Dissolved oxygen isotope modelling refines metabolic state estimates of stream ecosystems with different land use background

Dissolved oxygen (DO) is crucial for aerobic life in streams and rivers and mostly depends on photosynthesis (P), ecosystem respiration (R) and atmospheric gas exchange (G). However, climate and land use changes progressively disrupt metabolic balances in natural streams as sensitive reflectors of their catchments. Comprehensive methods for mapping fundamental ecosystem services become increasingly important in a rapidly changing environment. In this work we tested DO and its stable isotope (18O/16O) ratios as novel tools for the status of stream ecosystems. For this purpose, six diel sampling campaigns were performed at three low-order and mid-latitude European streams with different land use patterns. Modelling of diel DO and its stable isotopes combined with land use analyses showed lowest P rates at forested sites, with a minimum of 17.9 mg m−2 h−1. Due to high R rates between 230 and 341 mg m−2 h−1 five out of six study sites showed a general heterotrophic state with P:R:G ratios between 0.1:1.1:1 and 1:1.9:1. Only one site with agricultural and urban influences showed a high P rate of 417 mg m−2 h−1 with a P:R:G ratio of 1.9:1.5:1. Between all sites gross G rates varied between 148 and 298 mg m−2 h−1. In general, metabolic rates depend on the distance of sampling locations to river sources, light availability, nutrient concentrations and possible exchanges with groundwater. The presented modelling approach introduces a new and powerful tool to study effects of land use on stream health. Such approaches should be integrated into future ecological monitoring.

. Schematic illustration of effects on dissolved oxygen (DO) and its stable DO isotopes (δ 18 O DO ) by photosynthesis (P) and ecosystem respiration (R) with the respiration isotope fractionation factor (α R ), the gas exchange coefficient (k), stable water isotopes (δ 18 O H2O ) and water temperature (T) after Venkiteswaran et al. 27 . Horizontal and vertical dashed lines and the empty circle represent air saturated water (ASW) with DO saturation at 100% and δ 18  www.nature.com/scientificreports/ The MBH stream originates in the Czech Republic, where it is known as "Üjezdsky potok". The selected study site MBH-A is located within the forest "Rehauer Forst" and in the vicinity to the source region. The second study site of this stream system (MBH-B) was approximately 5 km downstream from MBH-A (Fig. 2d).
The WIS stream is located in the Franconian Alb, South Germany. It is dominated by karst lithology mostly of calcites and dolomites. The two selected study sites WIS-A and WIS-B were selected at distances of 2.9 and 3.3 km from the source, respectively (Fig. 2c).
The MOS stream is a tributary of the river Isar and flows in the Alpine foothills in Southern Bavaria. The study sites MOS-A and MOS-B were chosen between the cities of Munich and Freising. They are approximately 5 km apart from each other (Fig. 2b).
Field methods. In total six diel sampling campaigns of at least 28 h were carried out. In order to outline the strongest diel differences between metabolic rates all sampling campaigns were carried out in summer. The  www.nature.com/scientificreports/ upstream and downstream study site of each stream was sampled in 2-h time intervals. At MBH-A the end of the sampling was extended until 4 pm due to loss of samples that were collected between 10:00 and 12:00 a.m. All water samples were collected in the middle of each stream by syringe at 20 cm depth below the water surface, if possible. Before sample collection syringes were rinsed three times with sample water. Samples were then filtered through 0.45-µm disk filter (Minisart HighFlow PES, Sartorius AG, Germany). For oxygen and hydrogen stable isotopes of water (δ 2 H H2O and δ 18 O H2O ) samples were collected in 12-mL glass bottles. Samples for the 18 O/ 16 O ratios of dissolved oxygen (expressed as δ 18 O DO ) were collected in 12-mL Labco Exetainers™ (Labco Ltd. Lampeter, U.K). These were pre-poisoned with 20 µL of a saturated HgCl 2 solution in order to avoid secondary biological activity after sampling. The vials were completely filled and immediately capped using screw caps with a butyl rubber septum. Previous laboratory-internal tests showed negligible contamination by atmospheric O 2 when applying this sampling method.
Measurements of water temperature and DO were conducted in the field with a multi-parameter instrument (Multi 3620 IDS/3430 by WTW GmbH, Weilheim, Germany). All probes were calibrated at least once per day. One σ-repeat measurements of temperature was better than ± 0.1 °C and ± 2% for DO.
Discharge measurements are an important input to the PoRGy model and were determined with an electromagnetic currentmeter (SEBA Hydrometrie GmbH FlowSens) that was placed along a transect across the stream at two different depths (close to the surface and close to the bottom) with the 2-point-measurement method of Kreps 39 . Laboratory methods, stable isotope measurements. Water samples were analyzed for δ 18 O of DO with a modified method by Barth et al. 40 . The method couples an automated equilibration unit (Gasbench II) to a Delta V Advantage isotope ratio mass spectrometer (ThermoFisher Scientific, Bremen, Germany). The isolation of DO into a headspace relies on a helium extraction technique by Kampbell et al. 41 and Wassenaar and Koehler 42 . Prior to analyses, headspaces were automatically generated in each water-filled vial on the Gasbench II with an autosampler that was equipped with a double-hole needle. After headspace generation samples were placed for 30 min on a horizontal shaker that moved at a rate of 250 strokes per minute to mobilize all DO into the headspace. Subsequently, samples were placed back on the Gasbench II autosampler after a switchover to connect with the isotope ratio mass spectrometer. The headspace was then mobilized in a helium stream via another dry double-hole needle on the autosampler. The O 2 was separated by a CP-Molsieve 5 Å capillary column (25 m length Å ~ 0.53 mm OD Å ~ 0.05 mm ID; Agilent, Santa Clara, CA, USA). The purified O 2 was then transferred by continuous flow to the mass spectrometer. Laboratory air was used as an internal standard with a known value of + 23.88‰ 43 . Further details of the method are available in Köhler et al. 44 .
Water samples were analyzed for their δ 18 O H2O values by isotope ratio infrared spectroscopy (IRIS) that operates based on wavelength-scanned cavity ring-down time measurements (L2120-i, Picarro Inc., Santa Clara, CA, USA). Each sample was measured by nine injections of which the first three injections were discarded to exclude memory effects.
All water isotope measurements were normalized against two international reference materials named Vienna Standard Mean Ocean Water (VSMOW) and Standard Light Antarctic Precipitation (SLAP). This two-point calibration was controlled by a third laboratory reference water that was calibrated directly against VSMOW and SLAP.
Isotope results were reported in the standard delta notation in per mil (‰) versus VSMOW according to where 18/16 R s is the oxygen isotope ratio of heavy to light isotopes in the sample and 18/16 R r is the ratio in the standard (VSMOW, 0.0020052 45 ). All values were then multiplied by 1000 in order to convert them to permille (‰). Repeat measurements of field standards revealed a standard deviation of ± 0.2‰ (± 1 σ) for both, δ 18 25,[48][49][50][51][52] . Net gas exchange between water and atmosphere is controlled by the level of saturation, that in turn depends on water temperature and the atmospheric equilibrium constant k 53 . A comparatively small equilibrium isotope fractionation enrichment of atmospheric O 2 with a value of + 23.88‰ towards DO involves a temperature-dependent enrichment of about + 0.7‰ 43,54 . Data for each site were modelled with boundary conditions placed on each of the five parameters described above (P, R, k, α R , and δ 18 O H2O ). Metabolic rates were allowed to vary by two orders of magnitude, k was allowed to vary by a range of 50% wider than values calculated from stream velocity and depth [55][56][57][58] , α R was allowed to vary between 0.975 and 1.000, and δ 18  www.nature.com/scientificreports/ altered these constrained variables to find a minimum sum-of-squared errors between field data and model output. The r 2 values for field data and model output for both DO and δ 18 O DO values are a measure for the quality of fit. This approach is similar to the one described in Wassenaar et al. 28 .
In addition to the PoRGy model with the application of DO saturations and stable isotopes, we also tested the model performance based on DO concentrations alone in order to assess benefits of the combined approach.

GIS analyses.
For the analyses of the stream catchments the geographic information system (GIS) software ArcGIS Pro (https:// www. esri. com/ en-us/ arcgis/ produ cts/ arcgis-pro/ overv iew), version 2.7.2, was applied. A 30 m digital elevation map of Bavaria (https:// dwtkns. com/ srtm3 0m/), the stream network from Bayerisches Landesamt für Umwelt (LFU) and the stream gauging locations (Table 1) were used as inputs for creating the catchments of the three watersheds in this study. Catchments were delineated using the ArcSWAT tool (https:// www. arcgis. com/ index. htm). Land use patterns were extracted using the clip tool in ArcGIS from the CORINE land cover map for 2018 (https:// land. coper nicus. eu/ pan-europ ean/ corine-land-cover).

Results
A detailed overview of measured DO concentrations and saturations, water temperatures, nutrient concentrations, and δ 18 O DO and stable isotopes of water (δ 18 O H2O ) at the streams MBH, WIS and MOS is provided in the supplementary material (Table S1).

Water depths, flow velocities and water temperatures.
Measurements took place during low to medium water discharge at all sites [59][60][61] . Input values for water depths, flow velocities and water temperatures varied between the streams MBH, WIS and MOS, and upstream and downstream locations ( Table 2). Water depths and flow velocities were averaged over the river width from four sampling locations during the sampling period. Flow regimes were stable within the boundaries of measurement uncertainties. The maximum water depth was 0.  Table S1). Moreover, DO saturations of all study sites showed strong negative relationships with δ 18 O DO (r 2 = 0.86, p < 0.001). Times of minima of DO saturations did often not precisely match with times of maxima δ 18 The highest measured DO saturations and lowest measured δ 18 O DO occurred slightly before or after the solar noon period between 12:00 and 16:00 h (Fig. 3). However, due to the chosen sampling interval of 2 h, actual peaks Table 2. Overview of water depths, mean flow velocities and water temperature minima, maxima and averages of the upstream (A) and downstream (B) sampling sites at the streams Mähringsbach (MBH), Wiesent (WIS) and Moosach (MOS).

Date
Sample ID Water depth (m) Mean flow velocity (m s −1 ) Min. water temperature (°C) Max. water temperature (°C)  (Fig. 3).   (Fig. 3). The modelling approach was to adjust P, R, k, α R , and δ 18 O H2O to find a best-fit solution. This yielded average midnight-to-midnight P, R, and G rates from the modelled diel curves. Generally, modelled diel curves followed the general day-night course of the data and outlined timing of DO and δ 18 O DO minima and maxima as well as nighttime plateaus (Fig. 3). The goodness of fit (r 2 ) was best for the MOS stream with values ranging between 0.97 and 0.99 and varied between 0.63 and 0.89 at the other sites. Moreover, the modelled δ 18 O DO curves always showed a better fit with measured data than their corresponding DO curves. When running the model based only on DO saturations, model fits were similar compared to combined DO and δ 18 O DO modelling. Only site MBH-A showed a lower r 2 of 0.52 (supplementary material, Table S2).
Metabolic rates and ratios. For a better comparison of the different streams and study sites, the modelled midnight-to-midnight P, R and gross G rates (G*) and ratios are valuable indicators of the metabolic states of the three sites.
Lowest P rates were found at the MBH stream with a value of 18 mg m −2 h −1 at MBH-B, while the highest P rates were found at the MOS stream with a value of 417 mg m −2 h −1 at MOS-B (Table 4). Minimum and maximum respiration rates were detected at the same sites with an R rate of 230 mg m −2 h −1 at MBH-B and 341 mg m −2 h −1 at MOS-B. Modelled G rates ranged between 148 mg m −2 h −1 at WIS-A and 298 mg m −2 h −1 at MBH-A.
The relative importance of metabolic processes and gas exchange is best reflected by P:R:G ratios ( Table 4). The low primary productivity in the MBH stream compared to R and G yielded P:R:G ratios of 0.1:1.1:1 at MBH-A and MBH-B. At the WIS stream R was the main driver of the DO cycle with P:R:G ratios of 0.6:1.6:1 at WIS-A and 0.8:1.8:1 at WIS-B. Also, site MOS-A was dominated by R with a ratio of 1:1.9:1. In contrast, at study site MOS-B active photosynthesis (P) represented the highest DO fluxes on a diel basis. It was the only study site, where P exceeded R and G rates with a ratio of 1.9:1.5:1.
In contrast, the PoRGy model based solely on DO saturations yielded generally overestimated P, R and G rates. This effect was best visible at the low productivity sites MBH-A and B, with around 5 to 21 higher P rates and also significantly higher R and G rates (supplementary material, Table S2). This comparison with a single parameter approach of DO concentrations shows that the combined application of DO saturations and stable isotopes was much better in constraining P, R and G rates.
Ecosystem respiration and fractionation factor α R . In order to optimize the model to the data, the fractionation factor for respiration α R was allowed to vary within the range of known ecosystem respiration Table 4. Overview of modelled midnight-to-midnight photosynthesis (P), ecosystem respiration (R) and gross gas exchange (G*) rates and ratios, and gas exchange (G) coefficients (k) at the streams Mähringsbach (MBH), Wiesent (WIS) and Moosach (MOS) with respective upstream (A) and downstream (B) sites.

Study site
Midnight-to-midnight P rate (mg m −2 h −1 ) R rate (mg m −2 h −1 ) G* rate (mg m −2 h −1 ) P:R ratio P:G ratio P:R:G ratio k (m h −1 ) www.nature.com/scientificreports/  Table S1). ArcGIS analyses of the digital elevation map of Bavaria, the stream network, and land cover maps provided land use proportions within the catchments upstream from each sampling point. The land use types in these catchment parts were classified into the following classes:
Of these classes only the first four were relevant for the analyses conducted in the parts of the catchments studied.
The part of the MBH stream catchment that influenced the two sampling sites is mainly composed of forests and grassland with a low proportion of agriculture (Table 5). In contrast, the analyzed parts of the WIS stream partial catchment showed more pronounced anthropogenic influences with approximately 60% agriculture as the most frequently encountered land use type and about 30% forests. The upstream parts of the sampling points along the MOS stream showed the highest anthropogenic influences with agricultural and urban land use. Natural land use types such as forests and grasslands were less important in these upstream parts of the catchment (Table 5 and Fig. 4).
When excluding G rates with the assumption that they are hardly influenced by land use, the evaluated land use types and calculated stream metabolisms from the PoRGy model showed good relationships with P:R ratios. In general, higher proportions of forests yielded lower P:R ratios (Fig. 4d), whereas elevated proportions of agriculture caused higher P:R ratios (Fig. 4b). Other land use types such as urban areas (Fig. 4a) or grasslands (Fig. 4c) only showed a small exposure in the catchments and had much lower influences on P:R ratios.

Discussion
The overall performance of the PoRGy model with the application of DO saturations and additional stable isotopes was significantly better compared to tests when only DO concentrations were considered. This is because metabolic and G rates were better constrained especially at the less productive sites with higher contributions of G (Supplementary Material, Table S2). The literature describes various metabolism models that exclusively use DO concentrations 26,[63][64][65][66][67][68][69][70][71] . Although most of them were able to model reasonable metabolic rates in streams with lower G rates, more turbulent flow conditions have been suggested to be more challenging. Here, the additional application of stable DO isotopes is able to better constrain P, R and G rates 63 . Working with well-constrained k values is essential to estimate and assess vulnerabilities of stream ecosystems in rapidly changing environments 27 . Such well-constrained k values might also help to define denitrification rates (via open-channel methods) and greenhouse gas flux rates (e.g. nitrous oxide). Such constrained k values could thus improve our general understanding of the contribution of terrestrial environments to global warming 72,73 .
Several stream metabolism studies applied δ 18 O DO dynamics based on the same basic assumption of P and G as DO sources and R as a DO sink 29,30,67,74,75 . An advantage of the PoRGy model by Venkiteswaran et al. 24 is that it was also translated into Matlab code that is freely available and easily applied.  www.nature.com/scientificreports/ In general, the metabolic P and R rates found in our study sites lie between the rates found in other stream systems, with lower P rates at forested study sites and higher at more anthropogenically influenced streams with higher proportions of agricultural and urban land use [76][77][78][79] .
The diel sampling sites at MBH stream represented the most pristine stream with high proportions of forest and grassland within the catchment and little agricultural activity (Fig. 4c,d). At both study sites of the MBH stream DO undersaturation and δ 18 O DO values were above atmospheric equilibrium with a value of + 24.6‰ 43,54 . This indicated a constant heterotrophic state of the stream (Fig. 3) that was also reflected by low P:R ratios ( Table 4).
The upstream catchment part of site MBH-A had more than 60% forest coverage. It is also closest to the source. Therefore, increased shading by trees likely reduced solar radiation and resulted in typically low P rates. These low P rates were also observed in other forested headwater streams 37,80 . Simultaneously, high R rates are typically due to input of allochthonous material from high forest proportions in the catchment, which caused the constant DO undersaturation 37 . Among the streams, k was highest at the site MBH-A (Table 4). This likely resulted from turbulent flow conditions caused by stream bed roughness, shallow water depths and steeper stream slopes. These driving factors for k were also observed at other headwater streams 81,82 . Elevated k values that cause high G rates can in turn cause DO undersaturation due to R at night. Such combinations can generally dampen day-night amplitudes ( Figs. 1 and 5). However, particularly in this shallow headwater stream section, estimates of k also caused uncertainties in model outputs because turbulence and the possibility of air bubble inclusion may be underestimated 81,83 . This might have been one of the reasons for the poorer model fit at this study site (Fig. 3). Additionally, effects of shading by trees are not considered in the model which only includes latitude, altitude and daytime specific solar influences for the calculation of P. Another factor that may have caused higher uncertainties is that smaller water volumes of the stream can become affected more easily by short-term changes in P, R and G due to altered nutrient concentration, discharge, water temperatures and exchange with the soil water and HZ. In contrast, inputs of significant amounts of groundwater seem to be less probable because the bedrock in this area had a low permeability in this study area. These uncertainties can result in untypical day-time DO and δ 18 O DO patterns that were evident at study site MBH-A. These aspects render this site most challenging for modelling.
MBH-B was located outside the forested section. Therefore, less shading by vegetation may have enhanced photosynthetically active radiation (PAR) that in turn can trigger higher metabolic rates. However, minor effects on PAR could still have been caused by shrubs in the vicinity of the stream that may have caused partial shading of the stream water. The model yielded even lower metabolic rates with similar ratios when compared to the upstream sampling site (Table 4). However, the model fit was considerably better than at MBH-A. This implies that the higher metabolic rates observed at MBH-A may be due to model uncertainties and should be interpreted with caution. Nevertheless, the low modelled P:R ratios at MBH-B may also mark residual effects of the larger proportions of forests further upstream (Table 5, Fig. 4d). Additionally, nutrients such as nitrate and phosphate were very low at both study sites due to low agricultural land use proportions. This may have hampered photosynthetic DO production, even when PAR became higher at MBH-B.
The two study sites at the WIS stream were only 400 m apart from each other and therefore showed similar diel DO and δ 18 O DO ranges (Table 3, Fig. 3). Both study sites only approached DO saturations of atmospheric equilibrium around the time of the solar peak around 13:00 h. The remainder of the day, and especially during www.nature.com/scientificreports/ the night, both study sites became undersaturated in DO. Here the model results indicated a heterotrophic state of the stream with a dominance of R that outcompeted rates of G and P. Also note that sampling at the WIS stream was also influenced by alternating cloud cover with showers towards the end of the sampling event. These meteorological conditions may have reduced P as a result of reduced solar radiation by cloud shading while the rainfall could have increased turbidity in the water as a result of terrestrial sediment input from the surrounding area or from stream bed erosion 84,85 . Moreover, the precipitation events could have also increased R rates with more labile terrestrial carbon being washed into the stream. Subsequently, these inputs may increase heterotrophic respiration 64,86 . The higher proportion of agricultural land use found in the catchment of the WIS stream could enhance elevated inputs of nutrients which would increase P (Table 5). On the other hand, agriculturally worked soil could have caused more rapid increases in turbidity by enhancing mobilization of soil material. This would in turn have reduced P and may even outcompete effects of nutrient inputs 87,88 . Moreover, rainfall-related increased inputs of groundwater in the karstic region at the study sites during sampling may have affected P and R rates due to mixing 89 . These causes and effects could best be observed during the strong rainfall events at the end of the diel sampling, which showed a day-time atypical reduction of DO and an increase of δ 18 O DO. Both values indicate reduced P and increased R rates. This shows that unstable weather conditions can cause irregularities of diel patterns and render modelling of the stream sites more challenging. Despite such unstable weather conditions, metabolic processes were still more pronounced in the WIS stream than in the MBH stream. This was reflected by a larger diel range of DO and δ 18 O DO values (Fig. 5). Due to the proximity of both study sites to its spring, the observed dominance of R rates may also reflect a residual signal from upstream DO-undersaturated spring water that was also measured by van Geldern et al. 90 .
Although P rates at the WIS stream were much smaller than their corresponding R rates, they still were 2 to 7-fold higher than those in the MBH stream (Table 4). Nonetheless, these DO values never reached significant oversaturation. Also, the δ 18 O DO values were lower than + 24.6‰ during most of the day and only reached values higher than + 24.6‰ after sunset (Fig. 3). This shows the large effect of photosynthetic oxygen production with an average value of around − 9‰ via splitting of water molecules. This process then adds oxygen with a lower signal to the DO pool. This process seems to be important even in a R-dominated stream. A possible explanation for higher P rates can be inferred from a combination of PAR and nutrient availability. In contrast to the MBH stream with large proportions of forest in the catchment, the WIS stream catchment has more agricultural areas and grasslands (Table 5). This also implies that the upstream parts of the WIS stream are not fully shaded and only covered by smaller shrubs or few trees. This in turn allows higher PAR at the water surface. Additionally, at the WIS stream elevated nitrate concentrations were derived from the nitrate-rich spring and groundwater input from agriculturally influenced land use. This was combined with reduced nutrient attenuation potential of the karstic bedrock 90 . The elevated nutrient concentrations in combination with reduced shading likely fueled the photosynthetic activity during day time even in the proximity of the spring. This suggests that P rates may reach even higher values on a sunny day.
In this heterotrophic stream, G acted as an important parameter to counterbalance DO consumption at night. With stronger DO undersaturation G rates increased, until G and R rates became balanced during night. This balance helped to establish well-pronounced night plateaus of DO and δ 18 O DO at both study sites (Fig. 3).
In contrast to the above, study sites of the MOS stream showed differing day-night patterns of DO and δ 18 O DO and produced the highest metabolic variations among the three streams. The upstream study site, MOS-A, www.nature.com/scientificreports/ maintained heterotrophic conditions with DO undersaturation during night and most of the day with a low P:R ratio and a high R rate (Fig. 3, Table 4). Due to this constantly high oxygen demand, DO saturations were lowest at night when compared to the other streams. This undersaturation was mostly counterbalanced by G. Generally, k values were lowest at the MOS stream (Table 4). This may have been another reason for the low DO saturation plateau at night. Due to high R and low k, corresponding δ 18 O DO values reached their highest values when compared to the other two streams (Fig. 5). DO values only approached saturation, when DO input by high P and G was large enough during the day. The elevated P rates at this study site caused input of photosynthetic oxygen during the day with its distinct isotope signal. This is best reflected in the rapid shift from respirationdominated δ 18 O DO values above those of atmospheric equilibrium at night in relation to photosynthetic signals below + 24.6‰ during the day (Fig. 3).
Compared to the upstream MOS study site DO saturations and δ 18 O DO reached similarly high values during absence of light at MOS-B. This indicates that all DO consuming processes and G must have been on an equal level at both study sites. Generally, the downstream P rate was 2.6-fold higher when compared to the upstream study site MOS-A (Table 4). Therefore, this is the only study site, where P became the dominant part of the metabolism. This was also evident by a P:R ratio above 1. This active metabolism also caused more pronounced day-night curves with high P rates and DO oversaturation during the day and subsequently rapidly declining DO values during the night. Because R rates and k were similar at both study sites these elevated P rates were the main drivers of altered DO and δ 18 O DO curves during daytime.
Among the three studied streams, MOS showed the strongest anthropogenic impact by urban and agricultural practices. The river consists of a complex channel system with the MOS main stream as the integral collector. Both land use types generally relate to increased nutrient inputs such as nitrate, high biochemical oxygen demand, increased water temperatures and reduced shading 91,92 . The lower productivity found at study site MOS-A compared to MOS-B could stem from lower PAR because of partial shading by shrubs and trees at the study location and the stream water further upstream. Although at study site MOS-B was also surrounded by larger trees, which caused shading, the further upstream section only had sparse vegetation covers. Therefore, high DO by P may have been caused by residual upstream inputs. However, also increased input of DO-depleted groundwater or exchanges of groundwater via the hyporheic zone in MOS-A could have dampened DO curves and may have decreased T.

Conclusions
Diel measurements of DO saturations and associated δ 18 O DO values in three contrasting streams within the same climate zone each showed distinct diel curves that correlated with proportions of various land use forms in their catchments, proximities to their springs and weather conditions. These factors can in turn affect nutrient concentrations and PAR due to shading. The PoRGy model successfully matched diel field data to estimate important P, R and G rates. Here, the additional application of stable DO isotopes substantially improved the model output by constraining metabolic rates and k values. This modelling approach also enabled a synoptic view of different metabolic states of the stream sections. Uncertainties of model outputs can be attributed to tree canopy, alternating weather conditions and groundwater input. These factors were not directly considered in the model but likely influenced its outcomes.
Notably, all sites (except for one) were predominantly undersaturated in DO and confirmed heterotrophic states. This was marked by low P:R ratios. DO undersaturation combined with low productivity and elevated k values, caused atmospheric oxygen to become the dominant oxygen source in the MBH and the WIS stream. Stream sections with higher k values are considered more resilient to anthropogenic or climatic changes because elevated G serves as a reliable and constant oxygen input. In contrast, the two sites at the most anthropogenically influenced stream showed the highest metabolic DO turnover ranges. Here, larger DO amplitudes between upstream and downstream locations highlight effects of shading on the productivity of the stream during the day. Although estimated k values at these sites were lowest at the MOS stream, G was still sufficient to avoid severe DO depletion overnight. However, climatic changes with rising temperatures could further increase heterotrophy in the streams while lowering DO solubility and increasing respiration 20 .
The PoRGy model is a promising tool to determine ecological states of stream sections in an integral manner. It enables direct comparisons of the P, R and G rates. However, further testing should be performed at higher resolution and over longer time periods. This could be arranged by sampling via specialised autosamplers that isolate samples from atmospheric influences. Such higher frequency data would also help to increase the accuracy of the model. Moreover, model uncertainties could be improved by direct determinations of k and PAR.
Overall, our data provide robust early warning information for improved stream and river management to help mitigate effects of land use and climate change. Stream comparisons of this study yielded smaller differences within streams than comparison between streams. This seems to be mostly related to different proportions of land use. However, when comparing sites over the entire length of streams more pronounced differences may become obvious between sources and mouths of rivers.
The ecological importance of streams and rivers becomes increasingly recognized. In particular, human activities can severely affect water chemistry of river networks as integral reflectors of their catchments. Further applications of this technique at selected parts of rivers including tributaries and mouth sections would help evaluate entire catchments. In addition, diel changes of DO and its isotopes need to be further investigated in different seasons to establish better understanding of annual dynamics. Moreover, dynamic models that capture variable conditions such as high water stands or receding limbs of hydrographs, would improve the understanding of stream metabolism responses to a changing environment. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.