Composition of Stream Dissolved Organic Matter Across Canadian Forested Ecozones Varies in Three Dimensions Linked to Landscape and Climate

Abstract

Dissolved organic matter (DOM) is a key variable influencing aquatic ecosystem processes. The concentration and composition of DOM in streams depend on both the delivery of DOM from terrestrial sources and on aquatic DOM production and degradation. However, there is limited understanding of the variability of stream DOM composition at continental scales and the influence of landscape characteristics and disturbances on DOM across different regions. We assessed DOM composition in 52 streams at seven research sites across six forested ecozones in Canada in 2019–2022 using 26 indices derived from five analytical approaches: absorbance and fluorescence spectroscopy, liquid chromatography—organic carbon detection, Fourier-transform ion cyclotron resonance mass spectrometry, and asymmetric flow field-flow fractionation. Combined analyses showed clear clustering and redundancy across analytical techniques, and indicated that compositional variations were primarily related to three axes of DOM composition: (a) aromaticity, which was greater in low-relief, wetland-dominated catchments, (b) oxygenation, which was greater in colder and drier ecozones, and (c) biopolymer content, which was greater in lake-influenced catchments. Variability in DOM composition among research sites was greater than variability of streams within a site and variability over time within a stream. Forest harvesting and wildfire disturbances had no common influence on DOM composition across research sites, emphasizing the need for regional studies. Our study provides a broad understanding of the variability of stream DOM composition and its associations with landscape and catchment characteristics at a subcontinental scale, and provides key insights for the choice and interpretation of DOM indices from various analytical approaches.

Key Points

• We analyzed dissolved organic matter (DOM) composition in samples from 52 streams across 6 forested regions in Canada using multiple analytical techniques

• DOM composition varied in three dimensions: aromaticity, oxygenation and biopolymer content, linked to climate, wetlands and lakes

• Our subcontinental-scale assessment provides insights for data interpretation, monitoring program design and land management

Plain Language Summary

Dissolved organic matter (DOM) in surface waters influences water quality, aquatic organisms, and carbon cycling, but variability in its composition across different regions has not been extensively studied. We collected water samples from 52 streams across 6 different forested regions in Canada spanning from coast to coast, and analyzed them using five analytical approaches varying in complexity to characterize the composition of DOM in order to assess the differences in stream DOM among the forested regions, the environmental controls on DOM composition, and which approaches were most useful in our characterization. We found that many regions had distinct DOM composition, and climatic factors like mean annual temperature, the presence of wetlands and lakes explained most of the variations, but we were unable to detect any common effects of land disturbance. For assessing differences in DOM across regions, simple analytical approaches were as useful as the more complex approaches. Our findings are important for understanding the function of aquatic ecosystems, potential impacts of climate change and land management, and implications for drinking water treatment.

1 Introduction

The concentration and composition of dissolved organic matter (DOM) in surface waters influence ecosystem functions and serve as key drivers of drinking water treatment needs and challenges. For example, in aquatic ecosystems, DOM controls temperature and light penetration (Reitsema et al., 2018), protects organisms from harmful ultraviolet (UV) radiation (Williamson et al., 1996), governs the bioavailability and mobility of trace metals, nutrients and contaminants (Aiken et al., 2011; Cuss et al., 2020; Köhler et al., 2014; Van Leeuwen & Buffle, 2009), and can be an important energy source in aquatic food webs (Wetzel, 1995). Aquatic DOM is buried in sediments, transported to the oceans, or mineralized to CO₂ and emitted to the atmosphere, thus contributing to the global carbon cycle (Cole et al., 2007; Drake et al., 2018; Raymond et al., 2013). In drinking water treatment, DOM typically dictates chemical coagulant demand and serves as a precursor of disinfection by-products (DBPs) of health concern. If it is not sufficiently removed, it can foul membranes, contribute to unpleasant taste and odor, and increase the potential for bacterial regrowth in distribution systems (Emelko et al., 2011). The concentration and composition of DOM in streams is influenced both by the mixing of DOM from various terrestrial (allochthonous) and autochthonous sources (Thorp & Delong, 2002), and the biotic and abiotic degradation and transformation of DOM within aquatic networks (Kothawala et al., 2015; Ward et al., 2017). The spatial and temporal variability of stream DOM composition is thus linked to catchment characteristics, such as landcover, topography, size, and disturbances (Kothawala et al., 2014; Laudon et al., 2011). However, most studies of stream DOM composition focus on variability within specific physiographic regions, and less is known about controls on DOM composition among regions at continental scales, including influences of climate, geology, soil and forest types (cf. Creed et al., 2015; Jaffé et al., 2012).

DOM is a complex mixture of organic molecules that can be characterized using a variety of approaches that describe molecular size, class (proteins, carbohydrates, lipids, amino acids), structure, reactivity (aromaticity, hydrophobicity, polarity, acidity, functional groups), and humification (Nebbioso & Piccolo, 2013; Volk et al., 1997). Each approach for DOM characterization is associated with different cost and complexity. Relatively simple and low-cost approaches include biological and chemical oxygen demand (Erlandsson et al., 2008), elemental carbon (C), nitrogen (N) and phosphorous (P) ratios (Aitkenhead & McDowell, 2000), and indices of bulk properties derived from UV-visible (UV-vis) absorbance and fluorescence spectra (Coble, 1996; Fellman et al., 2010; Helms et al., 2008; Stedmon & Bro, 2008; Weishaar et al., 2003). The molecular and structural composition of the DOM pool can also be assessed using mass spectrometry (Stenson et al., 2002), field-flow fractionation (Guéguen & Cuss, 2011), and size-exclusion chromatography (Chin et al., 1994; Huber et al., 2011). These techniques allow for more detailed insights into the composition of DOM in aquatic systems and can help advance our understanding of the sources, diversity and reactivity of stream DOM. The use of multiple analytical approaches enables assessment of overlapping and unique information about DOM composition, which can inform the choice of approaches for future studies based on needs for both DOM interpretation and analysis cost/complexity (Stubbins et al., 2014).

Approximately 9% of the world's forested area is found in Canada, with distinct regions (known as ecozones) differentiated based on vegetation type, soils, geology, topography, and hydro-climate (Ecological Stratification Working Group, 1996). Each of these factors can potentially influence the concentration and composition of stream DOM. Vegetation type is known to influence the composition of DOM in soil litter leachates, for example, with distinct differences between coniferous and deciduous litter (Cuss & Guéguen, 2015; Jaffrain et al., 2007; Thevenot et al., 2010). As DOM moves through mineral soils, it generally exhibits a net change in composition associated with reduced molecular weight, hydrophobicity and aromaticity (Marín-Spiotta et al., 2014). The physical, chemical and biological processes altering DOM in soils are however affected by soil matter content, soil texture, mineralogy, moisture, redox and pH (Kalbitz et al., 2000; Kleber et al., 2015). Wetlands often deliver high concentrations of colored, aromatic DOM to streams (Kothawala et al., 2015; Laudon et al., 2011); however, regional differences in dominant wetland types (bogs, fens, marshes, swamps) with differences in soil types and hydrological connectivity may influence the role of wetlands for stream DOM composition across large scales. Lakes are known as hotspots for organic matter turnover, where DOM can be altered or lost via photo-oxidation, sedimentation and heterotrophic respiration, or produced by autotrophs, thus altering the concentrations and composition of DOM transported downstream (Evans et al., 2017). The effect of lakes on stream DOM may vary regionally, as water residence times and trophic status change the balance between DOM production and degradation (Larson et al., 2007; Vähätalo & Wetzel, 2004). Topography and surficial geology can also influence stream DOM by determining relative contributions to streamflow from near-surface and deeper groundwater flow paths (Battin et al., 2008; Jankowski & Schindler, 2019). Sedimentary rocks may introduce aged petrogenic DOM to streams (Stahl et al., 2021). Lastly, regional differences in climate result in differences in the water and energy balance, thus affecting production, degradation and delivery of DOM to streams. Precipitation and temperature may affect DOM in streams by influencing hydrologic connectivity of terrestrial DOM sources to surface waters, flushing of DOM from soils, aquatic processing through water residence times (Dawson et al., 2008; Kothawala et al., 2015), and microbial activity in soils and streams (Freeman, Evans, et al., 2001; Jankowski & Schindler, 2019; Pietikäinen et al., 2005). A comparison of stream DOM across climatic regions in the United States revealed compositional differences (Jaffé et al., 2008, 2012). The diversity of forested landscapes across Canada presents an excellent opportunity to assess regional physiographic differences and controls on stream DOM composition.

The quality of water originating in forested catchments is threatened by both natural and anthropogenic landscape disturbances, as well as by climate change (Baker, 2003; Emelko et al., 2016; Huntington et al., 2009; Lipczynska-Kochany, 2018). Forest harvesting and wildfire are the two most widespread disturbances in Canadian forests (Brandt et al., 2013); both can affect hydrology (Moore & Wondzell, 2005; Shakesby & Doerr, 2006) and downstream water quality and treatability (Carignan et al., 2000; Emelko et al., 2011; Hohner et al., 2019). The loss of vegetation and the surface organic layer after disturbance can alter both DOM sources and hydrologic processes that mobilize and deliver DOM to streams (Rhoades et al., 2019). Impacts on stream DOM concentrations and composition following disturbances can be short-lived, restricted to periods of high flow, or last for decades (Carignan et al., 2000; Emmerton et al., 2020; Webster et al., 2022; Yamashita et al., 2011). Climate change has been linked to browning of surface waters as a result of greater runoff and hydrologic connectivity in catchments, increased microbial decomposition of organic material and DOM production in soils (Kritzberg et al., 2019; Lipczynska-Kochany, 2018). While effects on stream DOM from disturbances and changes in climate have been evaluated at local scales (e.g., Dittman et al., 2007; Emmerton et al., 2020; Mistick & Johnson, 2020), less is known about commonalities of these effects across diverse physiographic settings.

Stream DOM concentration and composition and their linkages to landscape characteristics had not been systematically evaluated at the national scale in Canada. By analyzing DOM in samples from 52 streams located at seven watershed research observatories (henceforth referred to as research sites) across six major Canadian forested ecozones using several analytical techniques, our objectives were to: (a) describe differences in stream DOM composition among regions and assess whether major sources of variability are regional (among-site), local (within-site), or temporal (within-stream); (b) determine how DOM composition relates to various regional landscape and local catchment characteristics and disturbances; and (c) compare DOM information generated from different analytical techniques, to assess which metrics best describe key dimensions of DOM variability. As such, this study provides insights needed for our understanding of terrestrial-aquatic linkages and aquatic functions and for land management and drinking water treatability.

2 Materials and Methods

2.1 Research Sites, Stream Catchment Characterization, and Water Sampling

We collected water samples from creeks and rivers at seven research sites located in six Canadian ecozones (Figure 1, Table 1). An ecozone is the highest level in an ecological land classification, where geologic, landform, soil, vegetative, climatic, water, wildlife and human factors determine the broad characteristics of each ecozone on a sub-continental scale (Ecological Stratification Working Group, 1996). Most of Canada's predominantly forested ecozones were represented by a research site, and we henceforth refer to the research sites by their respective ecozone names (Figure 1). Broad differences in topography, vegetation and forest types, soils, surficial geology and climate among the seven research sites are summarized in Table 1.

Figure 1

Location of the research sites within Canadian ecozones.

Table 1. Climatic, Topographic, Geologic and Landcover Characteristics of Research Sites

Research site ID	Research site name	Topography^1	Vegetation zone^2	Dominant forest types^3	Dominant soil types^3	Surficial geology^3 (bedrock type)	MAT^4 (°C)	MAP^4 (mm)
Pacific Maritime 1	Kwakshua Watersheds Observatory^5,6	Hilly	Pacific Cool Temperate Forest	Coniferous	Ferro-humic podzol, folisol	Alpine complexes, colluvial blocks (igneous and metamorphic)	9.2	3,262
Pacific Maritime 2	Comox Valley and Victoria Water Supply Areas^7,8	Mountainous	Pacific Cool Temperate Forest	Coniferous, mixed	Humo-ferric podzol, dystric brunisol, ferro-humic podzol	Alpine complexes, till veneer/blanket, colluvial blocks (igneous and metamorphic)	9.7	1,456
Montane Cordillera	Southern Rockies Watershed Project^9	Mountainous	Cordilleran Cool Temperate Forest and Alpine Tundra	Coniferous, mixed, alpine/barren land	Dystric brunisol, gray luvisol, regisol	Alpine complexes, till veneer/blanket, colluvial rubble (carbonates and shale)	1.6^16	1,137^16
Boreal Plains	Utikuma Region Study Area^10	Flat to undulating	Boreal Forest & Woodland	Coniferous, mixed and broadleaf	Gray luvisol, mesisol, dystric brunisol	Till blanket, fine grained, glaciolacustrine (carbonates and shale)	1.7	462
Taiga Plains	Hay River Lowland and Wrigley areas^11,12	Flat to undulating; hilly to mountainous	Boreal Forest & Woodland	Coniferous, mixed and broadleaf	Eutric brunisol, organic cryosol, gray luvisol, mesisol, brunisolic turbic cryosol	Till blanket, till veneer, fine grained (glacio)lacustrine (carbonates and shale)	−2.8	388
Boreal Shield	Turkey Lakes Watershed^13	Hilly	Eastern Cool Temperate Forest	Broadleaf, mixed	Humo-ferric podzol	Till blanket, till veneer, glaciofluvial plain (metamorphic and igneous)	5.3	1,041
Atlantic Maritime	Pockwock Lake Watershed^14,15	Undulating	Eastern Cool Temperate Forest	Coniferous, mixed, broadleaf	Humo-ferric podzol, gleysol and mesisol	Till blanket, till veneer, rock with minor quaternary deposits (igneous and metamorphic)	6.5	1,513

• Note. ¹Gruber (2012); ²Baldwin et al. (2019); ³Marshall et al. (1999); a comparison of the Canadian soil classification with other soil classification systems is available in Soil Classification Working Group (1998). ⁴MAT = mean annual temperature, MAP = mean annual precipitation based on Canadian climate normals 1981–2010 data for the following climate stations: Addenbroke Island (Pacific Maritime 1), Courtenay Meadowbrook (Pacific Maritime 2), Fort Simpson A (Taiga Plains), Wabasca RS (Boreal Plains), Sault Ste Marie 2 (Boreal Shield), Pockwock Lake (Atlantic Maritime) (ECCC, 2022); ⁵Oliver et al. (2017); ⁶Giesbrecht et al. (2021); ⁷Jackson and Blecic (1996); ⁸Riddell and Bryden (1996); ⁹Silins et al. (2016); ¹⁰Devito et al. (2016); ¹¹Ecosystem Classification Group (2009); ¹²Quinton et al. (2019); ¹³Webster, Leach, Hazlett, et al. (2021); ¹⁴Gorham et al. (1998); ¹⁵Jutras et al. (2011); ¹⁶Montane Cordillera research site climate station data (Williams et al., 2019).

Between three and six streams from each research site were sampled, except for the Taiga Plains where 19 streams were sampled (Table S1 in Supporting Information S1). Each stream was sampled between two and 10 times in 2019–2021 to capture different seasons and flow conditions, with the exception of 16 Taiga Plains streams that were sampled only once in the summer of 2019 (Figure S1 in Supporting Information S1). Samples were usually collected during open-water conditions; under-ice samples were only collected at the Boreal Plains site.

Streams were selected to capture the variability in landcover and surficial geology within each research site, and at least one stream from each research site (except Pacific Maritime 1) had been extensively affected by either wildfire or forest harvesting in the last 30 years (Table S1 in Supporting Information S1). Catchment sizes varied between 0.05 and 1,260 km², with half of the catchments between 3.2 and 56 km². The largest catchments (>100 km²) were in the Boreal and Taiga Plains, the sites with the lowest runoff (Figure S1 in Supporting Information S1). Each catchment was further characterized with respect to climate, slope, stream order, and the proportions of wetlands, open water, various soil and forest types (Tables S1 and S2 in Supporting Information S1). Climate parameters along with soil and forest types were extracted at the ecodistrict level from the National Ecological Framework data set (Marshall et al., 1999). Climate variables included mean annual temperature (MAT) and climate moisture index (CMI), which was calculated as the difference between mean annual precipitation and potential evaporation (Hogg, 1997). Soil types were grouped into four broad categories: organic (humisol, mesisol, fibrisol, gleysol, organic and gleysolic cryosol), podzols (ferro-humic and humo-ferric podzols), brunisols/luvisols, and rock (bare rock and regosol). Forests were categorized as coniferous, mixed or broadleaf. For bedrock geology, we differentiated between sedimentary (Boreal Plains, Taiga Plains and Montane Cordillera) and igneous or metamorphic (Pacific Maritime, Boreal Shield and Atlantic Maritime). We classified catchments as “lake-dominated” when lake area was >4% of catchment area, and as “wetland-dominated” when wetland area was >20%. These cut-offs are similar to those used at local scales to assess the influences of lakes and wetlands on boreal stream water chemistry (Kasurinen et al., 2016; Laudon et al., 2011; St. Amour et al., 2005). Proportions of harvested (clear-cut or partial-cut) or burned areas were determined for each stream catchment based on publicly available, published or unpublished data (Table S1 in Supporting Information S1). Streams were classified as “disturbed” when >25% of the catchment area was burned or harvested in the last 30 years (Yamashita et al., 2011). There was no agriculture in any catchment, but some industrial development (roads, pipelines, well pads) existed in several Boreal Plains catchments, with the total developed area <5% of catchment area.

2.2 General Water Chemistry Analysis

Water samples for most analyses were collected in 60-mL amber glass bottles, unless stated otherwise below. These bottles were cleaned by soaking in 10% hydrochloric acid (HCl) for at least 24 hr and rinsed thoroughly with ASTM Type I (MilliQ^®) water, and/or combusted at 500°C for at least 4 hr. Samples for anion analysis were collected in 60-mL plastic bottles. All bottles were rinsed with sample water prior to filling. Samples were filtered in the field using 0.45 μm polyether sulfone (PES) syringe filters (with exception of several samples from the Taiga Plains, which were filtered using 0.7 μm GF/F filters), and then shipped in coolers with ice packs to the laboratory, where they were received within 1–5 days. Samples for elemental analyses, including dissolved organic carbon (DOC), total dissolved nitrogen (TDN) and cations (Na, K, Ca, Mg, Fe, Mn), were preserved with HCl. Samples were stored cool (4°C) prior to analysis.

The analyses of DOC, TDN, major ions and inorganic nutrients were performed at the Natural Resources Analytical Laboratory, University of Alberta. Concentrations of DOC and TDN were measured on a Shimadzu TOC-L_CHP Analyzer (Shimadzu Corporation, Jiangsu, China). Low DOC concentrations (<2 mg L⁻¹) were verified using the persulfate wet oxidation method with a detection limit of 0.05 mg L⁻¹ (Aurora 1030W TOC Analyzer, OI Analytical, College Road, Texas, USA). The carbon-to-nitrogen ratio (C/N) was calculated as the concentration ratio of DOC to dissolved organic nitrogen, estimated by subtracting NH₄⁺ and NO₂⁻ + NO₃⁻ from TDN. Ammonium (NH₄⁺), nitrate and nitrite (NO₂⁻ + NO₃⁻), soluble reactive phosphorus (SRP), chloride (Cl⁻) and sulfate (SO₄²⁻) were determined by colorimetry (Gallery Beermaster Plus Photometric Analyzer, Thermo Fisher Scientific Inc., Vantaa, Finland). Concentrations of dissolved sodium (Na), calcium (Ca), magnesium (Mg), iron (Fe), and manganese (Mn) were measured on a Thermo Scientific™ iCAP™ 6300 Duo ICP-OES Analyzer (Thermo Fisher Scientific Inc., Cambridge, United Kingdom). Detection limits are provided in Table S3 in Supporting Information S1. In addition to standard laboratory quality control checks, duplicates and MilliQ^® blanks were submitted for analyses as blind samples periodically to assess the error associated with the sampling and laboratory analysis. The pH was measured either in situ using different field pH meters (Boreal and Taiga Plains, and Atlantic Maritime sites) or in the laboratory at the University of Waterloo using a Fisherbrand™ accumet™ AB250 benchtop pH meter.

2.3 Analysis of DOM Composition

We used five analytical approaches to assess stream DOM composition, including UV-vis absorbance and fluorescence spectroscopy, Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR-MS), asymmetric flow field-flow fractionation (AF4), and liquid chromatography—organic carbon detection (LC-OCD). Each method yields indices of DOM composition (Table S4 in Supporting Information S1) that represent either direct or indirect measures of bulk DOM properties, such as aromaticity, degree of humification, hydrophobicity, average molecular mass, elemental ratios, or the relative abundance of specific DOM moieties (Chen & Yu, 2021; Nebbioso & Piccolo, 2013). Not all samples were analyzed using all approaches; out of the 229 samples, 219 were analyzed for UV-vis absorbance and fluorescence, 194 for AF4, 95 for LC-OCD, and 46 for FT-ICR-MS (Figure S2 in Supporting Information S1).

2.3.1 Absorbance and Fluorescence

The absorbance spectra between 200 and 700 nm were measured using a Shimadzu UV-1280 spectrophotometer (Shimadzu Corporation, Kyoto, Japan). From the absorption spectra we determined the absorbance at 254 nm (A₂₅₄; Dobbs et al., 1972), specific UV absorbance at 254 nm (SUVA), calculated by normalizing A₂₅₄ by DOC concentration (Weishaar et al., 2003), the ratio of absorbance at 250 and 365 nm (E2:E3; De Haan & De Boer, 1987), the spectral slope between 275 and 295 nm (S_275–295), and the ratio of slopes S_275–295 and S_350–400 (S_R; Helms et al., 2008). We assessed the need to correct SUVA for interference from iron absorbance (Poulin et al., 2014), but the correction did not change SUVA values or ranking among samples sufficiently to impact our results (Figure S3 in Supporting Information S1); thus, we report uncorrected SUVA values.

Fluorescence excitation-emission spectra were measured using a benchtop fluorometer (Aqualog^®, Horiba Scientific, Edison, New Jersey, USA) across excitation wavelengths of 250–480 nm (5 nm increments) and emission wavelengths of 280–500 nm (2.33 nm increments). Due to large variations in DOM concentrations and absorbance, samples with high absorbance were diluted with MilliQ^® water using a maximum dilution of 2× (Kothawala et al., 2013), and with integration times adjusted to be between 0.5 and 10 s. Water Raman signal-to-noise and emission calibration validations were performed upon every run. Parallel factor analysis (PARAFAC) was performed in MATLAB^® R2020a (The Mathworks, Inc.) using the drEEM-0.6.3 toolbox (Murphy et al., 2013). Sample excitation emission matrices (EEMs) were blank-subtracted, corrected for inner-filter effects, and Raman normalized. Noisy segments of EEMs (e.g., the region below the first order Rayleigh scatter) and outliers were removed, leaving 216 samples to develop the PARAFAC model, which was successfully split-half validated for five components (C1–C5) and matched in OpenFluor database (Murphy et al., 2013, 2014). Components were expressed as the percent contribution of each fluorophore to total fluorescence (in Raman units). Components C1 to C4 have been described as humic-like components, with C1 and C2 described as derived from terrestrial sources while C3 is described to be derived from microbial sources. Component C5 is associated with protein-like DOM (Table S5 and Figures S4 and S5 in Supporting Information S1).

Fluorescence data were also used to estimate the biological index (BIX), humification index (HIX) and fluorescence index (FI) using the drEEM-0.6.3 toolbox. The BIX is a proxy of the recent autochthonous production of DOM, and is calculated as the ratio between emission at 380 nm (β peak representing recently derived DOM), and the emission maxima between 420 and 435 nm (α peak representing highly decomposed DOM), at an excitation of 310 nm (Huguet et al., 2009). The HIX is a measure of the degree of humification (associated with lower H/C ratios in humified organic matter), and is the ratio of fluorescence intensity at 435–480 nm and the total of fluorescence intensity at 300–345 nm and 435–480 nm, at 254 nm excitation (Ohno, 2002). The FI is a ratio of fluorescence intensities at 470 and 520 nm, obtained at excitation of 370 nm (Maie et al., 2006); it differentiates between microbially derived sources of DOM (FI > 1.4) and terrestrially derived sources (FI < 1.4; McKnight et al., 2001).

2.3.2 Asymmetric Flow Field-Flow Fractionation

We used AF4 to measure the average molecular mass of DOM, reported as the molecular mass at the peak maximum of the fractogram (M_p). Samples were analyzed using a Postnova AF2000 Multiflow FFF fractionation system (Postnova Analytics, Salt Lake City, Utah, USA) equipped with a 300 Da PES membrane (Cuss et al., 2017). The instrument was calibrated over a molecular mass range of 0.69–20.7 kDa before and after every 10 samples using a mixture of bromophenol blue (Sigma-Aldrich, St. Louis, Missouri, USA) and polystyrene sulfonate (PSS) size standards (PSS-Polymer Standards Service—USA Inc., Amherst, Massachusetts).

2.3.3 Fourier-Transform Ion Cyclotron Resonance Mass Spectrometry

Another pivotal technique for DOM characterization used in our study was FT-ICR-MS (D’Andrilli et al., 2010, 2013, 2015; Kujawinski & Behn, 2006). Samples for the FT-ICR-MS analysis were collected in 60-mL plastic bottles and stored frozen. Prior to the analysis, samples were acidified using trace-metal grade HCl, de-salted and concentrated using solid phase extraction, and then passed through 100 mg Bond Elut PPL cartridges (Agilent Technologies) (Dittmar et al., 2008). Samples were eluted with 1 mL of methanol and injected into a Bruker 9.4 T Apex-Qe mass spectrometer (Bruker Daltonics, Billerica, MA, USA) with an Apollo II electrospray ionization source in negative mode using a flow rate of 120 μL hr⁻¹ to acquire 600 spectra scans. Molecular formula were assigned using the ICBM-OCEAN tool and the following constraints: C_1–50, H_1–200, O_1–50, N_0–4, S_0–2, P_0–1 (Merder et al., 2020). Oxygen-to-carbon ratio (O/C), hydrogen-to-carbon ratio (H/C), and modified aromatic index (AI_mod) were calculated from the molecular formula (Koch & Dittmar, 2006). Each formula was classified as Aliphatic (H/C ≥ 1.5), Aromatic (0.5 ≤ AI < 0.67), Condensed Aromatic (AI ≥ 0.67), and for H/C < 1.5 and AI < 0.5, Low-O Unsaturated (O/C ≤ 0.5) or High-O Unsaturated (O/C > 0.5). Aliphatic compounds include lipids, proteins and carbohydrates; unsaturated compounds include lignins and tannins; aromatic compounds include phenols; and the condensed aromatic class includes black carbon. Signal intensity for each formula was normalized and the relative abundance of compounds within each class was used to determine the proportion of each class in each sample. Sample H/C and O/C were also estimated as intensity weighted averages. It is important to note that the Bruker 9.4 T Apex-Qe mass spectrometer employed in this study was part of an extensive global interlaboratory exercise (Hawkes et al., 2020). This exercise confirmed the consistency and reliability of FT-ICR-MS results across various laboratories, and validated the robustness of our analytical approach and indices, ensuring that our findings are in line with internationally recognized standards for DOM composition analysis.

2.3.4 Liquid Chromatography—Organic Carbon Detection

Water samples for LC-OCD analysis were collected unfiltered in 1-L HDPE or LDPE bottles and shipped to the University of Waterloo. LC-OCD analysis used a Model 8 LC-OCD analyzer (DOC-Labor GmbH, Karlsruhe, Germany). A size exclusion column, weak cation exchange column on a polymethacrylate basis (Toyopearl HW 50S, 250 mm × 20 mm, 30 μm from TOSOH Bioscience) was used for separation. Two detectors—a nondestructive, fixed wavelength UV-detection (UVD 254 nm, type S-200, Knauer, Berlin, Germany) and an organic carbon detector (OCD, Huber & Frimmel, 1991)—were used for carbon detection and characterization after chromatographic separation. OCD and UVD calibrations were based on potassium hydrogen phthalate. For data acquisition and data processing a customized software program was used (ChromCALC, DOC-LABOR, Karlsruhe, Germany). Using LC-OCD, chromatographic DOM was separated into five size fractions based on the hydrodynamic radii (Huber et al., 2011). The biopolymer (BP) fraction is very high molecular weight (100,000–2,000,000 g mol⁻¹), hydrophilic, not UV-absorbing compounds, which include polysaccharides, amino sugars, polypeptides and proteins. Humic substances (HS) fraction consists of humic and fulvic acids with molecular weight of 400–1,100 g mol⁻¹. Building blocks (BB) fraction includes HS-like material of lower molecular weight (300–500 g mol⁻¹). All aliphatic organic acids with weights <350 g mol⁻¹ co-elute in the low molecular weight (LMW) acids (LMWA) fraction, and weakly charged hydrophilic or slightly hydrophobic compounds, like alcohols, aldehydes, ketones, amino acids, appear in LMW neutrals (LMWN) fraction. These fractions were expressed as a percent of the overall chromatographable DOC.

2.4 Statistical Analysis

Statistical analyses were performed in RStudio (RStudio Team, 2020). Data were processed and summarized using the R packages dplyr, tidyr, plyr, PerformanceAnalytics, corrplot, ggdendro, factoextra and tibble (Kassambara & Mundt, 2020; Müller & Wickham, 2020; Peterson & Carl, 2020; Wei & Simko, 2017; Wickham, 2011; Wickham & Henry, 2020; Wickham et al., 2020) and illustrated with the ggplot2 package (Wickham, 2016).

To assess variations in water chemistry and DOM composition among the sites, we performed a principal component analysis (PCA) using the prcomp function in the vegan package (Oksanen et al., 2019). Many variables did not meet the PCA assumption of linear correlations between the original variables; these variables were transformed using the bestNormalize function to reduce the nonlinearity (Peterson & Cavanaugh, 2020). Since not all samples were run using all analytical approaches, we first ran separate PCAs using different subsets of data. To combine DOM composition indices from all approaches in a single PCA, we reduced the data set to include average DOM composition indices across all sampling events for each stream. To maximize the number of streams in the PCA, we used the rfImpute function from the randomForest package (Liaw & Wiener, 2002) to impute the missing values for four streams (S3, S5, S16, and JMR) that had no FT-ICR-MS data. To assess which catchment characteristics (including size, landcover, soil type, climate, and disturbance) influenced DOM composition, PCA scores for the first three principal components were used in a random forest analysis performed using the randomForest function.

3 Results

3.1 General Water Chemistry

We found clear differences in general water chemistry among most research sites, despite substantial temporal variability for individual streams and spatial variability among streams within each research site (Figure 2a, Table S6 in Supporting Information S1). A PCA was performed using 211 samples and 11 general water chemistry variables, including pH, major ions, SRP, TDN and DOC concentration. There were strong correlations between concentrations of Ca and Mg, and of Fe and Mn, and thus only Ca and Fe were included in the PCA (Figure 2a). The PCA identified three components with eigenvalues >1 (Figure 2a, Figure S7 and Table S7 in Supporting Information S1). Streams from the Boreal Plains and Taiga Plains sites generally clustered together, with high PC1 scores (high DOC, Fe, TDN, Cl⁻ and SRP). Pacific Maritime 2, Boreal Shield and Montane Cordillera streams had low PC1 scores. The Montane Cordillera and some Boreal Plains and Taiga Plains streams had high PC2 scores (high NO₂⁻ + NO₃⁻, pH, SO₄²⁻, and Ca), while Atlantic Maritime and Pacific Maritime 1 streams had low PC2 scores.

Figure 2

General water chemistry (a) and DOM composition based on different analytical approaches (b)–(d) for stream water samples from seven research sites located in six Canadian ecozones. Symbols represent average principal component analysis scores for each stream, and whiskers point to individual samples, where multiple samples were available (i.e., temporal variability). Streams with catchments dominated by wetlands (squares) and lakes (triangles) are highlighted. The size of variable loadings (arrows) was increased for clarity.

The highest DOC concentrations were found in Boreal Plains streams, followed in order by Taiga Plains, Atlantic Maritime, and Pacific Maritime 1 streams (Figure 3, Table S6 in Supporting Information S1), although there was substantial temporal variation. The lowest DOC concentrations, typically <3 mg C L⁻¹, were measured in the Montane Cordillera, Pacific Maritime 2 and Boreal Shield streams. The concentration of DOC was positively correlated with TDN and Fe. Despite the overall correlation between DOC and TDN, the C/N ratio differed among research sites with the highest C/N in Atlantic Maritime streams and the lowest in Montane Cordillera streams (Figure 3b). The concentration of DOC was not correlated with pH. Concentrations of NO₂⁻ + NO₃ were high in the Montane Cordillera, Boreal Shield and Pacific Maritime 2 streams, which also had the lowest DOC concentrations (Figure 2a). Concentrations of NH₄⁺ and SRP were correlated and were highest in Boreal Plains and Atlantic Maritime streams (Table S6 in Supporting Information S1).

Figure 3

Comparison of stream DOC concentration and DOM composition for seven research sites across six Canadian ecozones. Concentration of DOC (a), C/N ratio (b), and SUVA (c) show individual measurements, while PARAFAC (d), FT-ICR-MS (e), and LC-OCD (f) results show average fractional DOM composition of various DOM moieties (data from streams with lake-dominated catchments excluded).

3.2 DOM Composition

3.2.1 Absorbance and Fluorescence

The highest SUVA was found in Pacific Maritime 1 (3.6–5.8 L mg⁻¹ m⁻¹) and Atlantic Maritime (3.6–5.2 L mg⁻¹ m⁻¹) streams, while the lowest SUVA was found in Montane Cordillera streams (1.0–3.6 L mg⁻¹ m⁻¹) (Figures 2b and 3c). Higher values (and greater range) of absorbance slope parameters were seen in Montane Cordillera for S_R, and Montane Cordillera and Boreal Plains for E2:E3 and S_275–295, and lower values in Atlantic Maritime and Pacific Maritime 1 (Figures S6a–S6c in Supporting Information S1). The highest BIX was measured in Montane Cordillera and several lake-dominated streams, and the lowest in Atlantic Maritime and Pacific Maritime 1. The HIX values were generally higher in Boreal and Taiga Plains, and lower in Pacific Maritime 2, Montane Cordillera and lake-dominated streams (Figures S6d and S6e in Supporting Information S1). PARAFAC component C1 had the greatest contribution to the EEMs (Figure 3d) and was especially high in Pacific Maritime 1, Atlantic Maritime, and non-lake dominated Boreal Shield streams, but relatively low for Mountain Cordillera streams. Component C4 had a similar pattern to C1 among research sites. Component C2 was especially high for Boreal Plains and Taiga Plains streams, while component C3 had the least variability among research sites. Component C5, associated with protein-like compounds, was highest in Pacific Maritime 2 and Mountain Cordillera streams and lowest in Boreal Plains and Taiga Plains streams (Figure 3d).

The PCA of absorbance and fluorescence data identified three components with eigenvalues >1 (Figure 2b, Figure S8 and Table S8 in Supporting Information S1). In this PCA, Atlantic Maritime and Pacific Maritime 1 streams and a wetland-dominated Boreal Shield stream had high PC1 scores (high SUVA, C1 and C4, low BIX, S_275–295), Boreal Plains and Taiga Plains streams had low PC2 scores (high C5, C4, low C2 and HIX), and Pacific Maritime 2 and Boreal Shield streams had high PC2 scores. Lake-dominated streams generally had relatively low SUVA and HIX, and high C5, C3, BIX, S_R, E2:E3, and S_275–295, and thus lower PC1 scores compared to other streams from the same research site (Figure 2b). Wetland-dominated streams generally had relatively high SUVA, HIX and C2, and thus high PC1 scores. Variability for individual streams showed that Atlantic Maritime streams had temporal covariance in PC1 and PC2 scores, while other streams had no such temporal covariance.

3.2.2 Asymmetric Flow Field-Flow Fractionation

The AF4 analysis showed that M_p did not vary among research sites (Table S6 in Supporting Information S1), but was lower in lake-dominated streams, which often had M_p < 1,125 Da. The highest M_p (>1,500 Da) was found in several Taiga Plains streams and a wetland-dominated Boreal Shield stream (Figure S5f in Supporting Information S1).

3.2.3 Fourier-Transform Ion Cyclotron Resonance Mass Spectrometry

The FT-ICR-MS analysis showed that most molecular formula were not unique to individual research sites; what varied among samples and research sites was primarily the relative abundance of specific formula (Figure 4). The Boreal Plains streams had the greatest number of unique formula, primarily compounds with high O/C, while the Pacific Maritime 1 and 2 and Montane Cordillera streams primarily had unique compounds with low O/C. The Pacific Maritime, Montane Cordillera, and Boreal Shield streams had a greater abundance of high H/C formula compared to Boreal Plains, Taiga Plains, and Atlantic Maritime streams (Figure 4, Table S6 in Supporting Information S1). The proportion of Aliphatic compounds was high in Montane Cordillera and Pacific Maritime 2 streams, while the proportions of Aromatic and Condensed Aromatic compounds were high in Atlantic Maritime, Pacific Maritime 1 and several Pacific Maritime 2 streams (Figure 3e). The Boreal Plains and Taiga Plains streams had the highest proportion of High-O Unsaturated compounds (Figure 3e).

Figure 4

Van Krevelen diagrams showing common and unique DOM compounds at each research site. Five fractions used in analysis are separated by black lines and labeled in panel (c) (adapted from MacDonald et al., 2021).

The PCA performed on FT-ICR-MS and AF4 data identified three components with eigenvalues >1 (Figure 2c, Figure S9 and Table S9 in Supporting Information S1). Similarities among research sites mimicked those found in the PCA using UV-vis absorbance and fluorescence data (Figures 2b and 2c). Wetland-dominated streams generally had relatively low PC1 scores, that is, high Aromatic and O/C. There was no consistent pattern for lake-dominated streams to have higher or lower PC1 or PC2 scores compared to streams within the same research site.

3.2.4 Liquid Chromatography—Organic Carbon Detection

Across all sites, stream DOM was dominated by the HS fraction (55%–80% on average), followed by BB (10%–14%) and LMWN (5%–9%) (Figure 3f; see Table S4 in Supporting Information S1 for acronyms). Proportions of BP and LMWA were generally <3%. The LMWA and LMWN fractions were relatively high in the Montane Cordillera and Pacific Maritime 2 streams (Figure 3f), and also in some Boreal Shield and Pacific Maritime 1 streams. The HS fraction was high in the Boreal Plains, Taiga Plains, Atlantic Maritime streams, and the wetland-dominated Boreal Shield stream, while the BB fraction was high in the Montane Cordillera (Figure 3f). Two lake-dominated Boreal Shield streams had the highest BP fractions and were excluded from Figure 3f.

The PCA of the LC-OCD data identified two principal components with eigenvalues >1 (Figure 2d, Table S10 in Supporting Information S1). The PC1 axis differentiated streams with high HS fraction (Taiga Plains, Atlantic Maritimes and wetland-dominated streams) and streams with high LMWA and LMWN fractions (Montane Cordillera, Pacific Maritime 2). The PC2 axis separated streams with high BB fraction (Montane Cordillera) and streams with high BP fraction (including several lake-dominated streams). Overall, there was more overlap among the research sites in the LC-OCD PCA plot compared to PCAs using other approaches (Figures 2b–2d), and wetland- and lake-dominated streams showed more consistent trends, similar to Aukes et al. (2021).

3.3 Correlations Among DOM Composition Indices

A correlation matrix using data from all stream samples was generated for the 26 DOM composition indices (qualitative indicators of DOM composition), along with DOC concentration and A₂₅₄ (quantitative measures) (Figure 5). Hierarchical clustering yielded four broad groups of DOM composition indices (Figure S10 in Supporting Information S1), with each group including DOM indices from at least three different analytical approaches. The first group of DOM indices was associated with humic, high-oxygen DOM compounds (HIX, HS, O/C, and High-O Unsaturated). The second group was associated with aromatic, low-nitrogen DOM compounds (M_p, SUVA, C1, C4, C/N, Aromatic and Condensed Aromatic). The third group of DOM indices was associated with autochthonous, LMW, aliphatic DOM compounds (Low-O Unsaturated, Aliphatic, H/C, LWMA, LMWN, BP and C5), while the fourth group included DOM indices that are likely associated with photodegraded, as well as microbially produced DOM compounds (BB, BIX, FI, C3, C2, E2:E3, S_R and S_275–295). The analysis showed that stream DOC concentration and A₂₅₄ were strongly positively correlated with DOM indices in the first group, weakly positively correlated with the second group, strongly negatively correlated with the third group, and weakly negatively correlated with the fourth group of DOM indices (Figure 5).

Figure 5

Correlation matrix of DOM indices used in the study. Circle color and size indicate Spearman's rank correlation coefficient (ρ). Insignificant correlations (p ≥ 0.05) are left blank. The groupings on the left are based on a cluster analysis (Figure S10 in Supporting Information S1). Text color for each index corresponds to the DOM analytical approach used.

3.4 Environmental Controls on DOM Composition

We combined all DOM indices into a single PCA (Figures 6a and 6b, Figures S11 and S12 in Supporting Information S1), which required us to use average values of DOM indices for each stream since not all DOM indices were analyzed for all samples. A total of 32 streams had all DOM indices analyzed for some samples, and an additional four sites had all DOM indices except indices from FT-ICR-MS analysis which were imputed for this PCA. The PCA had four components with eigenvalues >1; the first two components explained 52% and 20% of the variability, respectively; the third component explained an additional 10%, and the fourth only 5% (Table S11 in Supporting Information S1). Positive scores on the PC1 axis (i.e., high H/C, LMWA, BIX, Low-O Unsaturated, S_R and S_275–295) were associated with streams from the Montane Cordillera, Pacific Maritime 2, and a few lake-dominated streams. Negative scores on the PC1 axis (i.e., high SUVA, Aromatic, C/N, O/C, Condensed Aromatic, C1, HS and HIX) were associated with Atlantic Maritimes, Pacific Maritime 1, and a wetland-dominated Boreal Shield stream. Negative scores on the PC2 axis (i.e., high C2, High-O Unsaturated, E2:E3, S_275–295 and O/C) were associated with Boreal Plains and Taiga Plains streams. Positive scores on the PC2 axis (i.e., high C4, Aliphatic, C5 and LMWN) were associated with Pacific Maritime 2 and Pacific Maritime 1 streams. Low scores on the PC3 axis (i.e., low BP and C5, and high M_p) were primarily associated with lake-influenced streams.

Random forest analyses showed which climatic and catchment characteristics influenced each of the three principal components in the PCA (Figures 6c–6e). The components were influenced primarily by wetland soils (PC1), climate (PC2), and aquatic processes (PC3). Scores for PC1 decreased with greater presence of wetland soils, lower stream pH and lower slope (slope was negatively correlated with wetland abundance). Scores for PC2 increased with higher CMI and MAT, and larger presence of podzolic soils. Scores for PC3 decreased with greater presence of lakes, and to a lesser degree higher stream pH, lower slope, and greater catchment area, all of which were in turn related to the presence of lakes.

Figure 6

Principal component analysis (PCA) using averages for each stream for all DOM indices for the first three principal components (a–b). Points (circles/triangles/squares) represent parameter averages (where multiple samples are available) for each stream; open black circles indicate disturbed streams (% catchment area harvested or burned >30%). The scale of variable loadings (arrows) is increased for clarity. Random forest results using PCA scores for the streams with catchment characteristics as factor variables (c–e).

Disturbance was a poor predictor for each principal component in the PCA (Figures 6c–6e), regardless of whether it was expressed as a numerical variable (proportion of disturbed area in the catchment) or as a categorical variable (disturbed vs. undisturbed catchment). Disturbed streams plotted generally near the undisturbed streams within the same research site (Figures 6a and 6b).

4 Discussion

Our study compared DOM composition in streams draining forested headwater catchments (<1–1,260 km²) at seven research sites across six Canadian ecozones, each characterized by different climate, geology, soils and vegetation. Six of seven research sites included streams with forest harvesting or wildfire disturbances within the last 30 years (Figure 7). The combination of multiple analytical approaches allowed for discrimination among the research sites and individual streams. There was generally little overlap in DOM composition among most research sites. The spatial coverage of this study was restricted to one research site per ecozone (except Pacific Maritime ecozone), with only 3 to 19 streams per research site, which limited our ability for extensive characterization of spatial heterogeneity within each ecozone. Furthermore, the repeated sampling of individual streams may not have been sufficient to fully characterize the temporal variability in stream DOM composition (cf. McSorley, 2020; Oliver et al., 2017). Nonetheless, our study still provided clear evidence showing that regional landscape characteristics have strong associations with DOM composition across large geographical scales, as revealed by statistical analysis. Below we describe and discuss the differences in stream DOM composition among regions, the three main axes of variation in DOM composition, their links to catchment characteristics, and implications for our understanding of potential impacts from climate change, land use and disturbances.

Figure 7

Conceptual illustration of the differences in surficial geology and landcover among the ecozones, as well as observed variations in DOC concentration ([DOC]), DOM aromaticity () and O/C ratio.

4.1 Three Axes of DOM Composition

Our study used several approaches to describe DOM composition, yielding 26 indices. Many indices were strongly intercorrelated, and a majority of them differentiated between two main axes of DOM composition, while a few distinguished a third axis (Figures 5, 6, and 6b).

The first axis, which we refer to as the aromaticity axis, differentiated DOM with high aromaticity from DOM compounds that are known to be preferentially produced through autochthonous or photochemical processes (PC1 in Figure 6a). Hence studies interested in assessing DOM aromaticity can use indices such as SUVA (absorbance), HIX and PARAFAC component C1 (fluorescence), C/N (combustion oxidation analysis or FT-ICR-MS), HS (LC-OCD) or Aromatic and Condensed Aromatic (FT-ICR-MS). These indices increased with an increase in humified, plant-derived DOM of terrestrial origin, with high molecular weight and aromaticity (Kothawala et al., 2014; Marín-Spiotta et al., 2014). The DOM indices that were positively correlated with the aromaticity axis were S_275–295 and S_R (absorbance), BIX (fluorescence), LMWA (LC-OCD), and H/C and Low-O Unsaturated (FT-ICR-MS), which describe DOM of lower molecular weight and a lower degree of humification. Several indices on the aromaticity axis have been found to preferentially increase (S_275–295) or decrease (SUVA, HIX) during photodegradation (Helms et al., 2014), autochthonous production (Liu et al., 2020), and during selective sorption as DOM passes through mineral soil (McDonough et al., 2022).

The second axis, which we refer to as the oxygenation axis, differentiated humic DOM with high oxygen content from aliphatic, LMW DOM (PC2 in Figure 6a). Scores on the oxygenation axis decreased with an increase in E2:E3 (absorbance), PARAFAC component C2 (fluorescence), BB (LC-OCD), O/C and High-O Unsaturated fraction (FT-ICR-MS), and increased with higher PARAFAC components C4 and C5 (fluorescence), LMWN (LC-OCD), and Aliphatic fraction (FT-ICR-MS). High-O Unsaturated compounds are often described as tannins (D’Andrilli et al., 2015), which are abundant terrestrially derived compounds with large molecular size, polyphenolic structure, and varying reactivity (Kraus et al., 2003). Compounds with O/C > 0.9 are classified as sugar-like (Fellman et al., 2020). Higher O/C values were associated with the research sites characterized by higher DOC concentrations and A₂₅₄.

The third axis, referred to as the biopolymer axis, explained less of the overall DOM variability, but was associated with higher BP (LC-OCD) and PARAFAC component C5 (fluorescence), and lower M_p (AF4). Both the BP and C5 indices are associated with a greater fraction of biologically labile, autochthonous DOM, such as polysaccharides, proteins and amino acids (Huber et al., 2011; Kothawala et al., 2014). Biopolymers represent large DOM molecules, but were negatively correlated with M_p. However, given that M_p was most useful in distinguishing lake-dominated streams, which had lower M_p and larger BP fraction, this negative correlation indicates elevated biopolymer content with smaller average molecular mass in lake-dominated streams.

Each PCA on DOM composition using different analytical approaches (UV-vis absorbance + fluorescence, LC-OCD, and FT-ICR-MS + AF4) was able to differentiate DOM composition among research sites in a similar manner (Figures 2b–2d). In particular, the Boreal Plains and Taiga Plains streams cluster together in all figures, and across from them are the Pacific Maritime 2 streams. The Atlantic Maritime and Montane Cordillera plot at opposite ends from each other. The Pacific Maritime 1 streams are most similar to Atlantic Maritime and several Pacific Maritime 2 streams, while the Boreal Shield streams are similar to Montane Cordillera and some Pacific Maritime 2 streams. Thus, it appears that many goals of DOM characterization could be achieved using simpler UV-vis absorbance and fluorescence analyses instead of more complex and expensive techniques. At the same time, the more complex analytical techniques like FT-ICR-MS, LC-OCD and AF4 give us more detailed information about variations in DOM composition, including direct information on DOM size, elemental composition, and quantity of specific DOM fractions.

4.2 Comparison of DOM Composition Among Research Sites and Ecozones

Our seven research sites spanned six forested ecozones (Figure 7). The research sites were characterized by major differences in slope, where sites with flat terrain (Boreal and Taiga Plains) had extensive wetlands. Forests ranged from predominantly coniferous (Montane Cordillera, Pacific Maritimes, Taiga Plains) to mixed (Boreal Plains) and broadleaf (Atlantic Maritime, Boreal Shield), and soils from thick organic deposits (Boreal and Taiga Plains, Pacific Maritime 1) and gleysolic soils in depressions or poorly drained areas (Atlantic Maritime) to brunisols, luvisols (Boreal and Taiga Plains, Montane Cordillera) and podzols (Pacific Maritimes, Atlantic Maritime, Boreal Shield). Surficial geology varied from sedimentary bedrock with thick heterogeneous glacial deposits (Boreal and Taiga Plains), to fractured sedimentary bedrock with thick deposits in valleys or till veneers at higher elevations (Montane Cordillera) to poorly weathered or fractured igneous or metamorphic bedrock overlain by thin to absent glacial deposits (Pacific Maritimes, Atlantic Maritime, Boreal Shield). While the research sites and specific catchments included in this study do not represent the full variability of catchment characteristics within an ecozone (e.g., the differences between Pacific Maritime 1 and 2 discussed below), our data set allowed for a unique assessment of how landscape and climate influence stream DOM concentrations and composition.

Despite being located furthest apart, streams in Pacific Maritime 1 and Atlantic Maritime were found to have many similarities in DOM composition. Both research sites had moderate to high DOC concentrations and DOM with very high aromaticity and low to moderate O/C (Figure 7). Both research sites have relatively warm and humid climates, with coniferous forests on podzolic soils and undulating to hilly topography with a significant presence of wetlands in depressions. Streams at both research sites had relatively low pH and low concentrations of Ca and SO₄²⁻ due to the influence of wetlands and igneous bedrock, while the maritime influence led to elevated concentrations of Cl⁻ (Figure 2a). Similar landscapes and climates in other regions and countries seem to follow the broad trends that we observed in Canada. For example, a research site in Alaska further north along the Pacific coast had higher stream DOC concentrations than what we measured at Pacific Maritime 1 (D’Amore et al., 2015; Fellman et al., 2020). The Krycklan research site on the Scandinavian Shield, characterized by a maritime climate, and similar topography, soils, and wetland extent to Canada's Atlantic Maritime ecozone, has comparable DOC concentrations and SUVA (Kothawala et al., 2015).

Streams at the Pacific Maritime 2, Boreal Shield and Montane Cordillera sites showed similarities in DOM composition, but did not completely overlap (Figures 6a and 6b). These three research sites had several landscape characteristics in common, including shallow soils, few wetlands, and relatively steep terrain. They exhibited very low to moderate DOC concentrations and aromaticity, and low to moderate oxygenation. The exception was a wetland-dominated stream at the Boreal Shield with higher DOC concentration and aromaticity, and two Pacific Maritime 2 streams that were more similar to Pacific Maritime 1. Regional studies conducted at other research sites on the Canadian Shield suggest that our Boreal Shield research site may not be representative of the entire Boreal Shield ecozone. In particular, streams at our Boreal Shield research site (Turkey Lakes Watershed) drain catchments with fewer wetlands and more broadleaf forests than some other Boreal Shield sites and, for example, had lower DOC concentrations and aromaticity than streams at the Experimental Lakes Area (Aukes et al., 2021; Creed et al., 2008), and lower DOC concentrations than streams in boreal Quebec (Hutchins et al., 2017). Thus, although our results show clear distinctions in DOM concentrations and composition for most research sites, results for individual sites cannot always be extrapolated to the entire ecozone. At the same time, our study shows that similarities in landscape characteristics across different ecozones correspond to commonalities in stream DOM composition.

Lastly, streams in the Boreal and Taiga Plains had many similarities in DOM composition, suggesting that DOM composition at the research sites located in different ecozones may not always be distinguishable. These research sites have some of the highest DOC concentrations, moderate to high aromaticity, and the highest scores on the oxygenation axis (Figure 7). Lower stream DOM aromaticity at these research sites compared to other sites may be due to flocculation and consequent decrease in solubility of humic and fulvic acids in the presence of Ca²⁺ and Mg²⁺ in high concentrations, sourced from carbonate-rich mineral soils (Aiken & Malcolm, 1987; Römkens & Dolfing, 1998). No large differences in DOM composition were found, likely due to the similarities in landscape (including flat terrain, thick overburden on top of sedimentary bedrock, abundance of wetlands, primarily coniferous or mixed forests) and climate (lower MAT and precipitation than other research sites). The main difference between these research sites is the presence of permafrost. Some studies observe decreased connectivity in boreal peatlands with permafrost leading to lower DOC concentrations in streams (Frey & Smith, 2005; Olefeldt et al., 2014), which is consistent with our findings of lower DOC in Taiga Plains streams than in the Boreal Plains.

4.3 Environmental Controls on DOM Composition

Variability along the aromaticity axis was explained primarily by the proportion of wetlands (or wetland soils) in the catchment (Figure 6c). Wetlands had a common influence across ecozones, including elevated SUVA, HS, Aromatic/Condensed Aromatic compounds, and M_p, as indicated by the random forest results. Wetlands often covered more than 50% of Boreal and Taiga Plains catchments, and up to 50% of the Pacific Maritime 1 catchments. Due to reduced evapotranspiration and low storage capacity, wetlands are the runoff-generating areas in the Boreal and Taiga Plains (Devito et al., 2017). Dominance of surface and near-surface flow pathways intersecting organic rich soils in the at Boreal and Taiga Plains resulted in high DOC concentrations and aromatic DOM. Although the Atlantic Maritime catchments had fewer wetlands, high aromaticity may be related to a decline in acid deposition and a subsequent increase in solubility of DOM, and brownification seen in some surface waters in Canada's eastern provinces (Redden et al., 2021; Webster, Leach, Houle, et al., 2021). The type of forest (coniferous vs. broadleaf) was not a strong predictor of stream DOM composition along this axis, possibly due to convergence in DOM characteristics in the mineral soil (Thieme et al., 2019).

Variation along the oxygenation axis was explained by climatic factors (MAT and CMI) and podzols (Figure 6d). Podzolic soils that had some of the highest scores in random forest are formed in warm, humid climates through intense weathering. Hence, podzols, as a result of climate, have a common influence along this axis. The PC2 axis separated the two research sites with the lowest annual temperature and precipitation—Boreal Plains and Taiga Plains, which had the highest O/C—from the remaining sample locations, including the warmest Pacific Maritimes streams with the lowest O/C. Climate affects the rate of degradation of organic matter in soils (Thevenot et al., 2010). The Boreal and Taiga Plains streams are also characterized by the largest proportions of wetlands, and highest DOC concentrations. According to the enzymic “latch” theory, anoxic conditions in wetlands promote accumulation of polyphenolic compounds, like tannins (Freeman, Ostle, & Kang, 2001), although recent evidence suggests degradation occurs under both oxic and anoxic conditions (McGivern et al., 2021). Wetlands in higher latitudes have been found to have greater DOM carbohydrate abundance, potentially due to slower decomposition rates in cooler temperatures (Verbeke et al., 2022). Carbohydrates have a higher carbon oxidation state than aromatic compounds, which may explain higher export of oxygen-rich DOM via predominantly shallow flow paths in wetland-dominated Boreal and Taiga Plains catchments. Climate warming may accelerate decomposition of carbohydrates in wetlands at higher latitudes (Verbeke et al., 2022). Research sites with the highest scores on the oxygenation axis were the warmest and wettest—Pacific Maritime 1 and 2 (high H/C and oxygen-poor, LMW, aliphatic compounds), and to some extent Atlantic Maritime (dark-colored, aromatic DOM, but with lower oxygen content). Oxygen-rich DOM may also be lost at research sites with deep flow paths, like Montane Cordillera, as water percolates through till and fractured sedimentary bedrock, and oxidized and aromatic DOM is preferentially removed by sorption onto mineral particles, producing DOM with a strong aliphatic signal that mimics the effects of photodegradation and microbial DOM production (Hawkes et al., 2018; McDonough et al., 2022). Overall, the separation of research sites along the oxygenation axis aligns with the climate gradient, distinguishing between the relatively cooler and warmer sites. This likely indicates the effect of slowed DOM decomposition in soils in cooler climates, which leads to higher concentrations of oxygen-rich compounds like tannins and carbohydrates. The abundance of wetlands in cooler regions may enhance this effect.

Variation along the biopolymer axis was determined by lake influence (Figure 6e). The lake effect in our study was indicated by elevated BP fraction, BIX (especially Boreal Shield streams), PARAFAC component C5, S_R, E2:E3, and S_275–295, which are indicative of photodegradation and autochthonous DOM production (Helms et al., 2014; Kothawala et al., 2014; Sachse et al., 2005). Although the composition of DOM varied among the lake-dominated streams, likely reflecting the terrestrial DOM source, variable retention time and water chemistry in lakes and climatic factors (Kothawala et al., 2014; Kurek et al., 2023), aquatic processes including autochthonous DOM production and photodegradation appear to cause common shifts in DOM composition. We noticed that lake position in the catchment was also important. For example, at Pacific Maritime 2, one stream had a large lake in the upper reach, while another stream had a smaller lake just upstream of the sampling site (Government of British Columbia, 2021). Although the latter was not classified as lake-dominated (lake area <4% of catchment area), it exhibited a similar lake effect (greater proportions of LMW compounds, BP fraction, and PARAFAC component C5) due to the lake's proximity to the stream sampling site.

Temporal variations in DOM composition at a single stream can be large, and occasionally greater than variations among streams within the same research site (Figure 2). Variations appear to be the greatest in several streams where we measured low DOC concentrations (Montane Cordillera, Pacific Maritime 2), where a small addition of DOM from a different source may result in large changes to the proportions of different DOM compounds. Large changes in absorbance and fluorescence were also seen in several lake-dominated streams (Boreal Shield and Boreal Plains), where sunlight exposure, respiration and autochthonous DOM production during water residence may significantly alter DOM composition. Our study was not designed to explore controls on temporal variability in DOM, but rather to compare the magnitude of variability within research sites and streams to that among sites. Although there were large temporal shifts in DOM composition, differences among research sites were generally greater.

Our study showed that disturbances such as wildfire and forest harvesting had no large common influence on DOM composition and was not an important predictor for DOM composition. Hence, we found no common influence of disturbance on DOM composition across research sites similar to what we found, for example, for lake or wetland cover. The effect of disturbance on DOM composition was thus likely a smaller signal compared to the higher-order controls like climate and landcover. However, our study does not rule out the influence of disturbances on DOM composition, but suggests that detailed regional studies are required to understand impacts from specific disturbances in specific settings. Several reasons likely contributed to the lack of a common effect of disturbances on DOM composition in this study. For example, we arbitrarily picked a threshold of 25% disturbed area to classify our catchments as disturbed. The disturbances in our catchments differed in age: some occurred three decades ago (Boreal Shield), others as recently as 2019–2020 (Atlantic Maritime), and repeated harvesting took place in Pacific Maritime 2, where all forests are second growth. Although disturbed catchments can take many decades to recover, disturbances older than 30 years were not considered, thus missing much of the historical wildfire and harvest. In addition, the effects of disturbance are often best captured at high flows, where shallower flow paths are more likely to intersect with disturbed soils (Writer & Murphy, 2012); however, our sampling did not explicitly target storm events.

Another possible reason for the lack of a clear effect of disturbances on DOM composition is the difference in the type and magnitude of disturbances among the research sites and individual catchments. We treated forest harvesting (including clear-cut and partial-cut) and wildfires similarly, although they may produce different effects on DOM. Wildfires may lead to the loss of ground cover, decreased surface roughness, reduced infiltration due to soil crust formation and sealing (Larsen et al., 2009), increased fluvial erosion (Shakesby & Doerr, 2006), and a release of pyrogenic carbon (Myers-Pigg et al., 2015). The effect is a function of wildfire severity and extent, and catchment characteristics. For example, wildfires in the Boreal Plains ecozone had a small or no effect on stream and lake DOC concentrations due to low landscape-stream connectivity (Emmerton et al., 2020; Olefeldt et al., 2013). Higher DOC concentrations were measured in burned catchments in Montane Cordillera (Emelko et al., 2011). In comparison, forest harvesting has the potential to change water balance and flow paths, species composition, leaf litter quality and quantity, thus affecting soil organic matter inputs and DOM export (Yamashita et al., 2011). Some studies in the Pacific Maritime ecozone observed an effect of harvesting on DOM concentrations and composition (Mistick & Johnson, 2020), while others showed little effect (Bourgeois, 2021). In the Boreal Plains ecozone, where aspen forest regenerates rapidly after harvesting, harvest effects on water chemistry may be outweighed by seasonal variability (Petrone et al., 2016). Given these region-specific disturbance effects on DOM concentration and composition, it is not surprising that disturbance was not identified as a key control on stream DOM at this scale. Thus, our study further emphasizes the importance of regional studies to understand impacts of disturbances.

4.4 Implications

Our study suggests that it may be possible to anticipate common shifts in DOM composition across different ecozones, for example, in response to wetland restoration/degradation or reservoir creation/removal, or climate change. The oxygenation axis of DOM composition was primarily explained by climate variables (Figure 6d), and may thus be particularly sensitive to the effects of climate warming in the future, resulting in less oxygen-rich DOM in streams of northern regions. Further research may be necessary to understand how specific axes of DOM composition influence DOM turnover and aquatic functions, and vice versa.

Many of the streams in our study either are drinking water sources (Boreal Plains) or ultimately drain into drinking water sources (Pacific Maritime 2, Montane Cordillera, Atlantic Maritime, Taiga Plains). Drinking water treatability, in particular coagulant demand and formation of potentially harmful DBPs, has been linked to DOM composition (Matilainen et al., 2010). Coagulation has been shown to preferentially remove DOM characterized by high SUVA and molecular weight, low H/C and high O/C ratios, and fluorescent DOM at low emission wavelengths (Lavonen et al., 2015; Marais et al., 2019). Aromatic, HMW compounds are known as precursors of some DBPs (Marais et al., 2019; Parsons et al., 2004), although the relationships are not universal. Based on observed variations in DOM aromaticity and molecular size in our study, we may expect differences in treatability; however, they are hard to predict given the large differences in DOC concentrations among the region. Because DOM composition varies in more than one dimension, it may not always be sufficient to only measure SUVA to accurately predict shifts in treatability needs. Overall, the differences in stream DOM concentrations and composition we observed in our study suggest that drinking water treatability should be region-specific. As climate change may affect DOM concentrations and composition, drinking water treatability may also be impacted.

Our study also suggests that monitoring programs may benefit from including DOM composition indices associated with each of the three DOM composition axes. For example, in addition to SUVA even in simple monitoring programs, PARAFAC components may give insight into the oxygenation and biopolymer axes. Thus, similar to Jaffé et al. (2008), we found that UV-vis absorbance and fluorescence can provide the lowest effort DOM characterization relative to all three axes and place regional DOM composition in a larger context.

5 Conclusion

We analyzed DOM composition in a range of streams from seven research sites located in six forested ecozones in Canada sampled over the course of 2 years. Using 26 indices obtained from 5 analytical techniques, we found both distinct differences and some similarities in DOM composition among the research sites. Though widely disparate in their analytical approach, cost and ease of use, each technique (or a combination of techniques) produced comparable separation of the research sites. Therefore, it may be sufficient to use simpler approaches such as UV-vis absorbance and fluorescence to broadly characterize differences in DOM in surface waters for many research questions. Alternatively, the analysis of DOM composition can employ a hierarchical approach, beginning with simple indices to inform decisions about adding more advanced techniques.

Stream DOM composition broadly varied along three axes—the aromaticity, oxygenation, and biopolymer axes—which were explained by wetland coverage, climate, and lake presence, respectively. We were not able to detect a consistent effect of land disturbance on stream DOM composition across our research sites, which may be attributed to the study design, and the vast differences in climate and landscape characteristics among our research sites that conceal the potential effect of disturbances, suggesting the need for region-specific and disturbance-specific studies of wildfire and forest harvesting effects.

Our results indicate differences in DOM quantity among the research sites, as well as differences in DOM composition, including aromaticity, size and molecular composition, which in turn have the potential to affect chemical regime (e.g., pH, metal and nutrient transport), light penetration, thermal stratification, and thus create distinct habitats for aquatic plants and organisms. The differences in stream DOM composition will determine the rate of DOM turnover as water moves downstream. Our study provides a description of the range of DOM composition across Canadian headwaters, which will be useful for future studies attempting to put the aquatic DOM composition in their study regions into a larger context and to understand the likely controls on their DOM composition. Our findings will be useful for understanding the differences in drinking water treatability among different regions and the effects of disturbances and climate change on DOM composition in surface waters.

Acknowledgments

The research was supported by the forWater Network (forwater.ca), a pan-Canadian strategic research network funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) (NETGP-494312-16). We are grateful to the drinking water utility partners, who were integral to the forWater Network, and provided some of the impetus for this study. Sample collection on the Taiga Plains was funded through the NWT Cumulative Impacts Monitoring Program (CIMP#223).

Open Research

Data Availability Statement

Supporting data, tables and figures are provided in Supporting Information S1. Sample data used in the analysis are available in Orlova et al. (2023) and Emelko (2023).