a,b, The observed (a) and simulated (b) SST changes (background colour) and 850 hPa wind anomaly (arrows) both exhibit a pan-tropical dipole SST change, with warming extending from the tropical Atlantic to the Indian Ocean and the western Pacific, and cooling over the central–eastern Pacific. The observed trends (a) are estimated using the Sen’s slope method, from 1979 to 2012. c,d, The observed (c) and simulated (d) Walker circulation changes (arrows) and troposphere temperature anomalies (colour shading): the Indo-Pacific Walker circulation is enhanced. The vertical velocity is magnified by a factor of 750 to make its scale comparable to that of zonal wind. e,f, The observed (e) and simulated (f) ocean subsurface temperature anomalies (colour shading) for the tropical cross-section between 5° S and 5° N.
Data sets.
The UK Met Office Hadley Centre’s SST data set HadISST (ref. 31) was employed in this study to estimate the trend of the tropical SST from 1979 to 2012 (Fig. 1a), and the SST trend over the tropical Atlantic estimated by this data set was used to force the CESM and CAM4 models. The Kaplan Extended SST version2 (ref. 32), and the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST version 3b (ref. 33), were also used together with the HadISST to reveal the decadal relationship between the tropical Atlantic and the Indo-western Pacific.
The Ishii Subsurface Ocean Temperature Analysis34 was used to calculate the subsurface ocean temperature trends from 1979 to 2012 (Fig. 1e). The Global Precipitation Climatology Project35 (GPCP) data were used to estimate the trend in the tropical precipitation for the same period. The European Centre for Medium Range Weather Forecasts (ECMWF) Interim reanalysis36 (ERA-Interim) data were used to estimate the trend in the atmospheric circulation (Fig. 1a, c).
Model results from the Coupled Model Intercomparison Project37 (CMIP5) historical experiments were used to identify the relationship between the tropical Atlantic and the Indo-western Pacific decadal-mean SST.
Analyses methods.
Sen’s slope38 method is used to calculate the observed trends, with the confidence intervals estimated using the Mann–Kendall test39. We used Student’s t-test to calculate the confidence interval of the model responses.
CESM model experiments.
The National Center for Atmospheric Research (NCAR) coupled climate model, the Community Earth System Model40 (CESM1.06) was used in this study to investigate the response of the tropical climate system to an observed tropical Atlantic warming. We used F19_G16 (ref. 40) horizontal resolution, with ~2° resolution in the atmospheric component, and ~1° in the ocean component. We restored the tropical Atlantic temperature in the coupled model with an external heating within the mixed layer as follows:
where c is the heat content of sea water, D is the mixed-layer depth, Tr is the restoring target temperature, Tm is the model temperature at each time step, and τ is the restoring timescale, which was set as 20 days in this study.
The CESM response to the tropical Atlantic warming was calculated as the ensemble mean of 12 sensitivity experiments. In each experiment, we estimate the difference between a control run and a perturbed run. In the control run, the ocean temperature in the mixed layer of the tropical Atlantic (defined as the Atlantic Ocean between 20° S and 20° N, with linear buffer zones extending from 20° S to 30° S and from 20° N to 30° N) was restored to the model climatology. In the perturbed run, the tropical Atlantic SST was restored to the model climatology plus the observed 1979–2012 SST trend. We generated 12 ensemble members by slightly perturbing the external forcing around the observed SST trend (±0.1%, ±0.2%, ±0.6%, ±1%, ±1.4%, ±1.8% from the observed trend). Each simulation starts from the year-2000 initial condition of the CESM system and lasts for 30 model years. The first 5 years serve as a spin-up simulation and the results from year 6 to year 30 are used in the calculation. The ensemble mean of these simulations is then considered to be the CESM response to the observed trend of the tropical Atlantic SST.
CAM4 model simulations.
The NCAR atmospheric model, the Community Atmosphere Model version 4 (CAM4), was used in this study to identify the tropical atmospheric responses to the Atlantic SST trend from 1979 to 2012. CAM4 is the atmospheric component of CESM and is run with the same resolution. As we do with the CESM simulation, we estimate the CAM4 response by differencing the control runs (with the climatological SST forcing) from the perturbed runs (forced by the tropical Atlantic SST trend).
GFDL dry-dynamical-core simulations.
The spectral dry dynamical core of an atmospheric general circulation model41, developed at the Geophysical Fluid Dynamics Laboratory (GFDL), was used to investigate the evolution of the atmospheric response to a tropical Atlantic warming, in a primitive-equation dynamical system. The idealized model is initialized with the climatological background flow from the ERA-Interim reanalysis, averaged from 1979 to 2012. At each time step, an additional forcing that balances the model’s initial tendency associated with the climatological background flow was added to keep the model steady42, 43. This external forcing ensures that the model response at each time step is due only to the initial tropical perturbation. In the forced cases, a convective heating is added as an initial impulse over the tropical Atlantic. The model results at each snapshot could be interpreted as the evolution of the primitive-equation dynamics in response to the tropical heating (see ref. 43 for details).
Surface heat flux and WES effect.
The change of SST ∂T′/∂t satisfies a balance5, 6 between the oceanic dynamics Do and four surface heat flux components: solar radiation QS, long-wave radiation QL, sensitive heat flux QH, and latent heat flux QE, which can be expressed as:
where C is the heat capacity of the upper ocean, up to the depth of interest.
The latent heat flux QE can be further decomposed into an atmospheric forcing term QEa and an oceanic response term QEo,
The former is mostly sensitive to the surface wind anomaly (W′), and the latter serves as a Newtonian damping with respect to the ocean temperature change (T′). QEr refers to the residuals of the atmospheric forcing related to the relative humidity and stability effect and serves as a second-order factor6 in this study.
When the surface wind is reduced, according to the bulk formula, the evaporation will be suppressed. This effect thus increases the latent heat flux from the atmosphere to the ocean, warming the sea surface5.