The monthly mean SST data from the Optimum Interpolation SST version 2.1 OI-SST, produced by the Physical Sciences Laboratory of the National Oceanic and Atmospheric Administration (NOAA), are used as a reference dataset to validate the seasonal cycle of SST and of SST variability of the control run (introduced in the next section) and to compare with the ones from the different sensitivity experiments. These data are available at 0.25° horizontal resolution from September 1981 to present day. Monthly means of Chl a, derived from the daily gap-free product Copernicus-GlobColour , are used to compute the linear trend in Chl a pointwise in the tropical Atlantic for the period 1998–2024. This product is available at 4 km horizontal resolution from 4 September 1997 to present day.

To investigate the effect of Chl a on the equatorial Atlantic Ocean mean-state and its interannual variability, the NOAA-Geophysical Fluid Dynamics Laboratory Modular Ocean Model version 5 MOM5 was used. MOM5 is a free-surface primitive-equation model and uses z* rescaled geopotential coordinate. The model configuration for the control run (ICTP-MOM5-CTRL) corresponds to: 1° horizontal resolution with 50 vertical levels, subgrid mesoscale processes are parametrized with the Gent–McWilliams skew-flux closure scheme and submesoscale eddy fluxes according to . Vertical mixing is represented with a K-profile parameterization . The control run was integrated from January 1958 to December 2021 using the JRA-55-based surface dataset for driving ocean-sea-ice models JRA55-do. Additionally, ICTP-MOM5-CTRL includes a monthly climatology of Chl a, which is based on 8 d composites of Sea-viewing Wide Field-of-view Sensor (SeaWiFS) images taken from 1999 to 2001. This is the baseline climatology used across all sensitivity experiments described in the following. The climatological mean field of this dataset is shown in Fig. a. More details on the production of this climatology can be found in .

A caveat of forced ocean-sea ice configurations is the lack of two-way ocean–atmosphere coupling. In ICTP-MOM5, the ocean model is forced with the time-evolving JRA55-do atmospheric state, including winds at 10 m, shortwave and longwave heat fluxes, and near-surface (2 m) atmospheric temperature and humidity, which are used to calculate latent and sensible heat fluxes through bulk formulae. This forcing enables the model to realistically reproduce historical oceanic conditions over the simulation period. Additionally, the absence of SST restoring, together with the weak sea surface salinity restoring, prevent the artificial damping of intrinsic ocean variability. Nevertheless, prescribing the near-surface atmospheric state imposes some constraints on the simulated SST through the turbulent heat fluxes. Despite this limitation, showed that the impact of Chl a on the eastern tropical Pacific SST is qualitatively consistent between ocean-only and fully-coupled model experiments. Their results indicate that atmosphere-ocean coupling mostly amplifies the Chl a-induced mean change through a positive atmosphere feedback, but does not change the sign of the response. Moreover, the simulated response of ENSO in both fully-coupled and ocean-only model experiments was consistent . These findings suggest that forced ocean simulations provide a robust framework for investigating the influence of Chl a on the equatorial Atlantic variability. Yet, we acknowledge that the magnitude of the response may be lower than if a fully coupled system was considered.

In ICTP-MOM5-CTRL, the effect of the presence of phytoplankton on the light absorption is based on Chl a following . The total surface irradiance, I0, is split into three wavelength bands: the infrared (Iir), red visible (Ired), and blue/green visible (Ibg) with a light partitioning of Iir = 0.58 × I0, Ired = 0.21 × I0, and Ibg = 0.21 × I0, respectively. This leads to a shortwave penetration at depth z (I(z)) of: 1I(z)=Iir×e-kirz+Ired×e-kredz+Ibg×e-kbgz where kir, kred, and kbg are light attenuation coefficients. Following , kir = 2.86 m-1, whereas kred and kbg depend on Chl a as follows : 2kred=0.225+0.037×[Chl a]0.6293kbg=0.0232+0.074×[Chl a]0.674

found only small differences in SST between forced ocean simulations using depth-independent and depth-dependent profiles of Chl a. Therefore, for simplicity and consistency with previous studies, we have used depth-independent profiles of Chl a in all model simulations in this study. However, we note that this represents a limitation because the maximum Chl a concentration is usually found close to the nutricline in the tropical Atlantic Ocean .

To highlight the bio-physical effect on the equatorial Atlantic Ocean mean-state and interannual variability, we performed four sensitivity experiments identical to ICTP-MOM5-CTRL (see Sect. ), except for the prescribed monthly climatology of Chl a (Table ). In ICTP-MOM5-CHL0.01, the monthly climatology of Chl a is multiplied by 0.01 and aims at simulating a “clear-ocean”, i.e. without Chl a. In ICTP-MOM5-CHL0.5, the monthly climatology of Chl a is reduced by 50 %. In ICTP-MOM5-CHL1.5, it is increased by 50 %, while in ICTP-MOM5-CHL2.0, the monthly climatology is doubled. For clarity, the prescribed monthly climatologies of Chl a averaged over the ATL3 region are shown in Fig. b. As defined in Eqs. () and (), the use of these different climatologies of Chl a affects the e-folding depth scale of the light penetration in the visible red (1/kred) and in the visible blue/green wavelength bands (1/kbg; Fig. c) while the infrared (1/kir) is independent. The effect of Chl a is limited for 1/kred, as the e-folding scale is approximately 4 m in all simulations throughout the year (Fig. c). In contrast, the effect of Chl a is important for 1/kbg. In ICTP-MOM5-CHL2.0, the ATL3 seasonal cycle of 1/kbg varies around 14 m, whereas in ICTP-MOM5-CHL0.01 it varies around 40 m. In the following, we examine the biological effect by focusing on the difference between the low Chl a simulation (ICTP-MOM5-CHL0.01) and the control run (ICTP-MOM5-CTRL). Differences between the other runs and ICTP-MOM5-CTRL are documented in the appendix.

To obtain the detrended monthly mean anomalies, the monthly mean model outputs (and observational data, where applicable) are first linearly detrended point-wise over the study period January 1982–December 2020. Then, the monthly climatology of any variable is computed from the detrended data. Finally, the monthly climatology is subtracted to the detrended data to obtain the detrended monthly mean anomalies. Further, a 3-month centered running mean is applied to the detrended anomalies to attenuate the high-frequency variability. In order to preserve the length of the time series, the first and last values of the 3-month centered running mean are based on only two data points. The main regions of interest are the ATL3 region (3° S–3° N, 20° W–0°, red box on Fig. a) and the equatorial Atlantic (3° S–3° N, 40° W–10° E, blue box on Fig. a) where Atlantic Niños/Niñas occur. The maximum vertical temperature gradient (max⁡(∂T/∂z)) is used as a proxy for the internal thermocline depth. The mixed-layer depth (MLD) is diagnosed through a density threshold criterion of 0.03 kgm-3 increase from the reference value of surface potential density taken at 5 m depth .

To evaluate whether the variability (i.e. standard deviation of anomalies) or the mean state (time mean) differ between two simulations, we apply a nonparametric bootstrap method independently at each grid point. For the mean-state analysis, the statistic of interest is the difference between the time means of the two simulations, while for the variability analysis it is the difference between their standard deviations. Uncertainty in these differences is estimated using bootstrap resampling with replacement. For each simulation, the time indices are resampled independently 5000 times, and the statistics of interest are computed for each resampled dataset, resulting in a bootstrap distribution of the difference. From this distribution, we estimate a two-sided confidence interval using the 0.5th and 99.5th percentiles. Differences are considered statistically different from zero at the 99 % confidence level when zero falls outside this interval.

The response of the equatorial Atlantic Ocean mean-state to different monthly climatology of Chl a is investigated by comparing ICTP-MOM5-CTRL against the different sensitivity experiments introduced in Sect. . Relative to ICTP-MOM5-CTRL, equatorial Atlantic SSTs in ICTP-MOM5-CHL0.01 are significantly warmer between 2° S and 2° N and from 35° W to 2° E (Fig. a). In the ATL3 region, the climatological mean SST difference between the two simulations amounts to 0.15 °C. We also note a surface warming along the Angolan and Namibian coasts as well as in the Senegalo–Mauritanian upwelling system (Fig. a), which is consistent with results from . The ocean temperature differences between these two simulations are not restricted to the surface, as shown by the temperature section difference in the upper 250 m of the equatorial Atlantic (Fig. b). Compared to ICTP-MOM5-CTRL, ICTP-MOM5-CHL0.01 reveals a strong and statistically significant warming exceeding 2 °C within 50 m around the mean thermocline depth (Fig. b). This subsurface warming in the “clear-ocean” simulation leads to a deepening of the main oceanic thermocline and a weakening of the vertical temperature gradient around the main thermocline (Fig. c). A similar response is found when comparing ICTP-MOM5-CHL0.5 to ICTP-MOM5-CTRL (Fig. A1a–c), although with a smaller magnitude (Fig. d–f). Conversely, increasing the Chl a by 50 % and 100 % leads to a slight, though not statistically significant, SST cooling (Fig. g and j). Furthermore, the equatorial Atlantic Ocean temperature and its vertical gradient in ICTP-MOM5-CHL1.5 and ICTP-MOM5-CHL2.0 only show moderate changes, characterized by subsurface cooling, enhanced vertical temperature gradient around the thermocline, and a slight shoaling of both the thermocline and MLD (Fig. g–l).

The eastern equatorial Atlantic SST is characterized by a pronounced seasonal cycle, with maximum values in February–March–April (FMA), when SSTs exceed 28 °C, and minimum values in July–August–September (JAS), when the Atlantic cold tongue develops and SSTs reach 25 °C or lower . The amplitude of the seasonal cycle of SST is affected by the changes in vertical distribution of shortwave heating induced by the concentration of Chl a (Fig. d). Compared to ICTP-MOM5-CTRL, the amplitude of the ATL3 SST seasonal cycle (defined as the difference between the FMA and JAS seasonal means) is reduced by 14 % in ICTP-MOM5-CHL0.01, from 3.42 to 2.93 °C (Fig. d). This reduction mainly results from warmer SSTs in JAS (Fig. d), when the Chl a usually peaks in the ATL3 region (Fig. b). ICTP-MOM5-CHL0.5 also simulates a reduction where the ATL3 SST seasonal cycle amplitude decreases by 4 % relative to ICTP-MOM5-CTRL. On the contrary, ICTP-MOM5-CHL1.5 and ICTP-MOM5-CHL2.0 show only slight increases in the ATL3 SST seasonal cycle amplitude of 3 % and 1 %, respectively.

Comparing the ATL3 SST seasonal cycle from both control and sensitivity simulations to the one derived from OI-SST shows that ICTP-MOM5 simulations are able to capture the main features of the seasonal cycle of SST in the ATL3 region (Fig. d). However, a year-round warm SST bias remains, which is common in ocean and climate models in this region . We note that the idealized increase in Chl a leads to a reduction of the SST bias, particularly in JAS (Fig. d). However, this does not necessarily imply that the SST bias in this region results from an incorrect representation of Chl a, but rather highlights the sensitivity of SST to phytoplankton-induced changes in the vertical distribution of shortwave heating.

The surface warming in the equatorial Atlantic in the absence of Chl a is somewhat counterintuitive, as a reduced absorption of the shortwave heat flux near the surface would be expected to lead to local surface cooling, as observed in the open ocean and away from the equator (Fig. a). Such a counterintuitive response appears to be related to upwelling regions, as it has already been observed in the equatorial Pacific upwelling and in the Senegalo–Mauritanian and Benguela upwelling systems . We note that in most previous studies, the bio-physical effect was highlighted by showing the difference between the control run and the experiment (control minus experiment), whereas in this study the difference shown is experiment minus control. The dominant mechanism explaining the surface cooling has been shown to be the vertical redistribution of shortwave heating by Chl a, which enhances upper-ocean stratification and shoals the mixed layer, strengthening total upper-ocean divergent transport by weakening the partially compensating equatorward geostrophic transport. This leads to enhanced equatorial upwelling across the mixed-layer base. Our findings are therefore consistent with this mechanism but exhibit anomalies of the opposite signs.

In ICTP-MOM5-CTRL, the meridional ocean velocity averaged over the MLD shows divergence close to the equator driven by trade winds (Fig. a), leading to equatorial upwelling (Fig. c). In ICTP-MOM5-CHL0.01, this meridional divergence is reduced (Fig. b), leading to weaker equatorial upwelling across the base of the mixed layer (Fig. d and e). To understand this response we examine the steady zonal momentum budget integrated over the mixed layer, neglecting Reynolds stresses and assuming that baroclinic pressure gradients associated with horizontal density variations are small relative to the barotropic pressure gradient associated with sea surface height variations : 4My=-τxρ0f+gDf∂η∂x where My is the meridional transport, D the mixed-layer depth, η the sea surface height, τx the zonal wind stress, f the Coriolis parameter, and ρ0 is a reference density. The meridional transport can be decomposed into a wind-driven Ekman transport (first term on the r.h.s. of Eq. ) and a compensating geostrophic transport associated with the zonal SSH gradient integrated over the mixed layer (second term on the r.h.s of Eq. ). Because ICTP-MOM5-CHL0.01 and ICTP-MOM5-CTRL are forced with identical atmospheric forcing (See Sect. ), the reduced poleward transports within the MLD cannot be attributed to changes in Ekman transport. Additionally, the identical trade winds push warm, less dense waters to the west, leading to comparable large-scale zonal SSH gradients in both simulations (not shown). Hence, as zonal SSH gradients remain similar, differences in meridional transports must arise from changes in the MLD.

Under identical wind forcing, a deeper mixed layer can thus strengthen the compensating geostrophic transport, thereby reducing the net meridional transport divergence and weakening equatorial upwelling across the base of the mixed layer. Consistent with this mechanism, ICTP-MOM5-CHL0.01 features deeper MLDs than ICTP-MOM5-CTRL (grey contours in Fig. b, dotted line in Fig. d, and shadings in Fig. a and b). We note that changes in MLD in other simulations are less important Fig. c–e. The deepening of the mixed layer is likely due to increased vertical mixing associated with Chl a induced changes in the vertical distribution of shortwave heating. Ultimately, the reduced divergence leads to weaker equatorial upwelling into the mixed layer (Fig. d and e) and surface warming in the “clear-ocean” simulation (Fig. a).

Although outside the scope of this paper, we note that Chl a seems to exert a non-trivial effect on the vertical structure of the tropical cell. Indeed, the maximum vertical velocity (around 100 m at 20° W; Fig. c) occurs at the interface of the upper and lower branches of the tropical cell . Above this level, the poleward Ekman velocity is greater than the zonal pressure gradient, giving rise to a net poleward velocity (as previously discussed). The depth of the upper branch of the tropical cell seems to be related to the depth at which irradiance drops to 1 % of its surface value (around 100–150 m), often used to define the base of the euphotic zone. Below this interface, the zonal pressure term dominates, resulting in an equatorward velocity and closing the cell. The upward shift of the maximum vertical velocity (Fig. d) suggests that Chl a reduces the depth of the interface between the upper and lower parts of the tropical cell, increasing poleward velocities above the interface and reducing equatorward velocities below. In fact, the depth of the maximum vertical velocity in the ATL3 region is 10 m shallower in ICTP-MOM5-CTRL (54 m) than in ICTP-MOM5-CHL0.01 (64 m). In addition, the magnitude of the maximum vertical velocity is reduced by 6.4 % in ICTP-MOM5-CHL0.01 (4.4 × 10⁻⁶ ms-1) relative to ICTP-MOM5-CTRL (4.7 × 10⁻⁶ ms-1). We also observe that Chl a strengthens and shoals the core of the equatorial undercurrent (Fig. ). An in-depth analysis of whether velocities are modified by the altered vertical extent of each branch or partly due to a change in the total transport is left for a future study.

In addition to modifying the ocean mean-state, the presence of Chl a can also affect interannual SST variability. showed that, in the equatorial Pacific Ocean, forced ocean model simulations without, or with reduced, Chl a simulated weaker interannual SST variability than simulations with realistic Chl a levels. In the following, we assess the effect of different levels of Chl a on the equatorial Atlantic interannual SST variability (Figs.  and ).

ICTP-MOM5-CHL0.01 shows a statistically significant reduction in interannual SST variability in the central equatorial Atlantic, particularly between 30° W and 0° E (Fig. a). ICTP-MOM5-CHL0.5 (ICTP-MOM5-CHL1.5 and ICTP-MOM5-CHL2.0) also exhibits reduced (enhanced) interannual SST variability in the central equatorial Atlantic. However, these changes are not significant at the 99 % level (Fig. ).

The reduction in SST variability is also evident in the time series of the ATL3-averaged SST anomalies (Fig. c), with standard deviations of 0.31 °C for ICTP-MOM5-CTRL and of 0.27 °C for ICTP-MOM5-CHL0.01, corresponding to a 12.9 % reduction. We note that the interannual ATL3 SST variability in ICTP-MOM5-CTRL is underestimated compared to that of OI-SST, which has a standard deviation of the ATL3 SST anomalies of 0.38 °C over the period 1982–2020. This underestimation in the ocean model is likely due to the atmospheric forcing (JRA55-do), as shown by .

The effect of Chl a on the interannual temperature variability is not restricted to the surface. The ocean temperature variability in the upper 150 m of the equatorial Atlantic (3° S–3° N, 40° W–10° E) is also significantly reduced in ICTP-MOM5-CHL0.01 (Fig. b). As all simulations are forced with identical winds, thermocline depth variations are similar across simulations (Fig. ) and therefore cannot explain the changes in temperature variability. Instead, the reduced subsurface temperature variability in the vicinity of the thermocline is explained by the weakened vertical temperature gradient around the thermocline in ICTP-MOM5-CHL0.01 (Fig. c). Additionally, weakened equatorial upwelling across the basis of the mixed layer (Fig. d) also contributes to the reduced interannual SST variability in ICTP-MOM5-CHL0.01. Similar but weaker reductions are found when comparing ICTP-MOM5-CHL0.5 to ICTP-MOM5-CTRL (Fig. c and d). In contrast, and although these changes are not statistically significant, both ICTP-MOM5-CHL1.5 and ICTP-MOM5-CHL2.0 exhibit increased subsurface temperature variability in the equatorial Atlantic (Fig. e–h), consistent with the enhanced vertical temperature gradient in those simulations (Fig. i and l).

Next, we examine the response of the interannual ATL3 SST variability to varied levels of Chl a. The response is strongest in May–June–July and November–December, which correspond to the peak seasons for the Atlantic Niño and Atlantic Niño II (Fig. a), respectively. The seasonal cycle of the ATL3 SST variability is well captured by ICTP-MOM5-CTRL, although with too weak variability throughout the year (Fig. a). The response of the interannual ATL3 SST variability in May–June–July to prescribed Chl a is non-linear, with a quadratic fit explaining 98 % of the variance (Fig. b). The ATL3 interannual SST variability in May–June–July is of 0.345 °C in ICTP-MOM5-CTRL. Relative to this value, ICTP-MOM5-CHL1.5 and ICTP-MOM5-CHL2.0 show small increases of 2.29 % and 3.76 %, respectively, while ICTP-MOM5-CHL0.5 and ICTP-MOM5-CHL0.01 depict decreases of 3.27 % and 13.47 %, respectively.

The fact that changes in the mean state of Chl a affect the interannual SST variability in the tropical Atlantic Ocean is of particular interest, as a merged satellite product of Chl a (GlobColour) shows marked trends over the period 1998–2024 (Fig. c). Similar trends have also been reported in recent studies . The timeseries of the GlobColour Chl a averaged over the ATL3 region reveals a clear reduction in the yearly maxima, while the yearly minima remain stable (Fig. a). Comparing the seasonal cycle of the merged satellite product of Chl a averaged over the ATL3 region during 1998–2011 with the one obtained over 2012–2024 reveals that the largest decrease occurs during July-August-September, the season of maximum Chl a (Fig. b). These ongoing changes in Chl a in the tropical Atlantic Ocean underscore the need to improve our understanding of the relationship between Chl a and interannual SST variability in the tropical Atlantic.