Slowdown of the Walker circulation at solar cycle maximum

We first isolate spatially coherent patterns of annual-averaged tropical (30°S–30°N) SLP anomalies in the Indian and Pacific basins. Although the HadSLP2r dataset (Materials and Methods) spans the whole 20th century, we focus on the second half of the 20th century (1950–2013), for improved data quality, at the expense of a reduced sample length. The spatial pattern of the solar SLP regression coefficients (Fig. 1A) at lag +1 y has positive anomalies over the maritime continent and negative anomalies in the central and eastern Pacific, the latter demonstrating a chance probability of P < 0.1 according to the Cochrane–Orcutt method. This pattern, which indicates an overall relaxation of the climatological zonal SLP gradient, remains robust even over the whole period 1900–2013 or under different configurations of the MLR model (SI Appendix, Fig. S1) and is present in individual seasons (13). The ΔSLP difference between the equatorial (5°S–5°N) Indian/West Pacific (160°–80°W) and East Pacific (80°–160°E) (boxes in Fig. 1A), which serves as a proxy of the PWC strength (5), has a positive loading (24–28 Pa for a typical SC) both over 1900–2013 (P < 0.1) and 1950–2013 (P < 0.05) (SI Appendix, Fig. S2). This means that the climatological ΔSLP gradient slackens by about 13% from Smin to Smax, a response about an order of magnitude weaker than the reduction during typical El Niño events.

Positive ΔSLP anomalies imply weakened surface easterlies and this is evident in the wave- and anemometer-based sea surface wind (WASWind) surface wind observations from 1950 to 2011 (Materials and Methods), which show westerly wind anomalies of 0.7–1.2 ms⁻¹ over the western and central Pacific (arrows in Fig. 1A), with a Pacific-basin–wide (120°E–70°W, 5°S–5°N) mean reduction of ∼0.1 ms⁻¹ (P < 0.3). To study the vertical structure of the wind anomalies and in the absence of dense direct observations, we analyze equatorial zonal winds in a multireanalysis composite [mean of NOAA-CIRES 20th century (20CR), ERA 20th century (ERA20), NCEP/NCAR reanalysis (NCEP)] over the common period 1950–2010, but we note that similar signatures are found over the full 20th century (SI Appendix, Fig. S3). Despite some differences among the reanalyses, a statistically significant reduction of climatological equatorial easterlies is evident in the central Pacific, penetrating high into the troposphere up to ∼400 hPa, with anomalies exceeding 1 ms⁻¹ at 800 hPa (P < 0.05, Fig. 2A). The dipole of positive/negative wind anomalies straddling the climatological westerlies in the upper troposphere (e.g., 200 hPa, 180°–60°W) indicates an eastward displacement of the ascending branch of the PWC toward the central Pacific.

To validate these MLR results, we analyze time series of zonal wind anomalies averaged over the equatorial western and central Pacific troposphere (800 hPa, 140°E–160°W, black line in Fig. 3A) using a multichannel singular spectrum analysis approach (MSSA) that identifies quasiperiodic oscillations (19). The MSSA (nine channels with 15-y window) isolates an 11.9-y quasiperiodic oscillation that tracks TSI variations throughout the analyzed period with a correlation coefficient of 0.74 peaking at 1–2-y lag (Fig. 2E). The amplitude of the detected decadal oscillation in this region exceeds 1 ms⁻¹, consistent with the MLR signal, with prevalent westerly anomalies in Smax reversing to easterly anomalies during Smin years.

A weaker and eastward-shifted PWC at Smax results in an eastward migration of atmospheric convection and precipitation maxima toward the central Pacific as evident in the Global Precipitation Climatology Centre (GPCC) precipitation dataset (Materials and Methods) over the period 1950–2013 (Fig. 1B). Despite the limited spatial coverage of GPCC in the Pacific, the detected pattern is robust and is supported by the reanalyses and the independent GPCP v.2.3 global precipitation dataset (SI Appendix, Fig. S4).

The observed signals in Figs. 1 and 2 refer to 1 y after Smax (lag +1 y), when the strongest ΔSLP anomalies are detected in HadSLP2r and the reanalyses (SI Appendix, Fig. S5). This time lag is similar to that identified in previous studies (15, 18) and indeed our MLR analysis of the Kaplan SSTs over the period 1950–2013 (Materials and Methods) detects a significant SST warming of more than 0.4 K over the equatorial Pacific that maximizes at 1-y time lag (Fig. 1C). The spatial pattern of these SST anomalies is characterized by a stronger loading over the central Pacific and overall warming over the Indian Ocean, as shown in earlier MLR analyses (13, 18, 20).

In line with other studies, we propose that the observed SLP, wind, and precipitation anomalies are all associated with this SST warming. To highlight the bottom-up contribution of SSTs, we perform a six-member ensemble of an atmosphere-only global climate model for 1950–2013 (33 SCs) forced only at the surface boundary by time-varying SST and sea-ice concentration observations (AMIP ensemble, Materials and Methods), while deliberately keeping the solar irradiance and all other forcings constant. The SLP regression coefficients show good agreement with the HadSLP2r at lag +1 y, and significant positive/negative SLP anomalies are simulated over the Indian Ocean/West Pacific and the East Pacific, respectively, resulting in a positive ensemble mean ΔSLP of +41 Pa at Smax (Fig. 1D and SI Appendix, Figs. S2 and S5). Westerly surface wind anomalies (>1 ms⁻¹) are simulated between 150° and 170°W that extend up to 400 hPa into the equatorial troposphere in all individual AMIP runs, although the ensemble mean westerly anomalies are narrower than in reanalyses with an overall stronger magnitude (1.5 ms⁻¹ at 800 hPa), likely explained by differences in the extent and position of the climatological PWC (Fig. 2B and SI Appendix, Fig. S6). Consistent with precipitation observations, the upward branch of the PWC shifts eastward bringing significantly more rainfall (>2.5 mm d⁻¹) toward the central Pacific (Fig. 1E).

While the AMIP ensemble underlines the role of SST anomalies, this is insufficient evidence for the role of SC forcing on the PWC. Although the null hypothesis, that the observed responses simply reflect internally generated decadal variability, can be rejected in statistical terms with probability <0.1, the MLR analysis alone cannot establish causal linkages between SC forcing and warmer SSTs/weakened PWC per se. In fact, the detected patterns in tropical PWC, SSTs, and precipitation share common features with modes associated with internal Pacific variability (21), implying a possible signal contamination. To demonstrate a physical linkage to SC forcing, we perform a four-member, 100-y ensemble (∼40 SCs) using a coupled atmosphere–ocean model forced only by repeating the solar cycle 22 (1986–1996) variation in solar spectral irradiance (SOLAR ensemble, Materials and Methods).

Increased TSI at Smax in the SOLAR results in a widespread warming over the equatorial Pacific, with regression coefficients of up to +0.4 K that maximize at a lag of 1–2 y, in good agreement with the observations (Fig. 1I). The equatorial SLP responds to the positive SST anomalies by increasing over the Indian Ocean/West Pacific and reducing in the East Pacific and the strongest SLP signal is also found at 1–2 y lag (Fig. 1G and SI Appendix, Fig. S5). The ensemble mean ΔSLP change of +38 Pa per 1 W/m² increase in TSI is consistent with the observed change of about +24 Pa per 1 W/m² (SI Appendix, Fig. S2) and indicates a slower PWC. This is confirmed by the westerly surface wind anomalies between 150° and 170°W (0.5 ms⁻¹) that extend throughout the troposphere (0.7 ms⁻¹ at 800 hPa) in the western Pacific (Fig. 2C). SOLAR reproduces the pattern of observed zonal wind anomalies over the equatorial Pacific but the model simulates broader easterly anomalies over the Indian Ocean (30°–90°E), associated with overall broader simulated SLP and SST patterns. This is related to a common model bias in simulating tropical Pacific feedbacks, which also affects the SLP, precipitation, and SST responses to ENSO variability on interannual timescales (SI Appendix, Fig. S7) and is present in many other coupled atmosphere–ocean models (22). The spread of wind anomalies among individual ensemble members in SOLAR is low and all consistently show a weakened PWC with reduced easterlies in the western Pacific and a dipole of negative/positive anomalies in the upper troposphere (SI Appendix, Fig. S8). MSSA filtering of the ensemble mean SOLAR zonal winds at 800 hPa over the western and central Pacific identifies a quasiperiodic (∼10.3-y) fluctuation in phase with the specified TSI variability (∼10.5-y period) over 100 model years with a correlation coefficient 0.6 at lag +1 y (Fig. 2F). An eastward shift of the upward branch of the PWC in SOLAR brings more rainfall toward the central Pacific (Fig. 1H), with an amplitude and pattern comparable to the observations.

An eastward shift of the PWC following Smax has also been simulated in previous studies (17, 18), but as an additional measure to validate our model responses, we identify a similar pattern of westerly wind anomalies at Smax in the Community Earth System Model (CESM) Large Ensemble Project simulations that include all historical forcings (SI Appendix, Fig. S9). In summary, the SOLAR simulations reject the null hypothesis that the PWC response is due to internal decadal variability and instead provide supporting evidence that the observed weakening of the PWC is a genuine response to the increased TSI at solar maximum.

Previous analysis of CMIP5 historical simulations indicates that a starting point in understanding the mechanisms whereby the SC variability influences the PWC is found in the muted hydrological cycle mechanism (18). As direct observations of precipitation suffer from limited spatiotemporal coverage and homogenization issues (23), we turn to the reanalyses to examine global signals at lag +1 y in precipitation and water vapor content (WVC). Reanalyses generally indicate an atmospheric moistening at Smax with increases of 0.5% (P < 0.25), 0.7% (P < 0.05), and 0.4% (P < 0.4) for 20CR, ERA20, and NCEP, respectively (Fig. 3). Over the ocean only (land grid points masked out) the signals are 0.6% (P < 0.05), 0.74% (P < 0.05), and 0.57% (P < 0.15). This translates to a global-mean sensitivity of 6.3%/K, 7.7%/K, and 4.2%/K, under global surface warming (per 1 W/m²) of about 0.07, 0.09, and 0.1 K (Fig. 3). While the WVC is largely controlled by the C-C relation, global precipitation changes are constrained by atmospheric energetics and generally increase at a reduced rate in response to a warmer surface (24). The SC imprint on global precipitation is more uncertain than the WVC and reanalyses show insignificant positive anomalies at lag +1 (0.06%, 0.1%, and 0.32% for 20CR, ERA20, and NCEP, respectively, generally P < 0.6). However, the sensitivity of global precipitation changes to surface warming (0.94%/K, 1.15%/K, and 3.17%/K, for 20CR, ERA20, and NCEP, respectively) lies within the spread of responses in the AMIP individual runs (range −0.6–4.2%/K at lag +1 y), which show an ensemble mean increase of 1.92%/K. As expected, percentage increases of global WVC are larger than the corresponding precipitation signals in all AMIP runs, with an ensemble mean WVC sensitivity of 8.84%/K (ensemble range 6.1–11.6%/K). Comparable changes in global hydrology are also found in SOLAR at lag +1. The WVC increases significantly by 0.72% (P < 0.01 and range 0.4–1%), while global precipitation increases by merely 0.12% (P < 0.22 and range 0.07–0.2%), implying an ensemble mean sensitivity to the SC of 9.6%/K for WVC and 1.6%/K for precipitation.

Reanalyses and model results provide evidence for a stronger sensitivity of the global WVC increase versus global precipitation in response to the SC, which is consistent with SC signatures in global hydrology identified in the CMIP5 models and other attribution studies (18, 24). According to theory and sensitivity simulations of GHG forcing (2, 25), the dominant driver of global hydrological changes is the spatially uniform SST response while the SST pattern (longitudinal distribution of SST anomalies) is of secondary importance. To investigate whether this is also the case for solar forcing, we analyze an AMIP-ZM ensemble, which is identical to AMIP but forced only by zonally averaged SST anomalies from 1950 to 2013 superimposed on the SST climatology (Materials and Methods). In this way, the longitudinal pattern of SST response to SC forcing is removed and surface temperature anomalies over the Indian and Pacific Oceans are characterized by a zonally uniform warming, which barely exceeds 0.1 K (Fig. 1L). Analysis of this ensemble confirms that the global-mean WVC and precipitation increase at almost the same rates as in the AMIP (with differences arising mainly over land; Fig. 3). The muted response of the global hydrology weakens convection and tropical circulation as indicated by the westerly anomalies (up to 300 hPa) in the Far West Pacific (Fig. 2D). Westerly wind anomalies in AMIP-ZM are considerably weaker (<0.2 m/s, P < 0.4 at 800 hPA) than AMIP and show a wider ensemble spread because of the weaker magnitude of the specified SST gradients (SI Appendix, Fig. S10). For this reason, it is not surprising that AMIP-ZM shows insignificant ΔSLP anomalies (Fig. 1J and SI Appendix, Figs. S2 and S5). However, we detect a significant (P < 0.1) increase in precipitation which, like the circulation anomalies, is confined to the West Pacific.

Comparison of the AMIP versus AMIP-ZM simulations suggests that while the general response pattern may be determined by the zonal-mean warming, the amplitude of the surface precipitation and wind response as well as the extent of the eastward shift of the PWC, are probably associated with the longitudinal SST response pattern. It is therefore of interest to explore the cause of this SST warming pattern and how it further weakens the PWC.

It is not well understood why the Pacific warms preferentially in response to increased TSI (Fig. 1C). Previous studies have noted an ENSO-like pattern and suggested a dynamically excited mechanism involving a damped-resonant ENSO response triggered by direct TSI surface heating (9). However, the weak westerly anomalies found in AMIP-ZM suggest an alternative mechanism and here we propose that a Bjerknes feedback further weakens the PWC at Smax.

We investigate this hypothesis in a conceptual framework using a low-order atmosphere–ocean model of the Pacific that has been extensively used to study ENSO dynamics (26). The model solves for SST and thermocline depth anomalies in two boxes, representing the east and west equatorial Pacific, by taking into account the Bjerknes feedback that drives wind stress anomalies, which in turn cause equatorial upwelling in the eastern Pacific and zonal advection (SI Appendix). Effects of the SC are introduced as an 11-y sinusoidal modulation of the wind stress of 1% amplitude, roughly corresponding to the 10% reduction in surface easterlies simulated in AMIP-ZM. The model shows that the imposed westerly anomaly reduces the thermocline tilt between the West and East Pacific at Smax, resulting in eastward advection of anomalously warm water and positive SST anomalies in the East Pacific (SI Appendix, Fig. S11). The west–east SST gradient is thus weakened and hence the Bjerknes feedback reinforces the imposed westerly anomalies. The transition from Smax to Smin weakens and eventually reverses the wind anomalies to easterlies at Smin.

Results of this low-order ENSO model can only frame a theoretical background for understanding the observed PWC weakening given that the simulated response amplitude depends on the assumed coupling parameters. Nevertheless, the low-order model predicts a weakening of the equatorial Pacific thermocline tilt at Smax, and thus a lifted thermocline in the western Pacific that brings cold water to the subsurface. We find further evidence that the two-box model can be considered as a credible analog of the observed SC response by examining equatorial subsurface ocean temperature responses in the EN4 objective analysis for the period 1950–2013 (Fig. 4, Materials and Methods). MLR identifies negative temperature anomalies, indicating a shallower thermocline in the western Pacific at +1 y lag (−0.5 K per 1 W/m² TSI at 130 m, P < 0.3) and significant positive anomalies indicating a deeper thermocline in the eastern Pacific (>1 K, P < 0.05; 120°–160°W at 130-m depth). This means a reduced thermocline tilt at Smax as suggested by the low-order model. This observational evidence is supported by the SOLAR ensemble, which simulates cold water in the western Pacific subsurface, but does not show a stronger warming in the subsurface over the central and eastern sector (Fig. 4). This negligible change in the eastern Pacific thermocline likely explains the generally weaker SST warming in the equatorial Pacific at Smax (Fig. 1I).

The low-order model thus highlights the potential for a Bjerknes amplification of the weak westerly wind anomalies initially generated by the SC influence on global hydrology. In addition, the relaxation of the trade winds at Smax is likely to weaken evaporative cooling, which further weakens the surface winds. This wind–evaporation–SST feedback (27), not accounted for in the low-order model, could provide a source of additional warming in the West Pacific, as indicated by the pattern of Pacific SST anomalies in the Kaplan dataset and the SOLAR ensemble (Fig. 1 C and I).

Studies of the response to centennial increases in TSI have suggested a contrasting response, with reduced rainfall in the Pacific and stronger east–west SST gradients (28). However, this response is not supported by a recent multimodel comparison of the tropical impacts of a step-change doubling of CO₂ concentration versus a 2% increase in TSI (29). In those simulations, both forcings produced strikingly similar patterns of increased precipitation over the central Pacific and a relaxed east–west SST gradient, despite the fact that TSI predominately affects the surface, whereas GHGs perturb the radiative balance of the troposphere. In fact, the globally averaged precipitation is expected to increase at higher rates for the same surface warming under TSI than GHG forcing (24). Likewise, we find no evidence for divergent patterns in the Pacific response given that the detected patterns of increased rainfall, relaxed SST gradient, positive ΔSLP anomalies, and a weakened PWC in response to SC resemble modeled responses to 2% TSI and doubling CO₂, despite subtle differences in the timescale and the magnitude of the forcing (2% TSI roughly corresponds to 20× the SC amplitude).

By highlighting the importance of global hydrology in mediating SC responses in the tropical Pacific, our findings add confidence in the modeled predictions for a weakened PWC, driven by similar hydrological changes in response to GHG forcing. We are not suggesting, however, that observed SC signatures can be used as an analog for quantifying future PWC responses to GHGs. This is because we identify ΔSLP changes of about 300 Pa/K (24–28 Pa per 1 W/m² TSI associated with about 0.08 K per 1 W/m²) between solar maximum and minimum, which are considerably larger than the observed century-long ΔSLP trends (50–100 Pa/K scaled to the global-mean SST warming) (4, 5). Reasons for reduced sensitivity in the case of GHG forcing are likely to be found in the pattern of SST response in the Pacific, which is ill constrained in observations and/or cancellation from opposing mechanisms that may not be as effective on the 11-y SC timescale. For example, a long-term adjustment of the thermocline to GHG forcing has been found to reduce the effectiveness of the Bjerknes feedback (30), while our analysis, instead, highlights the importance of this feedback in reinforcing wind and SST anomalies and ultimately the PWC reduction. In summary, we propose a bottom-up pathway by which the SC influence in global hydrology induces weak westerly surface wind anomalies which are then amplified by the Bjerknes ocean–atmosphere feedback (and possibly also a wind–evaporation–SST feedback), leading to the further reduction and eastward shift of the PWC, positive SSTs, and more rainfall in the equatorial Pacific. This study suggests that SC forcing is an appreciable source of decadal variability in the Indo-Pacific and further studies are required to assess the potential to improve decadal predictability in this region.

We thank Dr. Fenwick Cooper for advice on statistical methods. We also thank three anonymous reviewers and the editorial team for constructive comments on the manuscript. All model simulations presented in this work were supported by the DeiC National HPC Center, University of Southern Denmark (SDU). The research was supported by the UK Natural Environment Research Council (NERC) through Grant NE/N010965/1 and the NERC National Centre for Atmospheric Science. M.F.K. and C.K. are grateful for financial support from Villum Foundation Grants VKR023114 and VKR010116.