From Iron Curtain to green belt: shift from heterotrophic to autotrophic nitrogen retention in the Elbe River over 35 years of passive restoration

We investigate changes to in-stream nitrogen retention and metabolic processes in the Elbe River between 1978 and 2020. We analyzed multidecadal time series data and developed a metabolic nitrogen demand model to explain trends in dissolved inorganic nitrogen (DIN) retention, gross primary production (GPP), and ecosystem respiration (ER) during a period of highly dynamic pollution pressures in the Elbe River (central Europe). Our findings reveal a marked increase in summer DIN retention and a decrease in winter DIN retention, establishing a distinct seasonal pattern. We identified three distinct periods in the Elbe's DIN retention dynamics: dominantly heterotrophic under high organic and inorganic pollution pressure (1980–1990), transition (1990–2003), and dominantly autotrophic with lower pollution (2003–2017). We link these changes to reduced industrial pollution, improved wastewater treatment, and a shift in the in-stream balance between heterotrophic and autotrophic processes. During the first period, high ER caused elevated metabolic nitrogen demands, primarily driven by heterotrophic processes. As pollution from industrial and wastewater emissions decreased, GPP rates increased and ER gradually declined, prompting a shift towards an autotrophic-dominated nitrogen retention regime. Our study indicates tight coupling of nutrient reduction from external sources and dominant processes of natural attenuation in large rivers, which needs to be considered for projections of recovery trajectories for sustainable water quality.

1 Introduction

Large river systems have been substantially impacted by anthropogenic pressures associated with economic development, including observed long-term trends of ecosystem degradation throughout much of the 20th century (Vörösmarty et al., 2015; Meybeck et al., 2018). However, ecosystem protection regulations promulgated in recent decades have supported the recovery of many river ecosystems (Minaudo et al., 2015; Westphal et al., 2019), including producing improvements in river metabolic regimes (Diamond et al., 2022; Jarvie et al., 2022) and reductions in dissolved inorganic nitrogen loads (Wachholz et al., 2022). Many studies have examined the terrestrial drivers of these multidecadal trajectories of river ecosystems (Ehrhardt et al., 2019; Dupas et al., 2018; Van Meter et al., 2017). However, little is known about the impact of these long-term changes on the in-stream processes. An important in-stream process that is susceptible to external pressures is in-stream retention of dissolved inorganic nitrogen (DIN), which plays a crucial role in watershed nitrogen (N) budgets. On a global scale, inland waters retain 60 % of the annual terrestrial nitrogen input (Schlesinger and Bernhardt, 2013). This important ecosystem function also helps protect downstream ecosystems from eutrophication (Bianchi et al., 2010) induced by highly reactive forms of nitrogen, such as nitrate and ammonium (Seitzinger et al., 2002). In-stream DIN retention is closely linked to other ecosystem functions, especially stream metabolism (Hall and Tank, 2003; Heffernan and Cohen, 2010). Long-term changes in metabolism (Arroita et al., 2019), nitrogen loading (Ballard et al., 2019), and nitrogen composition (Wachholz et al., 2022) have been observed in rivers. Here, we are interested in associated long-term patterns of in-stream DIN retention.

Here we use in-stream DIN retention as an overarching term for all processes that, temporarily or permanently, remove DIN from the water. These processes include assimilation by heterotrophic and autotrophic organisms and denitrification. In-stream DIN retention is performed by bacteria, algae, and macrophytes in the water column and in sediments (Deutsch et al., 2009; Middelburg and Nieuwenhuize, 2000) that either assimilate DIN into their biomass or use it for metabolic processes. The activity of these organisms is influenced by various environmental factors such as water temperature, residence time, and nutrient concentrations (Collos and Harrison, 2014; Rasmussen et al., 2011; Snell et al., 2019). While travel time (Bertuzzo et al., 2017) and water temperature (Sherman et al., 2016) are often assumed to be the primary controls of in-stream DIN retention, the composition of DIN can also play a significant role. For instance, NH₄ N is favored over NO₃ N by many algae and bacteria (Cejudo et al., 2020). If sufficient NH₄ N is available, the DIN uptake by unicellular algae can increase by a factor of 2–16 (Collos and Harrison, 2014). Understanding of DIN retention should also consider the relative contributions of algae, bacteria, and macrophytes, each of which has preferences for different DIN species (Bergbusch et al., 2021; Collos and Harrison, 2014), incorporates N into its biomass at different stoichiometric ratios (Diamond et al., 2022; Godwin and Cotner, 2018), and has different growth efficiencies (the ratio of consumed resources that are assimilated into biomass; e.g., del Giorgio, 1997).

While long-term trends in the drivers and correlates with in-stream DIN retention are relatively well known (e.g., Ballard et al., 2019; Wachholz et al., 2022; Diamond et al., 2022), their consequences for in-stream DIN retention itself are understudied. This leaves considerable uncertainty in long-term watershed N budgets, which are already uncertain due to hard-to-quantify phenomena such as time lags (Lutz et al., 2022). Therefore, we propose the following research question: how do the magnitude and dominant processes of in-stream DIN retention change in response to long-term changes in DIN composition and river trophic regime? To answer this question, we studied the Elbe River from 1978 to 2017. During this period, the Elbe River underwent a significant transition: before 1990, most of its catchment lay beyond the Iron Curtain and experienced significant chemical pollution, especially from heavy fertilizer use due to the agro-industrial revolution in the German Democratic Republic (GDR) after the 1960s (Bauerkaemper, 2004). Furthermore, large amounts of untreated wastewater from urban and industrial areas further polluted the stream (Netzband et al., 2002). However, after the GDR's collapse in 1989, industrial facilities closed, and wastewater treatment plants (WWTPs) were rapidly constructed in the 1990s following German reunification in 1991, resulting in decreased emissions from these sources and improved water quality in the Elbe (Adams et al., 2001). Parts of the Elbe's remaining floodplain are now located within the European Green Belt, with the aim to preserve its valuable functions for flood retention and biodiversity (Serra-Llobet et al., 2022).

Previous work suggests a shift from a heterotrophic system (primary production/respiration < 1) to an autotrophic system (primary production/respiration > 1; Doretto et al., 2020) following the reduction in riverine biological oxygen demand in response to wastewater treatment improvements after the GDR collapse (Lehmann and Rode, 2001). However, the concomitant changes in DIN retention during this period have not been examined.

Quantitative links between in-stream metabolism and nutrient retention have been described by many authors (Hall and Tank, 2003; Heffernan and Cohen, 2010; Kamjunke et al., 2021; Rode et al., 2016; Zhang et al., 2023). The N demand of gross primary production (GPP) and ecosystem respiration (ER) in an ecosystem can be estimated, subject to assumptions about growth efficiencies (the share of GPP/ER that leads to biomass growth), biomass C : N ratios, and photosynthetic/respiratory quotients (O₂ : C ratio during photosynthesis/respiration; Hall and Tank, 2003). As both autotrophic and heterotrophic microorganisms use DIN as their preferred N source (Rier and Stevenson, 2002), those assumptions allow us to link the dissolved oxygen (DO) and the DIN balance of a river segment. Other processes, however, disturb this link by influencing either DO or DIN but not both. Possible examples are other biological processes such as nitrification (retains DO but does not affect the DIN budget directly) and denitrification (removes NO₃ N but does not consume DO). However, physicochemical effects such as adsorption or desorption of NH₄ N also influence the DIN budget of a river segment without affecting the DO budget (Triska et al., 1994).

GPP in the Elbe is mostly caused by phytoplankton (Hardenbicker et al., 2014), and phytoplankton activity is closely linked to the in-stream N processes in the Elbe in recent years (Ritz and Fischer, 2019; Kamjunke et al., 2021). In-stream denitrification is assumed to be of lesser importance in the Elbe, at least after reunification (Ritz et al., 2017; Schulz et al., 2023). The strong decrease in NH₄ N concentrations after 1990 (Adams et al., 2001) suggests that nitrification and sorption processes play a minor role in the Elbe's DIN retention, as less NH₄ N is available. Before 1990, in-stream oxygen concentrations were low (Lehmann and Rode, 2001), and NH₄ N concentrations were high (Adams et al., 2001). We, therefore, expect that nitrification and denitrification occurred at relatively higher rates, with weaker coupling between DIN retention and metabolic processes.

To summarize, we hypothesize a strong coupling between metabolic processes and in-stream DIN retention (Heffernan and Cohen, 2010). We predict that this coupling will be weakened during periods of high pollution (such as during the late phases of the GDR before 1990) when processes such as denitrification, nitrification, and sorption influence the DO and N cycles independently.

To test our hypothesis, we quantified DIN retention using a two-station mass balance approach along an 110 km eighth-order segment of the Elbe with no noteworthy tributaries. Furthermore, we quantified changes in the trophic regime in the Elbe by estimating gross primary production (GPP) and ecosystem respiration (ER) using the single-station hourly oxygen mass-balance approach (Odum, 1956). We linked in-stream DIN retention to metabolic rates using stoichiometric constraints and assessed the relative importance of autotrophic and heterotrophic processes.

2 Data and methods

2.1 Overview

We used DIN and DO mass balances to assess the magnitude and responsible processes of in-stream DIN retention over 42 years, from 1978 to 2020. We inversely modeled DO concentrations using a maximum likelihood estimation method to quantify in-stream metabolic processes (GPP and ER) over time. We linked metabolism and in-stream DIN retention using a simple model based on stoichiometric constraints and growth efficiencies.

2.2 Study site

We studied the last 111 km of a 1094 km eighth-order river (the Elbe; Fig. 1a, b), between Schnakenburg (kilometer 474) and Geesthacht (kilometer 585), which was behind the Iron Curtain before 1990. The German Elbe is free flowing, meaning that for at least 474 km upstream of the studied segment, no dams are present. The studied segment has no noteworthy tributaries and is located 30 km downstream of the last larger tributary (Havel), which contributes between 10 % and 20 % of the Elbe discharge (Fig. S1). The annual mean discharge at the downstream station Geesthacht (kilometer 585) is 716 m³ s⁻¹ (IKSE, 2005). Lateral groundwater discharge was estimated to be < 0.05 m³ s⁻¹ km⁻¹ for the Elbe at stream kilometer 450 (Zill et al., 2023), corresponding to 5.5 m³ s⁻¹ for the entire segment or less than 1 % of the annual mean discharge, and was therefore not considered. We estimate the depth of the segment to be between 2.8 and 4.5 m and the width between 270 and 450 m during flow conditions between the 5th and 95th percentiles (see Sect. 2.5 for details). PO₄ P concentrations oscillated between 0.2 and 0.3 mg L⁻¹ before 1990 and between 0 and 0.1 mg L⁻¹ afterward (Wachholz et al., 2024). The aquatic productive season occurs between April and October, as indicated by increased chlorophyll-a concentrations, and without macrophytes primary production in the Elbe is assumed to be caused by phytoplankton (Hardenbicker et al., 2018).

Figure 1 Location of the river segment investigated (a, c) in Europe. The blue line marks the main stream of the Elbe, and the red outline represents the catchment. Vertical black lines mark the beginning and the end of the segment (sampling locations). Red diamonds in panel (c) indicate discharge gages. The former area of the German Democratic Republic (GDR) is represented by the gray shaded area. Intra-annual patterns of water temperature (b) and travel time (d). Circles show raw data, and red lines are multi-year 7-day running means for each day of the year.

2.3 Two-station mass balance

To quantify in-stream DIN retention, we use a two-station mass balance approach (e.g., Ritz and Fischer, 2019). All samples were collected and analyzed by the federal state authority of Lower Saxony. Nitrate samples were analyzed using liquid ion chromatography (DIN EN ISO 10304-1) and ammonium using flow analysis and spectrometric detection (DIN EN ISO 11732). It is not documented if and how the detection method changed over the period monitored. However, the time series exhibited no obvious break points (Fig. S2). The upstream site (Schnackenburg) is located at stream kilometer 474 and has weekly (1978–1991) and bi-weekly (1991–2021) water quality (NO₃ N and NH₄ N) time series available. Nitrite (NO₂ N) was not considered, as even in high nitrogen pollution periods, it made up between 1 % and 3 % of the DIN for both input and output (Fig. S3). The upstream discharge gage (Wittenberge) is located 21 km upstream of the sampling site (Fig. 1c). The downstream site (Geesthacht) at stream kilometer 585 has weekly NO₃ N and NH₄ N data from 1980 to 1988, bi-weekly until 2006, and monthly afterward. A data gap from 1988 to 1993 for the downstream site was filled with data from another sampling site located at stream kilometer 598 (Zollenspieker), as both sites exhibit very similar nitrogen concentrations (Wachholz et al., 2022, their Fig. S2). The downstream discharge gage (Neu Darchau) is located 50 km upstream of the sampling site used to estimate DIN load (Geesthacht station, Fig. 1c). However, the difference in catchment area between the sampling site and the gage is less than 3 % (Wachholz et al., 2022). We consider this by assuming a 10 % error in discharge measurements in our uncertainty propagation, while a previous mass balance study (Ritz and Fischer, 2019) reported errors ≤ 5 % in the Elbe, according to personal communication with the state authorities responsible for the measurements. The same study also reported an error of 10 % for the laboratory quantification of nitrogen compounds, according to the German standard procedures that were also used in this study.

2.3.1 WRTDS model

To reconcile differences in sampling dates between the upstream and downstream stations, daily loads of NO₃ N and NH₄ N were estimated using the Weighted Regressions on Time, Discharge, and Season (WRTDS) function of the R package Exploration and Graphics for RivEr Trends (EGRET; Hirsch et al., 2010). The WRTDS function uses a weighted regression approach to estimate daily loads and concentrations, accounting for non-linearity and non-stationarity in the relationships between the time, discharge, season, and concentrations over time. Measured and simulated daily concentrations showed very good agreement (R²>0.7 and percent bias < 3; see Fig. S4).

2.3.2 Retention metrics

Using the daily loads provided by the WRTDS function, we calculated in-stream DIN retention R_obs as

$$\begin{array}{}\text{(1)}& R_{\mathrm{obs}}=L_{\mathrm{in}}-L_{\mathrm{out}},\end{array}$$

where L is the DIN load [kg d⁻¹]. Uncertainty in R_obs was computed based on Gaussian error propagation (Sect. S2). We then calculate the relative retention, RR,

$$\begin{array}{}\text{(2)}& {\mathrm{RR}}_{\mathrm{obs}}={\displaystyle \frac{R_{\mathrm{obs}}}{L_{\mathrm{in}}}},\end{array}$$

and the area-weighted retention, U [kg d⁻¹ m⁻²],

$$\begin{array}{}\text{(3)}& U_{\mathrm{obs}}={\displaystyle \frac{R_{\mathrm{obs}}}{A}},\end{array}$$

where A [m²] is the bottom area of the river segment (Stream Solute Workshop, 1990). We calculate Robs, RR_obs, and U_obs for DIN. We corrected the L_out time series for travel time to align the inflow and outflow time series. The estimated travel times for the segment ranged from 19 to 52 h (Fig. 1d), but since loads were only available as daily means, we evaluated corrections in increments of single days (Table S1) and found that shifting L_out 1 d ahead of the L_in series led to the best fit between the inflow and outflow time series.

2.4 Channel geometry estimations

Calculating area-weighted retention rates and inverse modeling of in-stream metabolism for a river segment requires knowledge of the surface area, water residence times, and channel depth. The methods to obtain these estimates are summarized below but are described in detail and validated in Sect. S1 and Fig. S5. We used discharge-based transfer functions to estimate the geometrical parameters at different water levels (e.g., Booker and Dunbar, 2008). We estimated travel time τ using a transfer function proposed by Scharfe et al. (2009) for the German Elbe. For the channel area of the investigated segment, we established a transfer function based on discharge, which we parametrized with channel areas derived under different discharge conditions from Sentinel 2 images using a surface water detection algorithm (Normalized Difference Water Index). We employed a power law model for the channel depth based on data from Aberle et al. (2010).

2.5 Metabolism model

The single-station method for metabolism estimation in rivers uses DO data from a single monitoring station to estimate the rates of gross primary production (GPP) and ecosystem respiration (ER) in the river (e.g., Hall et al., 2016):

$$\begin{array}{}\text{(4)}& \begin{array}{rl}{\mathrm{DO}}_{t+\mathrm{\Delta }t}& ={\mathrm{DO}}_t+{\displaystyle \frac{\mathrm{GPP}}{{\mathrm{PPFD}}_{\mathrm{24}}}}{\mathrm{PPFD}}_t-\mathrm{ER}\mathrm{\Delta }t\\ & +{\displaystyle \frac{k_{\mathrm{600}}{\left(\frac{\mathit{\text{Sc}}}{\mathrm{600}}\right)}^{\left(\frac{-\mathrm{1}}{\mathrm{2}}\right)}\left({\mathrm{DO}}_{\max }-{\mathrm{DO}}_t\right)}{z}}\mathrm{\Delta }t,\end{array}\end{array}$$

where GPP and ER are the daily rates [mmol m⁻³ d⁻¹] of the respective parameters, PPFD₂₄ is the photosynthetic photon flux density for the day and PPFD_t per hour [µmol m⁻³ d⁻¹], k₆₀₀ is the gas exchange rate [m d⁻¹], Sc is the dimensionless Schmidt number for oxygen (Wanninkhof, 1992) that is calculated based on water temperature, DO_t and DO_max are the actual and maximum (at 100 % saturation) DO concentrations [µmol L⁻¹], and z is the channel depth [m].

2.6 Data preparation and metabolism estimation

Implementation of Eq. (4) requires hourly estimates of PPFD, DO, DO_max, z, and Sc. We interpolated diurnal DO concentrations by fitting sine functions to a time series of daily mean, minimum, and maximum values from 1978 to 2017, as described in Sect. S3. Simulated DO concentrations were validated with 2 years of hourly measured values showing the characteristics of a good fit (R²=0.96, RMSE = 0.42 mg L⁻¹). We estimated hourly solar radiation (as PPFD) based on the method proposed by Duffie and Beckman (2013), which is implemented in the Python package solarPy. We calculated DO saturation based on the method of Weiss (1970) using hourly air pressure data from the German weather service (DWD), Seehausen station (ID 4642), and daily mean water temperature measured together with the DO data. We estimated k₆₀₀ for the segment with a hydraulic equation from Raymond et al. (2012, their Eq. 7 of Table 2; see Sect. S4 for details), and we used maximum likelihood estimation (e.g., Van de Bogert et al., 2007) to fit parameter distributions for GPP and ER for each day of the time series. We used a limited memory Broyden–Fletcher–Goldfarb–Shanno algorithm, implemented in the Python package Scipy (Virtanen et al., 2020; function minimize (method L-BFGS-B)), to minimize the negative log likelihood between the modeled DO concentrations and the observed DO data (e.g., Van de Bogert et al., 2007). We constrained daily GPP and ER estimates to be between 0 and (−) 50 [g O₂ m⁻² d⁻¹] and use (−)10 [g O₂ m⁻² d⁻¹] as an initial estimate.

In particular, the use of estimated k₆₀₀ and the daily mean water temperature might introduce large uncertainties into our GPP and ER estimations. To quantify these uncertainties, we estimated the daily standard deviation σ of k₆₀₀ (based on the parameter variability described in Raymond et al., 2012), the daily water temperature (based on hourly data that are available some years), and DO_max (based on the propagation of the standard deviations from water temperature) for each day (see Sect. S4). We performed 100 bootstrap iterations for each day, drawing errors from normal distributions, N ∼ (0, σ), for the parameters k₆₀₀, DO_max, and daily water temperature. We reported the mean and standard deviations of the resulting GPP and ER estimations. We assessed the goodness of fit for the GPP and ER estimations by simulating DO concentrations using the mean daily GPP and ER rates and comparing those to observed DO concentrations, reporting daily R² and root mean square error values (Fig. S10). We further assessed the effects of the uncertainty in k₆₀₀, DO_max, and daily water temperature on estimated GPP and ER by performing a simplified one-factor-at-a-time sensitivity analysis for 2 exemplary years (see Fig. S11). The model implementation in Python can be found at github.com/alexiwach/MetabolismModelElbe. We also assessed the areal extent of the metabolic signal and found that it was in line with the mass balance analysis (see Sect. S4).

2.7 Estimating the N demand of metabolic processes

The demand for N caused by GPP and ER can be estimated based on the respective organisms' growth efficiency (GE), the photosynthetic and respiratory quotient (PQ and RQ), and the C : N ratio of their biomass (e.g., Hall and Tank, 2003). GE [–] describes the proportion of resources and energy captured by photosynthesis (GE_AUTO) or respiration (GE_HET) that is incorporated into new biomass (del Giorgio et al., 1997; Hall and Tank, 2003). PQ and RQ describe the ratio of O₂ produced/consumed per CO₂ consumed/produced (Berggren et al., 2012; Hall and Tank, 2003). These two concepts can be used to assess how much C is incorporated into biomass for any given GPP and ER rate. Via the C : N ratio of the biomass (C : N_HET and C : N_AUT), the N demand can then be estimated. Since autotrophic and heterotrophic microbes prefer DIN to other forms of nitrogen (Rier and Stevenson, 2002) and DIN is always available at concentrations > 1 mg L⁻¹, we interpret the N demand as DIN demand.

For autotrophic processes, we can formulate the following:

$$\begin{array}{}\text{(5)}& U_{\mathrm{AUT}}\left(t\right)=\mathrm{GPP}\left(t\right){\displaystyle \frac{\mathrm{1}}{\mathrm{PQ}}}{\displaystyle \frac{{\mathrm{GE}}_{\mathrm{AUT}}}{{\mathrm{C}:\mathrm{N}}_{\mathrm{AUT}}}},\end{array}$$

where U_AUT [mol m⁻² d⁻¹] is the DIN demand of the GPP rate [mol m⁻³ d⁻¹] and PQ [mol C mol⁻¹ O₂], GE_AUT is the autotrophic growth efficiency [–], C : N_AUT is the carbon to nitrogen ratio in the autotrophic biomass, and z is the channel depth [m] used to convert volumetric GPP to areal retention rates. Similarly, we can formulate the following for heterotrophic processes:

$$\begin{array}{}\text{(6)}& U_{\mathrm{HET}}\left(t\right)=R_{\mathrm{HET}}\left(t\right)\mathrm{RQ}{\displaystyle \frac{{\mathrm{GE}}_{\mathrm{HET}}}{{\mathrm{C}:\mathrm{N}}_{\mathrm{HET}}}};\end{array}$$

however, it is well known that the ER is not only caused by heterotrophic bacteria but also by autotrophs (e.g., Hall and Tank, 2003). A way to correct autotrophic respiration is to subtract the non-biomass-producing fraction of GPP from ER (Hall and Tank, 2003).

$$\begin{array}{}\text{(7)}& R_{\mathrm{HET}}\left(t\right)=\mathrm{ER}\left(t\right)-{\mathrm{GE}}_{\mathrm{AUT}}\mathrm{GPP}\left(t\right)\end{array}$$

Combining Eqs. (5), (7), and (8) allows us to estimate the complete metabolic N demand U_met as follows:

$$\begin{array}{}\text{(8)}& \begin{array}{rl}{U}_{\mathrm{met}}\left(t\right)& =(\mathrm{GPP}\left(t\right){\displaystyle \frac{\mathrm{1}}{\mathrm{PQ}}}{\displaystyle \frac{{\mathrm{GE}}_{\mathrm{AUTO}}}{{\mathrm{C}:\mathrm{N}}_{\mathrm{AUTO}}}}\\ & +\left(\mathrm{ER}\left(t\right)-{\mathrm{GE}}_{\mathrm{AUTO}}\mathrm{GPP}\left(t\right)\right)\mathrm{RQ}{\displaystyle \frac{{\mathrm{GE}}_{\mathrm{HET}}}{{\mathrm{C}:\mathrm{N}}_{\mathrm{HET}}}})z\left(t\right).\end{array}\end{array}$$

Calculating U_met based on GPP and ER rates therefore requires estimations of six parameters. For C : N_AUT, we used 7.3, which was reported for phytoplankton biomass in the Elbe, and we used PQ = 1 [mol C mol⁻¹ O₂] and GE_AUTO=0.5, which have been successfully used to explain assimilatory DIN uptake in the Elbe (Kamjunke et al., 2021). It is well known that the parameters C : N_HET, GE_HET, and RQ show strong variability across ecosystems (Godwin and Cotner, 2018; del Giorgio and Cole, 1998; Berggren et al., 2012). Since they all affect U_met, we evaluated three parameter combinations that would lead to low, intermediate, and high U_met values (Table 1).

Table 1 Parameter combinations for the estimation of the metabolic nitrogen demand. C : N_HET is the carbon to nitrogen ratio of heterotrophic organisms, GE_HET is the heterotrophic growth efficiency [–], and RQ is the respiratory quotient [mol O₂ mol⁻¹ C].

We compared U_met estimated from each of these parameter combinations to U_obs based on the mass balance approach. U_met as introduced here only considers the nitrogen demand directly associated with assimilation related to primary production or respiration. Other processes, e.g., denitrification, are not considered but are discussed later.

3 Results and discussion

3.1 DIN retention

The highest U_obs values during the entire time series were observed during regime 1 (1978–1990), oscillating between 1 and 2 g m⁻² d⁻¹ (Fig. 2a). Starting in middle of regime 1, U_obs decreased and oscillated mostly between 0 and 1 for the rest of the time series, with some negative values occurring during the winter in 3 years (2008, 2011, and 2018). During regime 1, U_obs peaked shortly before the aquatic growing season (days 106–122, April–May) and showed clear summer peaks (day 182, July) afterward. The relative retention, however, showed a consistent summer peak during the entire time series (days 190–220, July–August), while the amplitude increased in regimes 2 and 3 (19 % to 34 %; Fig. 2b). Annual mean loads decreased throughout all three regimes, from 0.44 to 0.18 kt d⁻¹, with a substantial decrease in annual minima around 1989 (Fig. 2c). Likewise, the share of NH₄ N from the DIN load declined from 36 % to 6 %. Concentrations of NH₄ N did not reach values > 1 mg L⁻¹ until later in regime 2.

Figure 2 Area-weighted (U_obs; panel a) and relative dissolved inorganic nitrogen (DIN) retention rates (RR_obs; panel b) estimated using a two-station mass balance approach. The three vertical black lines correspond to the major changes in U. Regime 1 shows a high mean (μ), a strong seasonal amplitude (SI) and peaks during spring (Φ is day 106). Regime 3 has a much lower μ and SI, while the day of the peak occurs during summer (Φ is day 182). Regime 2 represents the transition between the two regimes. The black lines represent a 30 d moving average value, and the shaded area around the black line shows raw values with a 90 % confidence interval. Panel (c) shows the DIN load (L) received by the investigated segment of the Elbe. The color represents the corresponding ammonium concentration. ${\mathit{\mu }}_{{\mathrm{fNH}}_{\mathrm{4}}}$ is the mean fraction of DIN that consists of ammonium. The white background represents the colder 6 months of the year (October–April) and the dark background the warmer 6 months.

Figure 3 Time series of daily dissolved oxygen saturation (DO_sat; a), daily gross primary production estimates (GPP; c), and ecosystem respiration estimates (ER; e). The black lines show the annual median DO, GPP, and ER values. Panels (b), (d), and (f) show the mean monthly patterns for DO_sat, GPP, and ER before and after 1990. μ_DO represents the mean DO concentrations and μ_GPPann and μ_GPPann the mean annual GPP and ER rates.

3.2 Metabolism

3.2.1 Oxygen saturation

The multidecadal pattern of DO in the Elbe showed distinct behavior before and after 1990, coinciding with German reunification and the fall of the Iron Curtain (Fig. 3a). Oxygen saturation before 1990 oscillated between 20 % and 70 % but increased rapidly after 1990, reaching supersaturation for the first time in 1991. Before 1990 there was no clear intra-annual pattern (Fig. 2b), but for the rest of the time series, DO_sat oscillated seasonally between ∼ 80 % and ∼ 180 %, peaking around day 180 and coinciding with the annual peaks of residence time, water temperature, and area-weighted DIN retention (Figs. 1b, c and 2a).

3.2.2 Gross primary production and ecosystem respiration

Similar to DO_sat, GPP showed a clear change around 1990, with low annual peaks (10–15 g O₂ m⁻² d⁻¹) before and higher ones (∼ 15–25 g O₂ m⁻² d⁻¹) in the following years (Fig. 3c). Compared to DO_sat, the timing of the GPP peak stayed similar throughout the time series (July; Fig. 3d).

While the seasonal pattern of ER stayed similar (peaks in May, June, and July) throughout the time series, annual ER rates were reduced to less than half after 1990 (Fig. 3e, f). Most apparent are higher ER rates during the winter months before 1990. In contrast to GPP, ER rates started to change as early as 1987.

Considering annual mean net ecosystem productivity (NEP = GPP-ER), the segment of the Elbe investigated was a net-heterotrophic system before 1990, which means more O₂ was consumed than was produced. High ER rates before 1990 led to a DO deficit of ∼ 2800 g O₂ m⁻² yr⁻¹. After 1990, reduced ER and increased GPP rates turned the NEP of the segment and led to an oxygen surplus of ∼ 240 g O₂ m⁻² yr⁻¹.

Figure 4 (a) Simulated and observed in-stream nitrogen retention during three periods. The different colored boxplots represent the three parametrizations of heterotrophic nitrogen demands described in Table 1. U_obs represents retention rates based on the two-station mass balance described in Sect. 2.3. (b) Pearson correlation coefficients for the three parametrizations of heterotrophic nitrogen demands with U_obs. (c) Monthly mean time series of U based on the best-fit parameter combinations for each period: U_HEThigh for periods 1 and 2 and U_HETlow for period 3. The colors indicate the share of the heterotrophic and autotrophic nitrogen demand. f_AUT is the fraction of U caused by autotrophic nitrogen demand [%].

3.3 Linking metabolism and DIN retention

The different parametrizations of heterotrophic nitrogen demand show diverging performance over the three investigated periods that correspond to the three regimes of U_obs shown in Fig. 2. Between 1978 and 2003, only the parametrization U_METhigh led to a similar distribution of U as was observed (Fig. 4a). During the last period (2004–2016), all parameter combinations led to U distributions that were similar to the observations. Pearson's R reveals poor correlation between simulated and observed U values before 2003, with the best results shown by the U_METhigh parametrization (Fig. 4b). This changes for the period 2004–2016, where overall better R values were achieved, and the best results resulted from the parametrizations U_METlow and U_METmean. The share of autotrophic nitrogen demand to total metabolic nitrogen demand f_AUT increased for all three parametrizations (Fig. 4c). The highest f_AUT values come from the parametrization with low heterotrophic nitrogen demand (U_METlow) and vice versa.

4 Discussion

4.1 DIN retention

We classified observed in-stream nitrogen retention U_obs under three regimes. Regime 1 had the highest retention, showed a pronounced seasonal amplitude, and peaked shortly before the beginning of the aquatic vegetation period during the colder half of the year (Fig. 2a). Peaks outside the warmer half of the year were not expected, as water temperature (Sherman et al., 2016) and residence time (Bertuzzo et al., 2017), considered the main drivers of in-stream DIN retention, are lower during this period. During regime 2, U_obs gradually decreased, reduced its amplitude, and shifted its annual peak into summer, where it remained in regime 3. Higher DIN retention rates during phases of higher DIN pollution, as observed in regime 1, have already been reported by Kelly et al. (2021). They also remarked on changes in the drivers of U_obs, which reached high values even in the cold season during phases of high pollution, a fact that agrees with our observations. We interpret regime 2 as a transition period, corresponding to the long-term improvements in wastewater treatment that followed German reunification in 1990 and took multiple years (IKSE, 2010). During this period, NH₄ N concentrations started to decrease first during summer, indicating changes in point source inputs (Wachholz et al., 2022). During regime 3, U_obs shows a stable pattern, with peaks coinciding with high water temperatures and residence times (Fig. 1b, c).

As neither temperature nor discharge exhibited noteworthy trends in seasonality in the Elbe during the studied period (Markovic et al., 2013; Mudersbach et al., 2016), we interpret this seasonality change as an indicator of a change in other driving processes, most likely in the abundance and metabolic activity of the responsible organisms. While the DIN input has a constant seasonal pattern, its magnitude and composition (share of NH₄ N) changed remarkably throughout the time series (Fig. 2c). As biota such as algae are known to form their assemblages according to environmental factors such as light, temperature, and nutrient availability (Snell et al., 2019), a biotic regime shift could have contributed to the U_obs changes.

4.2 Metabolism

4.2.1 Dissolved oxygen

Analyses of the oxygen saturation in the investigated segment of the Elbe revealed a constant saturation deficit, which was replaced by a strong seasonal pattern with pronounced supersaturation during summer (Fig. 3a) after 1990. It is understood that the oxygen budget of the Elbe after 1990 is controlled by primary production, which rapidly increased after 1990 (Lehmann and Rode, 2001; Petersen and Callies, 2002). The absence of supersaturation before 1990 could be related to high concentrations of suspended solids, which are known to limit GPP in rivers (Trentman et al., 2022) and which were observed to decrease in the Elbe following the improvements in wastewater treatment after 1990 (Hillebrand et al., 2018). However, toxic effects from industrial waste also seem plausible, as concentrations for many organic and inorganic pollutants rapidly decreased around 1990 (Adams et al., 2001). However, a lack of primary production would not explain the significant DO saturation deficit before 1990. From a mass balance perspective, only high rates of ER could have caused that phenomenon.

Further assessment of this requires estimation of the Elbe's metabolic processes, which we carried out using a maximum likelihood estimation.

4.2.2 Gross primary production and ecosystem respiration

GPP and ER estimations revealed an increase in GPP and a decrease in ER around 1990 (Fig. 3c, d). While the GPP increase was sudden and mostly affected summer peaks, ER started to decrease in the late 1980s and also shows drastically reduced rates during the winter months. Compared to DO_sat, the timing of the GPP peak stayed similar throughout the time series (∼ day 190 (July); Fig. 3d), which suggests constant drivers but some limitation before 1990. It is well known that light and flow regimes control the metabolism of rivers (Bernhardt et al., 2022), so the peak during high temperature and residence times is to be expected and suggests another limiting factor before 1990 (e.g., high light attenuation from substances or sediments delivered by WWTP effluents).

The higher rates of ER before 1990 could have been supported by high loads of organic pollutants from wastewater (Adams et al., 1996). Compared to the increase in GPP, the slower ER decrease could be explained by the gradual improvements in water treatment throughout the 1990s and early 2000s (Adams et al., 2001; Wachholz et al., 2022). However, it has to be acknowledged that ER started to change before 1990. This could have been caused by an early onset of the industrial collapse of the GDR, affecting some industrial wastewater emitters, or by improved wastewater treatment before the GDR collapse.

Another interesting observation is the high winter ER before 1990. One must consider that the term ER from Eq. (4) contains all processes that consume DO in the balance, including nitrification. NH₄ N was present in high concentrations in the Elbe River before 1990 (Fig. 2c). It is well known that nitrification can significantly contribute to oxygen depletion (Powers et al., 2017) and that nitrification rates during cold periods increase with NH₄ N concentrations (Cavaliere and Baulch, 2019). This correlates with our observation of high DIN retention rates outside the warmer half of the year, as nitrifying and denitrifying organisms also have nitrogen demands.

It has to be acknowledged that the metabolic estimations presented here are based on a few speculative assumptions. Nevertheless, the model showed overall good agreement with the observed DO data (Fig. S10). R² was generally the highest during spring (> 0.8) and lowest in winter (∼ 0.6). R² performance was similar for the period before and after 1990. Considering the root-mean square error, all seasons in both periods show values between 0 and 2.5 mg O₂ m⁻² d⁻¹, with the exception of summer after 1990 where the mean root mean square error is around 7 mg O₂ m⁻² d⁻¹. This is, however, the part of the year with the highest GPP and ER values, and since the R² in this period and season is > 0.7, we still consider that the model performance is good.

Our simple one-factor-at-a-time sensitivity analysis revealed that both GPP and ER are most susceptible to changes in k₆₀₀ (Fig. S11). Since this parameter was also estimated from a hydraulic equation and could not be verified for the studied segment, this introduces an unquantified uncertainty to our results. GPP, however, was affected very little by variability in k₆₀₀ compared to ER, and as similar summer GPP rates were found in the Elbe in other studies (e.g., Kamjunke et al., 2021), we assume that the GPP values and long-term changes have a high certainty. The absolute ER and k₆₀₀ values for the Elbe could not be independently verified. However, the reduction in ER rates due to wastewater treatment improvements seems plausible. If we assume k₆₀₀ to be solely driven by hydraulic factors, for which we found no indication of change, then an overall decrease in ER could be a likely mechanistic explanation for the observed changes in DO.

4.3 Linking metabolism and DIN retention

Despite its simple nature, the metabolic N demand model showed promising results in predicting mean DIN retention rates (Fig. 4a–c). Between 1979 and 1990, only the parametrization U_METhigh was able to mimic the distribution of observed DIN retention rates correctly. Only after 2003 did the model gain noteworthy predictive capabilities for the observed daily DIN retention rates (R≥0.7) for the parametrizations U_METlow and U_METmean, with the parametrization U_METhigh performing worse (R≈0.5). The overall observation of low model performance before 2003 can be interpreted as a weakened coupling between metabolic processes and in-stream DIN retention. As we hypothesized, processes such as denitrification, NH₄ N sorption, and nitrification can impact the DIN and oxygen balance differentially, weakening the link between both elemental cycles. While we have no independent information on the rates of denitrification, NH₄ N sorption, and nitrification in the studied segment before 1990, we argue that the circumstances indicate that they were likely higher than after 2000. With little NH₄ N present in the segment after the improvements in wastewater treatment, neither sorption nor nitrification could have occurred at high rates. Nitrification, on the other hand, is assumed to have been a major drain on the oxygen budget of the Elbe before reunification (Wachholz et al., 2024). Denitrification is assumed to be a minor part of the Elbe's DIN retention in recent years (Schulz et al., 2023), but it is known that lower DO concentrations in eutrophic rivers, as were experienced in the Elbe before 1990, can promote denitrification (Rysgaard et al., 1994). The DIN retention after 2003 is likely largely due to assimilation by heterotrophic and autotrophic organisms and, therefore, closely linked to the metabolic processes, as indicated by the improved model performance in this period.

Our findings also have implications for how much of the metabolic nitrogen demand comes from the assimilation by autotrophic organisms. The U_METhigh parametrization, which is the only one able to explain the high DIN retention values before 1990 (Fig. 4a), predicts that more than 90 % of DIN retention comes from heterotrophic nitrogen demands. Even if this parametrization is unrealistic and denitrification would have been a major part of the observed DIN retention, it seems likely that most DIN retention, at least before 1990, was caused by heterotrophic organisms. On the other hand, after 2003, U_METlow and U_METmean show the best performance and predict that between 50 % and 80 % of the observed DIN retention was caused by autotrophic organisms. We interpret this as a shift between an in-stream DIN retention regime dominated by heterotrophic activity to one dominated by autotrophic activity.

As in-stream processes are greatly influenced by the hydroclimate of the river (e.g., Bernhardt et al., 2022) and changes in its flow regime, as caused for example by dams (e.g., Aristi et al., 2014), these are competing drivers for the observed changes. At least as of 2013, no systematic trends in the Elbe's flow regime had been documented (Mudersbach et al., 2016). The Elbe's water temperature is rising (∼ 0.01 °C yr⁻¹) in concert with rising air temperatures, and a phase shift towards 2-week-earlier warming has been described (Markovic et al., 2013). Based on these findings, however, we find it unlikely that hydroclimatic changes caused the changes in in-stream processes described here.

Our results also highlight the importance of the parameters RQ, GH_HET, and C : N_HET, which, within reasonable bounds, can change heterotrophic nitrogen demand by a factor of 10 (Fig. 4a). GE_HET is often positively correlated with overall bacterial productivity or phosphorus concentrations (del Giorgio and Cole, 1998; Smith and Prairie, 2004). As both declined during the studied period, a decline in GE_HET, as suggested by the decreasing performance of the U_METhigh parametrization, seems plausible. For RQ, the quality of the available organic matter seems to be important, with small organic acids leading to especially high RQs (Bergren et al., 2012; Allesson et al., 2016). However, this was not assessed in this study. Overall, the variability in the literature values for GH_HET is 10 times higher than for RQ (Table 1). Understanding and constraining the variability in these parameters, especially GE_HET, is therefore paramount for a better understanding of carbon, oxygen, and nitrogen cycles in rivers. In our simple analysis, we ignored variability in the parameters C : N_AUT, GE_AUT, and PQ, which could explain the still-improvable fits of the metabolic nitrogen demand model after 2003.

4.4 Ecological implications

Long-term changes in GPP and ER can have a multitude of ecological implications for the river segment itself or for downstream ecosystems. For example, high ER and low GPP, as we observed in the Elbe before 1990, can lead to increased riverine CO₂ emissions (Attermeyer et al., 2021). Higher primary production in the Elbe after 1990 caused greater export of organic matter to the estuary during summer, which supported higher rates of estuarine ecosystem respiration, which in turn decreased DO concentrations (Amann et al., 2012) and reduced available habitats for fish species over extended stretches of the tidal zones in the estuary (Mann, 1996).

The changed seasonality of DIN retention also likely had an impact on downstream ecosystems. Even though absolute retention rates before 1990 were higher (Fig. 2a), only 20 %–25 % of DIN was retained during summer (Fig. 2b), when algal blooms are most likely to occur. After German reunification, the decreased DIN load (Fig. 2c) and increased in-stream retention led to less DIN being exported to the estuary during summer, decreasing the probability of DIN-induced estuarine algal blooms (Anderson et al., 2008).

5 Conclusions

Our study provides valuable insights into the long-term effects of inorganic and organic pollution reduction and the tight coupling with the ecosystem functions of DIN retention and metabolism within large rivers. The shift from a heterotrophic-dominated to an autotrophic-dominated DIN retention regime has implications for the carbon cycle and algal blooms in downstream ecosystems.

Our findings highlight the importance of considering different dimensions of integrative ecosystem functions (metabolism and DIN retention) when assessing long-term ecological changes in rivers, such as eutrophication. Dissolved oxygen concentration time series alone are sufficient to support our findings of a heterotrophic to autotrophic regime shift, but estimates of GPP, ER, and DIN retention were required to support a quantitative assessment of the magnitude and consequences of this shift. The discovery of decoupled responses of ER and GPP to improvements in water quality offers new insights into the timescales of aquatic ecosystem responses to changing external forcing and informs realistic estimates of the efficiency of nutrient management and the achievement of environmental objectives.