2.1 Study area

The LTER-MC sampling site (40°48.5′ N, 14°15′ E) is located in the Gulf of Naples (GoN) at two nautical miles from the coast, at a depth of approximately 75 m (Figure 1) between offshore and coastal dominions and is representative of a wider area of the GoN (Cianelli et al., 2017; Zingone et al., 2019). The GoN is a semi-enclosed bay that opens into the Southern Tyrrhenian Sea. It has an area of 870 km² and is surrounded by 195 km of strongly anthropized shores. The area is densely populated (approximately 3 million inhabitants), and the industry is a relevant source of nitrogen and phosphorus to GoN waters, mainly from food processing, leather, textile, and paper-making industries (Barbiero, 2001; Lofrano et al., 2015; Pagnotta & Barbiero, 2003). The Sarno River is the main river of the GoN, and a major source of freshwater, nutrients, and pollutants (mineral nitrogen, phosphate, organic matter, heavy metals, pesticides) from both agriculture and industry (De Pippo et al., 2006; Lofrano et al., 2015; Thiele et al., 2017). There are nearly 100 discharge points for urban and industrial wastewaters on the GoN coast, and 456 in the Sarno River (Adamo et al., 2009; Arienzo et al., 2001), most of which are not treated. Agriculture occupies a reduced extent but is locally intensive (De Vivo et al., 2006).

2.2 Sampling and analyses

Sampling started in January 1984 and was carried out fortnightly until September 1991. Then, a major break occurred in the time series until January 1995. Starting from 1995, the sampling was resumed at a weekly frequency. In this study, we analysed the data collected between January 1984 and December 2019 (1329 sampling events).

Samples for nutrients analysis (NH₄⁺, NO₃⁻, PO₄³⁻, and Si(OH)₄) were collected at ten depths (−0.5, −2, −5, −10, −20, −30, −40, −50, −60, and −70 m), and chlorophyll a concentrations (Chl a) were measured at seven depths (−0.5, −2, −5, −10, −20, −40, and − 60 m). Samples of the first 0.5 m are afterwards indicated as 0 m (surface). Temperature data were acquired by reversing thermometers from 1984 to 1991 and by means of multiparametric profilers from 1995 onwards. Salinity was determined using a salinometer until 2002 and then by a periodically calibrated CTD multiparametric profiler. Nutrient concentrations were analysed according to Hansen and Grasshoff (1983), using a segmented continuous-flow autoanalyzer. Chlorophyll a concentrations were determined by spectrophotometry (Strickland & Parsons, 1972) until 1991 and by a spectrofluorometer (Holm-Hansen et al., 1965; Neveux & Panouse, 1987) from 1995 onwards. All methods for sampling procedures, laboratory analyses, and instrumentations were described in detail by Sabia et al. (2019), who carried out a careful quality check of biogeochemical data and assigned them different quality flags. Herein, we will only use the best-quality data. Data used in this paper are available on Zenodo (https://doi.org/10.5281/zenodo.7924734).

Integrated mean values were calculated over the 0 to −70 m depth interval (0–60 m for Chl a) using the trapezoidal rule. Likewise, depth-integrated values for each of the intervals 0–5 m, 20–40 m, and 50–70 m were obtained for sampling events with at least one value in each interval.

Water density was calculated from temperature and salinity using the UNESCO equation of state (UNESCO, 1981) and the R package oce (Kelley & Richards, 2022). The stratification strength of the water column was calculated as the absolute value of the density difference between −5 m and −60 m, expressed in kg m⁻³ (Väli et al., 2013). The mixed layer depth (MLD) was defined as the thickness of the layer within which the density range was lower than 0.03 kg m⁻³ (de Boyer Montégut et al., 2004), going down from 10 m deep. The “summer” stratified period was defined as the period encompassing the July 15 in which the MLD is continuously <11 m. The beginning of the stratification was defined as the first week of the stratified period. The end of stratification was defined as the first week in which the MLD was >11 m, after the July 15. Events of lateral advection of fresher water were identified as sampling events in which the 0 m salinity was 0.2 lower than the 10 m salinity (Ribera d'Alcalà et al., 2004).

2.3 Numerical simulations

Numerical model simulations were performed using the Regional Ocean Modeling System (ROMS), a 3-D free-surface hydrostatic primitive-equation finite-difference model that is widely used by the scientific community to model temperature, salinity, and currents in the ocean for a wide range of applications. Simulations were developed for the Tyrrhenian Sea and downscaled to the GoN, covering the Campania coast with a 500 m horizontal resolution, on 30 vertical levels, and producing daily outputs from 2001 to 2016.

Kokoszka, Saviano, et al. (2022) compared the model with in situ data and found that the monthly and interannual variabilities of the velocity components of the surface currents in the GoN were accurately reproduced (average annual correlation: 0.63).

To compare the hydrography of the GoN with that of the Tyrrhenian Sea, we used the model and extracted the daily vertical profiles of T and S to follow the isopycnal levels at a deep location in the open area of the GoN (14°0′ E, 40°38.60' N, approximately 600 m deep, in the Bocca Grande area between Ischia and Capri).

2.4 N and P anthropogenic pressure

Because of the difficulty in identifying a clear watershed for the GoN outside of the Sarno River, we estimated the anthropogenic N and P sources in the entire province of Naples, whose administrative territory integrally encircles the GoN (del Giudice et al., 2001). We estimated N and P loads from the Sarno River, N and P production by the population (household sources) and the industry, and N and P surplus in croplands. We also estimated direct atmospheric N deposition onto the Gulf surface, as atmospheric N originating from fuel combustion can be a significant N source in urbanized environments (Jaworski et al., 1997; Paerl et al., 2002).

2.5 Climatological data

Monthly cloudiness, expressed as a percentage of cloud cover, was obtained from the Climatic Research Unit dataset TS 4.03 (Harris et al., 2020) for the area between 40°30′ N and 41°00′ N and between 14°00′ E and 14°30′ E.

2.6 Data analysis

3.1 Temperature and salinity patterns and water column structure

The grand medians of the temperature values are presented in Table 1. The temperature (T) began to increase in March and reached its highest value in August, with lower layers lagging behind by one (−20 m) and two (−60 m) months (Figure 2). Water column was homothermal from December (depth-integrated mean T ~17.10°C) to March (depth-integrated mean T ~14.06°C). At the interannual scale, median T ranged between 20.29°C (2003) and 18.10°C (2011) at the surface and between 15.52°C (2001) and 14.04°C (1987) at −60 m.

The grand medians of salinity (S) are presented in Table 1. Surface S strongly decreased in spring from March onwards, reached its lowest value in May (median = 37.25), and returned to its highest median value in September (median = 37.90). Consequently, the water column presented a strong S gradient in spring but was almost isohaline in winter. S widely fluctuated among years, with an apparent cycle of 5 to 6 years (Kokoszka, Le Roux, et al., 2022), characterized by the same phase at all depths, with a more pronounced amplitude at 0 m and decreasing with depth (Figure 2). During the cycle with the highest amplitude, surface S ranged between 37.91 (2006) and 37.42 (2010). In addition to cyclic fluctuations, salinity followed interannual trends of smaller magnitude (Figure 3; Table 2).

As in most coastal sites, recurrent events of lateral advection of fresher water (FW) from the coast were detected on 30% of the dates (see also Kokoszka, Le Roux, et al., 2022 for a detailed analysis). The frequency of these events was highest in May (~60% of the observations), and lowest in December–January (only in ~12% of the cases). The interannual variability was also considerable, with a minimum of 6% events in 1990 and a maximum of 49% in 2014. From 1995 to 2015, the frequency of FW input events increased by an average of 0.01% year⁻¹.

The two opposite vertical gradients of S and T and their variations drove the stratification, whose strength, quantified as the difference between bottom and surface densities, was on average 1.18 ± 2.43 kg m⁻³. Despite the frequent FW inputs, stratification, as previously defined, mainly followed the cycle of surface T, increasing in March, with a maximum in July and August (>3.18 kg m⁻³) and then decreasing until December. The stratification strength varied widely between years. The median density difference was 1.64 kg m⁻³ in the year with the strongest stratification (2013) and 0.43 kg m⁻³ in the one with the weakest stratification (1998). Stratification also displayed multiyear oscillations, with an overall decrease from the 2013 maximum to ≈0.7 kg m⁻³ in 2019.

The shape of the density profile also allowed the determination of the mixed layer depth (MLD) (e.g., Kokoszka, Le Roux, et al., 2022). Because of the intermittent FW inputs, the MLD monthly medians were always lower than the water column depth over the entire annual cycle (Figure 2). The average on the whole time series was 13.48 ± 21.32 m, with a peak in January (50 m), a continuous decrease until April (11.8 m), stable values from June to September (10–11 m), and a progressive increase from September to December (49.2 m). The deepest annual median MLD was 20.95 m in 2003 and the shallowest (11.27 m) in 1986. In the long-term, there was a progressive MLD decrease from 1995 onwards by 0.01 m year⁻¹.

The meteorological summer period, with MLD continuously lower than 11 m, began, on average, in the week 14 ± 4 (March–April) and ended in the week 39 ± 4 (September), with an average duration of 27 ± 3 weeks. The stratification seasonality changed over the study period. In the years 1984–1990, the stratification generally broke at week 43. After the 1991–1995 interruption, it ended much earlier (on average at week 37 in 1995–2002), subsequently the week of the stratification end increased by 0.2 week year⁻¹, reaching the highest value (week 43) in 2013 and 2018 (data not shown).

3.2 Nutrient concentrations and distributions

All nutrient concentrations displayed the highest values and strongest interannual variability at the surface (Figure 4).

Table 1 shows a clear NO₃⁻ minimum at −20 m, which persists almost year-round (Figure 4), with higher concentrations at the surface and at −60 m. The integrated mean was 0.45 ± 0.51 mmol m⁻³. The three layers exhibited different seasonal patterns. At the surface, NO₃⁻ concentrations followed an astronomical seasonal pattern, characterized by the highest concentrations in winter-autumn (up to 1.66 mmol m⁻³ in March) and a strong decrease in summer, reaching a minimum in July (0.15 mmol m⁻³). The lower concentrations at −20 m (<0.12 mmol m⁻³) were almost constant from April to September, to rise afterwards reaching the highest values in January–February (0.75 mmol m⁻³). At −60 m, concentrations were more stable (>0.4 mmol m⁻³), decreasing only in August and September (~0.25 mmol m⁻³). However, several spikes in nitrate concentrations were detectable in the bottom layer (50–70 m depth) in some years between May and August, when the water column was stratified, and the overlying intermediate layer (20–40 m depth) appeared more depleted in nutrients (Figure S4), thus excluding the possibility of vertical downward fluxes below that depth. Nutrient increases at the bottom can be related to local processes, such as resuspension/remineralization, or to larger-scale phenomena, such as lateral advection of deeper waters. Because no correlations were found with turbidity or particulate organic nitrogen, which indicate processes occurring at the bottom (data not shown), we focused our attention on the potential role of the circulation of subsurface water masses in the GoN by means of numerical simulations. More specifically, the depth of the 1029 kg m⁻³ isopycnal, chosen as representative of the upper limit of the Levantine Intermediate Water (LIW), was followed in the period 2001–2016 at an offshore location of the GoN, situated in the Bocca Grande area, where the main exchanges between the GoN and Tyrrhenian Sea occur. The time series displayed a strong variability (from 62 to 158 m), reaching the shallowest depths during the cold winters of 2004/2005 and 2005/2006 (Figure 5A). The mean nitrate concentrations in the period May–August were calculated for the bottom layer (50–70 m) and were plotted versus the average depth of this isopycnal for each year, showing an opposite significant correlation (p < 0.05) between concentrations and depths (Figure 5B).

The interannual variability of nitrate concentrations presented contrasting patterns between the first and second periods of the time series (Figure 4). Concentrations were particularly high in 1984–1990, especially at the surface (1.36 ± 1.32 mmol m⁻³) and much lower (50%–70% less depending on depth) in the second period (1995–2019). Over the period 1995–2019, concentrations fluctuated widely at the surface (median value 0.62 ± 1.06 mmol m⁻³) without any detectable trend (Figure 3); however, a phase of decrease in 1998–2005 followed by an increase from 2007 to 2019 was revealed by the rolling trend test (data not shown). A decreasing trend was also observed at −20 m depth until 2017 (−1.59% per year) to rise again in the last 3 years. On the contrary, NO₃⁻ concentrations displayed significant increases at −60 m (2.72% per year) and for the integrated values (1.07% per year). As a result, the annual NO₃⁻ concentrations became higher at −60 m than at the surface in approximately 25% of the years.

NH₄⁺ concentrations displayed a strong vertical gradient (Table 1), with a depth-averaged value of 0.47 ± 0. 31 mmol m⁻³. Seasonality at the surface was characterized by the highest concentrations in autumn and winter (reaching maxima in March) and considerably lower values in late spring–summer. At the depths of−20 and − 60 m, the highest values were observed from November to January, while integrated mean concentrations reached their highest values in January and February, decreased from March to August, and increased again from September to the end of the year (Figure 4). NH₄⁺ concentrations at the surface widely varied among years (Figure 4), without any long-term trends (Figure 3). Conversely, at −20 m, −60 m, and for the integrated mean value (0–70 m), the interannual variability was less pronounced, but a marked decrease was recorded between 1995 and 2015, with the strongest reduction recorded at −60 m (−2.24% per year). The decrease was particularly important between 1996 and 2009, followed by a slight increase in the period 2010–2015.

As for other nutrients, PO₄³⁻ concentrations displayed a decreasing vertical gradient (Table 1) with an integrated mean of 0.04 ± 0.03 mmol m⁻³. The seasonal variability was less pronounced than that of the other nutrients; at the surface, the concentrations were higher in the first part of the year and relatively low in the second part. At −20 m and − 60 m, and for mean integrated values, a similar pattern was discernible, characterized by small variations and a decrease in the summer months (July and August). High-quality data on PO₄³⁻ concentrations in the 1980s were scarce because the concentrations were close to the detection limit of the analytical method utilized, precluding a statistical comparison with the 1995–2019 period. In the period 1995–2002, concentrations fluctuated widely from year to year with very similar variations at all depths. From 2002 to 2019, the PO₄³⁻ concentrations decreased, with the fastest decreasing phase between 2002 and 2012. Calculated on the period 1995–2019, the rate of decrease varied between 2.19 and 5.14% per year, depending on the depth considered (Figure 3, Table 2).

Si(OH)₄ concentrations displayed a pattern similar to that of NO₃⁻ with a minimum at −20 m (Table 1) and an integrated mean of 1.44 ± 0.0.65 mmol m⁻³. By contrast, seasonal signals were observed at different depths. At the surface, Si(OH)₄ concentrations were the highest (>2 mmol m⁻³) in January and February, then they decreased reaching the minimum in August (0.86 mmol m⁻³), to rise again until December. At −20 m, concentrations rapidly decreased in the first months of the year (from 1.81 mmol m⁻³ in January to 0.84 mmol m⁻³ in March) and rose slowly and irregularly from April until December. At −60 m, concentrations decreased from January to March (1.30 mmol m⁻³) and then increased, reaching the highest values (around 2 mmol m⁻³) from August to October. Consequently, surface waters were depleted in Si(OH)₄ relative to the other depths during the entire summer but replete in winter. Si(OH)₄ concentrations strongly fluctuated between years at 0 m, leading to an alternation of years with the highest concentration at the surface and (more frequent) years with the highest concentration at the bottom. A significant trend for Si(OH)₄ concentrations was observed only at −20 m depth, characterized by a decrease of 0.86% per year between 1995 and 2019.

3.3 Chlorophyll a

The median values of Chl a concentrations over the entire period (Table 1) displayed a clear vertical gradient from the surface to the bottom and an integrated mean concentration of 0.54 ± 0.40 mg m⁻³. The annual cycle of Chl a presented two seasonal peaks, typical of temperate areas (Figure 6A). At the surface, Chl a showed the first significant accumulation in March, after the winter minima (December–February 0.63 mg m⁻³), and reached a peak in May (2.62 mg m⁻³). The Chl a concentration then decreased to a relative minimum in August (1.26 mg m⁻³). A secondary peak was detected in late summer-early autumn, with the highest surface concentration in September (1.62 mg m⁻³). At −20 m, −60 m, and for the integrated mean, the spring peak occurred earlier (in March), while May to September was a period of low concentrations. Another increase occurred in autumn, with the highest concentrations in October for the integrated mean (0.68 mg m⁻³), in November at −20 m (0.71 mg m⁻³), and in December at −60 m (0.26 mg m⁻³).

Chl a concentration was higher in the period 1984–1990 than in 1995–2019 at all depths (p < 0.05). Between 1984 and 1990, decreasing trends were observed at all depths, except −60 m (Figure 6B), ranging between −0.12 and −0.03 mg m⁻³ year⁻¹, depending on depth (Table 2). With the exception of the surface layer, Chl a concentrations varied by less than a factor 2 over the whole time series, with an increase in the final part of the series (+0.04 mg m⁻³ year⁻¹).

We independently analysed the two main seasonal blooms occurring in the GoN in spring (March–June) and autumn (September–December). In spring, the median Chl a concentrations decreased with depth and the integrated mean concentration was 0.71 ± 0.44 mg m⁻³. Fluctuations were quite strong at 0 m (Figure 6C), displaying differences of almost an order of magnitude (4.96 mg m⁻³ in 1984 and 0.66 mg m⁻³ in 1991). Spring Chl a concentration was significantly higher in the period 1984–1991 at all depths (p < 0.01), while it increased at the surface and as integrated mean (0.05 and 0.01 mg m⁻³ year⁻¹, respectively) from 1995 to 2019.

In autumn, Chl a concentration was less variable along the water column than in spring and displayed a mean integrated value (0.58 ± 0.39 mg m⁻³, Figure 6D) almost similar to the spring value. Autumn Chl a showed trends similar to those detected for the entire year, decreasing from 1984 to 1990 and increasing during the 1995–2019 period at all depths. In addition to the trend observed for the concentrations, the autumn bloom displayed a change in seasonality, assessed as the week of maximum concentration. Measured over the entire water column, the Chl a maximum value occurred 4 weeks later in the period 1997–2019 compared with the period 1984–1990 (week 44 ± 4 and 40 ± 6, respectively, p < 0.05). At −20 m, the maximum Chl a concentration occurred increasingly later over the 1997–2019 period (Figure 6E), and the steepest change was detected between 2000 and 2006. On average, the maximum Chl a at −20 m occurred in week 41 in the period 1984–1990 and in week 47.5 in 2007–2019, corresponding to a delay of 6–7 weeks.

3.4 Anthropogenic pressures and environmental drivers

From the vertical gradients of N and P, we inferred that nutrient inputs from land play a crucial role in the system. Therefore, we attempted to estimate the household, industrial, and agricultural mobilization of N and P in the territory insisting on the GoN (basically the province of Naples), being aware of the difficulty of quantitatively assessing the transfer functions. The analysis highlighted that most of the mobilized N was of industrial origin, while most of the P was of agricultural origin till 2010, when household P sources became dominant (Figure 7, Table 3).

N production due to household activities increased between 1984 and 1999, from 10,343 to 11,773 t-N year⁻¹, and subsequently decreased until 2019, returning to values similar to those in the mid-1980s. Household P production, which encompasses both physiological P excretion and the use of P-containing detergents, strongly decreased between 1984 and 1991, from an estimated 3409 t-P year⁻¹ to 1913 t-P year⁻¹, and to 1760 t-P year⁻¹ in 2019. Based on our calculations, changes were due to the modification of dietary habits (increasing consumption of proteins in the 1980s and the 1990s and decreasing afterwards) and to the strong reduction in the detergent P content between 1985 and 1995 (data not shown). Estimated data on industrial N and P production were not available for all years. However, a slight decrease was visible between 1981 and the 1990s (Table 3), and a strong decrease occurred between 2007 and 2011 (−23% in 4 years for both N and P, passing from 20,809 t-N year⁻¹ and 123 t-P year⁻¹ to 18,430 t-N year⁻¹ and 109 t-P year⁻¹). A slight increase occurred in the following years, without reaching the pre-2007 levels.

Based on a conservative estimation limited to depurated wastewater from coastal municipalities, at least 11,790 t-N year⁻¹ and 651 t-P year⁻¹ from point sources flowed into the GoN in the period 2010–2019 (Table 3).

Agricultural N and P surpluses fluctuated significantly from year to year, with major decreases in the 2000s. N and P surpluses decreased by 77% and 83%, respectively, between 2005 and 2013. In the successive years, the nutrient surplus remained low. This trend was linked to a massive reduction in the use of fertilizer, whereas the crop production decreased by a smaller proportion over the same period (Figure S4). Considering the soil retention, we estimated that the agricultural contribution to the GoN waters would have decreased from 4671 t-N year⁻¹ and 720 t-P year⁻¹ to 1158 t-N year⁻¹ and 237 t-P year⁻¹ between the 1980s and the 2010s.

Compared with the main anthropogenic pressures, the atmospheric N deposition (884 ± 172 t-N year⁻¹ directly deposited on the GoN surface in 2000–2019) and the discharge of N and P by the Sarno River (3463 ± 1147 t-N year⁻¹ and 193 ± 76 t-P year⁻¹) may be considered minor fluxes.

We then analysed the relationship between N and P mobilization on land and their concentrations in the water.

No significant relationship was found between industrial N production and the concentrations of ammonium or nitrate in the GoN. Conversely, positive correlations were found between the phosphate concentration and the sum of agricultural and point sources P loads in the GoN over 2000–2019 (r² = 0.28 at 0 m, 0.46 at −20 m, 0.37 at −60 m, and 0.46 on the integrated mean, p < 0.05), with agricultural loads explaining 99% of the variation.

Finally, we attempted to identify the main drivers of chlorophyll accumulation using in situ data. The chlorophyll concentration was modeled (Figure 8) independently for the whole year, the spring bloom (March–June), and the autumn bloom (September–December). The chlorophyll concentration was influenced by different mechanisms depending on the depth and time period considered. When averaged over the whole year, the NO₃⁻ concentration increased the chlorophyll concentration (explaining 20 to 33% of its interannual variation), while 19% to 39% of the variation was explained by the MLD (negative impact) or the lateral advection of fresher water (positive impact). It should be noted that the models including either MLD or lateral advection of freshwater had a very similar likelihood (measured by AIC; data not shown), highlighting the close correlation between the two variables.

When investigating the spring bloom, salinity explained most of the Chl a concentration at 0 m, whereas nitrates, temperature, lateral advection of fresher water, MLD, and cloudiness explained the concentration at other depths.

For the autumn bloom, the explanatory power of the models was low, but we identified the week of stratification end as explaining 23% of the integrated mean chlorophyll concentration. Lateral advection of fresher water and MLD explained together 29% of the chlorophyll at 0 m, and NO₃⁻ concentration explained 21% of the chlorophyll at −60 m.

4.1 Nutrient sources sustaining MC-LTER planktonic food web

The territory around the GoN hosts more than 3 million people who mobilize N and P through different activities, as summarized in Table 3. These numbers are computed using methods frequently adopted to infer mobilization and loads. Because these are indirect assessments based on several assumptions, they provide an order of magnitude rather than exact values. Considering the stocks mobilized in the entire province of Naples for the decade 2010–2019 as an upper bound, ~31 kt-N year⁻¹ and ~3 kt-P year⁻¹ could potentially be transferred to the GoN. These values are quite high, larger than the discharge of a large river such as the Ebro (26 kt-N year⁻¹, 0.8 kt-P year⁻¹) during the first decade of the millennium (Cozzi et al., 2019). Moreover, they would represent a significant terrestrial contribution also for the Tyrrhenian Sea as a whole, since the estimates for the riverine inputs amount to 88 kt-N year⁻¹ and 4.1 kt-P year⁻¹ in the 1990s (Ludwig et al., 2009), and the atmospheric depositions from 39 to 104 kt-N year⁻¹ and 12 to 40 kt-P year⁻¹ (extrapolating deposition data from Corsica to the basin; Desboeufs et al., 2018). A lower bound of 13 kt-N year⁻¹ and 0.9 kt-P year⁻¹ could be estimated by considering: 1. only the population living in the cities or towns on the coast (~40% reduction); 2. wastewater treatment, whose products are partly conveyed to the neighboring Gulf of Gaeta (~25% reduction); 3. soil retention (~50% and ~83% reduction in N and P, respectively). These values may be scaled up by a factor of 1.4 and 2.1 for N and P, respectively, considering the estimated mobilizations during 1980s (Table 3). The estimated ranges were reduced from 44 to 18 kt-N year⁻¹ and 7.1 to 1.9 kt-P year⁻¹, respectively. Phytoplankton growth and accumulation occur mostly within the 100 m isobaths, where LTER-MC is located (e.g., Cianelli et al., 2017; Uttieri et al., 2011). These areas account for ~30% of the Gulf surface (corresponding to ~260 km²). Assuming that the largest fraction of terrestrial inputs is utilized in these areas and that nutrient uptake occurs with a Redfield Ratio of C:N = 106:16 (Redfield et al., 1963), they would sustain an annual primary production of ≤674 to 283 g C m⁻² year⁻¹ or ≤ 519 to 141 g C m⁻² year⁻¹ depending on the element, N or P, which determines the carrying capacity. These numbers increased to ≤957–392 g C m⁻² year⁻¹ or ≤ 1116–299 g C m⁻² year⁻¹ for the first period of the time series. The derived minima were strikingly close to the observed value of primary production measured with the ¹⁴C incubation method in the first 4 years (1984–1988) of the time series, which ranged from 302 to 404 g C m⁻² year⁻¹ (Zingone et al., 1995). Other measurements (12, mostly in spring and summer) in the years 2004–2005 averaged 477 g C m ⁻² year⁻¹ (Santarpia, 2006), while during 2007–2008 (14 measurements), the average value was 237 g C m⁻² year⁻¹ (unpublished data), which is also close to the estimates based on the corrected nutrient load. This leads us to conclude that, with the caveat of the several assumptions made, the main source of nutrients for plankton at LTER-MC is inputs from land, certainly during spring and summer. This is also supported by the nutrient minima at −20 m, which suggests that upward diffusion from the layer below plays a minor role. Even during winter, when the vertical mixing supplies nutrients to the surface, they are utilized by phytoplankton in a minimal amount, and the episodic blooms are supported by nutrients supplied by runoff and are transported to the station with fresher water (Zingone et al., 2010).

4.2 Patterns and drivers of LTER-MC vertical dynamics

The presence of three layers that remain decoupled for several months copes well with what is known about the horizontal dynamics of the GoN and the impact of terrestrial inputs. In brief, the December to March period, when the water column is homogeneous because of recurrent convective and wind mixing events and the flushing with Tyrrhenian waters is more effective, is also a period with high terrestrial water inputs. Conversely, in summer, flushing is less frequent, although not completely absent, and runoff is weaker (Cianelli et al., 2015, 2017; de Ruggiero et al., 2016; Uttieri et al., 2011). The interplay between terrestrial inputs and flushing creates fluctuations in the water types (coastal, offshore, or mixed) at the study site LTER-MC. A typical sequence is observed within a few days to 2 weeks: (i) nutrient-rich runoff, (ii) phytoplankton response and biomass accumulation, and (iii) vertical redistribution or horizontal dispersion of nutrients and biomass. This sequence suggests that terrestrial nutrients are underutilized at LTER-MC in winter and possibly exported offshore together with the biomass accumulated near the coast, while they are much better utilized in summer and autumn.

When the three-layer system stabilizes, upon the start of stratification, the sporadic events of nutrient enrichment related to the intrusion of deeper offshore waters with LIW entrainment do not significantly contribute to an increase in phytoplankton biomass, while a subsurface maximum of Chl a (DCM) is almost absent at LTER-MC. In stratified oligotrophic waters, DCMs are typically located near the base of the euphotic zone, matching the top of the nutricline (Cullen, 2015; Estrada et al., 1993; Marañón et al., 2021). The nutricline and the DCM in the Tyrrhenian Sea reside around 65–70 m (Lavigne et al., 2015), which is very close to the bottom depth of LTER-MC, and the relatively high biomass at the surface in summer is again linked to waters of coastal origin (D'Alelio et al., 2015), indicating the absence of nutrient limitation at the surface, which prevents the establishment of a persistent DCM.

In autumn, the water entrained in the mixed layer is deprived of nutrients (Figure S4), and biomass accumulation is still dependent on terrestrial inputs or recycled nutrients.

4.3 The LTER-MC trophic regime

Notwithstanding the often conspicuous terrigenous inputs, the analysis of multidecadal nutrient data acquired at the LTER-MC station does not reveal a ‘heavily eutrophic’ condition at the sampling site. The observed nutrient concentrations are, on average, more similar to the values reported for other Mediterranean time series, for example, Blanes Bay (Gasol et al., 2016), Villefranche-sur-Mer (Vandromme et al., 2011), and the Balearic Sea (Fernández de Puelles et al., 2007) and lower than those of highly eutrophic sites, such as the Northern Adriatic Sea (Grilli et al., 2020) or the Venice Lagoon (Bernardi-Aubry et al., 2020). Also our direct and indirect estimates of annual primary production, despite some occasional peaks of high phytoplankton biomass (0.07% of samples above 10 mg Chl a m⁻³ and 2.28% above 5 mg Chl a m⁻³), are, on the average, significantly lower than those of some eutrophic gulfs, e.g., Izmir Bay in Turkey (Bizsel & Uslu, 2000). The spread of half-saturation constants for inorganic nitrogen uptake is wide and varies among chemical species, environments, and taxa (e.g., Mulholland & Lomas, 2008). Assuming 0.5 or 1.0 mmol m⁻³ as representative values, 40% of samples had a NO₃⁻ concentrations higher than 0.5 mmol m⁻³ and only 15% had concentration >1 mmol m⁻³. Applying the same thresholds to NH₄⁺, since ammonia is a relevant source of dissolved inorganic nitrogen in the GoN, the observed percentage is quite similar (>0.5 mmol m⁻³ in 40% and >1 mmol m⁻³ in 18% of observations, respectively).

In fact, dramatic effects related to eutrophication (hypoxia, HABs, and fish mortality) have never been reported in the GoN. Only some greenish seawater discoloration and turbidity events caused by diatoms and small flagellates were observed in the 1980s (Zingone et al., 2006 and references therein).

Physical dynamics mitigate anthropogenic impacts by periodically diluting land-derived inputs and phytoplankton biomass built up along the shore. A similar scenario was also observed in the adjacent Bay of Pozzuoli, characterized by a relatively good status of waters (Margiotta et al., 2020) in contrast to the high pollution of sediments (Cavaliere et al., 2021; Trifuoggi et al., 2017). The key role played by meteorological forcing and large-scale circulation in exacerbating or mitigating the effects of eutrophication has also been observed in different coastal areas of Southern Europe (Cozzi et al., 2019). Numerical studies and radar observations have confirmed that the flushing time of the inner part of the GoN can range from hours to a few days (Cianelli et al., 2015, 2017; de Ruggiero et al., 2016).

4.4 Seasonal and interannual variability

Notwithstanding the strong environmental variability characterizing the study area, robust seasonal signals were observed not only in the temperature and salinity data, as previously reported (Kokoszka, Le Roux, et al., 2022; Ribera d'Alcalà et al., 2004), but also for nutrients and Chl a.

The maximum nutrient concentrations at the surface are generally observed in February–May, in conjunction with the annual maxima of freshwater inputs (Kokoszka, Le Roux, et al., 2022) while the highest Chl a concentration occurs toward the end of this period (April–May), when the stratification increases, as observed in other temperate ecosystems (Cloern & Jassby, 2008; Lavigne et al., 2013).

The observed seasonality is rather constant over the study period for almost all the investigated variables, thus indicating a ‘stable’ system despite the high intermittence of terrestrial inputs. The regularity and stability of the seasonal cycle have been observed in phytoplankton (Longobardi et al., 2022) as well as in zooplankton communities (Mazzocchi et al., 2023). The recurrent seasonality patterns are partially explained by the key role played by astronomical forcing (e.g., annual cycle of solar radiation and photoperiod) at these latitudes, as also pointed out by Cloern and Jassby (2010) in a comparison study focused on phytoplankton variability. It is worth noting the variation in the timing of the autumn blooms. The LTER-MC displayed the maximum autumn values of integrated (0–60 m) Chl a in September–October during the period 1984–1990. In more recent years, this autumn peak is shifted to October–November (Figure S5) with a lower median concentration in September, and it expands below the surface because of the deepening of the MLD. However, as argued by Kokoszka, Le Roux, et al. (2022), the increasing salinity at the bottom, which is a common trend in the Tyrrhenian Sea (Fedele et al., 2022), enhances water column stability and requires more energy to disrupt the stratification of the whole water column. This suggests that autumnal phenology is delayed as a result of climate change, with possible impacts on coupling between trophic levels (Atkinson et al., 2015). In fact, the significant changes observed in zooplankton after 2011, particularly in autumn, and the increase in <5 μm phytoplankton concentrations (Mazzocchi et al., 2023), seem to confirm this hypothesis. Moreover, the interannual variability of the physical properties recorded at LTER-MC is modulated by larger phenomena observed at the basin scale, especially in the case of extreme events. For instance, the high temperatures recorded in summer 2003 were observed in all of the Mediterranean Sea (e.g., Pisano et al., 2020) causing well-documented mass mortality events (e.g., Garrabou et al., 2009; Lejeusne et al., 2010), as well as the low temperatures in winter 2004/2005 and 2005/2006, which modified the deep water in the Western Mediterranean Sea as a consequence of intense mixing (López-Jurado et al., 2005; Marty & Chiavérini, 2010; Schroeder et al., 2016).

The variations in phosphorus and nitrogen concentrations are affected by large-scale processes but also seem to be strongly influenced by local drivers. The decrease in the use of phosphate-containing detergents caused by EU regulations and the reduction of agricultural activities are the main causes of phosphate reduction in the GoN. However, these two processes were observed in different time frames. A comparison with data from older literature (Carrada et al., 1974, 1980; Zingone et al., 1990) suggests a decrease from the 1970s to the second part of the 1990s due to the regulation of detergents, which also had a visible effect in several European seas (Desmit et al., 2019; Moon et al., 2016; Solidoro et al., 2009). In the period 2002–2017, the reduction in agricultural activities had a large local impact on phosphate concentrations. Agriculture plays a minor role in other areas (Artioli et al., 2008; Malagó et al., 2019; White & Hammond, 2009) because P is strongly retained in the soil. It might play a key role in the the GoN because of the characteristics of the volcanic soils of the area (De Vivo et al., 2006; Maisto et al., 2017; Vingiani et al., 2015), which would poorly retain P (Hinsinger, 2001; Penn & Camberato, 2019). A reduction in phosphate concentrations in the 2000s was also observed in the northern Adriatic Sea due to low runoff caused by a deficit in precipitation (Cozzi et al., 2020; Giani et al., 2012) and in the surface layer (0-100 m) of the Western Mediterranean Sea offshore waters (Belgacem et al., 2021).

For N, the observed interannual variability is more complex. In the second part of the time series, the reduction in ammonia stock could be tentatively attributed to improved wastewater treatment, as observed in other settings (Grizzetti et al., 2012). For the nitrate concentrations, a decreasing trend was observed in the intermediate layer, whereas a significant increase was detected at the bottom. However, direct relationships between activity on land and concentrations at sea are not recognizable. The absence of relationship might be due to the industrial N mobilization index, which tracks inappropriately temporal variations as it does not take into account changes in industrial processes. It could also be due to not linear relationships that imply additional factors, such as climate, local hydrodynamic conditions, and the stock of nutrients previously accumulated (Grizzetti et al., 2012). In this context, the reduced flushing detected in the years 2010–2019 caused by different wind regimes (Kokoszka, Le Roux, et al., 2022) may also explain the increase in nitrate concentration over the last years of the time series.

Chlorophyll a concentration displayed strong interannual variations, as well as long-term trends of decrease (1984–1991) and increase (1995–2019). The decrease in the first part of the time series could be related to the estimated reduction in anthropogenic loads, as hypothesized for the nutrient concentrations. By contrast, the increasing trend in the period 1995–2019 is more difficult to explain because nutrient concentrations displayed decreasing trends. However, Chl a interannual variability is linked to local conditions and large-scale phenomena. The increase in lateral advection of fresher water may drive the observed variability, especially during the spring blooms, and may be attributed to rainfall or water circulation variability (Kokoszka, Le Roux, et al., 2022), as also observed in other Mediterranean coastal environments (Kotta & Kitsiou, 2019; Marchese et al., 2015; Polat & Terbiyik, 2013). As rainfall modifications are linked to large-scale atmospheric pressure oscillations (Caloiero et al., 2011; Kokoszka, Le Roux, et al., 2022), local biomass accumulation is indirectly connected to large-scale atmospheric phenomena. A positive trend in Chl a concentration was observed along the coastal areas of the Tyrrhenian Sea during 1998–2009 by satellite (Colella et al., 2016), which seems to confirm that the trends observed locally might be part of a larger-scale phenomenon.