmosphere Drives 2nd Year La Niña

La Niña persisted from 2020 to 2023, but its mechanisms are still unclear. In this study, atmosphere and ocean reanalysis and 100‐member initialized forecasts using a state‐of‐the‐art climate model were analyzed to identify factors contributing to the persistence of the first‐ to second‐year La Niña during 2020–2022. We found that North Pacific high pressure anomalies in the winter of 2020/2021 forced a negative phase of the Pacific meridional mode through the following spring, forming the broader structure of La Niña. The resultant broader La Niña pattern slowed down the recharge‐discharge process by Ekman transport, persisting La Niña. Ensemble forecast sensitivity analysis revealed that the meridional extent of La Niña explains its forecast spread, reaffirming the importance of La Niña spatial pattern. Advancing predictive understanding of 2020–2022 multi‐year La Niña can help to improve the extended seasonal forecast.

tends to be accomplished with the meridionally broad structure of ENSO, especially in ENSO decay phase (Iwakiri & Watanabe, 2022).Broad spatial pattern of ENSO works that ENSO-induced equatorial zonal wind maintains a symmetric structure about the equator; namely, the southward shift of zonal wind (Lengaigne et al., 2006) does not occur, bringing longer persistence.The North Pacific Meridional Mode (NPMM; Chiang & Vimont, 2004) and/or Inter-decadal Pacific Oscillation (IPO; Power et al., 2021) are occasionally concurrent with ENSO in phase, and then the apparent meridional scale becomes broader.Several previous studies reported the links between multi-year ENSO and NPMM (Ding et al., 2022;Park et al., 2020;Yu & Fang, 2018).Also, NPMM is sometimes forced by North Pacific Oscillation via seasonal footprint mechanisms (Vimont et al., 2003), suggesting that multi-year ENSO can be forced by the North Pacific atmospheric variability.
Triple-dip La Niña in 2020-23 is a good case study for an in-depth understanding of multi-year ENSO.Here, we analyzed atmosphere and ocean reanalysis data and then performed large-ensemble extended seasonal hindcasts to investigate the cause of the persistent La Niña condition.So far, most multi-year El Niño/La Niña studies are based on statistical analyses to observations and/or a long climate simulation driven by idealized forcing.The initialized extended seasonal hindcast experiments conducted in this study allow us to simulate events under realistic conditions which can be directly compared to the observations.Thus, we utilized a fully coupled forecast system as a tool for understanding the mechanisms and predictability of multi-year La Niña.
Data sets and model we used in this study are described in Section 2. In Section 3, the observed properties of the triple-dip La Niña in 2020-2023 are summarized.Section 4 described the results of forecast experiments to understand persistency.A summary and discussion are presented in Section 5.

Data and ENSO Event
We used monthly mean ocean reanalysis data of MOVE/MRI.COM-G3 (Toyoda et al., 2016;Usui et al., 2015) and atmospheric reanalysis data of JRA-3Q (Kobayashi et al., 2021).The monthly climatology defined from January 1991 to December 2020 was subtracted in advance.La Niña event is defined by October-February (ONDJF)-mean Niño 3.4 index, with a threshold of area-averaged SST anomaly over the central-eastern equatorial Pacific (170°W-120°W, 5°S-5°N) below −0.5°C.

Model and 100-Member Extended Seasonal Hindcasts
We utilized the Model for Interdisciplinary Research on Climate version 6 (MIROC6; Tatebe et al., 2019), which took part in the sixth Coupled Model Intercomparison Project (CMIP6; Eyring et al., 2016).The atmospheric component of MIROC6 has 81 vertical levels up to 0.004 hPa and a horizontal resolution of a T85 spectral truncation (approximately 1.4° gird).The ocean component has approximately 1° horizontal resolution in a tripolar coordinate system, and the latitudinal grid points are doubled near the equator to resolve equatorial ocean waves.The prediction system was constructed following the Decadal Climate Prediction Project of CMIP6 (Boer et al., 2016;Kataoka et al., 2020).The external forcing was set to CMIP6 historical experiment configuration up to 2014 and the Shared Socioeconomic Pathway 2-4.5 scenario later.MIROC6 is initialized by assimilating the observed monthly ocean temperature, salinity, and sea ice concentrations (Ishii & Kimoto, 2009;Ishii et al., 2006).Initialization is based on the simplified incremental analysis update (Bloom et al., 1996).The anomaly assimilation scheme is adopted to reduce the model drifts while sea ice concentration is full-field assimilation.The ensemble size of assimilation experiments increased to 50 members from 10 based on different initial states from historical experiments.The initial states of the atmosphere were taken from the JRA-55 (Kobayashi et al., 2015) instead of assimilation runs, which provide the realistic atmospheric variation.In addition, lagged averaged forecasting technique (Hoffman & Kalnay, 1983), using initial atmospheric states 1 day apart, was applied, and prepared 100 member initial states.
To focus on the persistence of the second-year La Niña in 2021/22, we conducted an 18-month long extended seasonal hindcast experiment starting from 1 November 2020.To remove a climate model drifts from predictions, monthly anomalies are defined by subtracting lead-time-dependent climatology during 1991 and 2020, calculated using hindcasts (Kataoka et al., 2020).

Observed Property of the Second-Year La Niña in 2021/2022
We first describe the phase transition of triple-dip La Niña in 2020-23.Figure 1 shows the phase space diagram, which helps us to determine whether this transition can be understood by the RO cycle.From December 2019 to January 2023, ENSO cycle was roughly on a track of clockwise rotation in the phase space.In the preceding year of the first-year La Niña in 2019, anomalous SST in the equatorial Pacific was relatively high and close to the El Niño state.Although some multi-year La Niña events are driven by extreme El Niño (Iwakiri & Watanabe, 2021b;Wu et al., 2019), second-year La Niña in 2020-22 is not associated with strong El Niño in the preceding year.
For understanding the persistence, the whole period from first-to second-year La Niña was divided into three terms (I to III in Figure 1b).In the term (I), the thermocline deepened, that is, recharge, and then the La Niña SST anomaly decayed.In the term (II), deepened thermocline neutralized and therefore the decay of La Niña paused.
In the term (III), remaining La Niña SST was re-intensified by the Bjerknes positive feedback after summer.These changes are consistent with the RO theory because ENSO SST evolution is preceded by changes in the thermocline depth.It is noted that inhibited La Niña decay in the term (II) was caused by the attenuated thermocline depth anomaly.Namely, the precondition of second-year La Niña was ready around summer.During the period from second-to third-year La Niña, negative SST persists even though mean thermocline depth deepened.
It means a strong decoupling between SST and thermocline depth, implying that a different mechanism works.We hereafter focus on the cause of second-year La Niña.
To identify the cause of thermocline depth change, we computed the upper ocean heat content (OHC) anomaly integrated over the equatorial Pacific (120°E−60°W, 5.5°S-5.5°N,surface to 500 m depth), named OHC eq , which is the alternative indicator of thermocline depth (Iwakiri & Watanabe, 2021a).We calculated OHC eq budget following equation.
where T is the ocean temperature.Ocean meridional current velocities are indicated v. ρ and C p are the seawater density and specific heat of the ocean, respectively.S is the cross-sectional area of the north/south boundaries.
The terms on the right-hand side represent the ocean heat transport across meridional boundaries, defining positive inward, named recharge rate.The recharge rate can be decomposed into geostrophic heat transport (GHT) and Ekman heat transport (EHT) contributions (Iwakiri & Watanabe, 2021a, 2022).Figure 2 shows the time evolution of Niño 3.4 index, OHC over the equatorial Pacific (named OHC eq , see methods), and its tendency for the period of first-to second-year La Niña.As shown for the thermocline depth (Figure 1), OHC eq decreased around the summer of 2021.OHC eq tendency is well explained by the meridional heat transport, which is recharge-discharge process.In addition, the anomalous heat transport is divided into geostrophic (GHT) and Ekman (EHT) components in Figure 2b.GHT is highly negatively correlated with ENSO SST, consistent with RO theory (Iwakiri & Watanabe, 2021a).It shows a positive tendency, geostrophic discharge, during multi-year La Niña.The attenuation of GHT recharge during early 2021 is caused by the weakened amplitude of La Niña; namely, the change in GHT is not the cause but the effect.On the other hand, EHT shows a negative tendency, implying that this term acts to slow down the ENSO phase reversal.The slowdown of the recharge process by EHT is the key for understanding the multi-year ENSO dynamics (Iwakiri & Watanabe, 2022).Therefore, we hypothesized that this EHT-induced discharge is plausibly forced by the meridionally broad scale of La Niña SST anomaly, especially in the northern part in the first half of 2021 (Figure S1 in Supporting Information S1).In fact, a negative phase of NPMM is observed in this period, contributing to the apparent ENSO latitudinal scale broader.

Predictive Understanding of the 2nd Year La Niña
To validate the hypothesis of the cause of La Niña persistence, we conducted 100-member extended seasonal prediction using MIROC6, initialized on 1 November 2020.MIROC6 is capable of reproducing higher-order properties of ENSO, such as phase asymmetry and occurrence frequency of multi-year ENSO, which tend to be underestimated in many climate models (Iwakiri & Watanabe, 2019, 2021b, 2022).Figure 3 shows the Hovmöller diagram of SST anomalies along the equator.The MIROC6 ensemble mean successfully predicted persistent equatorial Pacific cooling from first-to second-year La Niña during November 2020 to January 2022.While October-February (ONDJF)-mean Niño 3.4 index in 2021/22 was approximately −1.0℃ in the reanalysis, the predicted ensemble-mean anomaly was about half, indicating that the model underestimates the amplitude.This is partly due to a phase-locking bias in MIROC, the mature phase appearing one or 2 months earlier than observation.Also, our experiment did not include realistic biomass burning emissions in 2019/20 Australia.It may increase the La Niña amplitude and improve reproducibility (Fasullo et al., 2023).The large-ensemble hindcasts enable us to estimate the probability of La Niña with high confidence (Figure 3d).We found that 54% of members successfully predicted Niño 3.4 index in ONDJF(2021/22) below −0.5℃.Note that the ensembles are well spread around La Niña criterion, and therefore, we can identify factors that determine occurrence of second-year La Niña.We divided 100 members into two groups: 54 members that predicted second-year La Niña (called good member), and others that failed to predict (called poor member).
The difference in the anomalous Ekman transport that acts to counteract the recharge-discharge process between multi-and single-year ENSO mainly arises during the ENSO decay phase (Iwakiri & Watanabe, 2022).Therefore, we focus on the December-April (DJFMA)-mean in 2020/21 (Figure 4).Observations show, after the peak of first-year La Niña, negative SST anomaly that remains in the equatorial Pacific with a broad meridional structure (Figure 4a), and it is contributed by negative NPMM (Figure S1 in Supporting Information S1).Due presumably to this La Niña SST, positive sea level pressure (SLP) anomaly is found over the North Pacific, akin to the negative Pacific-North American (PNA; Wallace & Gutzler, 1981) pattern (Figure 4b).The ensemble mean of the good members reproduces well these anomaly patterns in SST and SLP with the lead time of 1-5 months (Figures 4c and 4d).A comparison between good and poor members could suggest the key for predicting secondyear La Niña (Figures 4e and 4f).Compared to the poor members, good members show a cooling signal in the northern side of the subtropical Pacific and a southward shift of the negative PNA pattern, which accompanies intensified subtropical easterlies in 10-30°N.The spatial coherence between the SST and wind anomalies in the subtropical North Pacific is consistent with the seasonal footprinting mechanism (Figure S2 in Supporting Information S1; Vimont et al., 2003), although the center slightly shifts westward from typical NPMM probably due to the North Pacific high shifted southward and westward.This westward shift decreases the wind speed and evaporation around the Philippine Sea, contributing to increased SST anomaly over there.The seasonal footprint mechanism generates and maintains negative NPMM.The combination of first-year La Niña and the negative phase of NPMM brings the latitudinally wider pattern of SST cooling in the tropical Pacific.The southward shift of the North Pacific high already appears in the peak season of first-year La Niña, while SST differences are relatively small, implying that the North Pacific atmosphere leads the subtropical cooling (Figure S3 in Supporting Information S1).The signals in anomalous SLP and 850 hPa wind are confined in the North Pacific, emphasizing the primary role of the North Pacific, although the SST signals in other basins might play a role as discussed later.The latitudinal position of the PNA pattern is highly affected by the atmospheric internal variability of the mid-latitude, such as a North Pacific oscillation (NPO; Di Lorenzo et al., 2008), because atmospheric initial conditions are basically identical across members.However, a moist linear baroclinic experiment shows that tropical SST also partly contributes to the position of the North Pacific high anomalies (Figure S4 in Supporting Information S1; Ding et al., 2022).
We quantify the sensitivity for ensemble prediction using the ensemble sensitivity analysis (Ancell & Hakim, 2007;Torn & Hakim, 2009).It evaluates the relationship between forecast spread and physical field that affects it.The ensemble forecast sensitivity was calculated as where J is the target variable, ONDJF(2021/22)-mean Niño 3.4 index (lead month approximate 12) and x i,τ is physical variable.The subscript i and τ are grid point and lead time, respectively.The left-hand side is forecast sensitivity derived from the regression.Large value of sensitivity implies large impacts on target variable.The results of the sensitivity analysis are shown in Figures 4g and 4h (signs are reversed for La Niña).We identified the negative high-sensitivity regions in the northern subtropical Pacific, indicating that the members which have cooler SST in this region tend to predict second-year La Niña.Namely, the broad meridional structure of firstyear La Niña is favorable for causing second-year La Niña to occur (Figure 4g).We also observe the positive high-sensitivity in SLP anomaly over the subtropical North Pacific, indicating that the members which have high pressure are effective to predict second-year La Niña (Figure 4h).These results are consistent with the comparison between good and poor members (Figures 4e and 4f).It supports the critical role of the negative NPMM driven by the North Pacific high in controlling the La Niña evolution.We also identified positive and negative high-sensitivity in SST and SLP around the tropical North Atlantic, suggesting an additional contribution to second-year La Niña (Ham et al., 2013;Hasan et al., 2022).
In the good members, OHC eq had a negative sign, implying shoaled thermocline; in contrast, the poor members had a positive sign in the summer of 2021 (Figure S5 in Supporting Information S1).After the peak of first-year La Niña, positive OHC eq tendency was significantly weak in the good members.The GHT-induced recharge was not significantly different between the good and poor members, but the EHT-induced discharge was significantly strong in good members, disrupting the recharge process.Therefore, Ekman transport brings recurrence La Niña in the hindcasts, consistent with reanalysis (Figure 2b).

Summary and Discussion
In summary, this study examined the mechanism for the transition from first-to second-year La Niña in 2021/22, which is a portion of the triple-dip event in 2020-23, using the atmosphere-ocean reanalysis and large-ensemble extended seasonal hindcast experiments.The SST transition from first-to second-year La Niña follows thermocline depth changes, consistent with the RO theory.However, the recharge process was disturbed by anomalous Ekman transport induced by the anomalous easterly -contributing to discharge-over the subtropical North Pacific.During the peak of first-year La Niña in 2020/21 winter, the North Pacific high pressure anomalies occasionally shift southward, which then excites the negative phase of NPMM in the following spring.The combination between first-year La Niña and negative NPMM brought a meridionally broad structure of La Niña.It maintains equatorially symmetric zonal wind, and thus the anomalous Ekman transport worked to discharge the OHC in the equatorial Pacific, resulting in the second peak of La Niña in 2021/22 winter.Note, the driving mechanism from the southward shifted North Pacific high to NPMM is a conventional seasonal footprint process, but NPMM to second-year La Niña is caused by the surface discharge process via the oceanic Ekman layer.These mechanisms were verified using the large-ensemble hindcast experiment initialized on 1 November 2020.The ensemble spread captures the occurrence of second-year La Niña, while ensemble members well spread with high confidence.A forecast sensitivity analysis based on 100 members revealed that the position of the North Pacific high pressure anomalies and the meridional structure of La Niña account for the ensemble spread, confirming the reanalysis results.The slightly different location of the North Pacific high is plausibly determined by internal variability of the extratropical atmosphere.However, tropical SST also partly contributes to the position of the North Pacific high anomalies.
A number of studies have proposed mechanisms for observed multi-year El Niño/La Niña, but they are not really verified using ENSO predictions.This study utilized a fully coupled initialized extended forecast as a tool for understanding the mechanisms of multi-year La Niña, half of the triple-dip event specifically, contributing to advance understanding of the ENSO irregularity.Predictive understanding is achieved to first-and second-year La Niña in 2020-22 through verifying and synthesizing the revealed mechanism: North Pacific atmospheric variability induced NPMM (Park et al., 2020;Yu & Fang, 2018), which widens the meridional scale of La Niña SST anomalies, and the resultant wind pattern acts to slow down the recharge process via EHT (Iwakiri & Watanabe, 2022).The precursors are also found on the tropical North Atlantic Ocean, but corresponding atmospheric signals are relatively weak.Atlantic Niño occurred in the summer of 2021 (Hasan et al., 2022).However, the difference in ocean condition arose before summer (Figures 1 and 2), and thus it may not be a primary cause that makes persistence, but the result (Tokinaga et al., 2019).Therefore, we conclude that predictability for the persistence from first-to second-year La Niña arises from the North Pacific.Based on the results, improving forecast accuracy of mid-latitude atmosphere and responses of subtropics and ENSO is expected to contribute to more skillful ENSO prediction.
Mechanism that explains the transition from second-to third-year La Niña is beyond the scope of this work.MIROC6 could not predict the third-year event 2 years in advance (Figure 3c).We identified that the transition from second-to third-year La Niña is characterized by a deepening of mean thermocline depth in 2022, suggesting that the RO-like process is not applicable.In addition, the meridional scale of La Niña is relatively narrow, especially on the northern side, in decay phase of second-year peak (Figure S6 in Supporting Information S1).Therefore, different persistence mechanisms worked between second-and third-year La Niña.Elaborating the full dynamics of triple-dip La Niña in 2020-23 needs further studies.MIROC6 prediction system.This work has been supported by MEXT program for the advanced studies of climate change projection (SENTAN) Grant JPMXD0722680395.TI is supported by a Grant-in-Aid from the JSPS Fellows (21J10287 and 23KJ2168).This report references MOVE/MRI.COM-G3 and JRA-3Q reanalysis data from the Japan Meteorological Agency.

•
The 100-member ensemble forecasts initialized in November 2020 by MIROC6 captured prolonged La Niña in 2021/2022 • The forecast sensitivity analysis to initial states of ensemble members revealed a critical role of the North Pacific high • North Pacific high in winter 2020/2021, causing negative North Pacific Meridional Mode spring, formed the spatially broader La Niña, resulting in long persistence Supporting Information: Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.(a) Phase space diagram of El Niño-Southern Oscillation during December 2019 and January 2023 in MOVE-G3 ocean reanalysis.The horizontal axis is Niño 3.4 index, and the vertical axis is the thermocline depth anomaly averaged over the equatorial Pacific (120°E−80°W, 5°S-5°N).The trajectory is smoothed with a 5-month running-mean.White markers denote December.(b) As in (a) but for the transition from first-to second-year La Niña (enlarged box area in (a)).Roman numerals and arrows indicate three characteristic terms (see text) and the direction of evolution.

Figure 2 .
Figure 2. (a) Time evolution of Niño 3.4 index (black) and OHC eq (blue) in reanalysis.(b) As in (a) but for OHC eq tendency (black), recharge rate (green), geostrophic heat transport (purple), and Ekman heat transport (orange).Niño 3.4 index and other time-series are smoothed with a three-and 5-month running-mean, respectively.The symbol (II) in (a) corresponds to the term indicated in Figure 1b.

Figure 3 .
Figure 3. (a) Hovmöller diagram of equatorial sea surface temperature anomaly (5°S-5°N mean) in reanalysis.(b) As in (a) but for ensemble mean of MIROC6's 100-member prediction initialized on 1 November 2020.Dots indicate anomalies statistically significant at the 99% confidence level.(c) Time evolution of Niño 3.4 index in reanalysis (black) and ensemble mean of MIROC6 prediction (blue).Blue shading represents ensemble spread (one standard deviation).(d) Probability density function of predicted Niño 3.4 index in ONDJF(2021/22) (corresponding to black box in b).The triangle marker is observed value.The shading and the corresponding fractional percentage indicate probability below −0.5°C of the Niño 3.4 index.

Figure 4 .
Figure 4. (a) Anomalous sea surface temperature (SST) and (b) anomalous sea level pressure (SLP) (shading) and 850 hPa wind (vector) in DJFMA(2020/21).(c), (d) As in (a) and (b), but for ensemble mean of good members in MIROC6's prediction.(e), (f) As in (c), (d), but for the difference between ensemble mean of good members and poor members.Ensemble forecast sensitivity of (g) SST anomaly and (h) SLP anomaly in DJFMA(2020/21) against ONDJF(2021/22)-mean Niño 3.4 index (Signs are reversed for La Niña).Dots and vectors are shown when shading and wind are statistically significant at the 95% confidence level, respectively.