The first large ocean-going oceanographic expeditions discovered OMZs in the Pacific, Atlantic, and Indian Ocean between the end of the 19th and the first third of the 20 century (Sewell and Fage, 1948, and references therein). Their occurrence was explained by the consumption of oxygen during the respiration of organic matter exported from the sunlit surface ocean and a sluggish horizontal renewal of water within the OMZ (Dietrich, 1936; Seiwell, 1937). Due to the main respiration depth of exported organic matter, OMZs mostly develop at water depths between approximately 100 and 1000 m (Suess, 1980), and oxygen concentrations within OMZs generally decrease with an increasing age of the water mass within the OMZ (Karstensen et al., 2008).

Sverdrup (1938) presented the first OMZ model showing that oxygen concentrations within the OMZ represent the balance between biological oxygen consumption and oxygen supply. Primary production and fluxes of oxygen across the air–sea interface are the sources of dissolved oxygen in surface waters. Vertical mixing and subduction of oxygen-enriched surface waters during the deep and mode water formation at high latitudes are in turn the main processes ventilating the interior of the ocean (McCartney, 1977; Sverdrup, 1938). Accordingly, Broecker and Peng (1982) introduced a model in which upwelling of oxygen-enriched deep water at the lower boundary and vertical mixing at the upper boundary serve to ventilate the OMZ, whereas the respiration of exported organic matter decreases oxygen concentrations.

More recent studies emphasize the influence of mesoscale eddies on the development of the OMZ (e.g., Chelton et al., 2011; Fassbender et al., 2018; Lachkar et al., 2016; Oschlies and Garcon, 1998; Resplandy et al., 2019). Mesoscale eddies emerge from baroclinic and barotropic instabilities related to the shear of horizontal currents and affect the vertical and lateral transport of water. This results in a patchiness of environmental conditions with complex and non-linear impacts on the OMZ (Fassbender et al., 2018; McGillicuddy, 2016). For instance, upward movements of water can increase the biological oxygen consumption by increasing nutrient inputs into the surface waters and thereby the biological production. Conversely, downward water movements could lower the biological production (Gruber et al., 2011) and increase oxygen concentrations in the OMZ additionally by increasing the supply of oxygen-enriched surface waters into the OMZ. In particular, stirring of oxygen by eddies along isopycnal surfaces has been suggested to modulate the intensity and distribution of low-oxygen waters in the ocean (Gnanadesikan et al., 2012, 2013). In the eastern tropical Atlantic and Pacific Ocean, recent work has highlighted the role of eddies in enhancing ocean mixing in regions of sluggish large-scale circulation, thus contributing to the ventilation of OMZs located there (Bettencourt et al., 2015; Brandt et al., 2015; Gnanadesikan et al., 2013). In this context, long-term changes in oxygen concentrations have been linked to changes in the intensity of eddy activity. For instance, Brandt et al. (2010) have shown that a reduction in filamentation and the strength of alternating zonal jets associated with mesoscale eddies between the periods 1972–1985 and 1999–2008 has contributed to a reduction in the ventilation of the OMZ located in the tropical north Atlantic. Eddy trapping in turn maintains properties of the trapped fluid over relatively long time periods. This favors the development of localized OMZs, which propagate laterally along with eddies as seen, e.g., in the open North Atlantic Ocean and off Peru in the Pacific Ocean (Bourbonnais et al., 2015; d'Ovidio et al., 2013; Fiedler et al., 2016; Karstensen et al., 2017; Schütte et al., 2016).

In the Atlantic Ocean and Pacific Ocean, hypoxic OMZs are associated with highly productive major eastern boundary current upwelling systems. In the Indian Ocean, the geographic setting prevents the development of such an upwelling system. However, a major monsoon-driven upwelling system emerges in the western Arabian Sea off the Arabian Peninsula, and a smaller one develops along the Indian southwest coast during the northern hemispheric summer (Fig. 3a). Initially described by Schott (1935), the upwelling system in the western Arabian Sea was subject to intense studies including the International Indian Ocean Expedition (IIOE) between 1959 and 1965 and the Joint Global Ocean Flux Study (JGOFS) with its field phase between 1994 and 1997 (e.g., Bauer et al., 1991; Brock et al., 1991; Bruce, 1974; Currie et al., 1973; Sastry and D'Souza, 1972; Wyrtki, 1973). However, contrary to expectations, the OMZ is most intense in the central and eastern Arabian Sea and not in the western Arabian Sea where the productivity is highest (Fig. 3c, Antoine et al., 1996; Naqvi, 1991). The offshore advection of upwelling-driven blooms, which increases the organic carbon export into the central Arabian Sea (Rixen et al., 2006), contributes to this eastward displacement of the OMZ, but monsoon-driven and seasonally varying physical oxygen supply mechanisms are assumed to be the main processes causing it.

Numerical model studies have shown that, on an annual timescale, mesoscale eddies and filaments dominate the vertical supply of oxygen to the OMZ and the lateral transport of ventilated waters into the central and northern Arabian Sea (Resplandy et al., 2011, 2012). In line with these numerical model studies and previous field work (Rixen and Ittekkot, 2005; Sen Gupta and Naqvi, 1984; Swallow, 1984), McCreary et al (2013) also highlighted the important role of vertical eddy mixing in the ventilation of the western Arabian Sea in addition to the inflow of oxygen-enriched Indian Ocean Central Water (ICW).

The ICW originates from convective mixing as Subantarctic Mode Water (SAMW) in the southern Indian Ocean (McCartney, 1977; Sverdrup et al., 1942). It enters the western Arabian Sea along with Timor Sea Water and the Subtropical Subsurface Water via the Somali Current (Schott and McCreary, 2001; Stramma et al., 1996; You, 1997). To a lesser extent, it is also carried into the eastern Arabian Sea along with an undercurrent, which compensates for the poleward-flowing West Indian Coastal Current (WICC, Fig. 4a, Schmidt et al., 2020; Shenoy et al., 2020; Shetye et al., 1990). Due to the much stronger inflow of ICW in the western Arabian Sea, the OMZ retreats eastwards in summer (Rixen et al., 2014). In winter the OMZ expands westwards due to weaker inflow of ICW and the seasonal reversal of the surface ocean circulation – counterclockwise during the winter monsoon and clockwise during the summer monsoon.

Data compiled by Acharya and Panigrahi (2016) indicate the response of the OMZ to the monsoon-driven seasonality in the Arabian Sea in more detail. These authors analyzed the Word Ocean Atlas 2013 (WOA13) and data obtained from the Global Ocean Data Analysis Project. Thereby, they took methodical biases into account, corrected the WOA13 data, and applied a 20 µM threshold to define the OMZ. Based on the corrected data set, these authors determined the areal extension and maximum thickness of the OMZ and calculated the mean oxygen concentration within the OMZ. According to their data, the OMZ has its lowest areal extension in summer (Fig. 5a) due to the inflow of ICW. However, since the shrinking horizontal extension of the OMZ is associated with a thickening, the OMZ also has its largest volume in summer, which in turn is accompanied by the lowest concentration of oxygen within the OMZ (Fig. 5b). The summer thickening of the OMZ is favored by the negative wind stress curl and the associated downwelling in the central Arabian Sea (Brock et al., 1991; Rao et al., 1989) and is coincident with the confluence of ICW that enters the Arabian Sea in the west and east as mentioned before. The low oxygen concentrations reflect in turn impacts of the enhanced upwelling-driven productivity and the associated increased oxygen consumption in the OMZ, which apparently exceeds the increased physical oxygen supply via eddy stirring and the inflow of ICW.

In order to test whether the monsoon controls the intensity of the OMZ via its impact on the biological production, satellite-derived primary production values were obtained from the ocean primary production website (http://www.science.oregonstate.edu/ocean.productivity/, last access: 15 August 2020). The data cover the period between 2002 and 2019 and were averaged seasonally for the Arabian Sea north of 10∘. In general, the obtained mean primary production rates were inversely linked to the OMZ oxygen concentrations (Fig. 6a), which supports the hypothesis that monsoon-driven changes in the biological carbon production and export exert a strong control on the intensity of the OMZ on a seasonal timescale. However, the autumn season appears to be an exception. During this season the OMZ oxygen concentrations were lower than expected from the low primary production rate, indicating an even lower physical oxygen supply. One explanation could be the strongly reduced inflow of ICW in combination with a low air–sea oxygen supply in response to the warming and the resulting enhanced stratification of surface waters after the upwelling season.

Similar to the Arabian Sea, upwelling-favorable winds also occur in the Bay of Bengal during the summer monsoon (Hood et al., 2017; Shetye and Shenoi, 1988). However, the productivity in the Bay of Bengal is lower than in the Arabian Sea and shows only a weakly pronounced seasonality (Fig. 3a, b). Nevertheless, sediment trap studies have shown that, despite a lower productivity, organic carbon fluxes into the deep Bay of Bengal are almost as high as those in the central and eastern Arabian Sea, due to a ballast effect associated with high loadings of lithogenic mineral material (Rao et al., 1994; Rixen et al., 2019b). Ballast minerals, supplied from land via rivers or as dust, protect organic matter against bacterial attacks by adsorbing organic molecules to atomic lattices (Armstrong et al., 2002) and accelerate the sinking speed of particles (Haake and Ittekkot, 1990; Hamm, 2002; Ramaswamy et al., 1991). Enhanced sinking speeds reduce respiration in shallower waters and thereby increase the flux of organic matter to deeper waters (Banse, 1990; Ittekkot, 1993). A stronger ballast effect, in addition to the lower primary production, was assumed to lower the oxygen consumption in the Bay of Bengal in comparison to that in the Arabian Sea OMZ (Al Azhar et al., 2017; Rao et al., 1994). High freshwater fluxes largely cause the lower oxygen consumption in the Bay of Bengal and also reduce the physical oxygen supply.

In addition to their role as supplier of ballast minerals, high freshwater fluxes from river runoff, but also from precipitation, form a buoyant low-salinity surface layer that isolates nutrient-enriched subsurface water and increases stratification (Kumar et al., 1996). The increased stratification weakens vertical mixing as well as upwelling and thereby the biological productivity and the physical oxygen supply. Furthermore, in addition to the reduced vertical oxygen supply, the inflow of oxygen-poor Arabian Sea Water also lowers the lateral oxygen supply into the Bay of Bengal OMZ. The inflow of Arabian Sea Water into the Bay of Bengal is marked by a broad salinity maximum that occurs below the low-salinity surface layer to a water depth of approximately 1000 m (Rao et al., 1994, and references therein).

The influence of eddies on the OMZ has been studied (Kumar et al., 2007; Prasanna Kumar et al., 2004; Sarma and Bhaskar, 2018; Sarma et al., 2018; Singh et al., 2015), but due to the pronounced interannual variability (Chen et al., 2012; Johnson et al., 2019) and the eddies' non-linear behavior, impacts of eddies on the Bay of Bengal OMZ are difficult to quantify. For instance, Sarma and Baskhar (2018) focused on anticyclonic eddies sampled by bio-Argo floats between 2012 and 2016 in the Bay of Bengal and showed that anticyclonic eddies are formed on the eastern side of the basin and propagate westward. They ventilate the layer between 150 and 300 m and weaken the OMZ. Such episodic injection of oxygen could be a mechanism that enhances the oxygen supply by reducing impacts of the strong stratification on vertical mixing. On the other hand, Sarma et al. (2018) showed that while anticyclonic eddies supply oxygen to the subsurface layer and hence weaken the OMZ, cyclonic eddies inject nutrients into the euphotic zone and thus enhance productivity and oxygen consumption at depth.

Today, the balance between physical oxygen supply and biological oxygen consumption is disturbed as indicated by the expansion of hypoxia, which is an increasingly common feature in coastal waters (Altieri et al., 2017; Diaz and Rosenberg, 2008). It is called the “spreading of dead zones” because their occurrence is often associated with mass mortality of fish and invertebrates (e.g., Weeks et al., 2002). Eutrophication and global warming mainly cause the spreading of dead zones by enhancing the production of organic matter and decreasing the oxygen supply due to a reduced solubility of oxygen in warmer surface waters. Since decreasing concentrations of dissolved oxygen have also been widely observed in the open ocean (i.e., beyond coastal systems) during the last 50 years, deoxygenation of the ocean is considered one of the main threats to pelagic ecosystems (Breitburg et al., 2018; Keeling et al., 2009; Schmidtko et al., 2017; Stramma et al., 2010a, b, 2008).

However, there are indications that the Arabian Sea and apparently also the Bay of Bengal OMZ was more intense in the recent past, i.e., prior to the Indian Ocean Expedition (IIOE, 1959–1965), than thereafter. For instance, Carruthers et al. (1959) described mass mortality of fish along the Arabian and Indian coast as well as in the central Arabian Sea at around 62.5∘ E and 9∘ N and identified oxygen depletion as the most likely trigger (Carruthers et al., 1959). This view was supported by a report on the occurrence of hydrogen sulfide from the northeastern Arabian Sea and off Oman at Ras al Hadd (Ivanenkov and Rozanov, 1961). Furthermore, hydrogen sulfide was also detected in the northwestern Bay of Bengal (Ivanenkov and Rozanov, 1961). These were the only reports on the occurrence of hydrogen sulfide in the northern Indian Ocean until Naqvi et al. (2000) discovered an anoxic event that developed along the western Indian coast off Mumbai in the late summer of 1999. Such strong events do not evolve every year (Gupta et al., 2016; Sudheesh et al., 2016), but their appearance shows that the spreading of dead zones in coastal regions does not spare the Indian shelf (Altieri et al., 2017; Diaz and Rosenberg, 2008; Diaz et al., 2019).

In contrast to these shelf processes, global, observation-based syntheses of OMZs (beyond continental shelves) reveal only a weak decrease in dissolved oxygen concentrations in the OMZs of the Arabian Sea and the Bay of Bengal in comparison to OMZs of the South Atlantic Ocean and the Pacific Ocean (Ito et al., 2017; Naqvi, 2019; Schmidtko et al., 2017; Stramma et al., 2008). The detailed analysis of all oxygen data available from the central Arabian Sea by Banse et al. (2014) ascribes this to opposing regional trends within the Arabian Sea. The authors analyzed oxygen data, which were measured between 1959 and 2004 in the Arabian Sea and in the depth range between 100 and 500 m. Biases caused by different analytical procedures were taken into account, and oxygen data were compiled for sub-regions within the Arabian Sea. The results showed that oxygen concentrations increased in the southern part of the Arabian Sea and declined in the central Arabian Sea. Follow-up studies also reported decreasing oxygen concentrations in the western and northern Arabian Sea (Piontkovski and Al-Oufi, 2015; Queste et al., 2018). In the northern Arabian Sea, dissolved oxygen concentrations in the surface mixed layer largely reflect the trend seen in the OMZ, as indicated by a compilation of dissolved oxygen data covering the period from the 1960s to 2010 (Gomes et al., 2014).

Since STOX data for the Arabian Sea were unavailable prior to 2007, the data compiled by Banse et al. (2014) do not resolve any changes in the oxygen concentrations below 0.09 µM (Fig. 2). Such low oxygen concentrations have recently been measured in the Arabian Sea and in the Bay of Bengal as well as in the OMZ of the eastern Pacific Ocean (Bristow et al., 2017; Jensen et al., 2011; Thamdrup et al., 2012). In the latter study, it was shown that nitrite accumulates in the water column at oxygen concentrations of < 0.05 µM. In the Arabian Sea OMZ, the accumulation of nitrite was first described during the John Murray expedition of 1933–1934 (Gilson, 1937). It is called the secondary nitrite maximum (SNM) and assumed to indicate active denitrification (Naqvi, 1991). The role of the SNM as an indicator of active denitrification is further supported by profiles of stable isotope ratios of nitrogen in nitrate (δ15NNO3) and nitrite (NO3-) concentration profiles (Fig. 7, Gaye et al., 2013; Rixen et al., 2014). Since denitrification increases δ15NNO3 in the water column due to the preferential uptake of the lighter 14NO3- (Cline and Kaplan, 1975; Mariotti et al., 1981), low nitrate concentrations correspond to high δ15NNO3 within the SNM. Assuming a similar response of the nitrogen cycle to low oxygen concentration in the Arabian Sea as in the eastern Pacific Ocean suggests that the Arabian Sea SNM is characterized by oxygen concentrations of about ∼ 0.05 µM and in turn that denitrification dominates the nitrogen cycle at such low oxygen concentrations. Nevertheless, an isotope tracer study indicated that the re-oxidation of nitrite to nitrate reduced the formation of N2 by 50 % to 60 % (Gaye et al., 2013), which implies an active competition between aerobic and anaerobic processes even at such low oxygen concentrations. However, since anaerobic denitrification dominates the competition, as indicated by the accumulation of nitrite, the depletion of nitrate, and the δ15NNO3 maxima, an oxygen concentration of ∼ 0.05 µM is seen as a threshold below which functional anoxia occurs (Fig. 2).

Naqvi (1991) was the first to use the SNM to map the spatial extent of functional anoxia in the Arabian Sea. His analysis was based on data obtained during the International Indian Ocean Expedition between 1959 and 1965 and cruises thereafter but did not include data from Ivanenkov and Rozanov (1961) and the JGOFS. The data of Ivanenkov and Rozanov (1961) indicate a more intense OMZ, including the occurrence of hydrogen sulfide as mentioned earlier and a larger extent of the SNM than calculated by Naqvi (1991), although there are doubts regarding the reliability of the older data (Sen Gupta and Naqvi, 1984). Comparison of JGOFS data, collected in 1994/1995, with those compiled by Naqvi (1991) shows that the SNM has expanded south- and westwards (Rixen et al., 2014). This implies, in accordance with decreasing oxygen concentrations, an expansion of the OMZ in the Arabian Sea, which might have started in the early 1990s. However, the reliability of this trend has been questioned by Naqvi (2019). While it is acknowledged that the data are too sparse to have unquestionable confidence in this trend, it is difficult to assess the doubts raised by Naqvi (2019), as these are based on data derived from sediment cores that do not cover the most recent past.

In the Bay of Bengal, a pronounced SNM has not yet been detected, although a recent study presented data from seven stations in the northern Bay of Bengal and revealed oxygen concentrations which partly drop below ∼ 0.05 µM at four sites (Bristow et al., 2017). Incubation experiments were carried out, but with one exception these failed to prove denitrification. At the one exceptional station, a denitrification rate of 0.9 nmol L-1 d-1 was measured, which falls much below denitrification rates of > 20 nmol L-1 d-1 as measured by Ward et al. (2009) in the Arabian Sea. However, these results indicate that the Bay of Bengal OMZ is on the verge of functional anoxia, with re-oxidation of nitrite to nitrate as yet preventing significant denitrification (Bristow et al., 2017). However, outbreaks of hydrogen sulfide as seen in the upwelling systems off Peru (Schunck et al., 2013) and Namibia (Weeks et al., 2002) have so far not been reported in the northern Indian Ocean during the last 50 years, other than in bottom waters on the Indian shelf as mentioned earlier. This implies that the physical oxygen supply and the biological oxygen consumption maintained hypoxic conditions and prevented persistent anoxia in the Arabian Sea and functional anoxia in the Bay of Bengal OMZ.

Based on the growing evidence that low concentrations of dissolved oxygen slow down the respiration of organic matter in the water column and thereby the biological oxygen consumption (Aumont et al., 2015; Cavan et al., 2017; Laufkötter et al., 2017; Thamdrup et al., 2012; Van Mooy et al., 2002), it has been hypothesized that an oxygen–related feedback mechanism stabilizes the Arabian Sea OMZ (Rixen et al., 2019a). This mechanism operates in the upper part of the OMZ, which hosts the seasonal thermocline but also affects the base of the OMZ and thereby its thickness as discussed before (Fig. 8).

The seasonal thermocline is the subsurface layer from which water is introduced into the euphotic zone via physical processes such as upwelling, vertical mixing, and eddy-driven transports on a seasonal timescale. Nutrients supplied by these mechanisms largely sustain the productivity of pelagic ecosystems and the associated export production (Eppley and Peterson, 1979). Hence, the seasonal thermocline is the main nutrient reservoir of pelagic ecosystems, and to fulfill this role the vast majority of the exported organic matter must be respired within the seasonal thermocline. Accordingly, the seasonal thermocline represents the main zone of respiration and, similar to soils on land, accommodates the nutrient recycling machinery of the pelagic ecosystem. Nutrient losses from the seasonal thermocline, via particle fluxes into the deep sea, denitrification, and lateral advection, must be compensated for by nutrient inputs in order to maintain the productivity (Rixen et al., 2019a). Nitrogen fixation, river discharges, and atmospheric deposition can be important nutrient sources, but in the Arabian Sea lateral inflow of water masses from the south via the cross-equatorial cell are the main source balancing nutrient losses from the seasonal thermocline (Bange et al., 2000; Gaye et al., 2013). Accordingly, a significant negative impact of denitrification on primary and export production and the associated oxygen consumption appear to be unlikely on seasonal to centennial timescales in the Arabian Sea.

The SNM, which occurs at water depths between 200 and 400 m in the central Arabian Sea and as deep as 500 m in the eastern Arabian Sea, divides the seasonal thermocline into an aerobic part at water depths between ∼ 40 and 200 m and an anaerobic part down to the base of the SNM (Fig. 8). The depth of the seasonal thermocline of approximately 300–400 m corresponds to the depth range of vertically migrating zooplankton as observed during the large summer bloom in the Arabian Sea (Smith, 2001) and roughly matches the water depth range from where subsurface water is introduced via upwelling into the euphotic zone in the western Arabian Sea (Brock et al., 1992; Rixen et al., 2000). Furthermore, nitrate concentrations, which decrease within the SNM, remain above 10 µM (Fig. 7), suggesting that supply of decomposable organic matter (rather than nitrate availability) limits denitrification, as also suggested by other studies (Bristow et al., 2016; Ward et al., 2009). A substrate limitation at a water depth of 400 to 500 m and the arrival of organic matter at sediment traps deployed in the deep sea at a water depth of 3000 m support the concept of export production that is divided between free (reactive) and protected (low reactivity) organic matter (Armstrong et al., 2002). This partition is based on the assumption that ballast-associated, protected organic matter is preferentially exported to deeper waters as fast-sinking particles, whereas the slow-sinking free organic matter is preferentially respired within the seasonal thermocline. If decreasing oxygen concentrations reduce the respiration of free organic matter, this prevents a further depletion of oxygen and the development of anoxia by reducing the oxygen consumption within the seasonal thermocline. The consequence is an increasing export of free organic matter out of the seasonal thermocline, which could enhance the oxygen consumption at the base of the OMZ. This would imply that low oxygen concentrations within the OMZ are accompanied by a thickening of the OMZ, which can be seen on a seasonal timescale by the negative correlation between mean OMZ oxygen concentrations and the maximum thickness of the OMZ (Fig. 6b).

The role of an oxygen-dependent, biologically driven feedback mechanism, counteracting impacts of the monsoon on the intensity of the OMZ, depends on the local oxygen consumption. If this is too low, remote processes instead of the local monsoon-driven carbon export rates need to be considered as forces controlling the intensity of the OMZ. The apparent oxygen utilization (AOU), in addition to mixing analyses, provides information that helps to estimate the role of the local oxygen consumption for maintaining the intensity of the OMZ in the Arabian Sea. The AOU represents the oxygen deficit caused by biological oxygen consumption and is calculated by subtracting the measured oxygen concentration from the temperature- and salinity-dependent oxygen saturation concentration. This approach is based on the assumption that the water mass of interest was saturated with respect to oxygen during its formation at the surface and, since then, the respiration of exported organic matter consumed oxygen within the water mass.

A mixing analysis based on data measured during the JGOFS in 1994/1995 reveals that oxygen deficits inherited from ICW contribute approximately 25 % to the AOU determined in the Arabian Sea OMZ (Rixen and Ittekkot, 2005). Accordingly, the respiration of organic matter produced in the Arabian Sea largely causes the low oxygen concentration in the Arabian Sea OMZ, which emphasizes the role of the monsoon-driven productivity in the Arabian Sea and the oxygen-dependent biologically driven feedback mechanism as local drivers for the OMZ development.

However, it should be noted that in the Arabian Sea, mean satellite-derived export production rates were too low to explain a high biological oxygen consumption considering a residence time of water within the OMZ of 10 years (Rixen and Ittekkot, 2005). The mismatch reflects uncertainties caused by the poorly constrained residence time of water within the OMZ and export production rates. Even though residence times of water within the Arabian Sea and Bay of Bengal OMZ of 10 and 12 years seem to be well accepted (Bristow et al., 2017; Olson et al., 1993), there are also estimates ranging from 1 to 51 years for the Arabian Sea OMZ (Naqvi and Shailaja, 1993; Sen Gupta and Naqvi, 1984). In the Arabian Sea, satellite-derived export production rates vary by a factor of 10 (Rixen et al., 2019b), and an even larger variability can be seen on a global scale. Global export production rates derived from satellite-data, numerical models extrapolated from sediment trap data and based on estimates of the biological oxygen demand vary approximately between 1.8 and 27.5 Pg C yr-1. In general estimates based on the biological oxygen demand, in line with inverse modeling studies, call for higher export production rates (Burd et al., 2010; del Giorgio and Duarte, 2002; Schlitzer, 2000; Schlitzer, 2002), whereas satellite and sediment trap data, as well as numerical model studies, suggest lower export production rates (Emerson, 2014). Nevertheless, considering estimates within the upper range (higher carbon export and longer residence times of water in the OMZ), the AOU determined in the Arabian Sea could be explained (Rixen and Ittekkot, 2005).

The OMZs in the northern Indian Ocean are melting pots collecting the influence of a variety of water masses with different origins and histories (e.g., Morcos et al., 2012; Schott and McCreary, 2001; You, 1997). Mixing analyses indicate that the Arabian Sea OMZ contains in addition to ICW also Arabian Sea Water (ASW) as well as Persian Gulf and Red Sea Water (RSW; Acharya and Panigrahi, 2016; Hupe and Karstensen, 2000; Rixen and Ittekkot, 2005). The Persian Gulf Water (PGW) is a dense and oxygen-rich surface water that subducts in the northern Arabian Sea and strongly contributes to the ventilation of the upper OMZ (Lachkar et al., 2019; Schmidt et al., 2020). Regional model simulations have shown that eddies control the transport and the spreading of the PGW into the Gulf of Oman (Lachkar et al., 2019; Queste et al., 2018; Vic et al., 2015). Projected future warming of the Persian Gulf can, in turn, lower the oxygen concentrations of the PGW and its sinking in the Gulf of Oman. The consequence is a drop of oxygen concentrations at depths between 200 and 300 m in the northern Arabian Sea (Lachkar et al., 2019). Such a reduced physical oxygen supply could explain the observed intensification of the Arabian Sea OMZ and the expansion of the SNM (see above). Furthermore, it implies that global warming impacts on the physical oxygen supply override effects of the biologically driven feedback mechanism.

Understanding the processes controlling oxygen consumption within OMZs, such as export production and the residence time of water, is still fraught with large uncertainties. Nevertheless, it seems that there are oxygen-dependent, biologically driven feedback mechanisms counteracting impacts of the monsoon on the intensity of the OMZ. This could explain the absence of persistent anoxia and functional anoxia in the Arabian Sea and Bay of Bengal OMZs. However, the recent expansion of the OMZ in the Arabian Sea and the first indication of denitrification in the Bay of Bengal OMZ indicate that there are other processes overriding effects of these feedback mechanisms. The postulated decrease in the physical oxygen supply caused by the inflow of warmer and hence more oxygen-depleted PGW agrees with the observed decreasing oxygen concentrations and expansion of the SNM in the Arabian Sea. In the Bay of Bengal, a response to global warming is more difficult to establish due to strong interannual variations in the intensity of the eddy activity. The fact that eddies affect both the supply of oxygen (through ventilation) and its consumption (through biological productivity) in a non-trivial manner increases the difficulties to adequately parameterize the effects of eddies on dissolved oxygen in coarse-resolution models.

On millennial and even longer timescales, sedimentary records have been used to trace changes in OMZ intensities. The δ15N values of particulate nitrogen in sediments are often used as tracers of OMZ intensity because they reflect major shifts in the pool of fixed nitrogen due to denitrification, as discussed before (Altabet et al., 1995, 1999; Brandes et al., 1998; Ganeshram et al., 1995). Locally, eolian and riverine nitrogen supply affects δ15N values (Kendall et al., 2007; Voss et al., 2006), but in the Indian Ocean sedimentary records of δ15N reflect the balance between denitrification and nitrogen fixation. Deep water nitrate has an average δ15N value of ∼ 5 ‰ (Sigman et al., 2005), but, due to denitrification, the δ15N of nitrate in the Arabian Sea increases to values > 17 ‰ (Fig. 7). Convective mixing, eddy pumping, and especially upwelling move nitrate-deficient water masses from the OMZ to the surface, so nitrate with high δ15N values is transported into the euphotic zone. After assimilation into biomass by phytoplankton, 15N-enriched particulate matter sinks through the water column to the seafloor where the signal of denitrification, and hence OMZ intensity, is preserved in sediments (Altabet et al., 1995; Gaye-Haake et al., 2005; Naqvi et al., 1998; Suthhof et al., 2001). Early diagenesis may raise sedimentary δ15N values by 2 ‰–5 ‰, and the diagenetic effect increases with water depth (Altabet, 2006; Tesdal et al., 2013). Nevertheless, the relative changes in δ15N in deep-sea sediments record variations in the OMZ intensity, while records from the continental slopes are subjected to negligible diagenetic enrichments so that they retain the signal of the nitrogen source (Altabet et al., 1999; Gaye et al., 2018).

A sediment core from the northern Bay of Bengal (Contreras-Rosales et al., 2016) indicates that the highest δ15N values (and thus the lowest OMZ oxygen concentrations) recorded in the core prevailed during the Holocene and the Last Glacial Maximum (LGM), with a δ15N range between 4.4 ‰ and 5.0 ‰ (i.e., in the range of the average value of deep ocean waters; see above). Therefore, denitrification in the past 21 000 years can be ruled out in the Bay of Bengal from a paleoceanographic perspective (Contreras-Rosales et al., 2016). The δ15N values at the core top (4.6 ‰) were similar to values in sediment trap materials (3.7 ‰–4.5 ‰) and were explained by a mixture of nutrients or suspended matter from the Ganges–Brahmaputra–Meghna river system with nitrate from subsurface water (Contreras-Rosales et al., 2016; Gaye-Haake et al., 2005; Unger et al., 2006). Enhanced δ15N values in the early Holocene to 6000 BP (BP means before present, where present means 1950) coincide with a stronger monsoon and were attributed to enhanced supply of nitrate from the subsurface, which has elevated δ15N compared to the depleted values of the riverine end-member (Sarkar et al., 2009). However, to our knowledge there is only one published sediment record from the Bay of Bengal spanning the entire Holocene (Contreras-Rosales et al., 2016), so we know nothing about the spatial variability within the basin.

In contrast to the Bay of Bengal, denitrification in the Arabian Sea has prevailed during the warm interstadials of the Pleistocene and during the entire Holocene, as can be discerned from sedimentary δ15N values > 6 ‰, with maxima of > 11 ‰ (Agnihotri et al., 2003; Higginson et al., 2004; Kessarkar et al., 2018; Möbius et al., 2011; Pichevin et al., 2007). Productivity increased with the onset of the Holocene as the summer monsoon strengthened and monsoonal upwelling off Somalia and Oman commenced and became a permanent feature of the Holocene Arabian Sea (Böning and Bard, 2009; Gaye et al., 2018). A rise of δ15N by at least 2 ‰ shows that an onset of upwelling immediately strengthened the OMZ and led to denitrification across the entire basin in the beginning of the Holocene (Böll et al., 2015; Gaye et al., 2018).

Furthermore, southward retreat of Antarctic Sea Ice is assumed to have reduced ventilation of the Arabian Sea OMZ through its influences on the formation of the oxygen-enriched ICW and associated meridional overturning circulation of the upper Indian Ocean (Böning and Bard, 2009; Naidu and Govil, 2010). A decline in δ15N by about 1 ‰ is found in the early Holocene until 6000 BP in high-resolution sediment cores from the western, northern, and eastern Arabian Sea and indicates that the OMZ weakened and became less persistent during this period (Fig. 9a, b). More vigorous upwelling, discernible from benthic foraminifera, may have led to a better ventilation of the basin as it is associated with an increased inflow of ICW from the south (Das et al., 2017). As discussed before, the inflow of ICW increases the physical oxygen supply and, furthermore, reduces the residence time of OMZ waters (Böning and Bard, 2009; Pichevin et al., 2007).

After 6000 BP, increasing δ15N values indicate a strengthening of the OMZ across the entire basin (Fig. 9a, b). It is hypothesized that a weaker ventilation is responsible for decreasing oxygen concentrations and that this could be due to reduced inflow of ICW, as it was blocked by the enhanced inflow of PGW and RSW since the sea level high stand at 6000 BP (Pichevin et al., 2007). Furthermore, a southward shift of the West Indian Coastal Current and the associated poleward undercurrent lowered the inflow of ICW and thereby the ventilation of the OMZ in the eastern Arabian Sea (Mahesh and Banakar, 2014).

In order to give an additional, model-based estimate of the OMZ evolution in the Indian Ocean, transient model simulations over the Holocene were performed with the global atmosphere–ocean Kiel Climate Model (KCM, Park et al., 2009) and the marine biogeochemistry model PISCES (Aumont et al., 2003).

In a first step, KCM was forced with transient orbital parameters and greenhouse gas concentrations from 9500 BP to present. In a second step, the PISCES model was forced with the ocean physical fields from the above KCM experiment in so-called offline mode. This model setup comprised a ventilation age tracer of the water masses (see Segschneider et al., 2018, for a more detailed description of the model components and experiment setup). While the oceanic 2∘×2∘ grid in this setup was refined to a meridional resolution of 0.5∘ near the Equator to allow a better representation of equatorial waves, the long integrations (9500 model years) required a coarse model resolution that is far from eddy resolving. The ballast scheme for the export of particulate organic carbon (POC) also neglected the lithogenic ballast effect, which is important in the Arabian Sea and Bay of Bengal as discussed before.

From these model experiments, temperature and oxygen fields have been analyzed and compared to sedimentary records mainly in the Arabian Sea. Here the model results were subdivided into areas corresponding to the binned sediment core regions specified in Gaye et al. (2018) (Fig. 9a; north: 62–68∘ E, 20–25∘ N; east: 68–75∘ E, 13–20∘ N; west: 54–60∘ E, 15–20∘ N; south: 48–55∘ E, 7–12∘ N). The simulated oxygen concentrations in the Arabian Sea are generally somewhat too high at the surface due to a cold bias of the KCM, but the observed near-surface gradients of oxygen concentrations are very well matched. However, in the deeper layers the model overestimates oxygen concentrations (not shown; see Supplement Fig. A.1c in Segschneider et al., 2018). As a result, simulated oxygen concentrations in the model Arabian Sea are nowhere low enough for denitrification to occur (denitrification sets in at 6 µM in the PISCES model, with a transition phase to full denitrification at lower oxygen concentrations). Moreover, no nitrogen isotopes are simulated in the current model version. Comparison to the δ15N data from the sediment cores is, therefore, restricted to a qualitative assessment.

The simulated oxygen concentrations (averaged between 200 and 800 m depth) show the lowest concentrations in the northern Arabian Sea (initially around 80 µM in the early Holocene, yellow curve in Fig. 9c). The concentrations are 10 µM higher in the western Arabian Sea (blue line) and a further 5 µM higher in the eastern Arabian Sea (red line), while they are much higher in the southern Arabian Sea (starting at 155 µM, grey line). Oxygen concentrations are fairly constant over the first 2500 years and then gradually decrease until the late Holocene. This decrease is strongest in the northern Arabian Sea (-20 µM) and quite similar in the western and eastern Arabian Sea (-10 µM). This is in qualitative agreement with the Holocene trends of δ15N data (Fig. 9b) that show the highest δ15N values (indicating strong denitrification and thus low oxygen) for the shallow northern core and lower δ15N for the western and eastern cores.

Simulated export production and water mass age in the Arabian Sea have been discussed for an earlier model experiment with the same model setup (but accelerated forcing) by Gaye et al. (2018) and in more detail for the global OMZs including the Indian Ocean for the model experiment analyzed here by Segschneider et al. (2018). While simulated export production in the Arabian Sea is fairly constant throughout the Holocene (Fig. 7 in Gaye et al., 2018), ventilation age is increasing throughout the Holocene concurrent with decreasing oxygen concentrations (Fig. 15 in Segschneider et al., 2018). This implies that changes in the ocean circulation and the associated inflow of oxygen-enriched ICW largely influenced the OMZ during the Holocene after the onset of upwelling at the beginning of the Holocene.

The δ15N sedimentary records reveal the difference in the late Pleistocene and Holocene history of denitrification in the Arabian Sea and Bay of Bengal. Oxygen concentrations in the Bay of Bengal never declined below the threshold of denitrification, whereas denitrification prevailed in the Arabian Sea during the warm interstadials and the entire Holocene. A data–model comparison shows that the age of the OMZ water mass increased after 6000 BP in both basins (not shown for the Bay of Bengal), coinciding with a strengthening of the OMZ and denitrification in the Arabian Sea. Based on the model results of constant export production and increasing water mass age, it is concluded that a reduced ventilation is responsible for decreasing oxygen concentration. The similar temporal evolution of observed OMZ intensity and modeled oxygen concentration in the Arabian Sea under orbital and greenhouse gas forcing thus indicates that the mid- to late Holocene OMZ intensification may be related to oceanic circulation rather than to local processes in the northern Indian Ocean. The progressive oxygen loss over the Holocene may thus be the result of orbital and greenhouse gas forcing in a qualitatively similar way to the much stronger variations simulated for LGM to mid-Holocene changes (Bopp et al., 2017).

For future climate projections we rely on Earth system models (ESM). Although these models reproduce large-scale features and global OMZ trends, they suffer from considerable mismatches between measured and model oxygen concentrations in the ocean (Bopp et al., 2013; Cabré et al., 2015; Oschlies et al., 2008, 2018). In comparison to observational data, they underestimate oxygen losses significantly (e.g., Oschlies et al., 2018, and references therein), and simulated volumes of OMZs differ considerably. Unresolved physical oxygen supply mechanisms, poorly constrained biological oxygen consumption rates, and their hardly known responses to global change cause these uncertainties (e.g., Oschlies et al., 2018; Segschneider and Bendtsen, 2013). Furthermore, feedbacks caused by the strong coupling of the marine oxygen and nitrogen cycles complicate long-term predictions (Fu et al., 2018; Oschlies et al., 2019).

Especially in the Indian Ocean, global coupled biogeochemical ESMs struggle to represent the OMZs (Fig. 10, Oschlies et al., 2008). In most ESMs the east–west contrast between the Arabian Sea and Bay of Bengal is opposite to what observations show, with most global models producing lower oxygen concentrations in the Bay of Bengal than in the Arabian Sea. Furthermore, half of the models cannot simulate hypoxic conditions in the Arabian Sea at all. A comparison of the thickness of the hypoxic layer in the northern Indian Ocean shows a disagreement among all models (Fig. 10). The maximum simulated volume (8.2 × 1015 m3, CESM1-BGC) is more than twice the hypoxic volume found from observations (3.1 × 1015 m3, WOA13). Moreover, this volume extends too far horizontally and does not cover the thickness of the observed OMZ in the Arabian Sea (Fig. 10).

To some degree, this problem may be attributed to the fact that ESMs are not tuned for the northern Indian Ocean. In addition, global models generally have coarser resolution to reduce computational costs and thus are far from eddy resolving, as for the KCM (results discussed in the previous section). Eddy transport is parameterized in the ESMs, but these still fail to represent the OMZs in the northern Indian Ocean. Even though the next generation of ESMs already targeted this problem, by providing high-resolution options including mesoscale processes in models used in the Coupled Model Intercomparison Project Phase 6 (CMIP6), there are only moderate improvements in subsurface oxygen representation (Kwiatkowski et al., 2020). The CMIP6 models still tend to overestimate oxygen concentrations in the Arabian Sea (Séférian et al., 2020).

The poor representation of the OMZs of the northern Indian Ocean in ESMs reduces the reliability of future projections of potential changes in the OMZs related to natural and anthropogenic forcing and thus their ecological impacts and possible feedbacks to climate change. Global models suggest a general decline of oxygen for the entire ocean, but there is no clear trend visible in the Indian Ocean (Oschlies et al., 2017). However, an older set of ESMs analyzed in Cocco et al. (2013) suggests a future decrease in oxygen in the subtropical Indian Ocean in the upper mixed layer and a small increase in the western tropical Indian Ocean. This increasing oxygen concentration is also seen in response to climate change in the RCP8.5 and RCP2.6 scenarios of the Coupled Model Intercomparison Project Phase 5 (CMIP5, Bopp et al., 2013). Specifically, Bopp et al. (2013) showed that a decrease in productivity is consistently simulated across all CMIP5 models and scenarios in the tropical Indian Ocean and that, by 2100, all models project an increase in the volume of waters with an oxygen concentration below 80 µM, relative to 1990–1999. This response is more consistent than that of the previous generation of ESMs, i.e., changes varying from -26 % to +16 % over 1870–2099 under the SRES-A2 scenario (Cocco et al., 2013).

However, for lower oxygen levels, there is less agreement among the CMIP5 models and also compared to observations regarding the volume of the OMZ (Bopp et al., 2013). Specifically, for the volume of waters < 50 µM, four models project an expansion of 2 % to 16 % (both GFDL-ESMs, HadGEM2-ES and CESM1-BGC), whereas two other models project a slight contraction of 2 % (NorESM1-ME and MPI-ESM-MR). For the volume of waters with an oxygen concentration < 5 µM, only one model (IPSL-CM5A-MR) is close to the volume estimated from observations and projects a large expansion of this volume (+30 % in the 2090s). These results for low-oxygen waters (oxygen concentrations of 5–50 µM) agree with those of Cocco et al. (2013), with large model–data and model–model discrepancies, and simulated responses varying in sign for the evolution of these volumes under climate change (Bopp et al., 2013).

Globally, the models agree on a negative oxygen trend over the 21st century driven by declining solubility of oxygen in surface waters through global warming (Resplandy, 2018; Schmidtko et al., 2017) and a reduced ventilation by changes in the ocean's circulation (Bopp et al., 2017). This holds true even though these models take into consideration the negative feedback caused by reduced tropical export production due to increased stratification of the upper water column (Fu et al., 2018; Palter and Trossman, 2018). Uncertainties and disagreements among the models arise from subtle differences in timing and magnitude of these opposing trends (Bopp et al., 2017). Waters with low oxygen saturation are particularly sensitive to impacts of climate warming (Fu et al., 2018) as well as vertical diffusivity that is parameterized by the mixing coefficient in the models (Duteil et al., 2012) and also mesoscale eddy transport and the lateral mixing coefficient (Bahl et al., 2019; Lachkar et al., 2016). Globally, reduced mixing across the mixed layer explains 75 % of the reduced subduction, but regionally changes in wind patterns that cause modulations in Ekman pumping and subduction are more important (Couespel et al., 2019). Thus, future trends in the northern Indian Ocean OMZs derived from the ESMs are highly uncertain, with projected potential increases or decreases in the volume of low-oxygen waters, depending on the model and the oxygen levels under consideration (Bopp et al., 2013; Cocco et al., 2013).

The OMZ in the Indian Ocean is the one we know least about, but it may also be the OMZ with the most complex dynamics in terms of forcing and variability. As discussed before, regional eddy-resolving modeling studies have been able to reproduce the OMZs, and thus they have helped us to better understand the interplay between physical and biogeochemical drivers (Lachkar et al., 2019; McCreary et al., 2013; Resplandy et al., 2012; Resplandy et al., 2011). Global models still struggle to reproduce the Indian Ocean OMZ. One explanation for this is the coarse resolution of these models; i.e., they cannot resolve the mesoscale and submesoscale processes that ventilate the subsurface waters, and they underestimate coastal upwelling during the monsoon seasons and, therefore, also primary production and biological oxygen demand. As a result, the oxygen trend in the tropical Indian Ocean remains unclear. However, in addition to poor representation of mesoscale and submesoscale features in global models, large uncertainties stem also from largely unknown biogeochemical and ecosystem responses to global physical changes.

There is a large body of evidence on the effects of hypoxia on macro-organisms. This includes reduced diel migration depths, vertical habitat compression, and shoaling distributions of fishery species and their prey (Breitburg et al., 2018). However, few reports exist on the effects of hypoxia on phytoplankton, the primary producers of the marine ecosystem. This is because it is generally understood that biological consequences of reduced oxygen concentrations are likely to be most notable for the 200–300 m layer, as these waters impinge on the euphotic zone (Stramma et al., 2010b). There have been several reports of coastal upwelling bringing up nutrient-enriched and hypoxic waters onto continental shelves stimulating production and increasing local biological oxygen demand (Stramma et al., 2010b).

In the Arabian Sea, the increasing expansion of hypoxia discussed in Sect. 2.4 both on the shelf and offshore is the predominant driver for a shift in the base of the pelagic ecosystem from mainly autotrophy before 2000 (Garrison et al., 1998, 2000; Smith et al., 1998a) to greater dependence on mixotrophy more recently (Gomes et al., 2014). A large-scale, ongoing study conducted by the National Institute of Oceanography, India, from 2003 onward, in support of India's ocean color program, first documented the appearance of extensive blooms of the green mixotrophic dinoflagellate, Noctiluca scintillans (Noctiluca).