Delineating Source Contributions to Stream Dissolved Organic Matter Composition Under Baseflow Conditions in Forested Headwater Catchments

Dissolved organic matter (DOM) composition in streams reflects the dynamic interplay between DOM sources, mobilization mechanisms, and biogeochemical transformations within soils and receiving water bodies. The information regarding DOM sources being mobilized during baseflow can improve our ability to predict hydrological and biogeochemical responses to environmental changes, with implications for catchment management strategies. The objective of this study was to characterize the spatial changes in DOM composition along a headwater stream in a small mountainous catchment during baseflow and to link the findings in‐stream with possible DOM sources in the catchment. DOM was monitored over 1.5 years at three sites in the Große Ohe catchment (Bavarian Forest National Park, Southeast Germany) which strongly varied in slope between upper and lower part of the catchment, using UV‐Vis absorption indicators of aromaticity (SUVA) and molecular weight (E2:E3) from high‐frequency probe measurements. Additionally, discrete samples were collected and analyzed by ultrahigh‐resolution Fourier transform ion cyclotron resonance mass spectrometry (FT‐ICR‐MS). At baseflow conditions, dissolved organic carbon (DOC) concentrations, a proxy of DOM amount, ranged from 1.5 to 4.7 mg L−1, but were similar along the stream. However, DOM quality exhibited clear spatial patterns, with overall high aromatic and low molecular weight DOM in the lower part of the catchment, which has a higher proportion of hydromorphic soils. Moreover, molecular data revealed that oxygen‐rich, aromatic compounds increased in their abundance at high DOC concentrations in both the steep upper and the flatter lower part of the catchment, with also additional input of oxygen‐depleted aromatic compounds identified in the lower part. In contrast, nitrogen‐rich aliphatic compounds were negatively correlated with DOC concentration, indicating a higher contribution of deep groundwater at low DOC concentrations.


Introduction
Dissolved organic matter (DOM) is a complex mixture composed of different classes of naturally occurring organic compounds (Jaffé et al., 2008). It plays a fundamental role in surface water chemistry, controlling metal bioavailability and mobility, as well as nutrient cycling (Stedmon et al., 2007;Tipping et al., 2002). Due to its heterogeneity and dynamic nature, estimation of DOM amount is based on dissolved organic carbon (DOC) measurements. DOC concentration and DOM quality are strongly connected in headwater streams (Peter et al., 2020). The amount of DOC is limited by the availability and mobilization of diverse sources, which in turn also impacts DOM quality in-stream. Subsequent turnover of terrestrial DOM in aquatic ecosystems can result in the release of CO 2 through mineralization and impacts primary productivity through light attenuation or nutrient release (Coble, 2007;Seekell et al., 2015). Spatial changes in in-stream DOM quality along the river network are hence a superposition of availability of sources and biogeochemical processes (Mosher et al., 2015;Peter et al., 2020). The river continuum concept, although not specifically addressing DOM, describes a decrease in diversity of solutes from headwaters to high-order rivers (Creed et al., 2015;Vannote et al., 1980). Furthermore, changes in baseflow DOC concentration and DOM composition in headwater streams can impact the macroinvertebrate and microbial communities in the aquatic systems (Baker et al., 2021;Parr et al., 2015), and the treatment of drinking water from downstream surface water reservoirs (Raeke et al., 2017).
The organic matter retained in soils represents an important source of allochthonous DOM for inland aquatic systems (Ågren et al., 2007;Kaiser & Kalbitz, 2012). The activation of DOM sources in the soil is governed by the catchment wetness state, which controls the water flow, groundwater levels, and the connection of different flow paths (Lambert et al., 2014). Furthermore, the quality of the mobilized DOM is a result of both soil properties (e.g., sorption-desorption processes) and microbial processing at different depths (Seifert et al., 2016;Steinbeiss et al., 2008). DOM stored in upper organic-rich layers of the soil is young, with a dominance of lignin-derived phenols and plant-derived carbohydrates, while most DOM in the deep layers of the soils is older and of microbial origin, being rich in nutrients such as nitrogen (N) and phosphorus (P) (Kaiser & Kalbitz, 2012). The relationship between hydrological conditions and DOM mobilization from different soil layers has been demonstrated during storm events as it represents a fast change in DOM sources (Fellman et al., 2009;Hood et al., 2006;Wagner et al., 2019). Yet, the main sources and mobilization dynamics of DOM during baseflow conditions over longer periods of time remain poorly characterized. The hydrological and biogeochemical knowledge of headwater stream characteristics during baseflow conditions provides valuable insights into changes in water sources, being especially important for the development of catchment management strategies, especially in light of future drought scenarios (Peralta-Tapia et al., 2015;Smakhtin, 2001). Moreover, low-order streams represent a particularly sensitive system to changes in DOM sources and dynamics due to their immediate interface with adjoining hillslopes and the high riparian soil-to-water area ratio (Aitkenhead et al., 1999;Gomi et al., 2002;Mosher et al., 2015).
DOC concentration and DOM quality are often monitored with spectroscopic techniques integrated in field deployed sensors (Hood et al., 2006;Jaffé et al., 2008). In headwater streams, spectroscopic sensors have been used to reveal temporal variability of DOC concentration and DOM quality during periods of drought, storm events (Broder et al., 2017;Strohmeier et al., 2013), and across seasons (Schwab et al., 2018;Werner et al., 2019). Absorption indices have emerged as an efficient way of characterizing the structure and reactivity of DOM (Ågren et al., 2008;Hansen et al., 2016;Jaffé et al., 2008). Among those indices, specific ultraviolet absorbance (SUVA) (see Weishaar et al., 2003) and absorption ratios (e.g., at 250-365 nm, (De Haan & De Boer, 1987)) can provide information of DOM aromaticity and molecular weight (MW), respectively. However, ultraviolet-visible absorbance (UV-Vis) and fluorescence measurements are limited to the bulk of the chromophore fraction of DOM, which provides only little information on changes in DOM composition. The state-of-art technique to study DOM chemical composition is Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) (Nebbioso & Piccolo, 2013). The ultrahigh-resolution of FT-ICR-MS allows the distinction of characteristic fingerprints in DOM composition on the molecular level that reflect different sources (Lechtenfeld et al., 2014;Raeke et al., 2017;Wagner et al., 2015) and transformation processes (Herzsprung et al., 2020). Moreover, the high resolution of FT-ICR-MS allows the identification of individual molecular formulas for molecules, which may also contain low-abundance heteroatoms like nitrogen and sulfur (Herzsprung et al., 2016). Nonetheless, the high costs and analytical effort SILVA ET AL.
of FT-ICR-MS analyses have limited the number of samples in previous studies, limiting further advances on the biogeochemical understanding of DOM dynamics (Fellman et al., 2009;Wagner et al., 2019). Therefore, the information provided by high temporal resolution spectroscopic UV-Vis and high mass resolution FT-ICR-MS analyses proves powerful for investigating changes in DOM quality.
The objective of this study is to characterize the spatial changes in DOC concentration and DOM composition along a low-order stream in a temperate, mountainous forested catchment in southern Germany with a focus on baseflow conditions. Previous work has shown that DOC export during rain events reflects varying contribution of different parts of the catchment (upper and lower) triggered by antecedent hydrological conditions and hydrological connectivity . Due to the strong gradient in catchment properties between upper and lower part (mean slope, proportion of hydromorphic soils, see below), we hypothesize that also during baseflow, the DOM composition in the steep (upper) part of the catchment is mainly driven by groundwater input, while in the flat (lower) part, also varying contribution from riparian topsoil shape the DOM composition in the stream. Thus, we also expect that DOM composition will be more variable over the year at the lower part of the catchment compared to upstream. Finally, we expect that due to the high dimensionality of FT-ICR-MS derived DOM composition data, even subtle changes in source contributions can be identified in a small headwater stream Section 2.5.3. To this end, continuous absorption data recorded by spectrophotometers and over 100 discrete samples analyzed by FT-ICR-MS, collected over 1.5 years in the Große Ohe catchment, were used to investigate the spatial and temporal DOM dynamics in unprecedented detail. Specific marker compounds of DOM sources within the catchment were used to support the delineation of source contributions to the mixed-signal observed in the stream.

Study Area and Sampling
Samples and data were obtained from the Bavarian Forest National Park (BFNP), which is located in Southeast Germany (Figure 1). The BFNP is the first designated national park in Germany and represents near-natural forests in central-European low mountain ranges. The study was carried out in two sub-catchments of the Große Ohe catchment (with 19.2 km 2 and mean terrain slope of 7.7°), named Markungsgraben (MG, with 1.1 km 2 and mean slope of 15.9°) and Hinterer Schachtenbach (HS, with 1.5 km 2 and mean slope of 7.4°), named as after their respective main streams. The whole hydrological system comprising MG, Kaltenbrunner Seige (KS, another small subcatchment), and HS includes an area of 3.5 km 2 . Elevation in the Große Ohe catchment ranges from 770 to 1447 m above sea level. The Große Ohe catchment has been monitored by the BFNP since 1977 (Beudert & Gietl, 2015) and two of the main monitoring sites are located within the studied area. The Große Ohe catchment is almost exclusively (98%) covered by forest, with Norway spruce (Picea abies, 70%), and European beech (Fagus sylvatica) being the dominant species . The geology of the area is dominated by biotite granite and cordierite-sillimanite gneiss . Overall, soils are mainly cambisols and podzols, but the proportions of hydromorphic soils vary considerably between 5% and 35% at MG and HS, respectively . At MG and HS, the mean annual precipitation (1990-2010) is 1600 and 1379 mm, respectively. Mean annual temperature is 6.2°C at a station 3.8 km east of the catchment outlet.
We further divided the study area into three distinct zones along the course of the streams: (a) MGL, lower Markungsgraben, located at the upper part of the Große Ohe catchment, (b) HSU, upper Hinterer Schachtenbach, and (c) HSL, lower Hinterer Schachtenbach, located at the lower part of the Große Ohe catchment and characterized as a floodplain area. In each zone, one continuous monitoring site for discharge (at MGL and HSL) and UV-Vis (at MGL, HSU, and HSL) and three sampling points for FT-ICR-MS analysis were established. Stream water samples were taken monthly between August and November 2018 and April and November 2019 (i.e., 11 sampling dates), summing up to 98 samples across different seasons. The seasonality in this study was defined according to the calendar. Samples were consistently collected at baseflow in the three defined zones, except for April 2019, where baseflow conditions were not met at HSL ( Figure S1, see below). However, during the sampling campaign in April 2019, baseflow conditions were identified at MGL and the DOC concentration of this sample (of HSL) was in the range of all samples, therefore this sample was also included in the further analysis. Next to the stream samples, three possible sources of DOM, outside the stream, contributing to the stream DOM were sampled over the period of study: shallow groundwater from a spring close to HS main stream (September 2018), riparian soil groundwater from a piezometer (October 2019), and riparian zone surface water (September 2019) from a puddle. The latter two both were located in the riparian zone of the floodplain area along the HS main stream. The piezometer was installed ∼65 cm deep in the soil with slots along the full depth and the puddle sample was collected in a small shallow pond resulted from the irregularities of the terrain that are filled with water during wet conditions (see Figure 1, for locations). The spring, piezometer, and puddle samples were representative of DOM sources and refer in the study as shallow groundwater, riparian soil groundwater, and riparian zone surface water, respectively.
The stream and source samples were collected in cleaned glass bottles (burned for 4 h at 400°C), filtered with polycarbonate track etched membrane filters (LABSOLUTE, 0.4 μm pore size), acidified with hydrochloric acid (HCl, ultrapure, Merk, Germany) to pH 2 and, stored in a refrigerator until analysis.

Discharge
Discharge was monitored at the outlet of the HS catchment ( Figure 1) during 2018 using pressure transducers (Solinst Canada Ltd., Georgetown, Canada), which measured the water level every 15 min, and regular measurements of flow velocity with an electromagnetic current meter (FlowSens, SEBA Hydrometrie GmbH, Kaufbeuren, Germany). With both data, a stage-discharge rating curve was calculated and used to derive the discharge values, following Kreps (1975). From 2019 on, discharge was monitored by the BFNP. Discharge data at the outlet of the MG catchment ( Figure 1) was retrieved from the Bavarian State Office for Environment database for the entire period (2018-2019) (https://www.gkd.bayern.de/en).
Discharge was monitored during the study period between June and November 2018 and between April and November 2019. The minimum of discharge over consecutive 5-day periods was calculated as described in Gustard and Demuth (2009) to separate baseflow and nonbaseflow conditions. Baseflow conditions were defined when the calculated baseflow was in the interval of the actual flow +5% ( Figure S1). The calculations were performed in R using the package "lfstat" (Koffler et al., 2016). For HSU zone, discharge was not recorded and, therefore, baseflow conditions at HSU were assumed when discharge at MGL and HSL were defined as base flow. Because of the limitation in the calculation of baseflow conditions at HSU, only the data recorded at identified simultaneous baseflow conditions at MGL and HSL were further analyzed. Rainfall and snowmelt events also occurred during the study period. Their effect on DOC mobilization is discussed in a separate paper by Blaurock et al. (2021).

DOC Concentration
DOC concentration of the grab samples was determined by high temperature catalytic oxidation (multi N/C 3100, Analytik Jena, Jena, Germany). Performance of the instrument was recorded by daily analysis of in-lab standard solutions.
Stream DOC concentration at the three defined zones (Figure 1) was estimated from UV-Vis spectrophotometers (spectro::lyser, s::can Messtechnik GmbH, Vienna, Austria) using the software ana::pro. The spectrophotometers measured the absorbance between 200 and 750 nm with increments of 2.5 nm, every 15 min. The spectrophotometers recorded data between June and November 2018 and between April and November 2019. However, during the monitoring period, there are data gaps due to instrument failure ( Figure S2, Table S1).
The window of the sensors was cleaned every two weeks and no drifts in the absorption caused by obstruction of the windows were detected. The estimated DOC concentration from the spectrophotometers was monthly checked and calibrated with DOC concentration in grab samples measured in the laboratory, with good fit of the calibration curve (Hoffmeister et al., 2020) (r 2 = 0.91, 0.98, and 0.95 at MGL [n = 31], HSU [n = 18], and HSL [n = 43], respectively; cf. Text SI for details). Samples used in the calibration of the sensors were collected over periods of base and event flow.

Absorption Indices
The average absorption coefficient between 700 and 750 nm was used to correct the absorption spectra due to baseline drift, temperature, scattering, and refractive effects (Coble, 2007) before the absorption indices were calculated. Specific UV absorbance (SUVA, L m −1 mg-C −1 ) was calculated as the ratio of DOM absorbance at 254 nm and the DOC concentration, and used as an estimator of DOM aromaticity (Weishaar et al., 2003). The ratio of DOM absorbance measured at 250 and 365 nm (E 2 :E 3 ) is inversely correlated with molecular weight of DOM and it was also used to characterize DOM apparent molecular weight (De Haan & De Boer, 1987). The spectral slope between 275 and 295 nm (S 275-295 ) was also calculated according to Helms et al. (2008) using the Rpackage "cdom" (Massicotte & Markager, 2016) and setting the reference wavelength to 285 nm. Although the data derived from the spectrophotometers (DOC concentration, SUVA, E 2 :E 3 , and S 275-295 ) have some gaps ( Figure S2, Table S1), the overall trend in DOC concentration and DOM quality could still be evaluated.

Solid-Phase Extraction
Solid-phase extraction (SPE) of filtered and acidified samples was performed according to Dittmar et al. (2008) and Raeke et al. (2017). A volume of 5-200 mL was extracted on 50 mg styrene-divinyl-polymer type sorbents (Bond Elut PPL, Agilent Technologies, Santa Clara, CA, United States) using an automated sample preparation system (FreeStyle, LC Tech, Obertaufkirchen, Germany). The sample volume was adjusted to load 200 μg carbon on the cartridges. The SPE-DOM was eluted with 1 mL methanol (Biosolve, Valkenswaard, Netherlands) and stored at −20°C until measurement. Aliquots of the methanol extracts (10 μL) from the SPE samples were evaporated under N 2 gas flow to complete dryness and subsequently redissolved in 10 mL ultrapure water (Milli-Q Integral 5, Merck, Darmstadt, Germany), and DOC concentration measured. The carbon-based extraction recovery was subsequently calculated to be 52 ± 10% (number of samples, n = 101) and is within the range of typical extraction efficiencies obtained for freshwater samples (Dittmar et al., 2008;Raeke et al., 2017).

Measurement
Immediately prior FT-ICR-MS analysis SPE-DOM samples were diluted to 20 ppm and mixed 1:1 (v/v) with ultrapure water (Milli-Q Integral 5, Merck, Darmstadt, Germany). Samples were measured using a solariX XR 12 Tesla FT-ICR-MS (Bruker Daltonik GmbH, Bremen, Germany) with dynamically harmonized analyzer cell equipped with an autosampler (infusion rate: 10 μL min −1 ) and electrospray ionization (ESI) source (Apollo II) in negative mode. Samples were measured in random order with negative ESI mode (capillary voltage 4.2 kV), 4 megaword time domain, and 256 coadded scans in the mass range of 147-1,000 mass-to-charge ratio (m/z). FT-ICR-MS data were initially recorded in magnitude mode and subsequently transformed to absorption mode using FTMSprocessing (v 2.2.0), which increased the resolution of mass spectra (Da Silva et al., 2020).
All spectra were internally calibrated with a reference mass list of known NOM masses (n = 188; 250 < m/z < 640) and the mass accuracy after linear calibration was better than 0.2 ppm (n = 101). Peaks were considered if the signal-to-noise ratio (S/N) was greater than 2. Reference DOM samples (Suwannee River fulvic acid, SRFA) and quality control samples (e.g., pooled aliquots of all the samples included in the run) were repeatedly measured to check instrument performance across measurement days. Extraction and solvent blanks were also included in all runs.

Data Processing
Molecular formulas were assigned to mass peaks within ±0.5 ppm in the range of 150-1000 m/z according to published rules using element ranges C 1-180 H 1-198 N 0-3 O 0-40 S 0-1 . All molecular formulas assigned to blank (solvent and extraction) peaks were removed from the samples. Multiple assignments accounted for less than 1% of all assigned peaks and were removed. Isotopologue formulas ( 13 C, 34 S) were used for quality control but removed from the final data set as they represent duplicate chemical information. Chemical constraints were further applied to the assigned molecular formulas: , −10 ≤ DBE-O ≤ 10 (Herzsprung et al., 2014), and element probability rules proposed by Kind and Fiehn (2007). Due to the high degree of spectral similarity between samples (47% of the mass peaks were present in at least 75% of all samples), detailed data evaluation was based on normalized intensities. Relative peak intensities (RI) were calculated based on the sum of intensities of all assigned peaks in each sample. The RI is a semi-quantitative measure of the contribution of formulas to the overall DOM in the sample, accessible by this method. The RI also allow the comparison of peak intensities across samples, with an increase in RI indicating an enrichment of the respective compounds contributing to the overall DOM. Throughout the manuscript, we refer to an assigned DOM molecular formula as compound, although a DOM molecular formula potentially represents multiple isomers (Han et al., 2021).
To calculate the probability of the existence of aromatic structures, the modified aromaticity index (AI mod) was used (Koch & Dittmar, 2006. The AI mod assumes that half of oxygen forms a C-O double-bond and is an index applied for identification of aromatic (0.5 > AI mod > 0.67) and condensed aromatic structures (AI mod ≥ 0.67) from FT-ICR-MS data (Koch & Dittmar, 2006. The intensity-weighted average ( wa ) of elemental ratios (H/C, O/C, N/C, S/C) and of AI mod was also calculated and used as aggregated DOM molecular descriptors. For molecular weight, the average m/z was used instead of intensity-weighted average because the highest peaks are located around 400 m/z.

Statistical Analysis
Analysis of variance (ANOVA) was used to test for differences in group means (DOC concentration, absorption quality indices, seasons, and aggregated DOM molecular descriptors) followed by a Tukey's test to determine which groups were significantly different. The t-test was used to test for significant differences in discharge values (two groups only). The coefficient of variation (CV) was also applied to describe the relative variability of absorption quality indices. The separation of sampling sites into zones was also evaluated by ANOVA and principal component analysis (PCA). PCA is a nonsupervised ordination method used to detect patterns in a multivariate data set. PCA was performed with the aggregated DOM molecular descriptors (H/C wa , O/C wa , N/C wa , S/C wa , m/z mean , and AI mod wa ) calculated from the shared formulas between the samples within each of the three zones (n = 98 samples in total). All variables were centered and scaled before the PCA was calculated. In order to analyze the spatial patterns of DOM composition, the relationship between RI of compounds and DOC concentration was tested in the subset of compounds shared within samples of the same zone and for each zone separately. Pearson's correlation was used to test whether the relationship between RI and DOC concentration was significantly (p < 0.05) positive or negative. To control the false discovery rate of multiple testing, p-values were adjusted using Benjamini and Hochberg "BH" method (Benjamini & Hochberg, 1995) before the nonsignificant correlated formulas were filtered out. To better compare the intensity distribution among DOM source samples, intersample rank was applied to the normalized intensity (RI) calculated for the shared compounds among the three samples (Herzsprung et al., 2012). All statistical analyses were performed using R (R-Core-Team, 2017).

Spatial Patterns of Mobilized DOM in the Große Ohe Catchment
Over the period of study (June-November 2018 and April--November 2019), baseflow conditions were identified at 26% and 25% of the time at MGL and at HSL, respectively (Table S1). In 14% of the time baseflow conditions were simultaneously identified in both locations and, based on this, baseflow conditions were inferred at HSU. The average DOC concentration retrieved from the spectrophotometers during baseflow conditions was very similar among MGL, HSU, and HSL (2.6 ± 0.3 mg L −1 , 2.6 ± 0.5 mg L −1 , and 2.6 ± 0.6 mg L −1 , respectively, Figure 2a), with no significant difference (ANOVA one way, p = 0.57). However, mean specific discharge was significantly (t-test, p < 0.001) higher at MGL (17.1 ± 15.8 L s −1 km 2 ) compared to HSL (14.1 ± 9.7 L s −1 km 2 ) ( Figure 2b). Moreover, at each continuous monitoring site, DOC concentration was not significantly different between summer and fall at baseflow conditions, while discharge was significantly different (ANOVA one-way, p < 0.001) between all seasons ( Figure S3). During spring, however, DOC concentration was significantly higher (p < 0.001) compared to the other seasons. In accordance with the high resolution DOC data from the spectrophotometers, DOC concentration in all grab samples used for FT-ICR-MS analysis was not significantly different between the three defined zones (ANOVA one-way, p = 0.07).
As indicators of DOM quality, SUVA and E 2 :E 3 were used. The range of SUVA values during the study period (from 4 to 8 L m −1 mg-C −1 ) was in the range as reported for other small forested catchments in Germany (e.g., Broder et al., 2017;Werner et al., 2019). The highest median SUVA and E 2 :E 3 values were recorded at HSL, followed by MGL and HSU (Figures 2c-2d). The coefficients of variation of SUVA and E 2 :E 3 were also higher at HSL (8.1% and 13.6%, respectively) as compared to HSU (5.9% and 7.5%, respectively) and MGL (7.1% and 6.6%, respectively). SUVA and E 2 :E 3 were significantly different between the monitoring sites (ANOVA one-way, p < 0.001). Regarding season, SUVA was only significantly different between fall and spring (ANOVA one-way, p < 0.001) and between fall and summer (ANOVA one-way, p < 0.001), while E 2 :E 3 was only significantly different between spring and fall (ANOVA one-way, p < 0.001) and between spring and summer (ANOVA one-way, p < 0.001) ( Figure S3). Spectral slopes as S 275-295 has also been widely used to assess DOM molecular weight (Helms et al., 2008). S 275-295 showed highest variability at HSU, but in general a similar trend was shown by S 275-295 and E 2 :E 3 ( Figure S4). Overall, DOM quality described by absorption indices showed higher variability, but also a more aromatic and low MW DOM in the HSL zone, which is less steep and covers a larger portion of hydromorphic soils, compared to the upstream zones (HSU and MGL).
Detailed molecular composition of the DOM from almost a hundred collected samples (11 sampling campaigns) during the study period (1.5 years) could be derived from the FT-ICR-MS analysis. In total, 23,712 unique molecular formulas were assigned to DOM compounds in this data set with a comparable number of compounds identified in each sample (Table 1). Yet, there is a significant difference in the RI of compounds shared between the three zones (ANOVA, p = 0.004). Tukey's test revealed that the RI of compounds at MGL was significantly different from HSL (p = 0.02), but not between HSU and HSL (p = 0.17) or between MGL and HSU (p = 0.29). Moreover, samples of the same zone have a high degree of compositional similarity SILVA ET AL.  between them (45% of the peaks are shared between all samples from the same zone, comprising >80% of the RI in each spectrum), with minor contribution of unique peaks to the overall DOM composition (Table 1). Therefore, in the further analysis, only the compounds shared between samples from the same zone were considered.
Nevertheless, PCA revealed subtle differences in aggregated DOM molecular descriptors between MGL, HSL, and HSU ( Figure 3). The first principal coordinate (PC 1) explains 44.5% of the total variation in the data set and the variables that most contribute to it are m/z mean , AI mod wa and H/C wa (Table S2). Accordingly, O/C wa , S/C wa , and N/C wa contribute the most to PC 2, explaining 31.0% of the total variation. Overall, PCA showed a slight ordination of samples from MGL to HSL. MGL samples tended to have higher N/C wa and S/C wa , while samples collected at HSL tended to have higher m/z mean and O/C wa (ellipses on Figure 3). HSU samples where distributed between HSL and MGL samples, indicating a transition between the steep and flat zone. Moreover, no apparent gradient for DOC concentration or sampling campaign evolving from the PCA was identified (result not shown). However, testing the difference in the aggregated DOM molecular descriptors used in the PCA via ANOVA resulted in no significant differences between zones (p = 0.54), suggesting that using less aggregated parameters, such as the distinct RI of the underlying individual compounds, may be better suited to reveal DOM composition differences along a small environmental gradient. Furthermore, the longitudinal profile of aggregated DOM molecular descriptors only show significant (negative) trend for N/C wa and S/C wa from upstream toward downstream and no apparent trend evolving from season ( Figure S5).

Molecular Composition of the Mobilized DOM Under Baseflow Conditions
Due to the high degree of compositional similarity among the samples of the same zone (even before the filter of shared samples), the significant difference in the RI of compounds between the zones and the gradient evolving from PCA, the relationship between RI of compounds and DOC concentration was tested for each zone separately. At HSL, the percentage of formulas with RI significantly (p < 0.05) correlated with DOC concentration (39%) was higher as compared to HSU (22%) and MGL (27%). From the total number of compounds for which RI was significantly correlated with DOC concentration, 12% were positively correlated in all three zones. Similarly, 13% were negatively correlated in all three zones. For all zones, the negatively correlated compounds were characterized by high H/C and low O/C (Figures 4a-4c), and were predominantly of the CHNO formula class (Figures 4d-4f). In contrast, the chemical characteristics of the positively correlated compounds differed between zones. At MGL, the positively correlated compounds with DOC concentration were oxygen-rich (high O/C ratio), while at HSU and HSL there was a second cluster of compounds characterized by both low O/C and low H/C. The positively correlated compounds at high O/C ratio were mainly of CHO formula class, while the cluster at low O/C and low H/C was characterized by CHOS and CHO formula classes ( Figure S6). This resulted in an overall larger contribution of CHOS formulas to the significantly correlating formulas at HSL as compared to MGL. Overall, formulas positively correlated with DOC concentrations were oxygen rich and less saturated, indicating that elevated stream water DOC values were associated with more aromatic and polar DOM.

Composition of DOM Sources in the Catchment
Three possible sources in the catchment contributing to stream DOM were sampled during the period of study as representative of shallow groundwater (spring), riparian soil groundwater (∼65 cm, piezometer) and riparian zone surface water DOM (puddle). Each DOM source was sampled one time and, therefore, there is an uncertainty associated with the temporal variability of these sources. However, the number of compounds shared between 2 (for shallow groundwater) and 3 (for each riparian soil groundwater and riparian zone surface water) samples of the same source and the distribution of the CV values of the RI of SILVA ET AL.

10.1029/2021JG006425
10 of 18 these compounds showed less variability than DOM in stream samples ( Figure S7). Thus, we assume here that at baseflow conditions the composition of DOM sources are more stable during seasonal changes than the composition of DOM in the stream, which is subject to varying hydrologic conditions. The source DOM samples also displayed some unique compounds (∼40% of all formulas assigned in each sample), which can represent characteristic fingerprints of those sources. However, as the unique compounds represent less than 10% of the total intensity in each mass spectrum and a detailed molecular characterization of DOM sources is outside of the scope of this study, the unique compounds identified in the source samples will not be discussed further. The variation in the DOC concentration between the DOM source samples was larger than the variation observed in the stream samples (Table 1). The DOC concentrations in the riparian zone surface water and riparian soil groundwater were higher than the ones sampled in the stream, while the shallow groundwater had the lowest DOC concentration of all samples during the period of study. Among the source samples, shallow groundwater had the largest number of formulas assigned, followed by the riparian soil groundwater and riparian zone surface water (Figure 5a). The formula class distribution in the DOM source samples was similar between them and the stream samples, with a predominance of CHO, typical for terrestrial DOM. However, the average m/z of the DOM source samples (shallow groundwater and riparian soil groundwater) was higher than the average in stream samples (Table 1).
In total, 4,440 compounds were shared between all three DOM source samples (accounting for 63% of all assigned formulas on average), representing 90% of total intensity (Figure 5b). The intersample rank analysis revealed distinct differences in the molecular composition between them (Figure 5c). The shallow groundwater DOM was characterized by the relative higher RI of compounds with high H/C and low O/C, being attributed as mostly aliphatic and saturated compounds. For the riparian zone surface water DOM, the relative most intense peaks were characterized by unsaturated and aromatic formulas having low O/C and low H/C values. The riparian soil groundwater DOM also had a relative intense signal in high O/C ratio and variable H/C ratio.
In total, 96% of the compounds positively/negatively correlated with DOC concentration in the stream samples (Figures 4a-4c) were also present in the three DOM source samples ( Figure 6). Most of the formulas with relative higher RI in samples from the riparian soil groundwater and from the riparian zone surface SILVA ET AL.
10.1029/2021JG006425 11 of 18 shared and unique formulas assigned to three DOM source samples shallow groundwater (blue), riparian soil groundwater (yellow) and riparian zone surface water (red), and (c) van Krevelen diagram of formulas shared between the DOM source samples (n = 4,440) colored according to the source in which the compound presented the relative highest relative intensity (i.e., lowest inter-sample rank) among the source samples. In (a) the value in the center is the total number of formulas assigned and in (c) the density contours include 25%, 50%, and 75% of the formulas in each sample.
water (48% and 49%, respectively) corresponded to the compounds positively correlated with DOC concentration in the stream samples at the three zones, while most of the compounds (94%) with relatively high RI in the shallow groundwater DOM agreed with the ones that were negatively correlated with DOC concentration (Figure 6).

Mobilization of DOM Under Baseflow Conditions
The variation in DOC concentrations over base and event flow recorded during the period of study is in the range reported in the literature for other German forested catchments (Beudert et al., 2012;Werner et al., 2019) and the baseflow concentrations are also similar to those reported for a forested catchment in the USA (Singh et al., 2015). The comparatively low DOC concentrations at baseflow conditions observed in our study are expected since, at baseflow conditions, DOM percolates through the mineral soil profile, which has generally low DOC concentrations (Inamdar, Finger, et al., 2011;Kaiser & Guggenberger, 2000). However, similar to northern catchments (Monteith et al., 2007), a significant increase in DOC concentration has been observed in the past 20 years in the Große Ohe catchment (Beudert et al., 2012). Moreover, it is difficult to isolate single factors driving this increase (Roulet & Moore, 2006). In this respect, a better understanding of the linkages between discharge and DOM dynamics is essential to assess changes in the terrestrial carbon cycle (Ledesma et al., 2015;Roulet & Moore, 2006).
In our study, DOC concentration showed no significant spatial variation (Figure 2a). However, the variability of DOC concentration and discharge is especially higher for HSL. Additionally, significantly higher DOC concentration and lower apparent MW were observed in spring for all zones ( Figure S3), indicating that the wetter conditions of the system may hydrologically connect DOM sources, especially from shallower layers of the soil, which are more conductive (Bishop et al., 2004;Ledesma et al., 2015) and export less microbially degraded DOM (Roth et al., 2019). However, at a lower temporal scale, no clear seasonal pattern was observed in the aggregated molecular descriptors of DOM composition ( Figure S5).
In contrast to the DOC concentration, DOM quality derived from the high-frequency UV-Vis data showed significant variability among the zones (Figures 2c-2d). However, the pattern observed in SUVA was not found in the AI mod wa values derived from the FT-ICR-MS data. The disagreement between SUVA and AI mod wa may be due to numerous reasons, like the high level of aggregation applied to the calculation of AI mod wa , the distinct frequency of sampling, or the fact that SUVA only captures the UV-active part of the SILVA ET AL.
10.1029/2021JG006425 12 of 18 DOM, while the mass spectrometry technique captures also the UV-inactive part. Considering the aggregated molecular descriptors derived from molecular information, the large number of chemical parameters were able to separate the zones according to N/C wa , S/C wa , O/C wa , and m/z mean in the PCA analysis. Samples collected at MGL showed overall higher N/C wa and S/C wa compared to HSU and HSL (Figure 3, Figure S5).
High N/C ratios in terrestrial DOM is used as a marker for DOM derived from microbial origin and often attributed to DOM mobilized in the deep layers of the soil (Kaiser et al., 2004;Malik et al., 2016). The sources of sulfur (S) in terrestrial and aquatic DOM are not fully constrained yet (Kaiser et al., 2001;Ksionzek et al., 2016). For example, the presence of S in DOM sampled in soil or surface water in forested catchments can be originated from atmospheric deposition (Vestreng et al., 2007), degradation of living tissues (Mangal et al., 2016), abiotic sulfurization (Poulin et al., 2017), and/or bacterial dissimilatory reduction (Kang et al., 2014). PCA revealed that the most pronounced compositional difference in our data set was that samples collected at the lower part of the Große Ohe catchment (HSL) represented DOM with a higher O/C wa ratio and higher m/z mean compared to the upper, much steeper part of the catchment (MGL). Although these differences are numerically small, they can be seen as significant given that DOM composition in rivers is quite similar around the world (Wagner et al., 2015). Moreover, the spectrophotometers also captured the signal of generally more aromatic DOM in the lower part of the catchment (Figure 2). High m/z, aromaticity, and O/C ratio indicate the input of fresh plant-derived material (Seifert et al., 2016). The pronounced difference in DOM composition between MGL and HSL and the more microbially processed DOM fingerprint in the upper part of the catchment also indicates that, in general, the water is flowing through deeper layers of the soil in the upper part of the catchment compared to the floodplain area. Since no in-stream or photo degradation processes are expected to substantially alter DOC concentration or DOM quality in our study area, as water residence time in the catchment is low, we conclude that the contribution of different DOM sources may be the main drivers for the observed changes in DOM quality and composition along the stream (Singh et al., 2015). The wetter conditions at the floodplain area may favor the connection and contribution of more superficial layers of the soil DOM in constrast to the steep terrain at the upper part of the catchment.

Identification of DOM Sources
In the first-and second-order streams, a similar number of compounds were assigned and no clear trend was observed in the unique compounds in the samples (Table 1). Consequently, shifts in RI rather than presence and absence better describe the DOM dynamics. The correlation between RI of compounds and DOC concentration showed that the amount and composition of DOM were connected even under baseflow conditions in the Große Ohe catchment (Figure 4). Whereas a negative correlation between DOC concentration and RI of compounds indicates a relative decrease in the contribution of these compounds to the overall DOM composition, an increase indicates enrichment of them. Compounds of which the RI was negatively correlated with DOC concentration were similar between all three zones, indicating their large contribution to the background DOC exported at low DOC concentrations. These compounds can be described as carboxylic-rich alicyclic molecules and lipid-like (Lam et al., 2007), but the high structural diversity present in DOM limits the assignment of these formulas to more refined molecular categories Patriarca et al., 2020). Moreover, the compounds for which RI is inversely correlated with DOC concentration in the stream samples are mostly representative of CHNO (Figure 4). This compound class has also higher contribution in the shallow groundwater sample (Table 1). At the three locations, most of the compounds negatively correlated with DOC concentration correspond to the relatively most intense compounds in the shallow groundwater sample ( Figure 6). The low DOC concentration, high nitrogen content, and high contribution of aliphatic compounds in the shallow groundwater sample are in line with the composition ascribed to groundwater DOM flowing through deep flow paths (Inamdar, Singh, et al., 2011;Kaiser & Kalbitz, 2012). The agreement between the negatively correlated compounds and the shallow groundwater is an indication that at low DOC concentration the stream water samples, independently of the location, may have higher contributions of deep groundwater inputs (Peralta-Tapia et al., 2015). The also higher m/z mean values in the shallow groundwater indicates that the low MW compounds become less abundant at deeper layers of the soil as a result of the consumption by microorganisms (Roth et al., 2019).
On the other hand, compounds positively correlated with DOC concentration changed between zones, indicating the contribution of distinct sources at different zones ( Figure 4). We assume that the similar pattern observed between HSU and HSL and distinct from MGL is not mainly caused by the flow of KS into HS, but rather a change in DOM sources from upstream to downstream parts of the catchment. In all zones, the compounds enriched at high DOC concentrations are mostly characterized by high O/C ratio and variable H/C ratio and may be attributed to lignin and tannin degradation products. These compounds are markers for DOM mobilized from superficial layers of the soil (Kaiser & Kalbitz, 2012;McDonough et al., 2020). In the MGL zone, the compounds significantly positively correlated with DOC concentration share compositional features with the compounds most characteristic for the riparian soil groundwater source sample (as observed from the intersample ranks, Figure 5). The riparian soil groundwater DOM source sample was characterized by the highest O/C ratio and relative high aromaticity among the DOM sources sampled, matching the description of riparian soil groundwater DOM, which is younger and less processed compared to deep groundwater (Ledesma et al., 2018;McDonough et al., 2020). Furthermore, the comparably high contribution of CHNO molecular formula in both the riparian soil groundwater and the shallow groundwater samples indicates that DOM is more biologically processed in both samples compared to the riparian zone surface water DOM source sample (Longnecker & Kujawinski, 2011). In addition, in the HSL and HSU, the positively correlated compounds also overlap with the characteristic compounds in the riparian zone surface water sample (low H/C and low O/C ratios) ( Figure 6). The riparian zone surface water DOM source was sampled in a puddle in the terrain, a feature which was only observed in the floodplain area in HSL. Thus, it is expected that their signal will not be found in the upper parts of the catchment. Moreover, the riparian zone surface water sample had the highest AI mod wa recorded and relatively high DOC concentration. In fact, these pools may represent a very important DOC source for small headwater streams, when hydrologically connected to the stream (Ploum et al., 2020). Moreover, all DOM sources were sampled at the lower part of the HS catchment, which may under represent the sources contributing at the upper part of the catchment, especially at MG. Overall, although DOC concentration was not significantly different between zones at baseflow conditions, the relationship between RI of compounds and DOC concentration is indicative of different sources of DOM being activated under baseflow conditions upstream compared to downstream.
Interestingly, formulas being enriched at higher DOC concentration at baseflow conditions reflect the findings discussed in the literature regarding rain event-based changes in DOM composition, with changes in DOM character from more aliphatic-like to more aromatic-like (Wagner et al., 2019). At comparatively higher DOC concentration during baseflow and peak discharge during storm events, more aromatic-and humic-rich DOM are exported to the stream compared to lower discharge values (Hood et al., 2006;Inamdar, Singh, et al., 2011). However, the contribution of terrestrial sources and amplitude of the changes in DOM composition during baseflow conditions are smaller compared to the ones reported for DOM mobilized during storm events.

Conclusions
Dominant trends in DOC concentration and DOM quality were captured by spectrophotometer probes in the stream, while discrete FT-ICR-MS data allowed to specifically characterize DOM molecular composition and link the findings from in-stream samples with specific DOM sources. Remarkably, nuances in DOM composition could be distinguished from the FT-ICR-MS data, highlighting the applicability of the ultrahigh-resolution technique for studying DOM dynamics. The detailed molecular level view of DOM composition could also elucidate the contribution of different DOM sources in the catchment to the mixed signal in-stream. As described in the River Continuum and River as Chemostat concepts, we were able to demonstrate the major contribution of terrestrial DOM sources and the importance of riparian and wetland areas for DOM dynamics at headwater streams even during baseflow conditions.
The information regarding DOM sources contributing to baseflow DOM concentration and composition along the stream is important in the construction of models of DOC export, improving our ability to predict hydrological and biogeochemical responses to environmental change and providing insights into changes in water sources. The elucidation of DOM sources in the Große Ohe catchment is also important given the increasing trend in DOC concentration in the catchment, the impact of DOC concentration in macroinvertebrate communities, and the possible changes in the hydrological conditions of the system in the context of climate change.

Acknowledgments
This work was supported by Rudolf und Helene Glaser-Stiftung in the frame of project "Influence of natural factors on concentration, quality and impact of dissolved organic carbon in the Bavarian Forest National Park" (Project No. T0083\30771\2017\kg). The authors thank the European Regional Development Funds (EFRE -Europe funds Saxony) and the Helmholtz Association for supporting the analytical facilities of the ProVIS Center for Chemical Microscopy within the Helmholtz Center for Environmental Research Leipzig. Jan Kaesler and Kai Franze are gratefully acknowledged for their assistance during FT-ICR-MS measurements and software development. We sincerely thank the Bavarian Forest National Park (BFNP) administration for the provision of the geographical and physiographic data, as well as together with the BFNP staff for the assistance with the installation of the field sensors and during the sampling campaigns.