Monsoon precipitation response to volcanic eruptions
We analyze seasonally averaged monsoon precipitation responses, using half-year seasons: May to October as boreal summer and November to April as austral summer (Materials and Methods). For each season, year +1 is defined as the season that commences after the volcanic eruption (Table 1). The means of years +1 and +2 precipitation anomalies show clear drying responses to volcanic eruptions in boreal and austral summer monsoon regions in the multimodel mean (MMM), with good intermodel agreement in the sign of the responses (Fig. 1, A and B). MMM time series of Northern Hemisphere (NH) and Southern Hemisphere (SH) mean terrestrial monsoon precipitation indicate that global monsoon drying is strongest at year +1 and the statistically significant (P < 0.01, based on a t test) drying persists for about 2 years after eruptions (Fig. 1, C and D). Although most of the models simulate a decrease in monsoon precipitation, a large intersimulation spread exists in its amplitude (Fig. 1, C and D). Note small but significant NH drying at year 0 (Fig. 1C), which includes a few months influenced by some volcanic eruptions by definition (Table 1).
El Niño–Southern Oscillation response to volcanic forcing and its impact on monsoon drying
Several studies based on past reconstructions and model simulations reported an El Niño–like response (i.e., equatorial eastern Pacific warming) to volcanic forcing within 2 years after large tropical eruptions (8–10, 14–18), and various mechanisms have been suggested including the oceanic dynamical thermostat (8), Maritime Continent (MC) drying (10, 17), a southward shift of the intertropical convergence zone (16), a recharge oscillation from initial La Niña–like cooling (9), as well as West African monsoon weakening (14). Here, we analyze the multimodel response of the El Niño–Southern Oscillation (ENSO) to volcanic forcing by using the Niño3.4 index, which is sea surface temperature (SST) anomalies over the equatorial eastern Pacific. The tropical mean (20°S to 20°N) SST is removed before analysis (the residual is referred to as RSST) to better identify ENSO variations under volcanically induced overall surface cooling (14, 18).
The MMM evolution of equatorial Pacific (5°S to 5°N averaged) RSST over the 5 years after volcanic eruptions are displayed in Fig. 2A. A weak La Niña–like cooling (also with weak model agreement) is seen over the eastern Pacific during autumn and winter immediately after the volcanoes erupted. The greater cooling over the eastern Pacific than central Pacific occurs mainly due to the shallow mixed layer depth in the eastern Pacific as examined by simple mixed layer temperature budget analysis (fig. S1) (9). Following La Niña, El Niño is initiated during the boreal spring of year +1, and it reaches its peak in the subsequent boreal summer [May (+1) to October (+1)] and winter [November (+1) to April (+2)] with good intermodel agreement. After the boreal spring of year +2, El Niño disappears and a La Niña–like pattern emerges over the central Pacific.
To analyze the influence of El Niño on the decrease in monsoon precipitation, the intersimulation relationship is examined using the Niño3.4 RSST index for boreal summer and winter of year +1 (Fig. 2, B and C). The results indicate a statistically significant (P < 0.01) intersimulation correlation between El Niño intensity and precipitation response for both seasons, which illustrates that one can constrain intersimulation spread of monsoon precipitation responses through understanding and reducing the spread of El Niño responses. Figure 3 (A and B) represents contribution of El Niño to precipitation responses to five volcanic eruptions for both seasons, estimated through linear regression analysis (Materials and Methods). El Niño after volcanic eruptions induces additional drying especially over South Asia, China, and Central America in boreal summer and over South America and Australia in the austral summer (red boxes in Fig. 3, A and B), which is found to be mainly through atmospheric circulation changes (fig. S2) (4, 19, 20). Furthermore, the regional averaged MMM precipitation decrease and the intersimulation spread are significantly reduced when removing the influence of El Niño from each model run based on a simple linear regression (P = 0.01 based on t test and P < 0.01 based on F test) (Fig. 3C). A similar result is obtained when considering global monsoon regions (Fig. 3D) (P = 0.03 based on t test and P = 0.05 based on F test), which supports robustness of the El Niño influence on precipitation decrease. This global monsoon drying response to volcanically induced El Niño is found to be overall similar in magnitude to that occurring during nonvolcanic years (i.e., responses to internally driven El Niño; fig. S3).
External and internal factors driving diverse El Niño responses
Previous single-model studies suggested that the occurrence of El Niño after volcanic eruptions largely depends on the intensity of the implemented volcanic forcing (8, 10) and the pre-eruption ocean conditions, which are generated internally (10, 14). Here, we quantify the contribution of each factor to the intersimulation spread in El Niño responses. First, to investigate the impact of the volcanic forcing intensity on the El Niño responses, we compare results from 10 CMIP5 models simulated with Ammann et al. (21, 22) forcing with those from 19 CMIP5 models simulated with Sato et al. (23) forcing (referred to as AM or ST, respectively) (table S1). The AM forcing is known to have greater stratospheric aerosol quantities than ST forcing (21) and hence give rise to greater reflection of incoming shortwave radiation at the top of atmosphere in the models that adopted AM forcing (referred to as AM-forced models) than in ST-forced models (fig. S4, A and B) (24). The two model groups, on average, simulate significantly different El Niño intensities after volcanic eruptions, with stronger El Niño in AM-forced models (P < 0.01, based on a t test) (Fig. 4A), which explains about 29% of intersimulation variance of the Niño3.4 RSST index obtained from all 29 models analyzed based on an analysis of variance (ANOVA; Materials and Methods). A further comparison of AM- and ST-forced models based on individual volcanic eruptions confirms the strong relation between the volcanic forcing intensity and the El Niño amplitude (fig. S4, C and D), which addresses concerns about different models between the AM- and ST-forced groups.
It is found that the different El Niño intensities are primarily associated with the difference in equatorial western to central Pacific westerly wind anomalies during El Niño development periods (fig. S5). Previous model studies suggested that MC drying due to rapid land surface cooling would have a crucial role in El Niño initiation through weakening the Walker circulation and inducing westerly wind anomalies (10, 17). During the transition period from the initial La Niña to El Niño [February (+1) to April (+1), early spring], strong MC drying and western-to-central equatorial Pacific westerly winds are observed (Fig. 4B). The MC drying starts even earlier than the early spring but with weaker amplitude due partly to La Niña–induced wetting, and a La Niña–like pattern induces easterly winds over equatorial eastern Pacific (fig. S6). The early spring MC drying (108°E to 140°E, 8°S to 4°N) has a strong intersimulation correlation with western-to-central Pacific westerly winds (120°E to 180°E, 5°S to 5°N), implying an important role of MC drying in the westerly wind burst (Fig. 4C). Furthermore, the early spring MC drying and westerly wind are significantly related to the subsequent peak El Niño response, supporting the important role of the MC drying mechanism for El Niño occurrence after volcanic eruptions in these models (fig. S7). Furthermore, AM-forced models have significantly greater MC drying than ST-forced models (P = 0.03, based on a t test) (Fig. 4D), which might induce stronger westerly winds and consequently stronger El Niño. Along with the MC drying, a westward downwelling oceanic Rossby wave and a reflected eastward downwelling oceanic Kelvin wave, which are generated from the initial La Niña, seem to help the El Niño initiation (9), but with no significant difference between AM- and ST-forced models (fig. S8). The significant differences between the two model groups are found only after January (+1) (fig. S8F) when the significantly different westerly winds begin to blow (fig. S5C).
To further test robustness of the significant influence of forcing difference, we have conducted a series of sensitivity tests. First, AM and ST forcing data are known to have some differences in meridional aerosol distributions for Agung and El Chichón eruptions (21), which may induce different El Niño responses (16, 17). When redoing our analysis using the other three volcanic eruptions only, results remain unaffected (fig. S9). Different ENSO amplitudes of individual models may contribute to the spread in El Niño responses (fig. S10, A and B). In this regard, excluding models with weaker ENSO amplitudes (fig. S10C) and using normalized Niño3.4 indices and MC precipitation based on each model’s climatology (fig. S11, A and B) do not affect main results. Results are also found insensitive to the use of models having minimum three ensemble members to remove possible pre-eruption ocean condition influences and also to the exclusion of three IPSL models that used solar constant change (fig. S11, C to H) (25). Results are similar when applying all criteria together (fig. S12). Regarding the MC drying mechanism, we also find that CMIP5 models have reasonable spatial representation over the MC and that results are not affected much by different land fraction thresholds (fig. S13). Nevertheless, there is still benefit in refining the resolution to better represent the complex coastlines and potential ocean dynamics around the MC.
Even when identical volcanic forcing is imposed, large intersimulation spread remains in the simulated El Niño responses after volcanic eruptions (Fig. 4A). We investigate the influence of internal variability (pre-eruption oceanic conditions) by using equatorial Pacific warm water volume (volume of water above 20°C isotherm), which has been suggested as a good precursor for El Niño development (14, 26, 27). As an earlier El Niño precursor, warm water volume over the equatorial western Pacific (120°E to 155°W, 5°S to 5°N, referred to as WWVW) is used (26, 27). Figure 5A shows lead-lag correlation between WWVW and Niño3.4 SST from observations. As shown in previous studies (26, 27), early winter [November (0) to January (+1)] Niño3.4 SST variability has maximum correlation with WWVW in August of the preceding year (−1). Similar correlations are obtained from May to April average Niño3.4 SST and leading WWVW. The observed strong relationship between ENSO and WWVW variabilities is well captured by CMIP5 models, representing importance of preceding year WWVW states for following year ENSO variability (Fig. 5B). Here, we use WWVW states at the month before the eruption and examine its impacts on post-eruption El Niño intensities using the 29 AM- and ST-forced models. Note that each group mean Niño3.4 RSST is removed before analysis to remove the influence of volcanic forcing difference. Although we composite five volcanic eruptions, substantial WWVW variabilities remain among AM- or ST-forced models. Results show a significant intersimulation correlation between WWVW and Niño3.4, indicating that oceanic conditions before the volcanic eruptions affect El Niño intensities after volcanic eruptions, explaining about 13.8% of the total intersimulation variance (Materials and Methods) (Fig. 5C). Similar results are obtained when considering individual volcanic eruptions (Fig. 5D), supporting the robust influence of the pre-eruption ocean state on El Niño responses. Results remain insensitive to the use of normalized El Niño intensity and the use of the whole 35 CMIP5 model ensemble (fig. S14, A to D). Also, results from a single model ensemble (CSIRO-Mk3-6-0), which provides 10 simulations, show the strong influence of pre-eruption WWVW on El Niño intensities (fig. S14, E and F), suggesting that WWVW is internally driven.
Our results provide important implications for understanding the observed climatic responses to volcanic eruptions. After the latest three volcanic eruptions, there was a large precipitation reduction in observations (28), especially at year +1 over austral summer monsoon regions (fig. S15, A and B). El Niño occurred after eruptions [based on RSST from two observations (29, 30); fig. S15C] (14, 18), contributing to the global monsoon drying, but several months earlier (fig. S15D) than CMIP5 simulations (Fig. 2A). An analysis of subsurface ocean temperature (31) indicates that warmer oceanic conditions at time of eruptions over the equatorial Pacific (fig. S15E) may be responsible for the observation-model discrepancy (32).