Drivers of multi-decadal nitrate regime shifts in a large European catchment

The River Elbe has a length of 1094 km (66.4% in Germany, 33.6% in the Czech Republic) and a catchment size of approximately 148 000 km² (IKSE 2005). Between 1949 and 1990, approximately 80% of the German Elbe catchment belonged to the German Democratic Republic (GDR) territories. The upper Elbe (until 96 km downstream of the Czech–German border) flows through the Czech Giant and Bohemian Forest mountain ranges, the middle Elbe (until 585 km downstream of the Czech–German border) passing through the north German lowlands, and the lower Elbe is the tidally influenced section to the North Sea (figure 2(a)). Its catchment is situated in the temperate climate zone and marks the transition between continental and maritime regimes (IKSE 2005). Mean annual temperature ranges from 1 ^∘C in the Czech mountains to 9 ^∘C in the German lowlands (IKSE 2005). Mean annual precipitation ranges from 1810 mm yr⁻¹ (upstream mountain ranges) to less than 450 mm yr⁻¹ (lowlands), with a catchment average of 628 mm yr⁻¹ (IKSE 2005). The discharge regime of the Elbe is pluvio-nival (Hübner and Schwandt 2018), with spring maxima and autumn minima (figure 2(c)). Mean discharge at the outlet in Geesthacht is 716 m³ s⁻¹ and 318 m³ s⁻¹ at the Czech–German border (IKSE 2014). While the Elbe River is free-flowing in the German part of the catchment, the Czech sub-catchment contains 59 reservoirs and 29 dams (Hübner and Schwandt 2018). The land cover is 56% agricultural areas, 30% forests, 8% artificial surfaces, and 3.5% wetlands and water bodies (EEA 2018, figure S1 (available online at stacks.iop.org/ERL/17/064039/mmedia)).

Zoom In Zoom Out Reset image size

With the onset of the industrial revolution in the late 19th century, chemical pollution of the Elbe increased significantly, primarily from wastewater originating from urban areas, industry, and mining (Petermeier et al 1996). In the early 20th century, many sewer systems were built across Germany to collect and discharge urban emissions (Blackbourn 2011). During the so-called agro-industrial revolution in the GDR between the 1960s and 1980s, the use of fertilizers increased rapidly (Bauerkaemper 2004, figure S2). By the end of the 1980s, large amounts of wastewater from German and Czechoslovakian cities and industrial areas reached the stream untreated (Netzband et al 2002). After the collapse of the GDR, the subsequent closing of many industrial branches and rapid construction of WWTPs (figure S2, bottom) in the 1990s led to a decrease in chemical pollution from point sources, which caused a rapid improvement of water quality in the Elbe (Adams et al 2001). The decreased pollution also led to changes in trophic state of the river, which was dominated by heterotrophic processes before German reunification and became autotrophic afterward (Lehmann and Rode 2001). In 1991, just after the German reunification, the European Union (EU) introduced the Nitrates Directive to tackle high diffuse source nitrogen inputs into water bodies, by observing and limiting mineral fertilizer and manure application for agricultural practices (European Commission 1991).

Starting in 2010 and initiated by the Water Framework Directive (WFD), many measures were undertaken or planned to achieve at least good ecological status for all European surface water bodies by 2027 (European Commission 2000). In the Elbe catchment, these measures included the restoration of floodplains, the connection of oxbow lakes to rivers, the removal of transverse structures, and the further reduction of point and diffuse source nutrient inputs (Damm 2013). Measures that targeted point source emissions showed fast results: for example, nitrogen inputs from wastewater were reduced from 4300 to 2400 tons per year between 1994 and 1998 and reached 336 tons in 2008 (IKSE 1998, 2010). On the other hand, diffuse source nutrient inputs remain, together with morphological alterations, the most critical pressures preventing good ecological status of surface water bodies in the Elbe catchment (IKSE 2010). Averaged over the years 2006–2010, 84% of nitrogen inputs into the Elbe river network stem from diffuse sources (IKSE 2018).

Despite all its alterations, the Elbe is a rare European case of a free-flowing river along 622 km of its main course (IKSE 2010). Its remaining flood plains are recognized as hotspots of biodiversity (Meyerhoff and Dehnhardt 2007). In its current ecological state, the Elbe is an essential contributor to nitrogen attenuation (attributed to phytoplankton assimilation, seston deposition, and denitrification (Ritz and Fischer 2019)), which can collectively attenuate up to 30% of the nitrogen inputs (Deutsch et al 2009, Ritz and Fischer 2019, Kamjunke et al 2021). The pollution-recovery trajectory of the Elbe summarized above is consistent with the experience of other major European and North American rivers over the last century (Billen et al 2007, Minaudo et al 2015, Van Meter and Basu 2017), although with distinct differences related to the German experience.

Nitrate, water temperature, and ammonia were measured at the tidal weir in Geesthacht (stream kilometer 586, figure 2(a)) weekly from 1954 to 1988, biweekly from 1992 to 2005, and monthly from 2006 to 2019. A five-year gap from 1988 to 1993 was filled with data from Zollenspieker (figure 2(a)), 12 km downstream of the tidal weir. Although the latter site is influenced by the tides, in 26 years of data overlap, the nitrate time series from both sites showed very good correspondence (root mean square error = 0.25 mg l⁻¹, percent bias of 1.1%, figure S3). Samples at Zollenspieker were collected weekly until 1991 and bi-weekly afterward. Daily mean discharge records from 1954 to 2019 were measured at the gaging station Neu Darchau (figure 2(a)), located ca. 50 km upstream of Geesthacht. The areal difference between the two sub-catchments (Neu Darchau and Geesthacht) is less than 3% (IKSE 2005, Mudersbach et al 2016).

3.3.1. Seasonality analysis

3.3.2. MSSM

To test our hypothesis, we developed the MSSM considering only mixing ratios of point and diffuse source components (a_d and a_p ) to explain in-stream nitrate concentration seasonality. The seasonal oscillations are, as for the time series analysis described above, approximated with sine functions. As any sine function can be expressed as a composite of multiple sine functions with the same period (Smith 2007), we express the total nitrate concentrations $\mathrm{NO}_3\!{-}\!\mathrm{N}$ as a composite of point and diffuse source nitrate:

$$\begin{align} \mathrm{NO}_3\!{-}\!\mathrm{N} &= a_d + b_d \mathrm{cos}(\omega(t-\phi_d) + a_p\nonumber\\ &\quad + b_p \mathrm{cos}(\omega(t-\phi_p) \end{align} \tag{ 2 }$$

where the subscripts p and d represent nitrate contributions stemming from point and the diffuse sources, respectively. The sine function parameters ($a_{no3}$, $b_{no3}$ and $\phi_{no3}$) can be derived using the following trigonometric equations (Smith 2007):

$$\begin{equation} a_{no3} = a_d + a_p \end{equation} \tag{ 3 }$$

$$\begin{equation} b_{no3} = \sqrt{( (b_d~\mathrm{cos}(\omega~\phi_d) + (b_p~\mathrm{cos}(\omega~\phi_p) )^2 + ( (b_d~\mathrm{sin}(\omega~\phi_d) + (b_p~\mathrm{sin}(\omega~\phi_p) )^2} \end{equation} \tag{ 4 }$$

$$\begin{align} \phi_{no3} = \frac{1}{\omega}~\arctan2 \left\{\frac{b_d~\mathrm{sin}(\omega~\phi_d) + b_p~\mathrm{sin}(\omega~\phi_p)}{b_d~\mathrm{cos}(\omega~\phi_d) + b_p~\mathrm{cos}(\omega~\phi_p)} \right\}. \end{align} \tag{ 5 }$$

The procedure of choosing values for the parameters a_p , b_p , φ_p , a_d , b_d , φ_d and estimating their uncertainties is described in detail in sections S1.3–S1.5. In summary, we base the diffuse source sine parameters b_d and φ_d on parts of the time series where diffuse source pollution is assumed to be responsible for most of the riverine nitrate. The parameters for point source nitrate are based on the seasonality of discharge, assuming constant effluent loads over the year with varying dilution by river discharge. a_d and a_p are conceptualized based on the above described, nitrate relevant socio-technological transitions (figure 1(a)).

The annual mean nitrate concentration at the catchment outlet generally increased from 2.7 mg l⁻¹ in 1954 to 4.2 mg l⁻¹ in 1992, with an interim peak of 3.5 mg l⁻¹ in approximately 1972 (figure 3(a)). After 1995, the annual mean concentrations declined, reaching the mid-1950s level of ca. 2.7 mg l⁻¹ in 2010. The double peaks in 1972 and 1992 are much less visible when ammonia and nitrate concentrations are summed (figure S5), which suggests variability in transformation between both nitrogen species. A possible explanation for this could be different rates of nitrification due to changes in oxygen concentrations in the river. Adams et al (2001) linked abrupt changes in nitrate concentrations of the Elbe around 1990 to this mechanism.

Zoom In Zoom Out Reset image size

However, in addition to these relatively well understood (Worrall et al 2014, Ehrhardt et al 2019, Westphal et al 2019) long-term changes in mean concentration, we also observed multi-decadal changes in nitrate seasonality (figures 3(a)–(c)).

The period 1954–1958 showed a low seasonal amplitude ($b_{no3}/a_{no3} \lt 0.2$) and highest concentrations during summer/fall ($\phi_{no3} \approx 250$). From 1958 to 1972, the seasonal amplitude remained low but $\phi_{no3}$ decreased to ${\approx} 150$. In 1972, the first regime shift in seasonality occurred, marked by simultaneously increasing winter/spring and decreasing summer/fall concentrations, gradually increasing $b_{no3}/a_{no3}$ and an abrupt change in $\phi_{no3}$ from ${\approx} 150$ to ${\approx} 70$. $\phi_{no3}$ changes abruptly from ${\approx} 150$ to ${\approx} 70$ but the $b_{no3}/a_{no3}$ shows only a gradual increase. The confidence interval of $\phi_{no3}$ also decreased strongly after 1972, with the earlier large uncertainty a consequence of the low seasonal amplitude (determining which day has the highest concentrations is difficult when concentrations are similar around the year). After the rapid changes around 1972, all seasonality metrics remain similar until ${\approx} 1990$, even though annual mean concentrations continued to rise, as described above.

We interpret the period from 1954 to 1990 as the pollution part of the trajectory (stages 2 and 3), containing the shift from higher concentrations in summer/fall (1954–1958) to constant concentrations throughout the year (1959–1972), and then maximum concentrations in winter/spring after 1993. Both stages are characterized by low seasonal amplitudes, which fits our hypothesis of a change from point to diffuse source dominated stage.

It is remarkable that before 1972, changes in annual mean concentrations did not propagate into seasonality changes (figure 3(a)). Following our hypothesis, this would mean that both point and diffuse source contributions changed almost in concert. Processes that preserve a given seasonal pattern even though nutrient sources change (e.g. agricultural nitrogen surplus figure S2) are not known to us.

After 1972, summer/fall and winter/spring concentrations diverged, indicating a change in the relative importance of nitrate sources. This period coincides with the agro-industrial revolution in the GDR, which included intensified fertilizer use (Bauerkaemper 2004) and thereby nitrogen surplus from agricultural areas (figure S2). This increase in diffuse source contributions is consistent with the observed increase in winter/spring concentrations during this period (figure 3(a)). The slight decrease on summer/fall concentrations was not expected by our hypothesis. Possible reasons could be changes in in-stream nitrification as described above or improvements in waste water treatment.

The distinct seasonality signatures observed during the period from 1990 to 2019 are consistent with the proposed pollution recovery stages 4 and 5, which also align with the expected timing of water quality improvements due to socio-economical changes associated with the German reunification. The seasonal amplitude $b_{no3}/a_{no3}$ increased from 0.2 to 0.5 between 1990 and 2005, indicating different trends in point and diffuse source contributions.

Also, beginning in approximately 1990 the summer/fall concentrations started a 15-year decreasing trend, which matches the vast improvement of wastewater treatment that succeeded the German reunification (IKSE 1998, Adams et al 2001). We interpret 1990 as the beginning of Stage 4, which is characterized by a decrease of point sources without a reduction of diffuse sources. Evidence of Stage 5, characterized by decreases in diffuse source contributions, did not appear for several more years, with decreases in Winter/spring concentrations beginning in approximately 1995 and continuing steadily until the end of the time series.

From calculated agricultural nitrogen surplus rates (figure S2) and the literature (Bauerkaemper 2004) we know that the collapse of the GDR economy had a strong and immediate impact on agriculture. However, the response of winter/spring concentrations appeared only some years afterwards. Similar lagged responses of nitrogen concentrations have been observed by many authors and are mostly linked to subsurface travel times (Wang et al 2013, Van Meter and Basu 2015, Worrall et al 2015, Ehrhardt et al 2019).

Another sign of recovery is the decrease in $\phi_{no3}$: before 1990, the annual maximum of nitrate concentrations $\phi_{no3}$ were very similar to the annual maximum of discharge (≈day 70 figure 3(c)) indicating a positive relationship between them. A causal explanation for this is that in-stream nitrate concentrations are driven by mobilization from nitrate source zones in the discharge producing part of the catchment (Musolff et al 2015). After 1990, $\phi_{no3}$ decreased to ${\approx}35$, coinciding with the annual minimum of river water temperature. Inverse relationships between nitrate and temperature have been reported e.g. by Van Meter et al (2020) and linked to controls of biological processes (e.g. denitrification, assimilation) on in-stream concentrations. In their study Van Meter et al (2020) found this behavior in less human-impacted catchments with large shares of forests and wetlands. Therefore this decrease in $\phi_{no3}$ can be interpreted as a change from a pure source-controlled seasonal regime to one in which concentrations are actively modulated by sinks, such as biological processes.

The MSSM captures the overall observed changes in nitrate seasonality and demonstrates that long-term changes in the mixing ratios of point and diffuse source are a plausible explanation for the observed seasonality changes. The mean model performance for the seasonal amplitude is roughly consistent over all time periods (RMSE 0.06–0.1, figures 4(a)–(c)) while the prediction of the phase improves with time (from 64 days to 16 days). This gradual improvement is expected, as we used data from 2010 to 2016 to define the seasonal parameters of the diffuse source contribution. While the observed seasonal amplitudes mostly lie within the confidence interval of the MSSM, they show a large scatter around the mean predictions. We assume this to be caused by inter-annual variability in the nitrate concentrations, which is not considered in the MSSM (scatter of observations in figures 4(d)–(f)). In the phase $\phi_{no3}$ however, we find considerable deviation of the predicted means before 1990. We find a consequent overestimation between 1965 and 1985 (figures 3(b) and (e)). Furthermore, the predictions show a gradual decrease of $\phi_{no3}$ before 1992, while the observed data show step changes in 1958 and 1973. As noted above, the 1958 step change could be intensified by our data analysis, as phase estimation is highly uncertain when the seasonal amplitude is low. The step change around 1972 is unlikely to be an artifact, as it is also visible in flow-normalized values from the WRTDS method (figure 3(a)).

Zoom In Zoom Out Reset image size

Our simple model approach did not explicitly include in-stream retention of nitrate, which is known to play a large role in the Elbe—at least after 2005 (Deutsch et al 2009, Ritz and Fischer 2019). In-stream processes were estimated to have attenuated up to 30% of nitrogen inputs to the Elbe during summer (Ritz and Fischer 2019), but it is unknown how this quantity changed over the pollution/recovery trajectory. As we base the diffuse source component on concentration data measured near the outlet after the year 2010, in-stream processing is implicitly included in our parameterization of diffuse source seasonality. While we did not include temporal changes of the in-stream retention, we tested how those changes would affect the observed seasonality parameters $\phi_{no3}$ and $b_{no3}/a_{no3}$. Our findings suggest that increased retention will lead to higher $b_{no3}/a_{no3}$ and lower $\phi_{no3}$ values (figure S8) which is consistent with the seasonal trends observed over large parts of the time series. However, the magnitudes of the observed $\phi_{no3}$ changes cannot be explained by realistic changes of in-stream retention (section S1.3.4). Therefore we conclude that, while changes in in-stream retention might have occurred, they can only contribute to but not cause the observed seasonal changes.

The findings of our study should be viewed with consideration of its limitations: In our hypothesis, we separated five distinct seasonality states of which we were able to identify four in our data. We speculate that the results might be different considering longer time series which are currently unavailable in this catchment. We were not able to validate the proposed point and diffuse source ratios as (a) no historical data on wastewater nutrient loads or their proxies are available for Germany and (b) even though agricultural nitrogen surplus may be estimated, it remains unknown which parts of that reach the river network. Also not considered here were potential time lags from subsurface transport, or mixing of nitrate from different compartments.

Independent from our hypothesis we showed that annual mean concentrations and seasonal behavior of riverine nitrate can be completely decoupled. Furthermore, our results showed that the stark changes in nitrate seasonality observed in the Elbe over the last 66 years can be explained by human actions. To our knowledge, those changes were unintentional and unnoticed, at least from an environmental management perspective. These changes in seasonality have potentially dramatic implications for ecosystems: For example, changes in the seasonality of nutrient inputs to Lake Eerie have been linked to the re-occurrence of cyanobacteria blooms (Stumpf et al 2012, Stow et al 2015). Correlations between algae blooms and seasonal nutrient availability have also been reported for Lake Taihu, China (Xu et al 2015) and Europes coastal zones (Romero et al 2013).

While we only focused on nitrate seasonality, we consider changes in other nutrients such as phosphorus and carbon likely, as their fluxes are also anthropogenically influenced (Grizzetti et al 2012, Asmala et al 2019). These changes may either buffer or enhance alterations in the overall riverine carbon, nitrogen, and phosphorus ratios with unknown consequences. Finally, we note that long-term changes in nutrient seasonality may also be expected to follow successive development stages in other catchments, with local adjustments to the conceptualisation. For example, the Thames (Worrall et al 2014), Seine (Billen et al 2007) and the Vltava River (Kopáček et al 2013) show similar pollution-recovery trajectories of annual mean nitrogen or nitrate concentrations to the Elbe case study. Their trajectories have been linked to changes in nutrient sources (Billen et al 2007, Kopáček et al 2013, Worrall et al 2014) but the changes in nitrogen seasonality that we would expect based on our findings have not been investigated. However, paths other than the pollution-recovery trajectory demonstrated by many developed countries are also possible. For example, Vörösmarty et al (2015) point out that Switzerland and Singapore took proactive measures early and thereby avoided vast pollution of their rivers.