Diel patterns in nitrate concentration suggest importance of microbial pathways for in-stream processing

Diel cycles in stream nitrate concentration represent the sum of all processes affecting nitrate concentration along the flow path. Being able to partition diel nitrate signals into portions related to different biochemical processes would allow to calculate daily rates of such processes that are urgently needed for water quality predictions. In this study we analyzed diel 10 nitrate patterns at three locations in a 5.1 km long stream reach draining a 430 km2, mainly forested but anthropogenically influenced catchment during one growing season. We tested if the observed diel variability in nitrate concentration resulted from upstream sources and subsequent downstream transport or emerged simultaneously along the stream. We determined time lags between monitoring sites by cross-correlation. We found that time lags were closer to zero than travel time estimation assuming plug-flow suggested and concluded that ubiquitous in-stream processes prevailed in the creation of diel variability. 15 To further analyze the diel nitrate signals we used k-means clustering to identify patterns in the diel portion of nitrate concentrations and interpreted the resulting clusters with regard to potential drivers and the calculated nitrate balance of subreaches. We found that 70% of all diel patterns were attributed to clusters negatively related to the diel course of insolation with highest nitrate amplitudes on warm and sunny days and low water levels. We argue that temporal shifts towards the remaining clusters are rather due to shifts in microbial nitrate processing than in photosynthesis-driven plant uptake. These 20 results suggest that the magnitude of microbial nitrate processing may be large compared to plant uptake.


Data collection 90
Nitrate concentration was measured at 15 minute intervals at the three monitoring sites using UV-Vis spectrometer probes (spectro::lyser, s::can Messtechnik GmbH, Vienna) from April to November in 2019. As only two spectrometer probes were available, one probe was periodically repositioned so that input and output concentrations of either the upper or the lower stream reach were measured. In addition to the in-situ measurements, biweekly grab samples were collected at eight locations along the studied stream reach, including the probe locations, to assess longitudinal concentration profiles and validate probe 95 measurements. Samples were analyzed using ion chromatography (Dionex ICS-1100, Thermo Fisher Scientific Inc., USA).
Stream temperature and water levels were continuously recorded at S3 (TD-Diver, Van Essen Instruments, Netherlands) at 15 minute intervals. Discharge was calculated using a rating-curve based on two local NaCl tracer injections during stationary flow conditions, data of which was provided by the local water agency, and one additional tracer test during elevated water levels on November 15 th 2019 (> 70 % of recorded water levels). In the latter tracer test we injected 33 kg of NaCl at S1 so 100 that both sub-reaches were covered. Solar radiation data was obtained from a climate station at the nearby (< 10 km) Loechernbach experimental site in Eichstetten.

2.3
Data analysis

Assessing the origin of diel nitrate variation
If diel nitrate variation originated from some upstream source and subsequent downstream transport, time lags should be 105 detectable between nitrate signals at adjacent monitoring points that correspond to solute travel times between these points. In order to exclude an external source for diel nitrate patterns we determined the time lags between nitrate signals by means of cross-correlation which is a standard method for determining temporal shifts between signals (Derrick and Thomas, 2004) and compared them to the tracer travel time (τa) and the nominal water residence time (τ) according to Kadlec (1994). While τa is the first moment of the tracer residence time distribution, τ is the ratio of reach volume and discharge. We determined reach 110 volume from water level recordings at S3 and observations of channel width. In order to account for variability in channel geometry, we estimated minimum and maximum values of τ using rough estimates of channel widths between 20 and 25 m in the lower sub-reach and 15 to 20 m in the upper sub-reach. Time lags obtained from cross-correlation were tested for difference from zero using two-sided t-tests.

Identification and characterization of diel patterns 115
We used k-means cluster analysis to identify and classify diel patterns in stream nitrate concentrations as done previously by Aubert and Breuer (2016). This method partitions a data set into a pre-defined number of k clusters by iteratively minimizing the within cluster sum of squares. We used the algorithms by Hartigan and Wong (1979) that is implemented in the 'stats' Rpackage (R Core Team, 2019). In order to ensure that the resulting clusters represented variability in diel cycles and not in nitrate background concentrations the analysis was done on the diel portion of the solute concentration signal, hereafter referred 120 to as residual concentration (Cres). Residual concentrations were obtained by subtracting a 24 hour centered moving average from the measured concentrations (Cobs) and smoothed by applying a moving average of 2 hours. One feature of the k-means method that introduces some degree of subjectivity is the determination of number of clusters k. We therefore tested cluster numbers ranging between 2 and 20 and determined the best partition by both assessment of explanatory benefit per additional cluster, also known as 'elbow method', and visual inspection of clusters. The elbow method was not clearly indicative, 125 however, we opted for six clusters as higher values of k did not produce new clusters in terms of timing but rather caused further splitting of existing clusters by amplitude.
In order to further characterize the identified diel patterns, we assessed environmental conditions during the occurrence of the respective clusters. Particularly, we compared daily average and amplitude of nitrate concentration, average of water levels 130 (hmean), solar radiation (Smean) and water temperature (Tmean). We further investigated the relationship of diel patterns to diel signals of potential drivers, i.e. insolation and water temperature. Diel nitrate cycles reflect the time-varying balance between nitrate inputs and producing and depleting processes, e.g. nitrate concentration increases the quickest when nitrate production https://doi.org/10.5194/bg-2020-429 Preprint. Discussion started: 26 November 2020 c Author(s) 2020. CC BY 4.0 License.
is most dominant. This means that potential drivers should be correlated to the rate of change of Cres, i.e. to its first derivative Cres. Correspondingly, we related Cres to the observed diel signals of insolation (S) and water temperature (T) by calculating 135 daily Spearman rank correlation coefficients. Another potential driver of diel solute cycles is discharge variation. In contrast to the above drivers of biochemical processes, discharge would directly affect solute concentration. For example, if stream water was diluted by rainfall, maximum discharge and minimum concentration would occur simultaneously and not be mediated by the rate of a process. We therefore expected a potential relationship of discharge with Cobs rather than with Cres.
Correspondingly, we assessed the potential effect of diel variation in discharge by correlating Cobs to water level recordings 140 (h), avoiding uncertainties from rating-curve extrapolation.

Assessing the relation of diel patterns to the nitrate balance
The influence of diel patterns on the stream nitrate budget was assessed by calculating both temporal and spatial net change.
As sub-reach balance we understand the difference between daily means of nitrate concentration at an upstream (Cup) and downstream (Cdown) monitoring site, thus exclusively representing changes that happen in the stream reaches between these 145 points: ( 1 )

Prevalence of in-stream processes in creation of diel patterns
The benchmark tracer injection at a water level of 15.4 cm and a corresponding discharge of 2.0 m 3 s -2 resulted in travel time 150 estimates of 2.0 h in the upper and 2.3 h in the lower sub-reach, while time lags of nitrate determined by cross-correlation were very variable. We considered lags between signals with strong cross-correlation more reliable than for signals with a weak or no correlation. Lags resulting from strong cross-correlation (r > 0.75) ranged between zero and travel time estimates from both the tracer injection and nominal water residence time (Figure 2), indicating that time lags were shorter than solute travel time.
Thus, time lags were usually too small to be considered the result of advective downstream transport of the concentrations 155 signal. At the same time, lags were statistically different from zero (both sub-reaches p<0.001 in two-sided t-tests).

Diversity in diel patterns
Data collection at the three monitoring sites resulted in 355 complete diel nitrate signals, almost all of which showed a diel pattern ( Figure 3). The cluster analysis resulted in 6 clusters that clearly differed in terms of amplitude and timing of minimum and maximum concentrations. 70% of the days were attributed to the clusters A (n=132) and B (n=115) which both reached peak concentration in the early morning and minimum concentration in the late afternoon, but the daily amplitude was higher 165 in cluster B. The remaining clusters were characterized by peaks around midday (cluster C, n=54), in the afternoon (cluster D, n=28) and around midnight (cluster E, n=21). The last cluster F did not include enough days (n=5) for a proper characterization.
The medians of clusters A to E roughly represented sine waves shifted in phase by a quarter of a period (0.5 π) or 6 hours in units of time. Correspondingly, the derivatives of the clusters preceded the residuals by another quarter of a period. Note that δCres was the rate of concentration change of the diel portion of the concentration signal (Cres) and not of observed 170 concentrations (Cobs). The change rate of diel residuals resembled the signal shape of potential drivers of diel patterns but may differ in absolute terms as it was determined from the detrended data. Particularly the fact that its sign switches between positive and negative does not imply that the direction of associated processes would also do so in reality.

Figure 3: Clusters found in diel residuals of nitrate concentration (Cres) and its derivative (δCres).
While the diel patterns are not clearly related to different concentration levels of nitrate, they are characterized by different environmental conditions (Fig. 4). The most distinct cluster B shows the highest solar radiation (median: 250.0 W/m²) and highest water temperature (median: 21.7 °C). The other clusters emerged during lower water temperatures (median: 15.2 °C) and variable solar irradiation. Daily average water levels were lowest and strongest confined in cluster B (median: 20.1 cm) 180 and highest in cluster F (median: 77.2 cm) with the remaining clusters representing intermediate flow conditions. Further assessment of the diel dynamics of potential drivers revealed that δCres was positively correlated with solar radiation in cluster D (Fig. 4e), and negatively in clusters A, C, and especially strongly in cluster B. Correlations of δCres with temperature appear in cluster C (negative) and cluster E (positive) (Fig. 4f). Correlations of nitrate concentrations with diel water level fluctuations  The relationships of the clusters with the different environmental variables are reflected in their seasonal occurrence (Fig. 5).
The general pattern is similar at all monitoring sites, but most clearly visible at the intermediate site S2 where the data set was most complete. From June to August the radiation-related cluster B dominates. In colder months in spring and fall the diel nitrate amplitude decreased and cluster A replaces cluster B. Cluster C appeared most often in early spring but continued to 195 play a minor role throughout the entire season. Cluster D, E and F are marginally present throughout the season with a short continuous block of cluster D and E at sites S1 and S2, respectively, in September.

Relation of nitrate clusters reach balance
No relation was found between clusters and ΔC. In the upper sub-reach (Fig. 6a), nitrate concentrations increased in almost all cases. Median nitrate surplus was 0.35 mg L -1 in cluster A and 0.49 mg L -1 in cluster B. In the lower sub-reach a deficit was observed for most days (Fig. 6b). Decreases in median concentration ranged between -0.09 (cluster C) and -0.54 mg L -1 (cluster E). In both sub-reaches, imbalance is most pronounced during low water levels (Figure 6c), i.e. low flow conditions, 205 and decreased for higher water levels.

Prevalence of in-stream processes in creation of diel patterns
We found that travel times calculated from the diel nitrate signals at the monitoring points were usually between zero and the estimated water travel time (Fig 2). However, lags only exceeded our estimate of minimum travel time in a few cases. These data points were often associated with a low cross-correlation and hence less reliable for time lag estimation. On the one hand, spatial heterogeneity in environmental conditions and stream properties may cause some transformation of solute signals 215 despite spatial synchrony in biochemical processes (e.g. start of photosynthesis may be delayed in temporary shaded areas), explaining lags scattering around zero. On the other hand, solutes cannot travel faster than water and travel time will increase with decreasing flow. Our estimates of minimum nominal water travel time can be considered conservative in the sense that plug-flow (maximum flow velocity) was assumed and no factors were included that may further delay solute transport when water levels decrease, e.g. reduced short-circuiting. Downstream transport of solute signals therefore fails to explain most of 220 our data. We therefore interpret our data to indicate primarily in-stream origin of diel nitrate cycles.
In-stream biotic control on nutrient biogeochemistry was also stated by Roberts and Mulholland (2007) in a forest stream in Tennessee, US. A simulation of longitudinal evolution of diel nitrate patterns by Hensley and Cohen (2016) showed that a distance of tens of kilometers can be required for nitrate concentration from a constant source to converge to a stable diel 225 pattern. During convergence, they observed that the timing of the daily nitrate minimum oscillated with a longitudinal period https://doi.org/10.5194/bg-2020-429 Preprint. Discussion started: 26 November 2020 c Author(s) 2020. CC BY 4.0 License. of about 11 km, corresponding to the mean distance travelled in 24 h, i.e. distance between extremes was 5.5 km. This distance is comparable to the distance between monitoring sites in our study (5.7 km from S1 to S3) and flow length from the main source area of the river Elz to the monitored stretch (≈ 45 km) was also comparable to the convergence distance observed by Hensley and Cohen (2016). It therefore seems plausible that longitudinal stability of diel nitrate patterns was not yet fully 230 reached at our monitoring sites and the observed differences in timing (time lags > 0 in Fig. 2) were due to longitudinal variability in the diel nitrate signal. However, convergence distance in such a model may depend on transport parameters and residence time in the simulated stream reach. Hence, further research is needed to investigate the influence of river hydraulics on the diel solute patterns.

Which processes may cause the observed diel nitrate patterns? 235
According to Nimick et al. (2011), diel nitrate concentrations reported in literature are usually characterized by a minimum in the early evening and a maximum just before dawn as represented by cluster A and B in our study, accounting for 69.6 % of all measured days (n=355). Such patterns are often attributed to nitrate uptake by primary producers. In our study the strong negative correlation between δCres and global radiation, which was observed in cluster B (and to a minor extend also in cluster A), points in the same direction. The interpretation of the remaining clusters is more complex. While cluster E and F are 240 strongly influenced by unstable discharge conditions and should not be interpreted in regard to instream processes due to non-stationary conditions, flow was stable in clusters C and D. These clusters were characterized by midday and evening concentration peaks, respectively. Possible explanations for the deviation of these clusters from the general pattern of morning peaks and afternoon minima are that either our assumptions on the diel course of plant uptake were violated (i.e. assimilatory uptake was not a function of insolation) or that variability in timing observed in these clusters was driven by microbial 245 processes rather than plant uptake.

Assimilatory uptake likely varies in intensity but not in timing
If plant uptake was the only control on diel nitrate patterns, variability in clusters would indicate temporal shifts in plant uptake that could either evolve from shifts in drivers of photosynthesis or from rate limitations caused by factors other than temperature or light availability. Drivers of photosynthesis did not undergo, despite a minor seasonal effect, systematic time 250 shifts in this study. Neither did canopy development reduce light availability as observed in other studies (Rusjan and Mikoš, 2010;Roberts and Mulholland, 2007) due to absence of a forest or large trees near the stream, nor did certain clusters represent a typical weather condition such as e.g. days with cloudy morning and sunny afternoon. If that would be the case, correlations of δCres with drivers of photosynthesis should have been more or less equal for all clusters. We found that such correlations were clearly different among the clusters (Fig. 4), indicating that photosynthesis-related plant uptake was not the only control 255 on nitrate processes and thus the formation of diel nitrate patterns.
https://doi.org/10.5194/bg-2020-429 Preprint. Discussion started: 26 November 2020 c Author(s) 2020. CC BY 4.0 License. Mulholland et al. (2006) determined nitrate uptake in two streams by tracer 15 N-NO3 addition and found that uptake at night was smaller than during the day but not zero. This finding was attributed to continued uptake by algae until photosynthate 290 reserves accumulated during daytime photosynthesis were depleted. If such reserves are indeed depleted during the night, the assumption of zero plant uptake at night remains valid. If, however, baseline plant uptake occurs at night, the relative importance of plant uptake would be underestimated.
In contrast to plant uptake, rates of opposed microbial processes (e.g. of nitrification and denitrification) may partially cancel 295 each other out. However, if microbial processing generally dominates over plant uptake, minor relative changes in microbial processing might strongly influence diel nitrate patterns and thus increase the likelihood to observe variability in such patterns.
In fact, dominance of denitrification over plant uptake has been found in other studies.  found that assimilatory uptake was responsible for about 20% of total nitrate removal in the Ichetucknee River (Florida, USA) and attributed the remainder mainly to denitrification. Recently, a similar ratio was found by Preiner et al. (2020) in three reaches 300 with different macrophythe density in the river Fischa (Austria).

Interpretation of individual clusters
The suggested dominance of microbial processes in in-stream nitrate processing has implications for the interpretation of the diel nitrate patterns observed in this study. The diel course of microbial nitrate processing rates has to adopt a shape complementary to plant uptake so that the combination of both rates (δCres in Fig. 3) can reproduce the observed concentration 305 patterns. Diel patterns such as those observed in cluster A are often attributed to primary production and associated plant uptake. The fact that diel change rate in concentrations closely reflected patterns in insolation as observed in this study and also in stream chamber experiments by Hensley and Cohen (2020) invites to this interpretation. However, the effects of plant uptake and denitrification cannot be separated, if both processes work synchronously. Increasing amplitudes from cluster A to B suggest a superposition of both nitrate depleting processes particularly during times of reduced flow velocity at low water 310 levels associated with cluster B. Then a larger fraction of water interacts with stream sediments where denitrification occurs.
Although similar diel patterns are often simply attributed to assimilatory uptake,  argued that denitrification may be promoted by algal exudates which are rich in labile organic matter and are released during photosynthesis (Wyatt et al., 2012). If such exudates are released by benthic algae, labile organic matter might relatively quickly diffuse into anoxic zones of river sediments and promote denitrification. 315 Depending on how quickly such zones in the sediment (hyporheic zone) are reached, the resulting peak in denitrification will lag behind the peak of primary production and produce a nitrate concentration pattern like that observed in cluster C. Shifts in diel nitrate patterns similar to those between cluster A and B on the one hand and cluster C on the other hand were observed at a seasonal scale by Rusjan and Mikoš (2010). They attributed this finding to inhibition of photosynthesis and associated and S2 and subsequent decrease towards S3 reflects the regional pattern of nitrate concentration in the groundwater monitoring wells measured by the state agencies. Regional groundwater wells upstream of S2 typically show higher concentrations than in stream water, while nitrate concentrations in wells downstream of S2 are mainly lower than stream water concentrations.
Groundwater influence was originally considered minimal in the study reach due to the presence of drainage ditches along the 355 outside of the levees on both sides of the stream, collecting lateral groundwater flowing to the stream. However, longitudinal sampling with high spatial resolution and infrared imagery suggested the presence of groundwater inflow zones in the lower sub-reach. The explanation of nitrate reach imbalances by unconsidered inflows is in line with the observation that in both reaches nitrate imbalance was greatest during phases of low flow when inflows of limited volume but strongly deviating concentrations cause maximum effects (Fig. 6c). 360

Conclusions
Our study shows that diel nitrate patterns recorded at three locations in a 5.1 km long stream reach largely resulted from inreach processes. Downstream advection of upstream concentration perturbations was ruled out as an explanation for diel patterns since time lags between monitoring sites determined by cross correlation were predominantly smaller than water travel time estimates. Analysis of diel patterns revealed that approximately 70% of all diel patterns were attributed to two clusters 365 that were negatively related to the diel course of insolation with highest nitrate amplitudes on warm and sunny days and low water levels. We suggest that these patterns were caused by synchronous denitrification and autotrophic nitrate uptake, relative importance of which is unclear and may vary according to environmental conditions. In the remaining clusters temporal shifts were evident that could be explained by temporal shifts in microbial nitrate processing but not by photosynthesis-driven uptake.
In these cases, microbial processing rates need to be higher than assimilation rates in order to reproduce the observed patterns. 370 In summary, our study suggests that varying dominance and synchronicity of autotrophic assimilation and microbial processes may cause different diel nitrate patterns in stream systems.