Oxygen is one of the most important chemical elements in the ocean, as marine organisms ranging from microorganisms to vertebrates use it for respiration. Its concentration is regulated by both circulation and biogeochemical processes. In general, high-latitude regions are characterized by high oxygen concentrations, whereas subsurface tropical regions are poorly oxygenated. Particularly in the eastern parts of the tropical oceans, the large export of organic material out of productive surface layers combined with sluggish circulation depletes oxygen levels at depth, resulting in the formation of large suboxic regions where the oxygen concentration falls below 20 mmol m-3 (Karstensen et al., 2008). The overall contrast between high latitudes and tropical shadow zones is generally well reproduced by ocean models (Bopp et al., 2013; Cocco et al., 2013; Cabré et al., 2015; Shigemitsu et al., 2017).

The temporal variability of oxygen concentrations in the ocean interior is, nevertheless, still poorly understood; this is particularly true for the eastern tropical Pacific Ocean, where oxygen concentrations are among the lowest in the world (Karstensen et al., 2008). A strong decrease in oxygen concentrations in these regions has been inferred from observations ranging from the 1960s to the 2000s (Stramma et al., 2008, 2012; Schmidtko et al., 2017). This deoxygenation is not correctly reproduced by current ocean models (Oschlies et al., 2017). It is not clear whether the changes in oxygen concentration are due to anthropogenic changes (Matear et al., 2003; Long et al., 2016; Ito et al., 2017), or are related to natural low-frequency climate oscillations such as the Pacific Decadal Oscillation (PDO) (Deutsch et al., 2014), which is a robust and recurring pattern of ocean atmosphere climate variability centered over the midlatitude north Pacific basin (Mantua et al., 1997).

The stronger trade winds that occur during a negative phase of the PDO cause a shoaling of the eastern thermocline of the tropical and subtropical Pacific Ocean (Miller et al., 1994) (see the Supplement Fig. S1 for an overview of the mechanisms controlling the oxygen levels in the eastern tropical Pacific Ocean). Deutsch et al. (2011, 2014) showed that the depth of the thermocline regulates the oxygen levels in the northeastern subtropical Pacific Ocean. In these regions, a shallower thermocline fosters low oxygen concentrations in the intermediate ocean as a larger amount of organic material is respired below the mixed layer (Deutsch et al., 2011). Simultaneously, in the tropical regions the zonal volume transport by the equatorial current system increases during a negative PDO event as does the meridional transport by the subtropical cells (STCs) (Hong et al., 2014), which connect the subtropics to the tropics (McCreary and Lu, 1994). The variability of the strength of the STCs forces the variability of the oxygen transport in the upper thermocline of the equatorial Pacific Ocean (Duteil et al., 2014a). Therefore, competition takes place between the increased oxygen transport and the increased respiration as primary production is enhanced by the increased nutrient supply. Finally, stronger trade winds also increase the subduction volume of the North Pacific Eastern Subtropical Mode Water (Qu et al., 2009) and the South Pacific Eastern Subtropical Mode Water (Luo et al., 2011). A subtropical increase in productivity related to an increase in the trade winds causes a negative oxygen anomaly in these mode waters, which is transported equatorward and leads to a delayed oxygen decrease in tropical regions as shown by Ridder and England (2014) in an Earth system model of intermediate complexity.

All of these studies highlight the potential effect of the PDO on oxygen concentrations and show that the processes at play are diverse and strongly region dependent. However, we still do not have a clear picture of the impact of the PDO on the suboxic regions and on the oxygen levels of the eastern Pacific Ocean. Indeed, the studies cited above focus either specifically on the suboxic regions of the northeastern Pacific region (Deutsch et al., 2011, 2014) or on the upper thermocline of the tropical Pacific Ocean (Duteil et al., 2014a). A caveat of the model used by Ridder and England (2014) is that the equatorial undercurrent (EUC) is poorly represented, as is the case for most coarse-resolution (lower than 0.5∘ at the equator) models (Karnauskas et al., 2012). The EUC indeed ventilates the suboxic regions, as shown by Cabre et al. (2015) and Shigemitsu et al. (2017) in a range of models as part of the Coupled Model Intercomparison Project 5 (CMIP5).

A precise understanding of the role of the PDO in setting oxygen levels is difficult in “traditional” 50-year hindcasts experiments (Deutsch et al., 2011; Ito and Deutsch, 2013; Duteil et al., 2014a) due to the superposition of a long-term climate trend and higher-frequency climate oscillations such as the El Niño–Southern Oscillation (ENSO) (Ito and Deutsch, 2013, 2016; Eddebbar et al., 2017). Ito and Deutsch (2013) showed that the temporal spectrum of oxygen concentrations in the northeastern tropical Pacific Ocean is characterized by a strong decadal variance which may partly arise from the “reddening” of the variability spectrum of the physical and biological drivers (Ito and Deutsch, 2010), suggesting, e.g., a contribution of ENSO to the decadal oxygen variability, in addition to the PDO.

Rather than performing a “hindcast” experiment, in this study we specifically isolate and assess the role of the phase of the PDO on the oxygen levels in the tropical eastern Pacific Ocean by forcing a coupled circulation–biogeochemical model using “typical” conditions characteristic of the negative and positive PDO phases. These idealized atmospheric forcings are derived from monthly averages of realistic winds and heat fluxes from the 1948–2007 COREv2 dataset (Large and Yeager, 2009). Our aim is to understand whether and by which processes (oxygen advection, diffusion, respiration) the PDO may be responsible for the observed oxygen decline in the suboxic regions of the tropical eastern Pacific Ocean.

This paper is organized as follows. Section 2 describes the construction of the “typical” PDO forcings and the configuration of the experiments that we perform. In Sect. 3, we assess the basin-scale circulation of our experiments. In Sect. 4, we present the difference in oxygen levels between a “typical” positive and a “typical” negative PDO phase. The mechanisms regulating the oxygen levels are described in Sect. 5. Temporal aspects are discussed in Sect. 6. In Sect. 7 we discuss the changes in the upwelling systems and the impact on suboxia. We then summarize our results in Sect. 8.

The NEMO ocean model version v3.6 (Madec, 2008) was used in the standard ORCA2 configuration. This configuration has been widely used in previous studies and constitutes the ocean component of the ISPL-CM5A model, which is part of the Coupled Model Intercomparison Project (CMIP5) effort (Dufresnes et al., 2013). Its zonal resolution is 2 degrees. The mean meridional resolution is 2 degrees outside the tropics and increases to 0.5 degrees close to the Equator. The resolution of ORCA2 is sufficient to realistically reproduce the EUC (Cravatte et al., 2007) and the subtropical–tropical connectivity (Luebbecke et al., 2008). The circulation model has been coupled to a six-compartment (nutrient, phytoplankton, zooplankton, particulates and dissolved detritus, oxygen) biogeochemical model. This model is described in detail by Kriest et al. (2010). It was adapted to the NEMO framework by Duteil et al. (2014a, b).

We constructed three atmospheric forcing datasets derived from the interannual 1948–2007 COREv2, 6 h temporal resolution, forcing dataset (Large and Yeager, 2009):

MEAN: (1) a low-pass filter was applied to remove the frequencies with a period shorter than 1 month. (2) The long term trend 1948–2007 (Yang et al., 2016) was removed. (3) The corresponding time steps of the individual annual forcings for the 1948–2007 period were averaged to reconstruct a climatological annually cycling forcing set. The difference between MEAN and the COREv2 “normal year” is the absence of high-frequency variability (<1 month) and the removal of the long-term trend implicitly contained in the “normal year”.

In PDO_PLUS the zonal wind speed decreases by about 0.2 to 0.5 ms-1 compared to MEAN in the mid-equatorial Pacific Ocean, where the winds are strongest (at least 8 ms-1) (Fig. 1b). It increases close to the eastern coast by up to 0.3 ms-1, where the winds are weaker (2 to 8 ms-1). A similar pattern was described by both Merrifield et al. (2012) and Zhou et al. (2017). The meridional wind speed decreases by about 0.2 ms-1 (Fig. 1c). The 10 m air temperature increases by 0.1 to 0.3 ∘C in the eastern Pacific Ocean and decreases in the gyres (Fig. 1d). PDO_MINUS presents the opposite pattern.

We spun up the model over 1000 years using the MEAN forcing dataset. We subsequently performed two experiments using the PDO_PLUS and PDO_MINUS forcing datasets, respectively, which we integrated for a period of 50 years: this integration period corresponds to the typical oscillation period of the PDO over the past 200 years (Mc Donald and Case, 2005).

The oxygen level below the euphotic zone is determined by the balance between consumption (respiration) and supply (transport). The supply is decomposed into advective, diapycnal and isopycnal diffusion terms. The analysis is based on the average of the 50 years of integration time.

As previously seen, changes in respiration play a significant role in setting the oxygen levels. Respiration either compensates for the changes in supply in the equatorial region or acts in synergy with the decrease in supply to deplete oxygen in the northern part of SUB20 in the PDO_PLUS experiment. While the changes in the oxygen supply and transport are primarily linked to changes in the structure of the interior ocean, the change in respiration is principally linked to the surface and upper thermocline productivity, which ultimately depends on the supply of nutrients to the mixed layer.

In this study we tested whether the PDO impacts the oxygen concentration in the eastern part of the Pacific Ocean. We use the forced ocean model NEMO coupled to a simple NPZD (Nutrient Phytoplankton Zooplankton Detritus) model. After spin-up, the NEMO-NPZD model was forced by “typical” PDO positive (experiment PDO_PLUS) and negative (experiment PDO_MINUS) conditions derived from the COREv2 atmospheric forcings. A PDO-positive phase is characterized by an approximate 5 % to 10 % decrease in the zonal and meridional wind speed over the Pacific Ocean, while the sea surface temperature increases by 0.2 ∘C. A PDO-negative event shows the opposite pattern. In agreement with observations (McPhaden and Zhang, 2002), the circulation of the tropical Pacific Ocean is more sluggish (by 5 % to 10 %) in the PDO_PLUS experiment compared with the PDO_MINUS experiment. After 50 years of integration time, the volume of the suboxic regions (oxygen lower then 20 mmol m-3) is 7 % larger and the oxygen concentration decreases by 5 % to 50 % in the suboxic regions in PDO_PLUS compared with PDO_MINUS.

Recent analyses of the observational datasets showed that the oxygen concentration has decreased by about 5 % in the eastern equatorial Pacific Ocean region since 1960 (Schmidtko et al., 2017; Ito et al., 2017). We show here that a PDO_PLUS PDO event lasting for 50 years may impact the suboxic region by at least a similar order of magnitude. The shift from a PDO-negative phase (prevailing conditions before 1975) to a PDO positive phase (prevailing conditions since 1975) may therefore explain a significant percentage of the large deoxygenation that has occurred over the last few decades.

The simulated suboxic regions are divided into equatorial (10∘ N–10∘ S), northern (10–30∘ N) and southern (10–30∘ S) parts. The oxygen levels of each subregion are constrained by different processes. In the equatorial part, the oxygen levels are set by advective processes (Cabre et al., 2015; Shigemitsu et al., 2017). In the PDO_PLUS experiment, the slowing-down of the equatorial current system (and more particularly of the equatorial undercurrent) decreases the supply of oxygen (Fig. 8). Simultaneously, the supply of nutrients decreases, leading to a decrease in productivity and respiration. In the eastern part of the basin (in the suboxic regions), the decrease of supply dominates the change in respiration leading to a net oxygen decrease. Inversely, the change in respiration is dominant in the mid-Pacific Ocean, highlighting the importance of the parameterization of the biogeochemical processes (Kriest et al., 2010; Kriest and Oschlies, 2015) and more particularly the response of the phytoplankton growth to a change in nutrient concentration.

The southern part of the suboxic regions is mostly constrained by isopycnal diffusion processes (Gnanadesikan et al., 2012). In the PDO_PLUS experiment, the supply of oxygen by isopycnal processes from the equatorial regions and the subtropical gyres to the tropics decreases compared to PDO_MINUS. In the northern part, the role of diapycnal processes is dominant. In the PDO_PLUS experiment, the oxygen decrease is caused by changes in the upwelling system, stimulating productivity and respiration.

The general slowdown of the large-scale circulation caused by a decrease in the intensity of the trade winds is not reflected in the upwelling strength (Narayan et al., 2010), which is mainly forced by local processes. The strength and seasonality of the upwelling constrain the amount of productivity and the subsequent respiratory oxygen consumption in the suboxic regions. A shift in the upwelling seasonality explains the larger respiration in PDO_PLUS compared with PDO_MINUS in the northern part of the suboxic region. This shift is potentially linked to the NPGO (DiLorenzo et al., 2008), a decadal climate oscillation which has a signature that is implicitly contained in the “typical” PDO forcing. Using two supplementary experiments where the PDO_PLUS and PDO_MINUS forcings are only applied in the mid-Pacific Ocean above the tropical Pacific low-oxygen areas, we highlighted that a change in the large-scale wind pattern constrains changes in the equatorial and southern oxygen levels, while oxygen changes in the northern part are constrained by local processes.

One of the largest limitations of our study is that the idealized PDO forcings were prepared using a relatively short period extending from 1948 to 2007 (the COREv2 dataset). The implicit contribution of statistically independent climate oscillations such as the NPGO (DiLorenzo et al., 2008) can therefore not be completely ruled out.

Finally, another limitation is the resolution of the model. The role of mesoscale activity has been previously demonstrated in the supply of oxygen to the suboxic regions of the eastern Pacific Ocean (Montes et al., 2014; Bettencourt et al., 2015; Vergara et al., 2016). The relative increase in the isopycnal diffusive supply in a PDO-positive phase suggests the importance of the role of mesoscale activity during this period.

Our study suggests that the shift from a prolongated (multi-decadal) negative to a prolongated positive PDO phase is accompanied by a decrease in the oxygen levels in the eastern tropical Pacific Ocean. Several multi-decadal PDO shifts have occurred in the last century. For instance, the period from 1943 to 1976 is characterized by a negative PDO phase, while a positive PDO phase occurred from 1977 to 2011 (Fig. 1a). Such shifts have occurred roughly every few decades during the last 1000 years, as shown by PDO reconstructions (MacDonald and Case, 2005). An open question is whether the succession of these alternate shifts may cause a change in the oxygen concentration on centennial to millennial timescales.

Our study can be compared to the study of Deutsch et al. (2011) (hereafter referred to as D2011). Based on a 1959–2005 hindcast experiment, D2011 showed that the global suboxic volume (O2 < 5 mmol m-3), 95 % of which is contained in the northeastern tropical Pacific Ocean (0 to 30∘ N, 140∘ W to the coast), is controlled by the depth of the thermocline that constrains the productivity and the amount of oxygen respired around the suboxic volume. The depth of the thermocline is strongly related to the strength of the PDO, which explains 24 % of the variability of the suboxic volume in the hindcast simulation of D2011. In D2011, a PDO-negative phase was characterized by a large extent of the suboxic regions resulting from enhanced respiration with only negligible effects of changes in oxygen supply. Our results differ from D2011 because the increase in the respiration in a PDO-negative phase (higher primary production) is more than compensated for by an increase in the oxygen supply due to advective/diffusive processes, which ultimately leads to elevated oxygen levels and smaller suboxic regions in a PDO-negative state. Considering that the models used by D2011 and those used in our study are similar with similar grid resolution, the balance between changes in the oxygen consumption (dominant effect in D2011) and the transport (dominant effect in our study) depends on (i) the response of primary productivity and export production to a nutrient increase, and (ii) the depth of the suboxic regions, as respiration changes are stronger, and more difficult to compensate for by oxygen supply changes, at shallower than at greater depths. While the biogeochemical model used in our study contains fully prognostic nutrient, phytoplankton, zooplankton and detritus fields, D2011 used a simple restoring model that diagnosed export production from restoring simulated against observed surface phosphate concentrations. Therefore, the model from D2011 did not account for possible PDO-driven changes in surface nutrient concentrations and instead likely overestimated the variability in export and respiration at depth.

While our modeling framework captures the general patterns of primary and export production reasonably well, it does not include an iron cycle; thus, despite displaying a well-tuned mean state, it may exhibit systematic errors regarding its sensitivity to environmental changes. Furthermore, the model's representation of the respiration processes is relatively pragmatic. In particular, our model lacks an explicit nitrogen cycle including anaerobic remineralization by denitrification under low oxygen conditions (Paulmier et al., 2009). Other limitations include, for instance, a simplistic parameterization of the attenuation of the flux of particulate organic matter which, in our model, neglects any dependence on temperature or oxygen (Laufkötter et al., 2017). Also not considered in the model is the diel vertical migration of zooplankton that actively transports material into the deep ocean (Bianchi et al., 2013). Anthropogenic activities impact the global biogeochemical cycles. In particular, atmospheric deposition of anthropogenic nitrogen and iron may partially relax the iron limitation in the tropical Pacific Ocean (Ito et al., 2016). Industrial fishing may affect the mortality rate of the zooplankton and possibly feed back on productivity and respiration (Getzlaff and Oschlies, 2017). Each of these `missing' processes may modulate respiration rates and may possibly be correlated with the state of the PDO. An important result of our study is that the PDO-induced changes in respiration are smaller than the PDO-induced changes in oxygen supply by a few percent in most of the eastern tropical Pacific Ocean. We show that this small imbalance integrated over a few decades results in a significant PDO-driven oxygen anomaly that may explain a large part of the observed oxygen decline over the past few decades in this region. Experiments including different biogeochemical parameterizations and processes will need to be performed to better assess the robustness of our results.