For this study, sediments and bottom water from four different locations in the inner and intermediate part of the Stockholm archipelago (see Almroth-Rosell et al., 2016; Fig. 1) were collected. The study sites are located in the basins Strömmen (central Stockholm), Baggensfjärden, Erstaviken and Ingaröfjärden and are characterized by a range of water depths and bottom water redox conditions (Figs. 2, S1 (in the Supplement); Table 1). Strömmen is located most proximately to the outlets of the sewage treatment plants close to the center of Stockholm, which presumably contributes to the relatively high total P concentrations in the bottom water (Fig. 3a). Baggensfjärden is the most restricted basin in this study (i.e., landlocked with narrow and relatively shallow connections to adjacent basins), leading to reduced water column mixing, culminating in low-O2 bottom waters annually in summer. Erstaviken and Ingaröfjärden have a more open connection with the Baltic Proper, leading to better water column mixing. Ingaröfjärden is the least restricted basin in this study and subsequently the most consistently well-oxygenated basin throughout the year. Extensive water quality monitoring of the study area (obtained from the SHARK database at http://www.smhi.se/klimatdata/oceanografi/havsmiljodata/marina-miljoovervakningsdata, last access: 10 April 2019; Swedish Meteorological and Hydrological Institute – SMHI, 2019) shows a clear inverse correlation between bottom water O2 concentrations and P and a positive correlation between bottom water O2 and N/P ratios (Fig. 3a, b). Bottom water O2 and nutrient concentrations follow a distinct annual pattern, with maximum O2 and minimum nutrient concentrations in winter. After winter, O2 gradually drops and nutrient concentrations gradually increase, reaching minimum and maximum values at the end of summer and in autumn, respectively, followed by a reset of the system (Fig. 3c, d). The difference in amplitude of the changes in O2 and total P concentrations between sites is likely the combined effect of differences in water column mixing and in recycling of P from the sediment associated with changes in bottom water O2.

Approximately 125 mg of the oxic sediment split was digested in a mixture of strong acids as described by van Helmond et al. (2018). The residues were dissolved in 1 M HNO3 and analyzed for their elemental composition by ICP-OES. Average analytical uncertainty based on duplicates and triplicates was <5 % for calcium (Ca) and <3 % for P. The calcium carbonate content (CaCO3 wt %) was calculated based on the Ca content measured by ICP-OES, assuming that all Ca was in the form of CaCO3.

Between 200 and 300 mg of the oxic sediment split was decalcified using 1 M HCl as described by van Helmond et al. (2018), after which dried and repowdered residues were analyzed for their carbon and nitrogen content with a Fisons Instruments NA 1500 NCS analyzer. Average analytical uncertainty based on duplicates was <2 % for carbon and <3 % for nitrogen. Organic carbon (Corg) and organic nitrogen (Norg) contents were calculated after a correction for the weight loss upon decalcification and the salt content of the freeze-dried sediment. For Baggensfjärden, Erstaviken and Ingaröfjarden, C and N contents were taken from van Helmond et al. (2020).

Between 50 and 100 mg of the anoxic sediment split was subjected to a sequential extraction procedure based on a combination of the procedures by Poulton and Canfield (2005) and Claff et al. (2010) to determine the different phases of sedimentary Fe (Kraal et al., 2017). Briefly, under O2-free conditions (1) 10 mL of 1 M HCl, pH 0, was added to extract (4 h) Fe(II) and Fe(III) minerals such as easily reducible Fe oxides (e.g., ferrihydrite and lepidocrocite), Fe carbonates and Fe monosulfides; (2) 10 mL of 0.35 M acetic acid, 0.2 M Na3 citrate and 50 g L-1 Na dithionite, pH 4.8, was added to extract (4 h) crystalline Fe-oxide minerals such as goethite and hematite; (3) 10 mL of 0.17 M ammonium oxalate and 0.2 M oxalic acid, pH 3.2, was added to extract (6 h) recalcitrant oxide minerals such as magnetite; and (4) 10 mL of 65 % HNO3 was added to extract (2 h) pyrite (FeS2). For all extracts, Fe concentrations were determined colorimetrically with the phenanthroline method, adding hydroxylamine hydrochloride as a reducing agent to convert all Fe3+ into Fe2+ (APHA, 2005). For the first step the absorbance before and after addition of the reducing agent was measured, in order to separate Fe2+ and Fe3+. The Fe concentrations of the Fe3+ fraction of the first step and the second step were summed and are henceforth referred to as Fe oxides. Average analytical uncertainty based on duplicates and triplicates was <10 % for all fractions.

Approximately 300 mg of the anoxic sediment split was subjected to a sequential extraction procedure (Burton et al., 2008) to determine sedimentary sulfur phases. Briefly, under O2-free conditions (1) 10 mL of 6 M HCl and 2 mL of 0.1 M ascorbic acid were added to dissolve acid-volatile sulfur (AVS, assumed to represent Fe monosulfides – FeS), and the released H2S was trapped in a tube filled with 7 mL of alkaline zinc acetate solution (24 h) and (2) 10 mL of acidic chromium(II) chloride was added to dissolve chromium-reducible sulfur (CRS, assumed to represent FeS2), and the released H2S was trapped in a tube filled with 7 mL of alkaline zinc acetate solution (48 h). For both fractions, the amount of sulfur in the zinc sulfide precipitates was determined by iodometric titration (APHA, 2005). Average analytical uncertainty, based on duplicates, was <7 % for both AVS and CRS.

Approximately 100 mg of the anoxic sediment split was subjected to a sequential extraction procedure following the procedure of Ruttenberg (1992), modified by Slomp et al. (1996), but including the exchangeable P step. Briefly, under O2-free conditions (1) 10 mL of 1 M MgCl2, pH 8, was added to extract (0.5 h) exchangeable P (Exch. P); (2) 10 mL of 0.3 M Na3 citrate, 1 M NaHCO3 and 25 g L-1 Na dithionite (CDB), pH 7.6, was added after which 10 mL of 1 M MgCl2, pH 8, was added, together extracting (8 and 0.5 h, respectively) the P-bound-to-Fe fraction, including Fe-oxide-bound P and vivianite (Fe-bound P; Nembrini et al., 1983; Dijkstra et al., 2014); (3) 10 mL of 1 M Na acetate buffered to pH 4 with acetic acid was added after which 10 mL of 1 M MgCl2, pH 8, was added, together extracting (6 and 0.5 h, respectively) authigenic Ca-P, including carbonate fluorapatite, hydroxyapatite and carbonate-bound P (Auth. P); (4) 10 mL of 1 M HCl, pH 0, was added to extract (24 h) P in detrital minerals (Detr. P); and (5) ashing of the residue at 550 ∘C (2 h) after which 10 mL of 1 M HCl, pH 0, was added to extract (24 h) P in organic matter (Org. P). The P content in the citrate-dithionite-bicarbonate extract was analyzed by ICP-OES. All other solutions were measured colorimetrically (Strickland and Parsons, 1972). Average analytical uncertainty, based on duplicates, was <7 % for all fractions. Total P derived from acid digestion and subsequent ICP-OES analyses was on average within 5 % of the summed P fractions derived from the sequential extraction.

Rates of benthic NO3--reducing pathways were determined using the whole-core isotope pairing technique (IPT) and parallel slurry incubations (Nielsen, 1992; Risgaard-Petersen et al., 2003). Bottom water from Niskin bottles collected at each site was used to fill the incubation chamber (approx. 30 L) and maintained at in situ O2 concentrations using compressed air and nitrogen gas mixtures. Small core liners (Ø 2.5 cm) were used to take subcores from the Gemini cores and were immediately transferred to the incubation tank so that all cores were submerged and stoppers were removed. Sodium 15N-nitrate solution (Na15NO3, 98 atom % 15N, Sigma Aldrich, final concentration ∼50 µmol L-1) was added to the water of the incubation tank, and cores were preincubated in the dark at the in situ temperature for 2 to 5 h. Three replicate cores were sacrificed by slurrying the entire sediment volume at approximately 0, 2, 5 and 8 h following preincubation. Sediment was allowed to settle for 2 min before samples for gas (12 mL Exetainers, Labco, UK, killed with 250 µL zinc chloride solution, 50 % w/v) and nutrients (10 mL, killed with 250 µL zinc chloride solution, frozen) were taken.

Sediment slurries were carried out in parallel to whole-core incubations. Briefly, a glass bead (0.5 cm Ø) was added to each 12 mL Exetainer, which was then filled with filtered (0.2 µm) helium-purged bottom water. Homogenized surface sediment (2 mL, 0–2 cm depth horizon) was added to each Exetainer, and vials were sealed. Exetainers were incubated on a shaker table in the dark at the in situ temperature for 8 to 12 h ensuring consumption of background NO3- and O2 before addition of 15N-substrates. Exetainers were divided into two treatments, amended with sodium 15N-nitrate or with sodium 14N-nitrite and 15N-ammonium chloride (each 100 µmol L-1 final concentration). Slurries were sacrificed at approximately 0, 5 and 10 h after substrate addition by injection of 250 µL of zinc chloride solution through the septum of Exetainers.

Analysis of 15N composition of N2 (and any nitrous oxide, N2O) was determined by gas chromatography isotope ratio mass spectrometry (GC-IRMS). A helium headspace was introduced to filled Exetainers, and gas samples were manually injected as described in Dalsgaard et al. (2013). Any 15N-N2O was reduced in a reduction oven and measured as 15N-N2. Determination of 15N in NH4+ was carried out by conversion of NH4+ to N2 with alkaline hypobromite iodine solution (Risgaard-Petersen et al., 1995; Füssel et al., 2012). Ammonium was extracted from sediment in slurry and whole-core samples by shaking for 1 h with 2 M KCl (1:1 sample KCl) before any NH4+analysis. The isotopic composition of the produced N2 was determined using a GC-IRMS as above. Recovery efficiency of 15NH4+ following the hypobromite conversion was >95 %.

Concentrations of NOx (NO3- + NO2-) in incubations were determined colorimetrically as described for pore water. For determination of total NH4+, samples were extracted with KCl as above and NH4+ concentrations were analyzed colorimetrically using the salicylate-hypochlorite method (Bower and Holm-Hansen, 1980).

Anammox and DNRA were detectable in slurry incubations, although both processes only played a minor role in NO3- reduction at most sites. However, they may have interfered to a minor degree with the IPT calculations. Thus areal rates of benthic N cycling processes were calculated according to Song et al. (2016) at all sites. The relative contribution of anammox to N2 production (ra) in slurries was calculated as in Song et al. (2013) using the average mole fraction of 15NH4+ in the total NH4+ pool (FA) as this was demonstrated to increase linearly over time.

Fluxes of NO3- and NH4+ were calculated using gradients (∼0–1 cm and ∼0–5 cm, respectively) of sediment pore water depth profiles and Fick's first law of diffusion. Porosity values were taken from the average porosities of the integrated depth horizons and diffusion coefficients from Schulz (2006).

At the end of autumn and during the winter dissolved O2 concentrations in the Stockholm archipelago peak, largely due to mixing of the water column and subsequent ventilation (Figs. 3c, d, S1). After winter, O2 concentrations decrease during spring and summer, following enhanced O2 consumption by degrading organic matter after the spring bloom and reaching minimum values at the end of summer and in autumn. The loss of O2 from the bottom water is further enhanced by reduced ventilation of deeper waters following intensified water column stratification as a result of formation or strengthening of the thermocline (Gidhagen, 1987), which at many locations in the Stockholm archipelago leads to hypoxia (Karlsson et al., 2010; Conley et al., 2011). In addition to nutrient availability, spring bloom intensity and water depth, hydrological restriction may contribute to low-O2 conditions. This is also reflected at our study sites, with Baggensfjärden being the most O2-depleted and restricted basin (i.e., landlocked with narrow and relatively shallow connections to adjacent basins) and Ingaröfjärden being the least restricted and, subsequently, the most consistently well-oxygenated basin throughout the year (Table 1; Figs. 1, 2, S1).

High dissolved O2 concentrations allow for the formation and presence of Fe oxides (Fig. 5) in the surface sediments that bind P (e.g., Slomp et al., 1996; Fig. 6). Low dissolved O2 concentrations, however, lead to the dissolution of Fe oxides in the surface sediments. The P associated with these Fe oxides can then be released into the water column again. This mechanism leads to P recycling in basins with strong (seasonal) contrasts in bottom water redox conditions, such as Baggensfjärden, where the sediments are a sink for P in the winter and a source of P in the spring and the summer (Fig. 3c), as also described previously for other basins in the Stockholm archipelago (Walve et al., 2018). Nevertheless, in year-round well-oxygenated basins, such as Ingaröfjärden, this seasonal P recycling is (nearly) absent (Fig. 3a). In such basins, deeper O2 penetration, which might partly be related to the presence of macrofauna (Fig. S3), leads to a thicker Fe-oxide-bearing layer (Fig. 5) and a larger and stable Fe-bound P pool (Fig. 6) and hence a larger enrichment of P in the surface sediments (Fig. 10). Besides Fe oxides, a major part of the surface sediment P pool consists of P in organic matter (Fig. 6), which, based on the C/N values close to the Redfield ratio (Fig. 5), is predominantly of marine origin. Part of the organic matter (and the P associated with it) is lost with depth (Fig. 6), because the most labile organic matter is degraded in the upper centimeters of the sediment, releasing the P associated with it into the pore water. For our study sites in the Stockholm archipelago we calculated that this surface sediment P pool, i.e., the P active in turnover as already suggested by Rydin et al. (2011), varies between 0.036 mol P m-2 at Baggensfjärden and 0.172 mol P m-2 at Ingaröfjärden (between ∼1 and 5 g P m-2, respectively; Fig. 10; Table 6). This is comparable to values found for previously studied sites in the Stockholm archipelago (1 to 7 g P m-2; Rydin et al., 2011; Rydin and Kumblad, 2019). The surface sediment P pool could, however, have been much larger for Strömmen, Baggensfjärden and Erstaviken if all of the FeS in the surface sediments had seasonally transformed into Fe oxides. The lack of such a transformation is likely linked to the high upward flux of H2S to the surface sediment (4.2 to 7.6 mmol m-2 d-1; Table 3). Besides the H2S flux, there is a relatively large efflux of NH4+ from the sediments into the bottom water (up to 1.4 mmol m-2 d-1; Table 5). Both the H2S and the NH4+ flux originate from decomposing organic-rich sediments at depth (Fig. 4). Upon aerobic oxidation, 2 mol of O2 is consumed per mole of H2S or NH4+ (e.g., Reed et al., 2011). Thus, the O2 demand resulting from these H2S and NH4+ fluxes is very high when compared to the diffusive flux of O2 into the sediment (3.1–13.8 mmol m-2 d-1; Table 2). As a consequence of the presence of H2S in the surface sediments and its high upward flux, in combination with reduced water column mixing and/or seasonally low-O2 bottom water conditions, FeS are formed and/or preserved (Fig. 5) and formation of a large(r) pool of Fe oxides and a Fe-bound P pool is hindered at Strömmen, Baggensfjärden and Erstaviken. At Ingaröfjärden, the well-mixed water column and year-round well-oxygenated bottom water allow for a deeper O2 penetration (Fig. S3), preventing the presence of H2S in the surface sediment despite its high upward flux (Table 3) and leading to a thicker Fe-oxide-bearing layer (Fig. 5) and a larger Fe-bound P pool (Fig. 6).

Absolute P concentrations in the sediments in the Stockholm archipelago (Figs. 6 and 10 in this study and in Rydin et al., 2011) are high (∼30 to 50 µmol g-1) in comparison with most other studied sites in the coastal zone of the Baltic Sea (generally <30 µmol  g-1; Jensen et al., 1995; Carman et al., 1996; Lenstra et al., 2018). The relatively low Corg / Ptot values in the top ∼2 cm, which are around the Redfield ratio (Fig. 5), show that the seasonal O2 depletion of bottom waters in our study area is not severe or long-lasting enough to cause substantial preferential regeneration of P relative to C (Algeo and Ingall, 2007; Sulu-Gambari et al., 2018). The combination of high absolute P concentrations and relatively high sedimentation rates leads to relatively high rates of P burial (Table 6; Fig. S5). Further research of P burial rates at additional locations in the Stockholm archipelago, including the impact of anthropogenic activities on sedimentation rates (e.g., near-shore construction and dredging) and of redeposition of sediments that have already undergone one or multiple diagenetic cycles (after resuspension due to, for example, land uplift; Jonsson et al., 1990; Bryhn and Håkanson, 2011) is required before these results can be extrapolated to the scale of the entire system. Furthermore, it remains unclear what parts of the Stockholm archipelago represent areas of net sediment accumulation (Karlsson et al., 2019; Asmala et al., 2019) and how much (and in what form) P is buried in euxinic parts of the Stockholm archipelago. Hence, our results cannot be directly used to resolve the apparent discrepancy between the model results of Almroth-Rosell et al. (2016) and Walve et al. (2018).

The constant concentrations of most P forms in the sediment below the clearly enriched surface sediments suggest there is generally little to no sink switching of sediment P forms (i.e., the transformation of relatively labile P reservoirs such as Fe-oxide-bound P and organic P to authigenic P minerals such as vivianite) in the Stockholm archipelago. The curved shape of the pore water HPO42- profiles indicates, however, that there is still some release of P into the pore water at depth, and we attribute this to slow degradation of organic matter. Both the detrital and authigenic (Ca-P) fractions are likely buried in the form in which they reached the sediment–water interface. The general dominance of P in organic matter and apatite (authigenic and detrital P; Fig. 6) at depth (representing permanent P burial, since the release of P from organic P is only minor) agrees with previous findings for organic-rich sediments in the Baltic Sea (e.g., Jensen et al., 1995; Carman et al., 1996; Mort et al., 2010; Rydin et al., 2011). By contrast, in the Bothnian Sea, Fe-bound P is a much more important P pool at depth (e.g., Egger et al., 2015; Lenstra et al., 2018). Evidence for potential sink switching is only found at Strömmen, which is characterized by a larger Fe-bound P pool at depth (Fig. 6). This larger Fe-bound P pool at depth contributes to the high P burial rate at Strömmen (Table 6; Fig. S5). Coastal sediments with a shallow SMTZ, relatively high inputs of Fe oxides and organic matter, and high sediment accumulation rates are prime locations for formation of vivianite-type minerals (e.g., Slomp et al., 2013; Egger et al., 2015). The presence of dissolved Fe2+ and decreasing dissolved HPO42- concentrations at depth at Strömmen (Fig. 4) in combination with elevated Fe-bound P in the lower part of the record (Fig. 6) may hence result from the formation of a vivianite-type mineral.

Denitrification is by far the dominant pathway of NO3- reduction at our study sites, accounting for ∼80 % to 99 % of total dissimilatory NO3- reduction (DNRA + anammox + (2 × denitrification)) at the time of sampling (Table 4).

The dominant role of denitrification in removing N and the gradient from inner to outer archipelago agrees well with regional models based on long-term monitoring data, which show the highest N removal capacity in the inner archipelago region (Almroth-Rosell et al., 2016; Edman et al., 2018). In the model of Almroth-Rosell et al. (2016), the inner archipelago, where Strömmen is located, annually removes approximately 3–5 times more N (∼8–12 t N km-2 yr-1) than the intermediate and outer archipelago sites (∼1–3 t N km-2 yr-1). Denitrification rates of both Baggensfjärden and Erstaviken are within this range (∼2.5 and ∼3 times lower than at Strömmen, respectively). However, despite Ingaröfjärden being located in a basin adjacent to Erstaviken (Fig. 1) and modeled as having an almost identical area-specific N retention capacity (Almroth-Rosell et al., 2016), denitrification rates were almost 20 times lower than those at Strömmen and ∼8 to 6 times lower than at Baggensfjärden and Erstaviken, respectively. As such, N removal rates between adjacent basins may be more variable than assumed by models. The differences in rates are likely related to lower organic matter inputs and subsequent lower sediment respiration rates as indicated by deeper O2 penetration at Ingaröfjärden (Table 2; Fig. S3). Suspended particulate organic matter may also be removed more quickly from Ingaröfjärden due to its more direct connection to the open Baltic Sea (Fig. 1), permitting more rapid water exchange and transport of particulate organic matter out of the basin than at Baggensfjärden and Erstaviken (Engqvist and Andrejev, 2003).

Given the minor contributions of anammox and DNRA in these sediments at the time of sampling, we focus predominantly on the control of heterotrophic denitrification in Stockholm archipelago sediments. Heterotrophic denitrification in sediments is limited by both the availability of NO3- and Corg.

In sediments, NO3- is supplied from overlying water and/or from nitrification in the surface layers (coupled nitrification–denitrification; Seitzinger, 1988; Seitzinger et al., 2006). The relative importance of the two NO3- sources to denitrification in coastal systems can be highly variable between locations and seasons (e.g., Seitzinger et al., 2006; Jäntti et al., 2011; Bonaglia et al., 2014). We observed a distinct positive correlation between rates of denitrification and bottom water NO3- concentration (Fig. 9), indicating a high capacity of the sediments to reduce riverine NO3- loads along the seaward gradient, as shown for other coastal systems of the Baltic Sea (Asmala et al., 2017). We additionally demonstrate that benthic nitrification provided the major proportion (∼55 %–90 %) of NO3- which was reduced in the sediments at all four sites (Table 5; Fig. 8), as has been demonstrated in previous studies and syntheses on coastal systems (e.g., see Seitzinger et al., 2006) and studies on the Baltic Sea (e.g., Silvennoinen et al., 2007; Bonaglia et al., 2014, 2017; Hellemann et al., 2017). One of the highest contributions of nitrification to NO3- production for denitrification (∼85 %) was measured at Ingaröfjärden. At this site, the lowest overall denitrification rates and bottom water NO3- concentrations were measured, despite the deep (18 mm) O2 penetration providing a large sediment volume for nitrification to occur (Table 2). This high O2 penetration may in part be due to fewer Corg inputs and thus a lower Corg content (Table 2), discussed in Sect. 4.2.1 and further below.

Inputs of Corg provide both a C source for heterotrophic processes (e.g., denitrification) as well as a source of NH4+ (from remineralization processes) for nitrification and subsequent NO3- production. In coastal sediments Corg is not thought to limit denitrification. However, in complex basin systems such as the Stockholm archipelago, and in the Baltic Sea coastal zone in general, differences in ventilation and retention times between basins (implying differences in vertical and lateral exchange of water and O2 and, hence, variations in bottom water O2) may mean that Corg inputs are more variable than assumed (see Sect. 4.2.1). Available Corg in Ingaröfjärden (Table 2) may be less labile than it is at other sites due to such variations in hydrology and bottom water O2, with the deep (18 mm) O2 penetration indicating a lower organic matter reactivity and sediment respiration compared to the other sites. Lower labile Corg availability will limit heterotrophic denitrification and may explain why anammox, an autotrophic process, is more dominant at this site (Table 5; Fig. 7). The presence of the invasive polychaete Marenzelleria (Table 2) may also reduce N removal at Ingaröfjärden and enhance the efflux and transport of NH4+ from sediments (e.g., Hietanen et al., 2007; Bonaglia et al., 2013), although it should be noted that the impacts of infauna on N cycling are notoriously complex (Robertson et al., 2019).

Sampling and experiments in this study were carried out in late winter (March), a period in the Baltic Sea when the water column is well mixed, with cold and well-oxygenated bottom waters and with persistently low organic inputs to sediments. However, conditions are of course not static throughout the annual cycle. Seasonal warming, stratification, phytoplankton blooms, and consumption and release of nutrients as seen in year-round monitoring data (Figs. 3d, S1) will have marked effects on sediment nutrient cycling. Year-round bottom water monitoring data collected at Bäggensfjärden show that NO3- accumulates annually in bottom waters during the autumn and winter months before being consumed during spring and summer by phytoplankton blooms (Fig. 3d). Hypoxic bottom waters develop over summer following bloom collapse and subsequent enhanced deposition of fresh organic matter and enhanced benthic respiration during summer and early autumn. Bottom water total N concentrations increase during summer in connection with the hypoxic events (Fig. 3d) due to enhanced benthic remineralization and subsequent NH4+ efflux from sediments.

Increased organic inputs following the spring bloom are likely to lead to increases in denitrification as the season progresses, as is commonly observed in coastal sediments (e.g., Piña-Ochoa and Álvarez-Cobelas, 2006; Jäntti et al., 2011; Bonaglia et al., 2014). Thus, a scenario would be assumed for the Stockholm archipelago that is similar to those for other estuaries, leading to higher rates of denitrification during spring and early summer and a reduction in autumn and winter as organic inputs subside (e.g., Bonaglia et al., 2014). Depending on the bloom intensity and organic matter inputs during spring, increased benthic respiration may lead to more reduced conditions in surface sediments as bottom water O2 is depleted. The availability of NO3- also declines under hypoxic–anoxic conditions due to NO3- consumption in the water column, lower O2 penetration and thus a reduced volume of surface sediment where nitrification can occur and due to the reduced efficiency of nitrification under low-O2 conditions. The resulting high C/N conditions may cause process dominance to shift from N removal by denitrification (or anammox) to retention by DNRA (e.g., An and Gardner, 2002; Burgin and Hamilton, 2007; Giblin et al., 2013; Algar and Vallino, 2014; Kraft et al., 2014), as has been repeatedly demonstrated in field, laboratory and model studies (An and Gardner, 2002; Algar and Vallino, 2014; Kraft et al., 2014; van den Berg et al., 2016; Kessler et al., 2018). Thus, under hypoxic conditions in summer and autumn, DNRA may become the dominant NO3--reducing process, altering the role of sediments from a NO3- sink through N2 production to a source via increased NH4+ release by DNRA.

While we have not assessed the NO3--reducing process over different seasons at these four sites, we have demonstrated that the microbial metabolic potential for DNRA is present through the detection of DNRA activity in incubations at all four sites (Table 5). We suggest that it is highly likely that DNRA contributes to NH4+ efflux at sites during sporadic bottom water hypoxia. Thus, the capacity for N removal by denitrification may be reduced during bottom water hypoxia, while the likelihood of N recycling by DNRA increases as shown in previous Baltic Sea studies (e.g., Jäntti et al., 2011; Jäntti and Hietanen, 2012; Bonaglia et al., 2014).