Understanding the recent increase in multiyear La Niñas

Since the start of the twenty-first century, five multiyear La Niña events (1998–2000, 2007–2008, 2010–2011, 2016–2017 and 2020–2022) have occurred, an unusually high clustering given that only ten multiyear La Niñas took place over the last century. Understanding why double- and triple-year La Niñas have become prominent recently, and whether they will become more common, is critical due to their outsized and persistent global impacts (for example, droughts, floods, heatwaves and hurricanes). Prior views emphasized persistence driven by massive upper-ocean heat discharge following super El Niño events, but several recent multiyear La Niñas did not follow strong El Niño, suggesting alternative mechanisms. At the same time, projections of future La Niña behavior remain uncertain due to model biases and ENSO nonlinearity. This study examines historical changes in La Niña properties, identifies mechanisms distinguishing single-year from multiyear events, and evaluates how western Pacific warming modulates their occurrence, with implications for prediction and future risk.

Previous hypotheses attribute multiyear La Niña persistence to processes including: large upper-ocean heat content discharge after strong or super El Niño; weakened recharge due to anomalous Ekman heat transport during decay; nonlinear delayed thermocline feedback; thermodynamic processes; and influences from mid-latitude variability or interbasin interactions. ENSO asymmetry in amplitude and duration is well documented, and El Niño flavors (eastern vs central Pacific) are sensitive to mean state changes. However, climate models (CMIP5/CMIP6) show substantial mean-state biases (e.g., cold tongue), inconsistent tropical Pacific trends, and difficulty reproducing ENSO asymmetry, La Niña persistence, and multiyear frequencies, limiting confidence in projections. Emerging evidence suggests that changes in the west-to-central Pacific SST gradient and associated trade wind/thermocline adjustments may modulate both extreme El Niño and multiyear La Niña occurrences by altering coupled feedback strengths and the location of El Niño onset.

Observational analysis: The study uses 1920–2022 datasets with linear trends removed. SST comes from HadISST1 and ERSSTv5 (averaged). Atmospheric fields are from ECMWF reanalyses (ERA-20C, ERA-40, and ERA5) with mean-state calibration over overlap periods to ensure temporal consistency. Land precipitation is from GPCC. Surface heat fluxes merge NOAA-CIRES 20CR (1921–2012) and NCEP/DOE R2 (1979–2022), calibrated over overlaps. Ocean reanalysis is primarily SODA v2.2.4 (1871–2008) extended with GODAS (2009–2022) and mean-state calibrated over 1980–2008. Event identification and metrics: La Niña years are defined when the ONDJF mean Niño 3.4 index (5°S–5°N, 120–170°W) is below −0.5 °C. Multiyear La Niña events last at least two consecutive La Niña years. Severity is quantified by accumulative intensity, the sum of ONDJF Niño 3.4 anomalies over all years of an event. The La Niña onset rate is defined as the cold tongue (5°S–5°N, 180–90°W) SST tendency from March (year 0) to October (year 0). Statistical significance is assessed using two-tailed Student’s t-tests. Event classification: Post-1970 multiyear La Niñas are categorized as: (1) SE2ML (super El Niño-to-multiyear La Niña) following super El Niño (1982–83, 1997–98, 2015–16), and (2) CPE2ML (central Pacific El Niño-to-multiyear La Niña) following CP-type El Niño whose initiation occurs in the western Pacific and warming extends eastward to peak with significant central Pacific anomalies (e.g., 1969, 1972, 2006, 2009, 2018–19). Mixed-layer heat budget: A mixed-layer temperature (MLT) tendency analysis decomposes contributions into zonal advective feedback (zonal heat advection by anomalous currents), thermocline feedback (vertical advection by mean upwelling associated with thermocline depth anomalies), upwelling feedback (vertical advection by anomalous upwelling), meridional heat advection (by mean currents), and surface heat flux residual. Analyses are performed over the equatorial central Pacific (5°S–5°N, 180–120°W) and eastern Pacific (120–80°W) for composites spanning the antecedent El Niño (year −1), La Niña onset (year 0), and persistence (year +1). Mean-state diagnostics: Changes in the west-to-central equatorial Pacific SST gradient (WP: 5°S–5°N, 130–170°E; CP: 5°S–5°N, 150–170°W), associated low-level easterlies (150°E–150°W), and thermocline depth are evaluated across historical periods. Model simulations: Ten CESM2 Large Ensemble (CMIP6 historical forcing, 1901–2013) members are analyzed to test observed relationships. The model’s representation of ENSO is evaluated; 288 La Niña events are identified. Relationships between WP background SST and the frequency of El Niño initiation in the west-central Pacific, between La Niña onset rate and accumulative intensity, and between 30-year mean WP SST and multiyear La Niña years are computed across early (1901–1930) and late (1981–2010) periods.

• Historical acceleration: Of 20 La Niña events (1921–2022), ten were multiyear. All five events during 1921–1945 were single-year, whereas five of six during 1998–2022 were multiyear; the increase in multiyear La Niña years is statistically significant (P = 0.01).
• Increased severity via persistence: Accumulative intensity (sum of ONDJF Niño 3.4 anomalies over an event) increased by 71%, from −1.17 °C (1920–1969, nine events) to −2.0 °C (1970–2022, ten events) with similar average yearly intensities (~−1.0 °C), indicating enhanced persistence. Four extreme events (≤ −2.3 °C): 1973–1975 (−3.9 °C), 1998–2000 (−3.5 °C), 2010–2011 (−2.4 °C), 2020–2022 (−2.8 °C).
• Two multiyear types and distinct mechanisms: Post-1970 multiyear La Niñas split into SE2ML (after super El Niño) and CPE2ML (after CP El Niño). SE2ML onset and persistence are dominated by thermocline feedback with notable meridional advection; CPE2ML onset and persistence in the central Pacific are dominated by zonal advective feedback, with thermocline feedback aiding persistence and upwelling feedback initiating far eastern Pacific cooling.
• Onset rate as a predictor of persistence: Multiyear La Niñas exhibit a substantially stronger first-year cooling tendency than single-year events. The March–October onset rate significantly correlates with accumulative intensity (observations: r = 0.64, P < 0.01) and with accumulative thermocline depth anomalies from April (0) to March (+1), indicating larger heat content discharge.
• Western Pacific (WP) relative warming link: The WP warmed relative to the CP, increasing the zonal SST gradient by ~21.2% (WP−CP: 1.37 °C in 1921–1940 to 1.66 °C in 2001–2020). Associated easterlies strengthened near the dateline and the thermocline shoaled by ~24 m (142 m to 118 m). This enhances zonal advective, thermocline, and upwelling-related feedbacks that favor (i) more SE and CPE antecedent events initiating in the west and (ii) faster transition to La Niña with greater persistence.
• Model support (CESM2 LE): Across 288 simulated La Niñas (1901–2013), WP warming increases the frequency of El Niño initiation in the west-central Pacific (r = 0.48, P < 0.01) and faster La Niña onset relates to larger accumulative intensity (r = 0.64, P < 0.01). The 30-year mean WP SST increased by ~0.24 °C on average (ensemble mean) from early to late 20th century. During 1901–1930, single-year (56) and multiyear (52) counts were comparable; during 1981–2010, multiyear events (91) nearly tripled single-year events (32) (P < 0.01). The number of multiyear La Niña years increases with mean WP SST (r = 0.52, P < 0.05).
• Double vs triple La Niñas: Both evolve similarly in first two years; triple events have larger onset rates and stronger CP–EP cooling and WP easterlies near the end of year two, with stronger zonal advective and thermocline feedbacks supporting persistence into a third year.

The findings address why multiyear La Niñas have recently become more frequent by identifying a strengthened coupling between the ocean and atmosphere tied to western Pacific relative warming. This mean-state change enhances zonal advective and thermocline feedbacks, increases the likelihood of El Niño initiation in the west-central Pacific (boosting both SE and CPE precursors), and promotes rapid transitions to La Niña with larger heat discharge and longer persistence. The onset rate emerges as a practical predictor of accumulative intensity and the likelihood that a post-CPE La Niña will persist into a multiyear event, thus improving early warning for severe impacts. Distinguishing SE2ML (thermocline-dominated) from CPE2ML (zonal advection-dominated) clarifies why different antecedent El Niño flavors can both lead to multiyear La Niñas via different feedback pathways. CESM2 ensemble results corroborate observed relationships, lending confidence despite limited observational samples. These insights imply that if WP relative warming continues, multiyear La Niñas—and their compound socioeconomic hazards—may become more common, providing emergent constraints for evaluating and improving climate models and projections.

Multiyear La Niñas have accelerated over the past century, with most occurring after 1970 and following either super El Niño or central Pacific El Niño events. Their defining characteristic is a rapid onset rate that presages persistence and greater accumulative intensity. Physical mechanisms differ by type: thermocline feedback dominates SE2ML, while zonal advective feedback dominates CPE2ML. Historical observations and CESM2 ensembles link these behaviors to strengthened west-to-central Pacific SST gradients and associated wind–thermocline adjustments under western Pacific relative warming. These results provide a predictive metric (onset rate) for extreme La Niña risk and suggest that continued WP warming will likely increase multiyear La Niña frequency and impacts. Future work should refine mechanistic understanding across basins and mid-latitudes, reduce model mean-state and ENSO asymmetry biases, and quantify anthropogenic contributions to WP relative warming to improve projections.

• Limited sample size: Only 20 La Niña events (ten multiyear) in 1920–2022 constrain statistical power and event diversity.
• Model biases: Many CMIP6 models struggle with cold-tongue bias, ENSO asymmetry, La Niña persistence, and reproducing moderate SE/CPE events originating in the west, limiting projection fidelity.
• Attribution uncertainty: The relative contributions of anthropogenic forcing versus internal variability to western Pacific relative warming and Pacific mean-state changes remain uncertain.
• Incomplete mechanisms: While onset rate is a strong predictor, not all CP El Niños lead to multiyear La Niña; other influences (mid-latitude variability, Indian/Atlantic interbasin interactions, annual cycle coupling, eastern Pacific mean-state changes) require further investigation.
• Dataset consistency: Despite calibration, merging multiple reanalyses and products may introduce residual uncertainties in mean-state and feedback diagnostics.