3.1 ENSO complexity in spatial structure: Three types of El Niño
Figure 3 shows the longitude-time evolutions of equatorial Pacific SST anomalies during the three types of El Niño in the observations and the TCWB1T model hindcasts at different lead times. In the observations, the EP El Niño (Fig. 3a) has warm SST anomalies originating off the South American coast during the spring of the first year, developing westward during the summer and fall, reaching its mature stage during the winter, and then retreating back toward the coast in the spring and summer of the second year. It is also noticeable that the largest EP El Niño SST anomalies of its lifecycle, from the developing to decaying years (i.e., 1st - and 2nd -years), are mostly confined to the tropical eastern Pacific. In contrast, from the developing to decaying years of the observed CP-I and CP-II El Niños (Figs. 3g, m), the largest SST anomalies are concentrated in the tropical central Pacific and very weak warm SST anomalies are in the tropical eastern Pacific.
The TCWB1T model hindcasts at lead-0 month (Fig. 3b, h, n) accurately reproduce the observed longitude-time evolution features of the three types of El Niño, as expected, but begin to produce discrepancies from the observations as the hindcast lead time increases. The EP El Niño hindcasts, with lead months increasing from lead-2 to lead-8 (Fig. 3c − f), show a strong tendency to underestimate the El Niño intensity. It is further noticed that the underestimation of the El Niño intensity is most severe in the eastern and far-eastern tropical Pacific regions, shortening the duration of EP El Niño in those regions. Therefore, the hindcasted EP El Niño exhibits a faster termination and phase transition to La Niña in the decaying year, with the model error intensifying as the hindcast lead time increases. This model deficiency is also clearly detected in Fig. 4 when we compare the temporal evolutions of the observed and hindcasted Niño indices, which include the Niño4, Niño3.4, Niño3, and Niño1 + 2 indices (cf. black rectangles in Fig. 1a). During the EP El Niño (Fig. 4a − d), it is evident from the figure that the hindcasted Niño indices largely underestimate the observed ones during the developing and decaying years (note. the hindcasted Niño indices shown are the mean index values averaged from the lead-0 to lead-8 month hindcasts). The underestimation from the EP El Niño hindcasts becomes more severe as the region quantifying Niño index moves from the tropical central Pacific to tropical far-eastern Pacific. For both the CP-I and CP-II El Niños, their longitude-time evolutions (Fig. 3i − l and 3o − r) show that the model hindcasts generally reproduce the major features of the two CP El Niño types from all lead times in terms of El Niño’s intensity and spatial structure. Despite this, some minor deficiencies exist. Firstly, the model hindcasts tend to slightly underestimate (overestimate) El Niño intensity during the CP-I (CP-II) El Niño, particularly in its development and mature stages. Secondly, the model hindcasts tend to simulate the decay of CP-I and CP-II El Niños slower than the observed, causing their phase transition to be delayed by 2–3 months. These minor model deficiencies for the CP-I and CP-II El Niño simulations are also well detected in the temporal evolutions of various Niño indices (Fig. 4e‒l; except for the Niño1 + 2 index whose values indicate the SST anomaly in the far-eastern equatorial Pacific).
The results above indicate that the TCWB1T model has a noticeable deficiency in the hindcasts of EP El Niño as it largely underestimates El Niño intensity during both the developing and decaying years. Meanwhile, in spite of some minor deficiencies, the model is able to realistically hindcast the observed evolution features of CP-I and CP-II El Niños with a lead time up to eight (8) months. To examine how the model error grows in the hindcasts, we compare in Fig. 5a‒c the hindcasted and observed Niño3.4 index values for all the three types of El Niño, from lead-0 to lead-8 month hindcasts. It is evident from the figure that, as the lead time increases, the hindcasted evolution of the index begins to deviate from the observed evolution for all three types of El Niño. However, when taking a closer look, the overall deviations are revealed to be the smallest for the CP-I El Niño, the second-smallest for the CP-II El Niño, and the largest for the EP El Niño. This result suggests that, for this particular climate model, the CP-I El Niño is the easiest El Niño type to be hindcasted, followed by the CP-II El Niño, while the EP El Niño is the most difficult type to be hindcasted. We then examine the growth of these hindcast errors by integrating the deviations between the hindcasted and observed Niño3.4 index values throughout the El Niño lifecycle (i.e., 1st -to-2nd years) for each lead time, which can be simply formulated as
The bar-charts in Fig. 5d − f, which display the error growth in the hindcasts of the three types of El Niño from lead-0 to lead-8, further support the previous suggestion that the EP El Niño has the largest simulation errors while the CP-II El Niño has the second largest and the CP-I El Niño has the smallest. It is also found from Fig. 5d − f that the hindcast errors for the EP and CP-II El Niños grow continually exceeding 6°C as the lead month increases, with the former being relatively faster than the latter after the 3-month lead. However, the hindcast errors for the CP-I El Niño grow very slowly and reach a plateau without exceeding 6°C until the 8-month lead. Collectively, it can be concluded that the TCWB1T model has the best performance when hindcasting the CP-I El Niño, the second-best performance for the CP-II El Niño, and the worst performance for the EP El Niño.
Then, what is responsible for the model’s worst performance on the EP El Niño hindcasts? The annual cycle plots of tropical Pacific climatological SSTs in the observations (Fig. 6a) and the TCWB1T model hindcast mean (averaged from lead-0 to lead-8 month; Fig. 6b) give us a clue as to why the model largely underestimates the EP El Niño intensity. Compared to the observations (Fig. 6a), the model largely overestimates the climatological SSTs over the tropical eastern Pacific (i.e., east of 120°W) (Fig. 6b). This warm bias of the model is well revealed in the difference between the hindcasted and observed climatological SSTs (Fig. 6c). It is also evident from the annual cycles of the tropical eastern Pacific climatological SSTs (Fig. 6d) that the warm bias tends to be stronger during the second half of the calendar year (i.e., July to December) than the first half (i.e., January to June). This overall strong warm bias in the model indicates that the simulated mean thermocline in the tropical eastern Pacific is too deep. The model’s deeper mean thermocline in the tropical eastern Pacific can weaken the sensitivity of the overlying SSTs to wind variations due to the weakening of the so-called thermocline feedback in that region (An and Jin 2001). When the mean thermocline is deeper (as in the model hindcasts), the strength of the surface wind stress-forced ocean wave that induces thermocline variation should also become weaker since the downward transfer of surface wind stress momentum for the geostrophic adjustment is mostly consumed above the deeper thermocline (cf. An and Kim 2017)―a weakening of the thermocline feedback. It is well known that the SSTs in the tropical eastern Pacific are primarily influenced by the subsurface ocean dynamics related to the thermocline variation and its coupling with surface winds (cf. Philander et al. 1996; Yu and Mechoso 1999). As a consequence of the abovementioned processes, the model tends to largely underestimate El Niño intensity over the tropical eastern Pacific (Figs. 3 and 4) and has the worst simulation performance for EP El Niño among the three types of El Niño (Fig. 5).
Furthermore, we investigate to understand why the model has deficiency in simulating climatological SSTs over the tropical eastern Pacific. Climatologically, the tropical Pacific Ocean exhibits zonal contrast in SSTs between the western Pacific warm pool and the eastern Pacific cold tongue. A colder cold tongue SST can be established for at least two reasons. One is related to the strength of the Pacific Walker circulation, whose surface easterly winds can cause the cooling of the cold tongue SST by upwelling cold water from deeper layers. The other is related to the strength of the South Pacific high, a semi-permanent anticyclone located in the southeast Pacific near the west coast of Peru and northern Chile (see the letter ‘H’ in Fig. 7a), which can drive cross-equatorial surface winds over the tropical eastern Pacific to then upwell cold subsurface ocean water which cools down the overlying SSTs. Figure 7a, which displays the observed climatological atmosphere and ocean structures over the Pacific basin during fall (i.e., August to October, when the cold tongue SST climatologically marks its lowest value), clearly illustrates the abovementioned Pacific conditions, including the strengths of the Pacific Walker circulation (manifested by the easterly trade winds and the zonal SST contrast) and South Pacific high (manifested by the southeastern Pacific anticyclone and the northward component of surface winds). From the model hindcast mean (Fig. 7b) and its difference from the observations (Fig. 7c), it is evident that the simulated strengths of the Pacific Walker circulation and South Pacific high are weaker than the observations. As shown in Fig. 7c, the difference is characterized by surface westerlies along the equatorial Pacific (i.e., an indication of a weaker Pacific Walker circulation) and a cyclonic circulation over the southeastern Pacific (i.e., an indication of a weaker South Pacific high). These weaker-than-observed Pacific Walker circulation and South Pacific high in the model can both cause weaker cooling of the cold tongue SST, thus producing a strong warm bias over the tropical eastern Pacific.
Further attributing these model deficiencies to a particular cause is difficult due to the strong atmosphere‒ocean coupling nature of the tropical eastern Pacific (Mechoso et al. 1995). Small model deficiencies in the atmospheric or oceanic model component can be largely amplified by this coupling, thus easily affecting the coupled climate model performance. Yet previous studies have pointed out that climate models’ inability to properly simulate the marine stratus clouds over the southeast Pacific can influence the strength of the South Pacific high through radiative cooling (cf. Mechoso et al. 1995; Philander et al. 1996). A majority of contemporary climate models still have difficulty properly simulating marine stratus clouds over the southeastern Pacific as well as the resultant equatorial upwelling processes over the tropical Pacific (e.g., Gordon et al. 2000; Large and Danabasoglu 2006).
3.2 ENSO complexity in temporal evolution: Multi-year evolution events
To evaluate whether the TCWB1T model can or cannot hindcast the temporal complexity of ENSO, we choose to focus on analyzing the model hindcasts of the atypical multi-year evolution events. Those are the multi-year El Niño which occurred in 1986‒1988 (simply, the 1986/87/88 El Niño) and the multi-year La Niña which occurred in 1998‒2000 (simply, the 1988/99/00 La Niña) (cf. Figure 2b). Both are re-intensified events whose 2nd -year SST anomaly peak intensity is larger than their 1st -year peak intensity. We firstly compare in Fig. 8 the observed and hindcasted longitude-time evolutions of equatorial Pacific SST anomalies during the life cycle (from 1st to 3rd years) of these multi-year evolution events. For the 1986/87/88 El Niño in the observations (Fig. 8a), its warm SST anomalies emerge and develop mostly in the tropical central Pacific during the summer and fall of 1986 and then peak in the following winter to materialize the 1st -year of its multi-year evolution. While the 1st -year El Niño persists into the spring of 1987 to start the 2nd -year evolution, warm anomalies also emerge separately from the tropical eastern Pacific. The latter processes become dominant for the development of the 2nd -year evolution of this event, forming the largest warm anomalies in the tropical eastern Pacific. This re-intensified multi-year El Niño in 1986/87/88 can, therefore, be assumed to be a CP El Niño pattern during its 1st -year evolution and an EP El Niño pattern in its 2nd -year evolution. This sequence of SST evolution is similar to that observed during the 2014/15/16 multi-year El Niño (Paek et al. 2017; Kim and Yu 2021). For the model hindcasts, despite the slightly underestimated El Niño intensity, the lead-0 month hindcast (Fig. 8b) adequately reproduces the observed multi-year evolution of the 1986/87/88 El Niño by capturing its 1st - and 2nd -year El Niño peaks. However, with the lead months increasing from lead-2 to lead-8 (Fig. 8c − f), it is found that the model fails to develop the 2nd -year El Niño during 1987‒1988. The hindcasted event, instead, tends to become a single-year El Niño with its peak in the winter of 1986/87 before transitioning into a La Niña afterwards. Figure 9a, which displays the Niño3.4 index evolutions during 1986/87/88 El Niño in the observations and the individual model hindcasts, also supports this model’s failure in hindcasting the multi-year evolution of the 1986/87/88 El Niño.
Next, we extend our analysis to evaluate whether the model hindcasts have an ability to forecast the multi-year evolution of 1986/87/88 El Niño. To achieve this, we plot the same Niño3.4 index evolutions as in Fig. 9a but using the individual model hindcasts that are launched at different starting months from December0 to December1 (note. the lead-0 month hindcast values were utilized as starting months). As depicted by thin curves with various colors and markers in Fig. 9b, the results clearly indicate that the model has no chance to forecast the 1986/87/88 El Niño’s multi-year evolution. Once they are launched at any starting month, the initial Niño3.4 index values, with the lead months increasing, have an immediate downward slope toward a neutral or negative ENSO phase, failing to capture the re-intensification of the 1986/87/88 El Niño. As shown in Fig. 10, the horizontal maps of seasonally-averaged SST and wind anomalies during the 1st -year winter, following spring, and summer for the 1986/87/88 El Niño allow us to find why the model hindcasts fail to capture the re-intensification of the 1986/87/88 El Niño. For the observations (Fig. 10a − c), it is easily noticed that the El Niño, after its 1st -year peak in winter, persists and re-intensifies during the following spring and summer to give rise to its 2nd -year peak. Meanwhile, during the 1st -year winter (Fig. 10a), we also notice that there is a meridional band of warm SST anomalies extending from the subtropical northeastern Pacific to the tropical central Pacific accompanied by southwesterly wind anomalies in between (see black parallelograms in Fig. 10). This meridional band of anomalous SST warming in the subtropical northeastern Pacific is also known as a positive Pacific Meridional Mode (PMM; Chiang and Vimont 2004). The positive PMM, which is often induced during El Niño as a result of El Niño’s teleconnection to the extratropics (Stuecker 2018; Fang and Yu 2020a), is able to maintain for several seasons and spread its warm anomalies into the central equatorial Pacific via a wind − evaporation − SST feedback mechanism (Xie and Philander 1994), enabling to develop another El Niño (Yu et al. 2010; Yu and Kim 2011; Amaya 2019; Kim and Yu 2020, 2021). As shown in Fig. 10a − c, the abovementioned tropical‒subtropical climate interactions within the Pacific basin (simply, the Pacific basinwide interactions) are indeed at work in the re-intensification of the 1986/87/88 El Niño. For the model hindcasts (Fig. 10d − f), differently from the observations, the positive PMM during the 1st -year winter, which is represented by the anomalous SST warming and southwesterly wind in the subtropical northeastern Pacific, suddenly disappears and is replaced by anomalous SST cooling and northeasterly wind during the following spring and summer. This indicates that the Pacific basinwide interactions are absent in the hindcasts; hence, the model hindcasts fail to capture the re-intensification of the 1986/87/88 El Niño.
When analyzing the model’s ability to hindcast the observed multi-year La Niña event that occurred from 1998 − 2000 (i.e., the 1998/99/00 La Niña; Fig. 8g), we find that the hindcasts generally reproduce the multi-year evolution of 1998/99/00 La Niña, not only for the lead-0 month hindcast (Fig. 8h), but also for hindcasts at lead-2, 4, 6, and 8 months (Fig. 8i − l). The Niño3.4 index evolutions shown in Fig. 9c, d confirm that all individual model hindcasts successfully simulate the 1998/99/00 La Niña’s multi-year, re-intensification evolution feature (Fig. 9c) regardless of the launching months (Fig. 9d). Why, then, is the model able to successfully simulate/forecast the 1998/99/00 La Niña evolution in stark contrast to the model’s failure to simulate the 1986/87/88 El Niño evolution? We find in Fig. 11, which uses the same horizontal maps as in Fig. 10 but for the 1998/99/00 La Niña, that the model reproduces the observed Pacific basinwide interactions in the La Niña hindcasts. During the 1st -year winter, following spring, and summer of the 1998/99/00 La Niña, the model properly simulates the observed negative PMM with anomalous SST cooling and northeasterly wind in the subtropical northeastern Pacific (Fig. 11a − c vs d − f; see black parallelograms in Fig. 11). This indicates that, contrary to the El Niño simulations, the TCWB1T model is able to capture the Pacific basinwide interactions activated by the 1st -year La Niña during 1998‒1999 to successfully simulate the re-intensification of the 2nd -year La Niña during 1999‒2000.
One question remains as to why the La Niña event can activate the Pacific basinwide interactions in the model to produce a multi-year evolution pattern but the El Niño event cannot. In this regard, we find in Fig. 12a, b that the simulated climatological SSTs in the tropical central Pacific are too cold compared to the observations during spring, which is the season when the PMM are most active (Chiang and Vimont 2004). The cold bias can provide more favorable conditions for La Niña to activate the Pacific basinwide interactions but not for El Niño. This is because, as the cold bias lowers climatological SSTs closer to the convective threshold temperature (~ 28°C, Sud et al. 1999; see magenta contours in Fig. 12a, b), the La Niña-related cold anomalies can easily bring down SSTs to below the threshold value over the tropical central Pacific and switch off the deep convection. The induced large anomalous cooling can then excite an anomalous anticyclone over the subtropical Pacific, including the PMM region, activating a negative phase of the Pacific basinwide interactions, whose cold anomalies can later intrude into the equatorial Pacific to start the 2nd -year La Niña in subsequent seasons. The bar-charts representing the climatological spring SSTs averaged over the tropical central Pacific (cf. black rectangles in Fig. 12a, b) in the observations and the model hindcast mean (Fig. 12c) further confirm that the model’s cold bias is statistically significant. Although the El Niño-related warm anomalies can also strengthen deep convection over the tropical central Pacific to produce an anomalous heating in that region, the heating is relatively small compared to the cooling induced by a La Niña event that completely turns off the climatological deep convection. Also, since most CP El Niño events have weak intensities (e.g., Kao and Yu 2009; Lee and McPhaden 2010), the anomalous heating they induce is weak and provides less favorable conditions for El Niño to activate the Pacific basinwide interactions. The physical mechanism explaining the La Niña-favored activation of the Pacific basinwide interactions has been well documented in Fang and Yu (2020a).