Projected Increase in Fast-Growing and Slow-Dissipating El Niño Events in the 21st Century

Future changes in the seasonal evolution of El Niño - Southern Oscillation (ENSO) during the onset and decay phases have received little attention by the research community. This work investigates the projected changes in the spatio-temporal evolution of El Niño events in the 21 st Century (21C) using a large ensemble simulation of a couple general circulation model under anthropogenic forcing. Here we show that El Niño is projected to (1) initiate sooner in boreal spring, (2) to grow at a faster rate, (3) to persist longer over the eastern and far eastern Pacific, and (4) to have a broader impact on remote teleconnections. Significant changes in the tropical Pacific mean state, dominant feedback processes, and a projected increase in stochastic westerly wind burst forcing largely explain the fast growing and slow dissipating El Niño in the late 21C. Important implications of these findings are that the global climate impacts are projected to become more significant and persistent, owing to the extended persistence of El Niño.

The seasonal El Niño evolution is analyzed using a mixed layer heat budget, which includes the dominant feedback mechanisms for ENSO growth. Figure 2 depicts the composite evolutions of heat storage, three feedback terms (i.e., thermocline, zonal advective, and Ekman feedbacks), and a residual term in the same order as they appear in Eq. 1 (see Methods). Note that the largest differences between 20C and 21C are in the amplitude rather than spatiotemporal structure.
Specifically, it is clear that the thermocline, zonal advective, and Ekman feedbacks are greatly enhanced (points labeled A in Fig. 2) in the central equatorial Pacific during the developing phase in boreal summer and fall (Year 0), readily explaining the faster rise in heat content ( Fig. 2 green contour) and the faster growth of SSTAs during the growth phase (Fig. 1d). During the late decay phases in April-July (Year +1), the Ekman feedback is enhanced in the central and eastern equatorial Pacific between 150°W and 100°W (point labeled B in Fig. 2). A similar increase in the Ekman feedback is also found during the peak phase in boreal winter. Interestingly, the thermocline feedback is reduced in the central equatorial Pacific between 170°E and 140°W during the decay phase in boreal spring (point labeled C in Fig. 2). Overall, the residual term mostly constitutes a damping of the heat content anomaly.
To further investigate what dynamical processes are responsible for the projected changes in feedbacks and thus El Niño growth, we repeat the computations of the feedback terms in Eq. 1 (Methods) by either fixing the mean state (i.e., overbar) or the anomalies (i.e., primes) to that of the 20 th Century (see supporting information and Figs. S3 and S4 therein). The projected changes in the thermocline feedback are mostly explained by an increase in the vertical gradient of anomalous temperature ( ′ in Fig. S3 and Fig. S4d) rather than the projected reduction in mean upwelling ( ̅ in Fig. S3 and Fig. S4g). In contrast, the zonal advective feedback changes are dominated by an increase in the anomalous zonal current ( ′, Fig. S3 and Fig. S4e) rather than changes in the mean zonal temperature gradient ( ̅ , Fig. S3 and Fig. S4h). Whereas, the projected increase of the Ekman feedback can be explained by both contributions from anomalous upwelling ( ′, Fig. S3 and Fig. S4f) and an increase in the mean stratification ( ̅ , Fig. S3 and Fig. S4i It is important to investigate what mechanism could explain the tendency to have a shift towards stronger eastern Pacific El Niño events in the late 21C (Fig. 1). The delayed thermocline feedback is one of the major transitioners of strong El Niño events 33 . For example, the negative thermocline feedback is enhanced in the central Pacific west of 140°W during the decay phase ( Fig. 2). Analysis of the projected changes of each component of the thermocline feedback suggests that while the mean upwelling is weakened due to weaker trade winds, ′ is projected to increase (Figs. S3 and S4). These changes are potentially due to either (1) larger thermocline depth anomalies associated with stronger wind anomalies in the future (Fig. S5) and/or (2) a stronger dependence of ′ on thermocline depth anomalies. Here we found that the mean thermocline is projected to deepens (shoals) in the eastern (western) Pacific and sharpen (e.g., increased ̅ ) during the 21C (Fig. S6), strengthening the thermocline feedback.
In summary, the projected changes in feedback terms during El Niño developing and decay phases, as shown in Fig. 2, readily explain why El Niño events in the late 21C develop sooner (i.e., grow at a faster rate) and persist longer (i.e., dissipate at a slower rate) in the eastern and far eastern Pacific, but decay earlier in the central Pacific. In the next sections, we further explore the effects of Pacific mean-state changes and wind anomalies on the spatio-temporal El Niño evolution.

c) Changes in Westerly Wind Bursts
Previous studies have stressed the role of stochastic forcing in modulating ENSO growth, variability, and predictability. For instance, the stochastic optimal forcing pattern, the noise forcing pattern prone to lead to ENSO growth, is consistent with the spatial structure associated with observed Westerly Wind Bursts (WWBs) 34  This is a topic of future study.

Remote Impacts
southern and southwestern U.S. The enhanced cooling is a result of increased clouds and precipitation over the southern U.S, with changes in the teleconnections being generally consistent among climate models 56 . The projected enhancement of the 20C teleconnection into the 21C is consistent with previous work, which showed that sufficiently warm and persistent SSTA in the far eastern equatorial Pacific is required to excite teleconnection patterns that influence rainfall over the western U.S 30 . Over South America, El Niño's remote effects are also projected to increase, with enhanced warming over the northern two thirds of the continent and drier Amazon basin. Over Australia, there is a projected increase in the El Niño warming continent-wide, as well as more drying of the northwestern region associated with enhanced upper-level convergence and anticyclonic circulation (Fig. 4b).
It is also worth examining the projected changes during the onset and decay phases of El Niño, when significant changes in SSTAs and precipitation are projected to occur (Fig. 1).
Future changes in El Niño's remote impacts during the onset phase (June-July-August in Year 0) are mostly over the Southern Hemisphere (Fig. S7). This is consistent with enhanced SSTAs and precipitation over the tropical Pacific (Fig. 1), the associated upper tropospheric teleconnection response, and the notion that the Southern Hemisphere ENSO response leads the ENSO peak in the tropics. During the decay phase (e.g., January-February-March in Year 1, Fig. S8), El Niño's remote effects mimic those of the peak phase, with enhanced future teleconnections to the surface temperature and precipitation responses over Australia and North and South America, again consistent with enhanced tropical Pacific SSTAs and convection.

Concluding Remarks
The main objective of this work is to better understand the dominant drivers that modulate the projected changes in the spatio-temporal evolution of El Niño events during the 21C. Our  Further studies, using a diverse set of models and emergent constraints, will be needed to assess the robustness of the results reported here.

Observations and Model Data
The results presented here rely on observations and model simulations. radiative forcing of the CMIP5 design protocol 68,69 . Each ensemble member has a distinct climate trajectory due to differences in the atmospheric initial conditions. Thus, differences among ensemble members are solely due to internal variability.

El Niño event definition
In this study, we define an El Niño event following the method used at the NOAA Climate

Mixed Layer Heat Budget
The mixed layer temperature tendency is driven by the thermocline, zonal advective, and Ekman feedbacks as described by the right-hand-side integrands in Eq. 1, plus a residual term.
Here, denotes the mixed layer temperature, and and are the zonal and vertical components of velocity, respectively, which are evaluated in an Eulerian reference frame on the model grid.
= 75 is the mixed layer depth, taken here to be constant throughout the tropical Pacific 10 .
Note that we repeated the analysis using different mixed layer depths, ranging from H = 30m to 100m and the results were consistent. The entrainment velocity is taken to be the vertical velocity at the base of the mixed layer. Bars and primes represent the climatological mean and monthly anomalies, respectively, computed independently for the 20C and 21C. The residual term serves mostly to damp the heat content anomaly and is composed of the surface net heat fluxes, meridional heat advection, diffusive heating, unresolved sub-grid scale heat fluxes, sub-monthly scale heating, and also the zonal and meridional advection terms not listed in Eq. 1 (i.e., ̅ ′ ; 65,66 .

Wind-forced equatorial oceanic Kelvin waves
To identify WWBs, we first remove the 91-day running-mean climatology from the daily zonal wind stress. Then, the anomalous wind stress is averaged between 2.5S and 2.5N.
Westerly wind stress anomalies exceeding +0.03 Nm -2 , with a minimum zonal fetch of 500km, and a minimum duration of 3 days are defined as WWBs 72,73 . Since the influence of WWBs on equatorial dynamics depends on the amplitude, fetch, duration, and probability of occurrence of the forcing, a single parameter is defined here which encompasses all those aspects into one index. This index is summarized by Eq. 2 and is referred to as the downwelling Kelvin wave forcing, and thus is directly linked to El Niño evolution 74 . Eq. 2 considers the amplitude, zonal fetch, and temporal duration of the WWBs, summarizing their integrated impact on ENSO. Here, WWB(x,t) represents the wind stress anomaly associated with WWBs, which is a function of longitude x, which is centered at xo, and time t. Note that the Kelvin wave forcing is the integral of the WWB forcing from the central longitude xo to the eastern boundary xe along the characteristics (i.e.,