Multiple linear regression models to predict the ... - ASLO - Wiley

LIMNOLOGY and

Limnol. Oceanogr. 00, 2018, 00–00

OCEANOGRAPHY

C 2018 The Authors Limnology and Oceanography published by Wiley Periodicals, Inc. V

on behalf of Association for the Sciences of Limnology and Oceanography doi: 10.1002/lno.10820

Multiple linear regression models to predict the formation efficiency of triplet excited states of dissolved organic matter in temperate wetlands Andrew J. McCabe

, William A. Arnold

*

Department of Civil, Environmental, and Geo- Engineering, University of Minnesota-Twin Cities, Minneapolis, Minnesota

Abstract The source and composition of dissolved organic matter (DOM) dictates light absorption in surface waters. Sunlight absorption by chromophoric dissolved organic matter (CDOM) forms reactive intermediates and drives global organic carbon processing. Triplet excited states of CDOM (3CDOM*) are primary reactive intermediates formed by sunlight absorption by CDOM. 3CDOM* also produce secondary reactive intermediates, including radicals and reactive oxygen species, which are active in biogeochemical pathways. The efficiency of 3CDOM* formation (apparent quantum yield, AQYT) depends on DOM composition, especially DOM molecular weight. This dependence may arise from the greater probability of forming intra-molecular chargetransfer (CT) complexes in high-molecular weight DOM that inhibit 3CDOM* formation. There are few examples that demonstrate this in field samples. In this report, vegetation, general hydrology, and watershed characteristics for 39 temperate wetlands, which are critical sources of high-molecular weight DOM, from the United States were defined and related to DOM composition and AQYT. The DOM bulk composition was assessed using absorbance and fluorescence spectroscopies. AQYT was estimated under simulated sunlight using the probe 2,4,6-trimethylphenol. Relatively high AQYT values (7%) were observed in wetlands with long hydroperiods and > 50% cropland watershed land cover compared to wetlands with >50% forest watershed land cover (< 1–4%). Low molecular weight DOM (E2/E3 > 7 and SUVA254 < 3 L mg-C21m21) and autochthonous DOM (b/a > 0.7) had relatively high AQYT estimates ( 10%), indicating that allocthonous, high-molecular weight compounds produce 3CDOM* less efficiently than autochthonous DOM. The CT theory of DOM light absorption and internal light-screening offer mechanistic explanations for these trends.

Larson et al. 2014; Yu et al. 2015). In headwater systems with surrounding shrubs, trees, or emergent vegetation, inputs of DOM are dominated by terrestrial sources, such as plant matter and soil organic matter (allochthonous DOM). In contrast, the composition of DOM in systems with large open water regions and long water residence times is dominated by exo-metabolites from photoautotrophs and pelagic heterotrophs as well as products of photo- and microbial€ nster and Chro st degradation (autochthonous DOM) (Mu 1990; Ogawa 2001; Raymond and Spencer 2015). Chromophoric dissolved organic matter (CDOM) is the fraction of DOM that absorbs ultraviolet and visible light energy (Del Vecchio and Blough 2004; Nebbioso and Piccolo 2013). The interaction of CDOM and sunlight is a principal process controlling DOM transformation and mineralization. On a global scale, the rate of photomineralization of dissolved organic carbon to dissolved inorganic carbon (0.4–1.7 Pg-C yr21) is comparable to the mass transport rate of riverine carbon to the ocean ( 0.26 Pg-C yr21) (Mopper et al. 2015). Controlled studies of photo-exposed DOM consistently show a loss of absorbance and aromatic moieties, the

Naturally occurring dissolved organic matter (DOM) is a complex mixture of components including aromatics from terrestrial plants and biomacromolecules from primary and secondary aquatic producers (Wilson 1987; Stenson et al. 2003; Nebbioso and Piccolo 2013). Wetlands are integral sources of DOM to the hydrosphere and function as natural reactors in DOM transformation (Watanabe et al. 2012; Raymond and Spencer 2015). The relative inputs of terrestrial and aquatic DOM, and thus DOM composition, changes from headwaters and isolated wetlands (terrene wetlands) to wetlands associated with lake basins and rivers (lentic and € nster and Chro st 1990; lotic wetlands, respectively) (Mu

*Correspondence: [email protected] Additional Supporting Information may be found in the online version of this article. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

1

McCabe and Arnold

Triplet CDOM in temperate wetlands

formation (also called the apparent quantum yield, or AQYT) in natural systems. To date, predictions have relied on the spectral properties of DOM, usually the ratio of light absorbance at 250 nm to the absorbance at 365 nm (E2/E3; typical range E2/E3 5 4– 12) (Dalrymple et al. 2010; McCabe and Arnold 2016; McKay et al. 2017). This relationship is purported to exist because absorbance in the UV region is suggestive of light absorption by discreet chromophore (quantified by the absorbance at 250 nm) and absorbance in the visible and near-UV region is suggestive of CT complex formation (quantified by the absorbance at 365 nm). High E2/E3 ratios are interpreted as high light absorption by discreet chromophores relative to CT complexes. Past work has suggested that AQYT is negatively correlated with DOM molecular weight (McKay et al. 2016; Maizel and Remucal 2017b), however, DOM photodegradation and reduction of carbonyl moieties to hydroxyl moieties only modestly changes AQYT (Golanoski et al. 2012; Sharpless et al. 2014; McCabe and Arnold 2017). Despite our growing understanding of the photo-physics of DOM light absorption and 3CDOM* formation, models are lacking that specifically relate environmental variables, such as landscape condition and DOM source, to AQYT to allow large-scale prediction of AQYT. The central hypothesis of this study is that watershed and landscape conditions influence factors on the individual wetland scale, such as DOM composition, that dictate 3CDOM* formation efficiency. It is expected that allochthonous DOM, characterized by high molecular weight, high aromaticity, and high specific rate of light absorption, will have relatively low AQYT because of the greater probability of forming intra-molecular CT complexes (instead of 3CDOM*) and greater internal light-screening of 3CDOM* precursors. Thus, wetlands with watersheds dominated by trees and shrubs are expected to have lower AQYT than wetlands with watersheds dominated by open water regions. Mechanistic models that are detailed enough to capture regional variability in DOM composition and photoreactivity, however, are poorly constrained (Mopper et al. 2015). In this work, we identify key landscape and DOM compositional variables that correlate with 3CDOM* formation efficiencies across the ecological gradient of Minnesota, U.S.A. We explore several factors that control the molecular composition of DOM and CDOM in aquatic systems: temperature and precipitation (Chen et al. 2013; Kellerman et al. 2014), water residence times and n et al. 2016; water flow paths (Spencer et al. 2010; Catala Tiwari et al. 2017), and watershed land cover/use and soils (Mattsson et al. 2005; Wilson and Xenopoulos 2009). The results of this work identify readily measurable spectral properties of DOM and landscape variables that allow robust prediction of AQYT. Accurate estimates of AQYT enable prediction of formation rates of secondary reactive oxygen species and phototransformation rates of DOM biomarkers (lignin phenols and amino acids) and organic pollutants.

release of inorganic carbon and nitrogen, and the formation of refractory aliphatic compounds (Del Vecchio and Blough 2002; Kujawinski et al. 2004; Stubbins et al. 2010; Mopper et al. 2015). Products of DOM photodegradation typically have reduced molecular mass (Stubbins et al. 2010) and lower conjugation (Gonsior et al. 2009). The chromophores of DOM are categorized into two overlapping groups that correspond to their excited state structure: (1) discreet chromophores which form singlet and triplet excited state species upon light absorption and (2) chargetransfer (CT) complexes which form between closely associated donor and acceptor groups. Discreet chromophores include olefins, carbonyls, and aromatic moieties (Kujawinski et al. 2004; Gonsior et al. 2009; Turro et al. 2010). Singlet excited states of CDOM form when ground state electrons absorb light energy and are promoted to high-energy atomic orbitals, maintaining their original electronic spin state. Triplet excited states of CDOM (3CDOM*) form when electrons in the singlet excited state relax to relatively low-energy excited states, undergoing an electron spin flip. CT complexes are excited state species in which excited electrons are shared between donor and acceptor moieties forming an excited state structure that is energetically stable relative to the individual moieties in the ground state and excited state (Turro et al. 2010). In DOM, donors are identified as reduced organic groups, such as phenols and hydroquinones, and acceptors are identified as partially oxidized organic groups, such as aromatic ketones and quinones (Sharpless and Blough 2014). While direct observations of 3CDOM* have not been made definitively (McNeill and Canonica 2016), observations of singlet oxygen (1Dg) production from photo-exposed CDOM offers strong evidence for the existence of 3CDOM* (Zepp et al. 1977; Cooper et al. 1988). The precursors of 3 CDOM* are likely aromatic ketones which is based on evidence that alkyl-substituted phenols (common triplet probes) are degraded at comparable rates to model aromatic ketones in the presence of photo-exposed DOM (Canonica et al. 1995) and selectively removing carbonyls in DOM decreases the rate of alkyl-substituted phenol photodegradation (Golanoski et al. 2012). Further, aromatic ketones are prevalent moieties in DOM based on high-resolution molecular characterization (Baluha et al. 2013). 3 CDOM* are particularly important in biogeochemical processes because they are potent oxidants and they form secondary reactive oxidants, such as radicals and reactive oxygen species (Zepp et al. 1977; Parker and Mitch 2016). 3 CDOM* are involved in the photodegradation of soluble organic pollutants (Remucal 2014), pathogen inactivation (Kohn and Nelson 2007), and the phototransformation of lignin phenols and free amino acids (McNally et al. 2005; Boreen et al. 2008; Waggoner et al. 2017). Despite their prominent role in several photodegradation pathways, there are limited models to predict the efficiency of 3CDOM* 2

McCabe and Arnold


Experimental

characteristics are summarized in Supporting Information Table S2 (Section 1). Watershed analyses were completed using ArcMap v. 10.4.1 using the hydrology toolbox (ESRI 2016). All shapefiles and raster datasets used in this analysis are freely available online at the Minnesota Geospatial Commons (https:// gisdata.mn.gov/accessed 11 October 2017). Watershed areas were delineated using 30-m resolution digital elevation model (DEM) data from the Minnesota Department of Natural Resources that was preconditioned by burning-in known natural and altered watercourses from the Minnesota Pollution Control Agency to a depth of 2 m below the DEM surface. Average watershed slope (reported as rise/run), average watershed flow path length, and the ratio of the average flow path length to the average slope (L/G; a proxy for the average watershed water residence time) are reported for each watershed (McGuire et al. 2005). Watershed land cover was determined by clipping watershed areas to 15-m resolution land cover data for Minnesota from 2013 (Rampi et al. 2016). Relative land cover for six characteristics are reported: (1) impervious cover, (2) wetland land cover, (3) forest land cover, (4) grassland land cover, (5) cropland land cover, and (6) open water land cover. Land cover definitions are summarized in the Supporting Information Table S3 (Section 1). Weighted-average watershed soil composition (percent sand, silt, and clay, and percent organic matter) for the top 50-cm was calculated from data freely available online from the Natural Resources Conservation Service Web Soil Survey (websoilsurvey.nrcs.usda.gov/accessed 11 October 2017). Ten-year (2004– 2014) average net primary productivity rates (NPP, kg-C m22 yr22) over the watershed areas were calculated using data available from the Land Processes Distributed Active Archive Center (lpdaac.usgs.gov accessed 11 October 2017) (Running 2015). Twenty-one of the sampled wetlands were assigned human disturbance scores (HDS) following the protocol of the Minnesota Pollution Control Agency (Genet and Bourdaghs 2006; MPCA 2016). The HDS is a qualitative measure of land use and wetland perturbations that accounts for the cumulative anthropogenic impacts on a wetland system. HDSs are empirically derived for individual sites using four ratings (best, moderate, fair, and poor) for five factors: (1) buffer landscape disturbance, (2) immediate landscape influence, (3) physical habitat alteration within the immediate landscape, (4) hydrological alteration, and (5) general chemical pollution factors. The HDS ranges between 0 (low or no human influence) and 100 (substantial human influence) and allows comparison of relative anthropogenic influences between wetland systems (MPCA 2014).

Sample collection A total of 113 samples and eight field duplicates were collected from 39 wetlands with varying hydrologic regimes and surrounding land cover between August 2014 and October 2015 throughout Minnesota, U.S.A. (Fig. 1). Ecosystem classes included the Boreal White Spruce Forest and Woodlands (BWSFW; 3 wetlands), Laurentian-Acadian Northern Hardwood Forest (LANHF; 17 wetlands), Laurentian Pine-Oak Barrens (LPOB; 1 wetland), Northern Tallgrass Prairie (NTGP; 8 wetlands), Central Mixedgrass Prairie (CMGP; 1 wetland), North-Central Interior Maple Basswood Forest (NCIMBF; 6 wetlands), and North-Central Interior Oak Savanna (NCIOS; 3 wetlands). Coordinates for each wetland are given in the Supporting Information Table S1 (Section 1). The landscape in Minnesota generally varies from grasslands (dominant species: Andropogon gerardii) in the south and northwest to deciduous and coniferous forests (dominant species: Acer saccharum, Betula spp., Quercus spp., and Pinus spp.) in the northeast, with gradients in land use, from native vegetation to cropland and high intensity development. The mean annual temperature and precipitation follow a gradient from the north to the southeast: 2.28C and 46 cm to 9.48C and 81 cm, respectively. Temperature and precipitation data for each site were taken from webpages maintained by the National Oceanic and Atmospheric Administration (ncdc.noaa.gov/accessed 11 October 2017) and the Minnesota Department of Natural Resources (dnr.state.mn.us/climate/accessed 11 October 2017), respectively (Supporting Information Table S1). Grab samples were collected in 1-L polycarbonate bottles (Corning or Nalgene) that had been pre-rinsed in 10% ACS grade HCl, thoroughly rinsed with ultrapure water, and autoclaved. Samples were transported on ice and were vacuumfiltered with 0.7-lm glass-fiber filters (pre-combusted at 5508C, Millipore) within 6 6 3 d of sample collection. Samples were then filtered with 0.2-lm Omnipore filters (no preconditioning, Millipore). Filtered samples were stored at 48C in the dark until used. Wetland classification and watershed characteristics The dominant terrestrial ecological setting of the wetlands were defined using NatureServe’s standardized system (NatureServe 2009; Sayre et al. 2009). Wetlands were classified using the U.S. Fish and Wildlife Service classification system which is based on wetland size, vegetation, hydroperiod, hydrogeomorphic class (landscape position), and surficial hydrologic connectivity (Cowardin et al. 1979; FGDC 2013; Tiner 2014). Classifications were taken from the U.S. Fish and Wildlife Service National Wetlands Inventory (USFWS 2017; fws.gov/wetlands/accessed 11 October 2017) and the Minnesota National Wetlands Inventory Update (dnr.state.mn.us/eco/wetlands/nwi_proj.html accessed 11 October 2017). Definitions and abbreviations of the wetland

Water chemistry and DOM optical measurements Measurements of pH, specific conductance (SC), anion concentrations, concentrations of dissolved organic ([DOC]) and inorganic carbon ([DIC]), and UV-visible absorbance spectra are described in the Supporting Information, Section 2. 3

McCabe and Arnold


Fig. 1. Map of sampling locations in Minnesota, U.S.A. The abbreviations for the ecosystem classes of the sites are given beside the study areas (NTGP 5 Northern Tallgrass Prairie, NCIOS 5 North Central Interior Oak Savannah, NCIMBF 5 North Central Interior Maple Basswood Forest, CMGP 5 Central Mixedgrass Prairie, LANHF 5 Laurentian-Acadian Northern Harwood Forest, LPOB 5 Laurentian Pine-Oak Barren, BWSFW 5 Boreal White Spruce Forest and Woodland). Coordinates for each site are given in Supporting Information Table S1.

4

McCabe and Arnold


m22 and 2.0 W m22, respectively, measured using a broadband PMA2100 radiometer with PMA2110-WP (UVA) and PMA2106-WP (UVB) detectors. These UVA and UVB intensities are comparable to maximum sunlight intensities at 458N. The average maximum (at 1 : 00 PM) over 16 sunny days in July 2016 on the University of Minnesota-Twin Cities campus was UVA 5 45 W m22 and UVB 5 2.0 W m22. An actinometer solution of 6.8 lM p-nitroanisole (97%, Sigma-Aldrich) and 5.7 mM pyridine ( 99.0%, Sigma-Aldrich) was used to estimate the spectral irradiance of the lamp (Ik, mol-photons L21 s21) between k 5 275–400 nm following the procedure in Sharpless et al. (2014) with the p-nitroanisole/pyridine quantum yield relationship reported in Laszakovits et al. (2017) (Supporting Information Fig. S1A, Section 3.1). The spectral irradiance, Ik, was used to calculate the total rate of light absorption by the samples between k 5 275–400 nm in the solar simulator (Ra, mol-photons L21 s21) using Eq. 1:

Excitation-emission matrices were collected with a Horiba Aqualog using a 1-cm quartz cell, with a 2 or 3 s integration time with 1, 3, or 5-nm excitation wavelength intervals (Gilmore et al. 2012). Spectral corrections were made using the drEEM toolbox in MATLAB (Murphy et al. 2013; MathWorks 2014). Raman scattering signals were corrected by subtracting the matrix spectra of Milli-Q water, inner-filter effects were corrected following Parker and Barnes (1957) (samples were diluted with Milli-Q water to obtain an absorbance measurement of < 0.6 at 254 nm if necessary to ensure accurate application of correction factors), and spectra were normalized to the area of the water Raman scattering peak at an excitation of 350 nm. The optical parameters used to assess DOM source and composition were: (1) the specific UV absorbance at 254 nm (SUVA254, L mg-C21 m21), (2) the ratio of absorbances at 250–365 nm (E2/E3), (3) the ratio of fluorescent emissions at peak b (also called peak M) to peak a (also called peak C), (4) the humification index (HIX, see Supporting Information, Section 2), (5) the fluorescence index (FI, the ratio of ratio of emission intensities at 470–520 nm at an excitation of wavelength of 370 nm), and (6) the ratio of fluorescent emissions at peak C to peak A (C/A). SUVA254 is a direct proxy for DOM molecular weight and aromaticity (reported in decadic units) (Chin et al. 1994; Weishaar et al. 2003). E2/E3 is an inverse proxy for DOM molecular weight (e.g., as DOM molecular weight increases, E2/E3 tends to decrease), and it may serve as a proxy for the degree of DOM photobleaching (De Haan and De Boer 1987; Sharpless et al. 2014). b/a is a proxy for autochthonous or recently produced DOM (Wilson and Xenopoulos 2009), HIX is a proxy for the degree to which fluorescence emissions red-shift as DOM is degraded in litter layers and soils (Ohno 2002), and both FI and C/A are proxies for terrestrially or microbially derived DOM (Cory et al. 2010; Hansen et al. 2016). Calculations of these optical parameters are given in the Supporting Information, Section 2. It must be emphasized that these parameters offer only proxy information on DOM composition because different sources of DOM may not necessarily give a unique signal. While they do not definitely identify DOM source, they offer signatures of a probable source. Using information from several different parameters can alleviate this ambiguity.

Ra 5

400nm X

Ik ð1–10–ak z Þ;

(1)

k5275nm

where ak (m21) is the decadic absorbance and z is the effective light path length through the quartz test tubes. An experimentally determined value of z 5 1.12 3 1022 m for 13 3 100 mm test tubes is given in Leifer (1988), and used here to compute Ra. Because the irradiance throughout the test chamber spatially varies (Supporting Information Fig. S1B, Section 3.1), duplicate test tubes were spaced apart to approximately obtain an average irradiance between the duplicates. Photodegradation experiments with 2,4,6-trimethylphenol The AQYT (mol mol-photons21) and the rate of formation 3 of CDOM* (Rf,T, M s21) were estimated in all collected wetland samples using 2,4,6-trimethylphenol (TMP, 99%, Acros Organics). The photodegradation of TMP was measured in all collected water samples at an initial TMP concentration of 4 lM. Previous work has demonstrated that this concentration of TMP is appropriate for measuring 3CDOM* reactivity (Canonica and Freiburghaus 2001). TMP undergoes a one-electron oxidation reaction with 3 CDOM* (E8TMP 5 1.22 V) (McNeill and Canonica 2016). Because 3CDOM* possess one-electron reduction potentials ranging between E8*51.4 V–1.9 V SHE (3CDOM*/CDOM2•), TMP theoretically samples the entire distribution of 3CDOM* reduction potentials (McNeill and Canonica 2016). Other triplet probes such as cis- or trans-1,3-pentadiene (Zepp et al. 1985) or trans,trans-2,4-hexadienoic acid (Grebel et al. 2011) react with 3CDOM* via an energy transfer reaction, requiring at least 250 kJ mol21. 3CDOM*, however, occupy a distribution of excited states energies ranging between 100 kJ mol21 and > 250 kJ mol21 (McNeill and Canonica 2016), thus, alkene probes only sample a subset of 3CDOM* energy levels. Further, triplet state energies and triplet reduction potentials are not correlated (McNeill and Canonica 2016).

Photochemistry experimental Solar simulator All photochemical experiments were performed in duplicate in quartz test tubes (13 3 100 mm, Ace Glass) held at a 308 angle from horizontal in an Atlas Suntest CPS1 solar simulator equipped with a xenon arc lamp and a 290-nm cutoff filter (some low-intensity light between 275 nm and 290 nm does reach the experimental samples). The intensity of the lamp between k 5 300–800 nm was set to 350 W m22 and the temperature of the experimental solutions was maintained 308C by blowing ambient air at 208C through the test chamber. The UVA and UVB intensities were 37 W 5

McCabe and Arnold


order rate constant for the reduction of TMP1• by reduced moieties in DOM. This reduction mechanism changes the assumed model for TMP photodegradation in Eq. 2. Instead of accounting only for TMP loss, kobs,TMP also includes a reformation mechanism (Eqs. 8 and 9).

TMP was selected in this study because it is assumed that TMP samples the maximum concentration of 3CDOM*. The rate of TMP photodegradation was modeled as pseudofirst order. Under dilute DOC concentrations ([DOC] < 5 3 1024 mol-C L21), the pseudo-first order rate constant, kobs,TMP (s21), is estimated as the product of the second order rate constant for the reaction between TMP and 3CDOM* (kT,TMP, M21 s21) and the steady-state concentration of 3CDOM* ([3CDOM*]ss, M) (Eqs. 2 and 3) (Canonica et al. 1995). kobs;TMP 5kT;TMP ½3 CDOM ss Rf;T ½3 CDOM ss 5 0 kq 1kT;TMP ½TMP0

d½TMP 5–kobs;TMP ½TMP (8) dt d½TMP 5–kT;TMP ½3 CDOM ss ½TMP1kred TMP•1 ss ½DOC (9) dt

(2)

Where [DOC] is used as a proxy for the concentration of DOMred. Substituting for the steady-state concentration of TMP1• gives Eq. 10, which simplifies to Eq. 11.

(3)

Where [3CDOM*]ss is the ratio of the rate of Rf,T (the rate of 3 CDOM* formation) to the rate of 3CDOM* loss as estimated by the sum of the rate of energy transfer to O2 (k0q , s21) and the rate of reaction with TMP (kT,TMP[TMP]0, where [TMP]0 is the initial concentration of TMP in units of M). The rate constant, k0q , was estimated as 5.0(62.5)3105 s21 (Foote 1976; Canonica et al. 2000), assuming the second order rate constant for the energy transfer from 3CDOM* to O2 is kq 5 2(61)3109 M21 s21 and the dissolved O2 concentration in the reaction solutions was 2.4–2.8 3 1024 M at approximately 208C and 101 kPa. The rate constant, kT,TMP, was estimated from experiments in which kobs,TMP was measured as a function of [TMP]0 (5–750 lM) in a subset of four samples selected to be representative of the range in wetland classes and watershed land covers (Al Housari et al. 2010; Golanoski et al. 2012; McCabe and Arnold 2017). To model these experiments, Eq. 3 was substituted into Eq. 2 and subsequently linearized (Eq. 4) and normalized by Ra (Eq. 5). k0q 1 ½TMP0 1 kobs;TMP Rf;T Rf;T kT;TMP k0q 1 1 5 ½TMP0 1 fTMP AQYT AQYT kT;TMP 1

5

kred kT;TMP ½3 CDOM ss ½DOC kred ½DOC1kox (10) k ox (11) kobs;TMP 5kT;TMP ½3 CDOM ss kred ½DOC1kox

kobs;TMP 5kT;TMP ½3 CDOM ss –

McCabe and Arnold (2017) showed that the inhibition factor of TMP photodegradation (IFTMP) defined originally by Canonica and Laubscher (2008) can be used to correct kobs,TMP for DOM-induced inhibition according to Eq. 12. kcobs;TMP 5

kT;TMP

kox

kred

TMP1• 1DOMred !TMP1DOM1•

(12)

IFTMP quantifies the ratio of the rate of triplet-induced oxidation of TMP with DOM present (i.e., accounting for reduction of TMP reaction intermediates back to the parent TMP structure) to the rate of triplet-induced oxidation without DOM present (Eq. 13).

(4)

IFTMP 5

(5)

kox kred ½DOC1kox

(13)

IFTMP was measured in a subset of water samples using the model 3CDOM* sensitizer, 4-carboxybenzophenone (CBP, 99%, Sigma-Aldrich). The subset of water samples was selected to give a range of watershed land covers and DOC concentrations. The inverse of IFTMP shows a linear dependence with [DOC] according to Eq. 14.

Where fTMP is the 3CDOM* quantum yield coefficient (kobs,TMP/Ra, L mol-photons21) (Canonica et al. 1995; Sharpless et al. 2014) and AQYT is the apparent quantum yield for 3 CDOM* formation (Rf,T/Ra, mol mol-photons21) measured with TMP. Eq. 5 is a linear relationship with y 5 1/fTMP, x5[TMP]0, slope 5 1/AQYT, and intercept 5 k0q /(AQYTkT,TMP). At [DOC]>5 3 1024 mol-C L21, the photodegradation of TMP may be inhibited by the reduction of the reaction intermediate, TMP1•, via reduced moieties in DOM (DOMred) (Eqs. 6 and 7). TMP13 CDOM !TMP1• !TMPox

kobs;TMP IFTMP

1 kred 5 ½DOC11 IFTMP kox

(14)

This linear model was used to estimate IFTMP in all collected water samples based on [DOC]. A full description of the experimental procedures and derivation of Eq. 13 is provided in the Supporting Information, Section 3.4. Values of kobs,TMP were normalized by IFTMP to correct for DOM-induced inhibition. With an estimate for kT,TMP, AQYT was estimated in all collected water samples using a single measurement of kobs,TMP at [TMP]054 lM corrected using IFTMP (Eq. 15).

(6) (7)

Where kox (s21) is the pseudo-first order rate constant for the oxidation of TMP1• by O2 and kred (M21 s21) is the second 6

McCabe and Arnold

AQYcT 5


0 kobs;TMP kq 1kT;TMP ½TMP0 IFTMP Ra kT;TMP

variables was 90%. The assumption of homoscedasticity (randomness of the error) for all fit models was assessed using residual plots and q-q plots. Standard errors (SE), p values, and x2 values (a measure of the effect size for a single predictor term) of the fit coefficients are reported. Separate models were fit that included either HDS or soil characteristics (% organic matter, % sand, % silt, and % clay) or both because data were missing for some sites (18 sites were missing HDSs and 3 sites were missing soil characteristics). Both HDS and soil characteristics were consistently removed during variable selection, however, so the separate models are not presented. Model criterion are compared between four models: (1) E2/E3 as the only predictor variable for AQYcT (model 1), (2) DOM composition parameters (SUVA254, E2/ E3, FI, HIX, b/a, and C/A) as predictors (model 2), (3) wetland and watershed characteristics defined in “Wetland classification and watershed characteristics” section as predictors (model 3), and (4) predictors from both models 2 and 3 included (model 4).

(15)

Values that have been corrected using IFTMP are denoted with a superscript “c”: kcobs;TMP , AQYcT , and Rcf;T . Solution preparation, high-pressure liquid chromatography detection methods, and additional experimental details are in the Supporting Information, Section 3. Statistical analyses All statistical analyses were performed in MATLAB. Kruskal-Wallis analysis of variance with post-hoc Dunn tests were used to compare surface water chemistry and descriptors of 3CDOM* photochemistry between the wetland classifications described Table 1. Spearman rank correlations coefficients (q) and associated p values were computed to qualitatively identify statistically significant bivariate trends between AQYT and climate conditions, watershed characteristics, wetland water chemistry, and DOM composition. The stepwiselm function in MATLAB was used to build multiple linear regression (MLR) models to identify variables that could reliably predict AQYT. MLR models have been used extensively in previous studies of DOM to predict organic carbon concentration and composition from landscape conditions (Sobek et al. 2007; Wilson and Xenopoulos 2009; Fasching et al. 2016). Predictor variables were added to and removed from the models according to the Bayesian Information Criterion (BIC). In variable selection, BIC balances the goals of model goodness-of-fit and parsimony by optimizing a log-likelihood function of the included parameters (i.e., the maximum likelihood is computed that given the fitted parameters the observations of the variables would occur) and penalizing the score for inclusion of additional predictor variables (Neath and Cavanaugh 2012; Aho et al. 2014). Predictor variables were added to the model if the change in BIC was < 0 and removed if the change in BIC was > 0.01. Models were specified to begin with only an intercept term and to add or remove terms including linear, quadratic, and interaction terms until BIC was minimized. Other fitting scenarios were considered, such as specifying models to begin with the full model including quadratic, interaction, and linear terms and removing terms until BIC was minimized or specifying models to only include linear terms. Given the relatively low ratio of samples to possible predictor variables (121/42 5 2.9), however, beginning with the full model typically resulted in rank deficient models (i.e., models that included too many predictors with too few observations to confidently estimate coefficients) and building models up from only an intercept model consistently resulted in minimum BIC. Strong bivariate correlations between predictor variables were identified prior to building the MLR models using a correlation matrix. A single variable was selected in cases when the correlation coefficient (r2) between predictor

Results and discussion Data Key to assessing the effect of DOM source and composition on AQYT is having enough disparity among wetlands that trends can be ascertained. Land cover, which is a driver of DOM source, was variable in the studied wetlands. The fraction of impervious cover ranged from 0% to 42%, wetland cover from 0% to 74%, forest cover from 1% to 88%, grassland cover from 0% to 48%, cropland cover from 0% to 76%, and open water from 0% to 27% (Supporting information Table S5, Section 4). Processing time of the DOM on the landscape is driven by slope (3–23%) and L/G (0.5–187 km; Supporting Information Table S5). NPP ranged from 0.398 kg-C m22 yr21 to 0.795 kg-C m22 yr21 with one outlier with a value of 0 (Supporting Information Table S6, Section 4). The DOC concentration in the wetlands ranged from 1.8 mol-C/L to 32.7 3 1024 mol-C/L (Supporting Information Tables S7 and S8 and Fig. S2, Section 5). The spectral parameters, E2/E3 (4.29–11.5), SUVA254 (1.15–8.98, decadic L mg-C21 m21), and b/a (0.382–0.818) also showed a wide range of values, indicating that variations in molecular structure (and thus light absorbing properties) of the DOM are present (Supporting Information Tables S9 and S10 and Figs. S3 and S4, Section 5). The units, mean, median, and range of these and other predictor variables used to develop the MLR models for AQYcT are summarized in Tables 1 (climate, watershed characteristics, and wetland characteristics variables) and 2 (water chemistry and DOM composition variables). These variables were selected because they were expected to relate to DOM source, composition, and concentration. Some variables (e.g., pH and SC) could affect intermolecular interactions of the DOM. Variables were ultimately included or eliminated 7

McCabe and Arnold


Table 1. Summary of climate conditions, watershed characteristics, and wetland classifications. Predictor variable

Unit

Mean

Median

Minimum

Maximum

Latitude (WGS1984)

DD

45.80837

46.36391

44.15559

47.71671

Longitude (WGS1984)

DD

293.69671

293.13634

296.15157

291.09639

Climate Conditions Mean Annual Temp. (MAT)*

8C

5.6

5.1

2.6

8.5

Mean Specific Year Temp. (MSYT)*

8C

6.1

6.2

1.5

9.2

Mean Specific Month Temp. (MSMT)* Specific Month Temp. (SMT)*

8C 8C

13.7 13.9

9.8 10.7

5.1 4.8

24.1 23.5

Mean Annual Precip. (MAP)†

cm

69.8

71.5

61.6

75.7

Mean Specific Year Precip. (MSYP)† Mean Specific Month Precip. (MSMP)†

cm cm

79.4 6.9

79.4 6.2

57.4 4.4

112.1 10.1

Specific Month Precip. (SMP)†

cm

6.8

5.5

0.7

20.9

Watershed Characteristics Area

km2

6.78

0.86

0.01

97.6

fraction

0.056

0.023

0.000

0.42

Wetland‡ Forest‡

fraction fraction

0.23 0.31

0.20 0.32

0.00 0.007

0.74 0.88

Grass‡

fraction

0.13

0.078

0.000

0.54

Cropland‡ Open Water‡

fraction fraction

0.15 0.054

0.001 0.034

0.000 0.000

0.76 0.27

Slope‡

fraction

0.097

0.084

0.025

0.23

Flow path/Slope (L/G) Elevation

km m

23.7 393

7.7 387

0.5 265

187 536

Impervious

‡

Human Disturbance Score (HDS), n 5 21

–

44.8

52.0

9.5

79.0

Net Primary Production (NPP)§ Soil Organic Matter,k n 5 36

kg-C m22 yr21 fraction

0.555 0.130

0.585 0.079

0.000 0.004

0.795 0.689

Sand,k n 5 36

fraction

0.441

0.432

0.111

0.870

Silt,k n 5 36 Clay,k n 5 36

fraction fraction

0.413 0.147

0.423 0.103

0.086 0.044

0.756 0.297

Wetland Classifications Predictor Variable Terrestrial Ecosystem¶

Number of wetlands per classification NTGP 5 8, NCIOS 5 3, NCIMBF 5 6, CMGP 5 1, LANHF 5 17, LPOB 5 1, BWSFW 5 3

Cowardin System#

L 5 6, P 5 33

Cowardin Class (Vegetation)# Cowardin Water Regime (Hydroperiod)#

FO 5 2, SS 5 7, EM 5 11, AB 5 2, UB 5 17 B 5 7, C 5 5, F 5 12, G 5 7, H 5 8

Hydrogeomorphic Class**

Terrene 5 27, Lentic 5 8, Lotic 5 4

Hydrologic Connectivity**

Vertical 5 16, Outflow 5 15, Throughflow 5 8

* ncdc.noaa.gov/ † dnr.state.mn.us/climate/ ‡ Rampi et al. (2016) § Running (2015) k websoilsurvey.nrcs.usda.gov/ ¶ Sayre et al. (2009) # fws.gov/wetlands/ ** dnr.state.mn.us/eco/wetlands/nwi_proj.html Abbreviations for wetland characteristics are (1) Terrestrial Ecosystem: NTGP 5 Northern Tallgrass Prairie, NCIOS 5 North Central Interior Oak Savannah, NCIMBF 5 North Central Interior Maple Basswood Forest, CMGP 5 Central Mixedgrass Prairie, LANHF 5 Laurentian-Acadian Northern Harwood Forest, LPOB 5 Laurentian Pine-Oak Barren, BWSFW 5 Boreal White Spruce Forest and Woodland; (2) Cowardin System: L 5 lacustrine and P 5 palustrine; (3) Cowardin Class: FO 5 forested, SS 5 scrub-shrub, EM 5 emergent, AB 5 aquatic bed, UB 5 unconsolidated bottom; and (4) Cowardin Water Regime: B 5 seasonally saturated, C 5 seasonally flooded, F 5 semi-permanently flooded, G 5 intermittently exposed, H 5 permanently flooded. Refer to Supporting Information Table S2 (Section 1) for definitions of each wetland characteristic.

8

McCabe and Arnold


Table 2. Summary of water chemistry and DOM composition of the wetland samples. Predictor variable

Unit

Mean

Median

Minimum

Maximum

pH

2log10 (M)

6.9

8.1

5.2

9.7

Specific Conductance

lS cm21

353

321

3

1133

Dissolved organic carbon ([DOC]) Dissolved inorganic carbon, ([DIC])

31024 mol-C L21 31024 mol-C L21

10.5 29.5

9.3 31.0

1.8 0.1

32.7 77.8

Absorbance at 440 nm (a440)

Decadic, m21

1.88

1.20

0.20

15.5

Absorbance at 350 nm (a350) Absorbance at 254 nm (a254)

Decadic, m21 Decadic, m21

9.23 41.5

6.10 33.6

1.40 6.30

71.5 245

Specific UV-absorbance (SUVA254)

Decadic, L mg-C21 m21

3.18

2.96

1.15

8.98

E2/E3 (a250/a365) Fluorescence Index (FI)

– –

6.60 1.56

6.42 1.56

4.29 1.41

11.5 1.81

Humification Index (HIX)

–

0.885

0.885

0.784

0.954

Biological Index (b/a) C/A

– –

0.636 0.514

0.656 0.514

0.382 0.240

0.818 0.632

inhibition of TMP photodegradation is minimal. This model should be used cautiously with DOM samples collected from ecologically distinct areas because we observe that IFTMP measured in prairie pothole wetlands do not follow the general trend (DOM samples from prairie pothole wetlands are likely more oxidized than the DOM from the other samples due to relatively long surface water residence times). Equation 16 was used to estimate IFTMP in all collected water samples. The average IFTMP value applied in this sample set was 0.78 (range 5 0.52–0.94), which resulted in a 33% increase on average in kobs,TMP (Supporting Information Fig. S6B, Section 6). To define kT,TMP the quantum yield coefficient, fTMP, was measured as a function of [TMP]0. As expected from Eq. 5, a c positive linear relationship is observed between (fTMP )21 and 3 [TMP]0, corresponding to increased CDOM* scavenging at elevated [TMP]0 (Supporting Information Fig. S7, Section 6). The average 6 95% confidence interval of kT,TMP was 2 6 1 3 109 M21 s21 (RSD 5 37%, n 5 19, Supporting Information Table S11, Section 6). This is the expected order of magnitude for the reaction between 3CDOM* and TMP (Canonica et al. 2000) and is in good agreement with previous estimates: 2.5–10 3 109 M21 s21 (Golanoski et al. 2012).

based on rational and model criteria (see “Multiple Linear Regression Models” section). The complete dataset is available online at the Data Repository for the University of Minnesota (https://conservancy.umn.edu/). TMP photooxidation TMP photodegradation experiments were carried out for 50 min in the solar simulator, with an average (6 standard deviation) half-life of 27(6 17) min. Repeated experiments and experiments with field duplicates showed that measured pseudo-first order rate constants between the experiments were < 4% different (Supporting Information Fig. S5, Section 6). To estimate AQYT from the photodegradation of TMP, both IFTMP and kT,TMP needed to be experimentally determined (see Eq. 15). To define IFTMP, experiments were performed using CBP as a model 3CDOM* sensitizer with and without DOM present. These experiments showed that the photodegradation rate of TMP was inhibited by DOM, and the magnitude of this inhibition was directly related to the concentration of DOC (Supporting Information Fig. S6A, Section 6). Experimental results using samples from this study were combined with data from stormflow samples (McCabe and Arnold 2017) and prairie pothole wetland surface waters to formulate a simple linear model to estimate the IFTMP as a function of [DOC]: 1 IFTMP

50:028ð60:010Þ½DOC11:01ð60:13Þ;

Trends in AQYcT Estimates of AQYcT in all wetlands ranged between 0.2 3 22 10 mol mol-photons21 and 14 3 1022 mol mol-photons21 with an average (6 standard deviation) of 4.2(62.4)31022 mol mol-photons21 (Supporting Information Tables S12 and S13 and Fig. S8, Section 7). This average AQYcT equates to f cTMP 5148(685) L mol-photons21, which is a typical range observed for CDOM (Sharpless et al. 2014; McCabe and Arnold 2016; McKay et al. 2017). Experimental precision in the measurement of AQYcT as assessed from the field duplicates was 16% (n 5 8). Trends between AQYcT and climate

(16)

where values in parentheses are 95% confidence intervals and [DOC] is in units of 3 1024 mol-C L21 (adjusted r250.760; p value < 0.001; Supporting Information Fig. S6A, Section 6). This model shows that the photodegradation of TMP is inhibited by 50% at [DOC]535 3 1024 mol-C L21 (42 mg-C L21) and that at [DOC]53 3 1024 mol-C L21 (3.5 mg-C L21) the IFTMP is not statistically different from 1, suggesting that below this concentration the DOM-induced 9

McCabe and Arnold


between wetland characteristics and water chemistry and DOM compositions are summarized in Supporting Information Table S17 (Section 8). The experimental precision of the DOM composition measurements as assessed from field duplicates was 27% (n 5 8; 95% of the data was 12%). Like AQYcT , E2/E3 (a proxy for DOM molecular weight) and b/a (a proxy for microbial or autochthonous sources of DOM) are moderately and negatively correlated with latitude (q 5 20.4758 and 20.5521, respectively, both p values < 0.001; Supporting Information Table S14, Section 8) and longitude (q 5 20.5915 and 20.5092, respectively, both p values < 0.001). Moving south to north and west to east, DOM increases in molecular weight and decreases in microbial or autochthonous character. SC and pH also show strong negative correlations with latitude (q 5 20.7328 and 20.7638, respectively, both p values < 0.001) and longitude (q 5 20.5639 and 20.6494, respectively, both p values < 0.001). DOC concentration, however, does not show strong spatial trends (latitude: q 5 20.1774 and longitude: q 5 20.2172). E2/E3 and b/a are strongly and negatively correlated with relative watershed forest land cover (q 5 20.6242 and 20.5480, respectively, both p values < 0.001). Both DOM parameters are also moderately to strongly correlated with watershed cropland and open water land cover (range in q 5 0.3462–0.5937, all p values < 0.001; Supporting Information Table S16, Section 8). SUVA254 (a proxy for DOM aromaticity and molecular weight) and HIX (a proxy for soilderived DOM) decrease with relative open water land cover (q 5 20.5016 and 20.4515, respectively, both p values < 0.001). There are not statistically significant trends between relative impervious cover and any proxy for DOM composition. HDS, however, does weakly correlate with FI and b/a (q 5 0.3955 and 0.3901, respectively, both p values < 0.001), suggesting that human disturbance on a wetland may promote autochthonous DOM production. This conclusion is supported by previous observations that DOM in streams draining cropland has relatively high microbial character compared to streams draining forests or wetlands (Wilson and Xenopoulos 2009). Like E2/E3 and b/a, pH and SC both negatively correlate with relative forest land cover and both are positively correlated with relative cropland land cover. DOC concentration only weakly correlates with relative cropland land cover (q 5 0.3348 and p value < 0.001). In general, DOM from terrene wetlands with seasonal hydroperiods and either forested or scrub–shrub vegetation have highly aromatic DOM with high average molecular weight (SUVA254 > 5 mg-C21 m21 and E2/E3 < 5) with high HIX (> 0.9; suggesting soil organic matter as the primary source of DOM to these wetlands) and relatively high DOC concentrations (10 3 1024 mol-C L21 to 20 3 1024 mol-C L21; Supporting Information Table S15, Section 8). These wetlands also tend to have relatively low pH (pH < 7; likely due to build-up of organic acids) and low SC (< 300 lS

parameters, watershed characteristics, and wetland characteristics are discussed in the following paragraphs. Wetland classifications and watershed characteristics There is a general spatial trend between AQYcT and both latitude and longitude, where AQYcT monotonically decreases south to north (448N to 488N; q 5 20.5247 and p value < 0.001) and west to east (–968W to 2918W; q 5 20.5308 and p value < 0.001), suggesting that regional climate and ecosystem variables influence AQYcT (Supporting Information Table S14, Section 8). AQYcT shows systematic trends with wetland characteristics (Fig. 2 and Supporting Information Table S15, Section 8). It is observed that wetlands from prairie ecosystems have relatively high estimates of AQYcT (average range 5 5.8 3 1022 mol mol-photons21 to 6.4 3 1022 mol mol-photons21, respectively) compared to wetlands from forested ecosystems (average range 5 0.23 3 1022 mol mol-photons21 to 4.7 3 1022 mol mol-photons21). Wetlands with the highest observed AQYcT (> 5 3 1022 mol mol-photons21) are characterizted as lacustrine or lentic systems with unconsolidated bottoms, permanently flooded hydroperiods, and throughflow surficial water regimes (Fig. 2). In contrast, wetlands with the lowest observed AQYcT (< 4 3 1022 mol mol-photons21) are characterized as palustrine or terrene systems with forested or scrub–shrub vegetation with seasonal hydroperiods and vertical surficial water regimes (Fig. 2). AQYcT is moderately and positively correlated with relative open water (q 5 0.3715 and p value < 0.001) and cropland (q 5 0.4764 and p value < 0.001) land cover. In contrast, it is negatively correlated with relative wetland (q 5 20.3133 and p value < 0.001) and forest (q 5 20.6299 and p value < 0.001) land cover (Supporting Information Table S16, Section 8). There are not strong rank correlations between AQYcT and HDS (an empirical measure of the level of human disturbance on a wetland) nor relative impervious land cover (Supporting Information Tables S16 and S17, Section 8). These trends show that autochthonous DOM efficiently produces 3 CDOM*, while allochthonous DOM produces 3CDOM* less efficiently, possibly because of competition with CT complex formation or screening of 3CDOM* precursors. These trends also suggest that DOM sourced from impervious surfaces, such as biolabile aliphatics (Chen et al. 2016) which are not expected to be photoactive, do not directly influence 3 CDOM* formation. Further, HDS is a poor indicator of 3 CDOM* formation efficiency likely because different anthropogenic influences, such as urban/commercial vs. cropland, can have different influences on DOM composition, yet both influences would be captured equivalently with HDS as high human disturbance. Water chemistry and DOM composition Spearman rank correlation coefficients between watershed characteristics and water chemistry and DOM composition are summarized in Supporting Information Tables S15 and S16 (Section 8). Kruskal-Wallis analysis of variance results 10

McCabe and Arnold


Fig. 2. Trends in AQYcT by wetland characteristics. Wetlands are grouped by (A) terrestrial ecosystem, (B) Cowardin system (size), (C) Cowardin class (vegetation), (D) Cowardin water regime (hydroperiod), (E) hydrogeomorphic class (landscape position), and (F) hydrologic connectivity. The p value represents the result of a Kruskal-Wallis analysis of variance test comparing the wetland groups. Group pairs listed below the p values are groups that are statistically different at the indicated significance level according to a post-hoc Dunn-Sidak test. Abbreviations for wetland characteristics are (A) NTGP 5 Northern Tallgrass Prairie, NCIOS 5 North Central Interior Oak Savannah, NCIMBF 5 North Central Interior Maple Basswood Forest, CMGP 5 Central Mixedgrass Prairie, LANHF 5 Laurentian-Acadian Northern Harwood Forest, LPOB 5 Laurentian Pine-Oak Barren, BWSFW 5 Boreal White Spruce Forest and Woodland; (B) L 5 lacustrine and P 5 palustrine; (C) FO 5 forested, SS 5 scrub-shrub, EM 5 emergent, AB 5 aquatic bed, UB 5 unconsolidated bottom; (D) B 5 seasonally saturated, C 5 seasonally flooded, F 5 semi-permanently flooded, G 5 intermittently exposed, H 5 permanently flooded. Refer to Supporting Information Table S2 (Section 1) for definitions of each wetland characteristic.

cm21). DOM from wetlands with semi-permanent and permanent hydroperiods with aquatic beds or unconsolidated bottoms tend to have DOM with relatively low aromaticity

and low average molecular (SUVA254 < 3 L mg-C21 m21 and E2/E3 > 6) with high b/a, indicating an autochthonous source of DOM. 11

McCabe and Arnold


AQYcT shows strong monotonic relationships with SUVA254 (q 5 20.6520, p value < 0.001), E2/E3 (q 5 0.8255, p value < 0.001), and b/a (q 5 0.814, p value < 0.001; Fig. 3). These trends between AQYcT and DOM bulk composition are consistent with previous reports (McCabe and Arnold 2016; McKay et al. 2017). AQYcT does not correlate with [DOC] (q 5 0.0488, p value 5 0.5949), but it does correlate with both pH and [DIC] (q 5 0.7258 and 0.5467, respectively, both p values < 0.001; Supporting Information Fig. S9, Section 8). This trend may indicate bias in the measurements of AQYcT (pH range 5 5.2–9.7; average 5 6.9; median 5 8.1). That is, a higher fraction of TMP (pKa510.9; Shiu et al. 1994) may exist in the phenolate form in samples with relatively high pH, resulting in relatively fast rates of TMP photosensitized-oxidation. Literature evidence suggests, however, that within the range of pH 7–9, a pH effect on the rate of TMP photooxidation is relatively minor (Faust and Hoigne 1987; Canonica et al. 1995). There is also evidence that pH has little effect on 3 CDOM* quenching (Wenk et al. 2013) and little impact on efficiencies of formation or steady-state concentrations 3 CDOM* (Maizel and Remucal 2017a). Because 80% of the samples fall within the range pH 7–9, this suggests that the observed trend between pH and AQYcT is more likely due to differences in other variables that covary with pH, for examples, strong correlations exist between pH and E2/E3 (q 5 0.760, p value < 0.001) and b/a (q 5 0.734, p value < 0.001). We also observe a moderately strong correlation between AQYcT and SC (0.6248, p value < 0.001). While ionic strength (empirically measured as SC) does influence the photo-physics of 3CDOM* (Parker et al. 2013; Glover and Rosario-Ortiz 2013), our sample set covers a fairly low SC gradient (range 5 3–1133 lS cm21; sea water is 503 higher in SC) and we expect that the trend between SC and AQYcT is also due to other variables that covary with SC. AQYT shows strong bivariate correlations with several landscape variables and variables describing DOM composition. These trends are consistent with the hypothesis that relatively low values of AQYcT are observed in wetlands with high inputs of reduced polyphenolic DOM, ultimately sourced from plants. This observation is consistent with CT complex formation between polyphenols and aromatic ketones, which are precursors of 3CDOM*. As DOM is processed during long water residence times, reduced polyphenols are lost, and other products are produced leading to more favorable conditions for 3CDOM* formation. A single independent variable, however, is insufficient to capture the observed variability in AQYcT . In the following section, the results of stepwise MLR as a technique to select those variables that show the highest correlation with AQYcT , and thus have the most predictive power.

Multiple linear regression models MLR models were built to identify the minimum number of key variables that could adequately explain the observed variability in AQYcT . Prior to building the MLR models, two pairs of variables were found to correlate with each other: (1) mean specific month temperature (MSMT) with specific month and year mean temperature (SMT) and (2) watershed area with L/G. MSMT was selected over SMT because of higher data availability, and L/G was selected over watershed area because L/G is directly related to average water residence times (McGuire et al. 2005). Concentration dependent water quality variables (pH, SC, [DOC], [DIC], and absorption coefficients) were excluded from the models on the basis that they are not expected to mechanistically influence AQYcT (see previous discussion in “Water chemistry and DOM composition” section). Terrestrial ecosystem classifications were excluded as possible predictor variables because these classifications are designated based on iso-climate, lithology, and land forms, each of which are considered in the models with other individual predictors. Coefficient estimates and selection criterion for models 1 (E2/E3 as the only linear predictor of AQYcT ) and 4 (DOM composition, climate, wetland classifications, and watershed characteristics as predictors) are reported in Table 3. Coefficient estimates and selection criterion for models 2 (DOM composition parameters as predictors) and 3 (climate, wetland classifications, and watershed characteristics as predictors) are reported in the Supporting Information Table S18 (Section 9). Plots of predicted values of AQYcT vs. observed values of AQYcT are shown in Fig. 4 and residual analyses are shown in Supporting Information Figs. S10 and S11 (Section 9). In general, model 4 was the best performing model, accounting for 97% of the variance in AQYcT . Model 2 accounted for slightly less of the total variance in AQYcT (90%), and models 1 and 3 accounted for comparably similar levels of variance (79% and 75%, respectively). Based on model BIC, model 4 also had the lowest value (–854.2), suggesting it is the simplest model with the highest explanatory power and highest consistency with the observed data. Model 2 had a lower BIC than model 1 (–777.1 vs. 2726.4), indicating that inclusion of additional DOM composition variables improved the consistency of the model with the data. Model 3 had the highest BIC (–691.5), indicating that while climate, wetland classifications, and watershed characteristics can account for a relatively high level of variance, this model is the least consistent with the observed data. In comparing models 1 to 2 and models 3 to 4, inclusion of DOM composition variables improved the explanatory power of the model and model consistency with the data (Fig. 4). Below we further examine the influence of predictor variables selected in model 4 and compare the MLR relationships between AQYcT and climate, wetland classifications, and watershed characteristics to those identified with Spearman rank correlations and Kruskal-Wallis analysis of variance. In 12

McCabe and Arnold


Fig. 3. Trends between AQYcT and DOM composition parameters. (A) SUVA254, (B) E2/E3, (C) b/a, (D) HIX, (E) FI, and (F) C/A. The Spearman rank correlation coefficient (q) is given with the associated p value.

13

Table 3. Model criterion and coefficient estimates for models 1 and 4. The response variable, AQYcT , has units of mol mol-photons–1 and is expressed as a fraction. Model 1 Model criterion Adj. r2

0.7923

BIC* RMSE†

2726.4 0.009955

SSE‡

0.01120

SSR§ n

0.04318 115

Coefficients Intercept E2/E3

Estimate

SEk

20.04801

4.301 3 1023

5.930 3 10220

24

1.398 3 10240

0.01334

p value

6.389 3 10

x2¶ – 0.79

Model 4 Model criterion Adj. r2

0.9663

BIC RMSE

2854.2 0.004485

SSE

0.001871

SSR n

0.06852 121

Coefficients

Estimate

SEk

p value

Intercept

20.8875

0.1210

8.206 3 10211

–

Wetland Cropland

0.6761 20.01210

0.1476 0.003344

1.436 3 1025 4.815 3 1024

0.005710 0.003456

x2¶

Open Water

0.02797

0.01267

0.02977

0.001106

Slope L/G

20.2835 0.001270

0.06792 3.374 3 1024

6.739 3 1025 2.921 3 1024

0.004692 0.003764

25.697 3 1024

1.071 3 1024

7.165 3 1027

0.007798

NPP Terrene

1.581 0.01985

0.1965 0.003652

2.723 3 10212 4.372 3 1027

0.01821 0.01590

Lentic

0.02438

0.003255

3.919 3 10211

20.01218 20.01987

0.002292 0.004195

7.334 3 1027 7.797 3 1026

0.007781 0.006123

0.9024

0.1045

1.596 3 10213

0.02101

b/a Wetland:NPP

0.2939 20.2801

0.06319 0.06768

1.091 3 1025 7.673 3 1025

0.005895 0.004608

Wetland:E2/E3

20.01266

0.004231

0.003536

0.002274

Wetland:HIX Slope:E2/E3

20.4870 0.06250

0.1372 0.01143

6.095 3 1024 3.790 3 1027

0.003312 0.008261

L/G:Elevation

27.428 3 1027

3.215 3 1027

0.02306

0.001240

L/G:NPP NPP:HIX

20.001552 21.131

4.953 3 1024 0.1829

0.002314 1.667 3 1028

0.002519 0.01063

NPP:b/a

20.7362

0.08309

5.290 3 10214

0.02214

0.03562 6.631 3 1027

0.005318 1.344 3 1027

1.594 3 1029 3.527 3 1026

0.01253 0.006669

20.1107

0.02685

8.173 3 1025

0.004568

3.667 3 1025

0.005088

Elevation

SUVA254 E2/E3 HIX

E2/E3:b/a Elevation2 NPP2 SUVA2254

24

9.527 3 10

24

2.197 3 10

The notation “A:B” corresponds to an interaction term such that the predictor variables, A and B, are multiplied together. * BIC 5 Bayesian Information Criterion. † RMSE 5 Root mean square error. ‡ SSE 5 Sum of squares of the error (unexplained variance). § SSR 5 Sum of squares of the regression (explained variance). k SE 5 Standard error of the estimated coefficient. ¶ x25the relative variance explained by an individual predictor term, computed as (SSeffect–dfeffectMSE)/(MSE 1 SST), where SSeffect5sum of squares of the effect, dfeffect5degrees of freedom of the effect, MSE 5 mean square error, and SST 5 total sum of squares (SSE 1 SSR).

McCabe and Arnold


Fig. 4. Plots of model fits of AQYcT vs. the observed values of AQYcT . Error bars indicate the 95% confidence interval for the predicted value of AQYcT . The dashed line is a 1 : 1 line.

model 4, 12 predictor variables of AQYcT were selected with 26 coefficient estimates. The influences of each predictor variable included in model 4 on AQYcT are shown graphically in

Fig. 5, and these trends are discussed in the following sections in the context of DOM composition and the formation of CT complexes. 15

McCabe and Arnold


cover, assuming lignin polyphenols are photodegraded (with minor contribution from microbial degradation) during long surface water residence times (Benner and Kaiser 2011), thus lowering the rate of light absorption and increasing AQYcT . Photodegradation, however, cannot completely account for the positive relationship between AQYcT and open water land cover because 3CDOM*-precursors and CDOM are photodegraded at comparable rates, which translates to effectively no change in AQYcT (Sharpless et al. 2014; McCabe and Arnold 2017). Logically, either moieties that inhibit 3 CDOM* formation must be preferentially removed and/or 3 CDOM*-precursors must be preferentially formed. It is possible that microbial action (through the formation of refractory metabolites; Ogawa 2001) may account for the observed increase in AQYcT with relative open water land cover, but other processes (such as coagulation and sorption; Aufdenkampe et al. 2007) could also be operative and further study is required to clarify the dominant process. The relationship between NPP and AQYcT includes several interaction terms and a quadratic term:

Climate, watershed characteristics, and wetland classifications Of the predictor variables included in model 4, eight variables describe wetland classifications and watershed characteristics: hydrogeomorphic class, relative wetland, cropland, and open water land cover, slope, L/G, elevation, and watershed NPP (Table 3). No climate related variables were selected likely because the climate gradient across the study areas was not strong enough to observe possible influences. The coefficient associated with relative wetland land cover includes linear and interaction terms: coeff wetland 5 0:6761–0:2801NPP–0:01266E2=E3–0:4870HIX (17) At average conditions of NPP, E2/E3, and HIX (NPPave5 0.555 kg-C m22 yr21, E2/E3ave56.60, and HIXave50.885), AQYcT is positively related to relative wetland land cover (Fig. 5A), however, the magnitude of this relationship may become negative with increasing watershed NPP, E2/E3, or HIX. The contrasting influence of wetland land cover on AQYcT suggests that differing wetland classes throughout a watershed (Fig. 2) and/or the location of wetlands in a watershed exert different influences on AQYcT . The coefficient associated with relative cropland is linear: coeff cropland 5 20:01210

coeff NPP 5 1:58120:2801Wetland21:552 3 1023 L=G 21:131HIX20:7362b=a20:1107NPP

On average, AQYcT decreases as NPP over the watershed increases (Fig. 5G). The magnitude of this relationship becomes more negative as any of the interaction terms (wetland land cover, L/G, HIX, or b/a) increase. Because NPP effectively quantifies net inputs of fresh vascular plantderived DOM, it is consistent with our hypothesis to find that AQYcT is negatively related to NPP. The influences of the variables describing landscape position (slope, L/G, elevation, and HGM class) on AQYcT reflect balances between inputs of vascular plant-derived DOM and DOM aging. The coefficient relating average watershed slope to AQYcT is linear, with dependency on DOM composition:

(18)

This coefficient indicates that per a 1% increase in watershed cropland, AQYcT decreases by 0.01 3 1022 mol mol-photons21 (Fig. 5B), which is in contrast with the positive Spearman rank correlation between relative cropland and AQYcT . Based on Wilson and Xenopoulos (2009), which reported greater autochthonous character of DOM in streams draining cropland, and the positive rank correlations between b/a and both relative cropland land cover and AQYcT , it was expected that cropland would positively influence AQYcT . It is plausible that a negative relationship between cropland and AQYcT could occur given that annual crops contribute lignin phenols to drainage water (Hernes et al. 2013; Eckard et al. 2017), which lead to greater light absorption and lower triplet formation efficiency. It is further plausible that the positive rank correlation between cropland land cover and AQYcT is caused by a higher occurrence of lacustrine or lentic systems with > 50% cropland land cover. The coefficient relating relative open water land cover to AQYcT is positive and linear: coeff open

water

5 0:02797

(20)

coeff slope 5 20:2835 1 0:06250E2=E3

(21)

With an average E2/E3, the AQYcT increases by 0.13 3 1022 mol mol-photons21 for a 1% increase in watershed slope (Fig. 5D). This relationship remains positive for most of the range of observed E2/E3 values, only becoming negative at E2/E3 4.5. Wetlands with high slopes are expected to have lower inputs of lignin polyphenols because organic soil layers are likely shallower than in flat topographies (Sobek et al. 2007). The coefficient relating L/G to AQYcT is linear:

(19)

coeff L=G 51:270 3 1023 27:428 3 1027 Elevation

For a 1% increase in relative open water area, AQYcT increases by 0.03 3 1022 mol mol-photons21, which is consistent with the Spearman rank correlation between relative open water and AQYcT (Fig. 5C). It is consistent with our hypothesis that AQYcT is positively related to relative open water land

21:552 3 1023 NPP

(22)

On average (Elevationave5393 m and NPPave50.555 kg-C m22 yr21), AQYcT increases by 0.01 3 1022 mol mol-photons21 for a 16

McCabe and Arnold


Fig. 5. The influence of each predictor in model 4 on AQYcT . The predicted trend from model 4 is shown as a solid black line, and 95% confidence intervals for the predicted values are shown as dashed lines. Where interaction terms exist, average predictor values are used to illustrate the “average” trend across range of the plotted predictor variable.

17

McCabe and Arnold


decrease with increases in watershed wetland land cover, but it would not become negative across the range of observed wetland land covers. These observations are consistent with our hypothesis that high molecular weight DOM (high SUVA254 and low E2/E3) is more likely to form CT complexes between phenol donors and ketone acceptor species due to proximity (Sharpless and Blough 2014) and thus lower AQYcT . The coefficient relating HIX to AQYcT includes both linear and interaction terms:

1-km increase in L/G (Fig. 5E). As either elevation and NPP increase, the influence of L/G on AQYT may become negative. These interactions suggest that increasing watershed residence times leads to increases AQYcT , but inputs of fresh plant-derived organic matter negates this effect. The modeled relationship between elevation and AQYcT is quadratic: coeff elevation 5 25:697 3 1024 27:428 3 1027 L=G 16:631 3 1027 Elevation

(23)

coeff HIX 5 0:902420:4870Wetland21:131NPP

With average L/G (L/Gave523.7 km), AQYcT decreases with elevation, indicating that wetlands at higher elevations tend to have relatively low AQYcT . It is probable that wetlands with relatively high elevation are headwater wetlands with short water residence times have DOM inputs dominated by plant-derived lignin polyphenols. Thus, leading to the favorable conditions for the formation of CT complexes. Finally, the influence of HGM class on AQYcT tends to follow the relative surface water residence times of these systems: lotic < terrene < lentic. AQYcT in wetlands classified as terrene and lentic will be 2.0 3 1022 mol mol-photons21 and 2.4 3 1022 mol mol-photons21 greater relative to lotic wetlands, respectively. DOM aging during relatively long surface water residence times, including the loss of reduced polyphenols and possibly the production microbial exometabolites, leads to less favorable conditions for CT complex formation and explains the positive relationships between AQYcT and HGM class.

On average (Wetlandave50.23 and NPPave50.555 kg-C m22 yr21), AQYcT increases by 2.1 3 1022 mol mol-photons21 for a 0.1-unit increase in HIX (Fig. 5K). As the fraction of wetlands and the average NPP of the watershed increase, this relationship becomes weakened and may become negative. HIX is a qualitative proxy for soil-derived organic matter (SOM), which is complex mixture of plant and microbial exudates and various metabolites (Schmidt et al. 2011; Lehmann and Kleber 2015). In systems with relatively low inputs of fresh plant-derived organic matter (quantified via NPP and relative wetland land cover), it is probable that SOM has low concentrations of reduced lignin phenols relative to microbially oxidized metabolites, suggesting that DOM from degraded SOM confers relatively high estimates of AQYcT . The coefficient relating b/a to AQYcT also includes both linear and interaction terms:

DOM composition Four of the predictor variables selected in model 4 describe DOM source and composition: SUVA254, E2/E3, HIX, and b/a. The terms relating SUVA254 to AQYcT are linear and quadratic with no interaction terms: coeff SUVA 5 20:01218 1 9:527 3 1024 SUVA254

(26)

coeff b=a 5 0:293920:7362NPP10:03562E2=E3

(27)

On average, AQYcT increases by 1.2 3 1022 mol mol-photons21 for a 0.1-unit increase in b/a (Fig. 5L), which qualitatively agrees with the positive Spearman rank correlation between AQYcT and b/a. Because of the interactions between b/a, NPP, and E2/E3, the influence of b/a on AQYcT may be strengthened or weakened depending on the magnitude of changes in NPP and E2/E3. The contrasting influences of b/a on AQYcT likely arises because it qualitatively describes both autochthonous and fresh plant-derived DOM (Hansen et al. 2016). Under high watershed NPP, it is probable that b/a qualitatively reflects inputs of vascular plant-derived DOM, whereas under high E2/E3, typically observed with long surface water residence times, b/a describes production of autochthonous DOM. Each predictor variable in model 4 can be designated into one of three groups depending on their influence on AQYcT : (1) always or usually positive (relative open water land cover, average slope, HGM class, and E2/E3), (2) always or usually negative (relative cropland land cover, average elevation, NPP, and SUVA254), and (3) positive on average, but may be negative (relative wetland land cover, L/G, HIX, and b/a). In general, we observe negative trends between AQYcT and

(24)

Across the range of SUVA254 values observed, there is a negative relationship between SUVA254 and AQYcT (Fig. 5I), which qualitatively corresponds to the Spearman rank correlation between SUVA254 and AQYcT (q 5 20.6520, p value < 0.001). The coefficient relating E2/E3 to AQYcT includes both linear and interaction terms: coeff E2=E3 5 20:0198720:01266Wetland10:06250Slope 10:03562b=a (25) On average, AQYcT increases by 0.59 3 1022 mol mol-photons21 for a 1-unit increase in E2/E3 (Fig. 5J), which is comparable to the slope of model 1 (1.3 3 1022 mol molphotons21 increase in AQYcT for a 1-unit increase in E2/E3). The magnitude of this relationship may become more positive as either watershed slope or b/a increase, or it may 18

McCabe and Arnold


allochthonous DOM and the production of autochthonous DOM during long surface water residence times in controlling AQYcT . These relationships deserve further investigation to evaluate if they are consistent in other aquatic systems with varying spatial scales.

variables that describe DOM aromaticity and high inputs of plant-derived organic matter (watershed NPP and SUVA254) and positive trends between AQYcT and variables that describe aging DOM and surface water residence times (HIX, E2/E3, open water land cover, and L/G). These observations suggest that (1) vascular plant-derived organic matter inhibits AQYcT and (2) either the degradation of plant-derived organic matter removes the inhibition or microbial metabolism and/or production of autochthonous DOM forms highly efficient 3CDOM* precursors. It is consistent with our hypothesis that relatively low values of AQYcT are observed with high inputs of fresh vascular plant-derived DOM because we expect lignin polyphenols to form CT complexes with 3CDOM* precursors. It is further consistent that AQYcT is relatively high in samples with aged DOM (degraded from its biological origin), because the content of reduced lignin phenols is expected to decrease with DOM aging. A recent report using solvent and temperature adjustments to perturb DOM absorbance and fluorescence suggests that CT complexes do not form in DOM (McKay et al., 2018). While the hypothesis of CT complex formation has well-founded precedent (Sharpless and Blough 2014), mechanisms other than formation of CT complexes may contribute to the observed trends in AQYcT , including a simple screening mechanism of 3CDOM* precursors by nontriplet forming chromophores, a disproportionate increase of nontriplet forming precursors with increasing broadband light absorption, or a complete oxidation reaction between 3CDOM* and reduced polyphenols (McNally et al. 2005; McKay et al. 2017). These mechanisms may be possible alternatives to the CT hypothesis, or they may be active in combination CT complex formation. That said, the trends observed here clearly demonstrate that fresh allochthonous DOM suppresses the formation efficiency of 3CDOM*, and this efficiency increases as DOM is processed within wetlands and their corresponding watersheds.

References Aho, K., D. Derryberry, and T. Peterson. 2014. Model selection for ecologists: The worldviews of AIC and BIC. Ecology 95: 631–636. doi:10.1890/13-1452.1 Al Housari, F., D. Vione, S. Chiron, and S. Barbati. 2010. Reactive photoinduced species in estuarine waters. Characterization of hydroxyl radical, singlet oxygen and dissolved organic matter triplet state in natural oxidation processes. Photochem. Photobiol. Sci. 9: 78–86. doi: 10.1039/b9pp00030e Aufdenkampe, A. K., E. Mayorga, J. I. Hedges, C. Llerena, P. D. Quay, J. Gudeman, A. V. Krusche, and J. E. Richey. 2007. Organic matter in the Peruvian headwaters of the Amazon: Compositional evolution from the Andes to the lowland Amazon mainstem. Org. Geochem. 38: 337–364. doi:10.1016/j.orggeochem.2006.06.003 Baluha, D. R., N. V. Blough, and R. Del Vecchio. 2013. Selective mass labeling for linking the optical properties of chromophoric dissolved organic matter to structure and composition via ultrahigh resolution electrospray ionization mass spectrometry. Environ. Sci. Technol. 47: 9891– 9897. doi:10.1021/es402400j Benner, R., and K. Kaiser. 2011. Biological and photochemical transformations of amino acids and lignin phenols in riverine dissolved organic matter. Biogeochemistry 102: 209–222. doi:10.1007/s10533-010-9435-4 Boreen, A. L., B. L. Edhlund, J. B. Cotner, and K. McNeill. 2008. Indirect photodegradation of dissolved free amino acids: The contribution of singlet oxygen and the differential reactivity of DOM from various sources. Environ. Sci. Technol. 42: 5492–5498. doi:10.1021/es800185d Canonica, S., U. Jans, K. Stemmler, and J. Hoigne. 1995. Transformation kinetics of phenols in water: Photosensitization by dissolved natural organic material and aromatic ketones. Environ. Sci. Technol. 29: 1822–1831. doi: 10.1021/es00007a020 Canonica, S., B. Hellrung, and J. Wirz. 2000. Oxidation of phenols by triplet aromatic ketones in aqueous solution. J. Phys. Chem. A 104: 1226–1232. doi:10.1021/jp9930550 Canonica, S., and M. Freiburghaus. 2001. Electron-rich phenols for probing the photochemical reactivity of freshwaters. Environ. Sci. Technol. 35: 690–695. doi:10.1021/ es0011360 Canonica, S., and H.-U. Laubscher. 2008. Inhibitory effect of dissolved organic matter on triplet-induced oxidation of aquatic contaminants. Photochem. Photobiol. Sci. 7: 547– 551. doi:10.1039/b719982a

Conclusions Stepwise MLR was used to identify environmental-level and DOM composition variables that reliably predict AQYcT in temperate wetlands. In instances where several possible predictor variables independently covary, MLR allows for the simulated experiment where possible co-variables may be artificially controlled to understand the underlying relationships between the response variable and individual predictor variables. The presented models have applications in the development of large-scale mechanistic models of DOM photodegradation and pollutant fate. We identified key watershed variables that show strong correlations with AQYcT , such as elevation, slope, L/G, and relative open water and wetland land cover. These predictor variables highlight the importance of inputs of fresh vascular plant-derived DOM (allochthonous DOM) as well as the processing of 19

McCabe and Arnold


Catalan, N., R. Marce, D. N. Kothawala, and L. J. Tranvik. 2016. Organic carbon decomposition rates controlled by water retention time across inland waters. Nat. Geosci. 9: 501–507. doi:10.1038/ngeo2720 Chen, M., N. Maie, K. Parish, and R. Jaffe. 2013. Spatial and temporal variability of dissolved organic matter quantity and composition in an oligotrophic subtropical coastal wetland. Biogeochemistry 115: 167–183. doi:10.1007/ s10533-013-9826-4 Chen, H., Z. Liao, X. Gu, J. Xie, H. Li, and J. Zhang. 2016. Anthropogenic influences of paved runoff and sanitary sewage on the DOM quality of wet weather overflows: An EEM-PARAFAC assessment. Environ. Sci. Technol. 51: 1157–1167. doi:10.1021/acs.est.6b03727 Chin, Y.-P., G. Aiken, and E. O’Loughlin. 1994. Molecular weight, polydispersity, and spectroscopic properties of aquatic humic substances. Environ. Sci. Technol. 28: 1853–1858. doi:10.1021/es00060a015 Cooper, W. J., R. G. Zika, R. G. Petasne, and A. M. Fischer. 1988. Sunlight-induced photochemistry of humic substances in natural waters: Major reactive species, p. 333–362. In Aquatic humic substances. American Chemical Society. Cory, R. M., M. P. Miller, D. M. McKnight, J. J. Guerard, and P. L. Miller. 2010. Effect of instrument-specific response on the analysis of fulvic acid fluorescence spectra. Limnol. Oceanogr. Methods 8: 67–78. doi:10.4319/lom.2010.8.0067 Cowardin, L. M., V. Carter, F. C. Golet, and E. T. LaRoe. 1979. Classification of wetlands and deepwater habitats of the United States. United States Fish and Wildlife Service. Dalrymple, R. M., A. K. Carfagno, and C. M. Sharpless. 2010. Correlations between dissolved organic matter optical properties and quantum yields of singlet oxygen and hydrogen peroxide. Environ. Sci. Technol. 44: 5824–5829. doi:10.1021/es101005u De Haan, H., and T. De Boer. 1987. Applicability of light absorbance and fluorescence as measures of concentration and molecular size of dissolved organic carbon in humic Lake Tjeukemeer. Water Res. 21: 731–734. doi:10.1016/ 0043-1354(87)90086-8 Del Vecchio, R., and N. V. Blough. 2002. Photobleaching of chromophoric dissolved organic matter in natural waters: Kinetics and modeling. Mar. Chem. 78: 231–253. doi: 10.1016/S0304-4203(02)00036-1 Del Vecchio, R., and N. V. Blough. 2004. On the origin of the optical properties of humic substances. Environ. Sci. Technol. 38: 3885–3891. doi:10.1021/es049912h Eckard, R. S., B. A. Pellerin, B. A. Bergamaschi, P. A. M. Bachand, S. M. Bachand, R. G. M. Spencer, and P. J. Hernes. 2017. Dissolved organic matter compositional change and biolability during two storm runoff events in a small agricultural watershed. J. Geophys. Res. Biogeosci. 122: 2634–2650. doi:10.1002/2017JG003935 ESRI. 2016. Environmental Systems Research Institute: ArcMap 10.4.1.

Fasching, C., A. J. Ulseth, J. Schelker, G. Steniczka, and T. J. Battin. 2016. Hydrology controls dissolved organic matter export and composition in an Alpine stream and its hyporheic zone. Limnol. Oceanogr. 61: 558–571. doi:10.1002/ lno.10232 Faust, B. C., and J. Hoigne. 1987. Sensitized photooxidation of phenols by fulvic acid and in natural waters. Environ. Sci. Technol. 21: 957–964. doi:10.1021/es50001a008 FGDC. 2013. Classification of wetlands and deepwater habitats of the United States, Second Edition (FGDC-STD-004– 2013). Wetlands Subcommittee, Federal Geographic Data Committee, and U.S. Fish and Wildlife Service. Foote, C. S. and W.A. Pryor [ed.]. 1976. Free radicals in biology, Volume II. Academic Press. Genet, J. A., and M. Bourdaghs. 2006. Development and validation of indices of biological integrity (IBI) for depressional wetlands in the temperate prairies. Minnesota Pollution Control Agency. Gilmore, A., R. Hurteaux, S. FitzGerald, and A. Knowles. 2012. Moving towards a technical specification for fluorescence excitation-emission mapping and absorbance analysis of colored dissolved organic matter. WIT Trans. Ecol. Environ. 160: 295–306. doi:10.2495/DN120261 Glover, C. M., and F. L. Rosario-Ortiz. 2013. Impact of halides on the photoproduction of reactive intermediates from organic matter. Environ. Sci. Technol. 47: 13949– 13956. doi:10.1021/es4026886 Golanoski, K. S., S. Fang, R. Del Vecchio, and N. V. Blough. 2012. Investigating the Mechanism of Phenol Photooxidation by Humic Substances. Environ. Sci. Technol. 46: 3912–3920. doi:10.1021/es300142y Gonsior, M., B. M. Peake, W. T. Cooper, D. Podgorski, J. D’Andrilli, and W. J. Cooper. 2009. Photochemically induced changes in dissolved organic matter identified by ultrahigh resolution Fourier transform ion cyclotron resonance mass spectrometry. Environ. Sci. Technol. 43: 698– 703. doi:10.1021/es8022804 Grebel, J. E., J. J. Pignatello, and W. A. Mitch. 2011. Sorbic acid as a quantitative probe for the formation, scavenging and steady-state concentrations of the triplet-excited state of organic compounds. Water Res. 45: 6535–6544. doi: 10.1016/j.watres.2011.09.048 Hansen, A. M., T. E. C. Kraus, B. A. Pellerin, J. A. Fleck, B. D. Downing, and B. A. Bergamaschi. 2016. Optical properties of dissolved organic matter (DOM): Effects of biological and photolytic degradation. Limnol. Oceanogr. 61: 1015– 1032. doi:10.1002/lno.10270 Hernes, P. J., R. G. M. Spencer, R. Y. Dyda, B. A. Pellerin, P. A. M. Bachand, and B. A. Bergamaschi. 2013. DOM composition in an agricultural watershed: Assessing patterns and variability in the context of spatial scales. Geochim. Cosmochim. Acta 121: 599–610. doi:10.1016/j.gca.2013.07.039 Kellerman, A. M., T. Dittmar, D. N. Kothawala, and L. J. Tranvik. 2014. Chemodiversity of dissolved organic 20

McCabe and Arnold


topography on catchment-scale water residence time. Water Resour. Res. 41: W04002. doi:10.1029/2004WR003657 McKay, G., K. D. Couch, S. P. Mezyk, and F. L. Rosario-Ortiz. 2016. Investigation of the coupled effects of molecular weight and charge-transfer interactions on the optical and photochemical properties of dissolved organic matter. Environ. Sci. Technol. 50: 8093–8102. doi:10.1021/ acs.est.6b02109 McKay, G., W. Huang, C. Romera-Castillo, J. E. Crouch, F. L. Rosario-Ortiz, and R. Jaffe. 2017. Predicting reactive intermediate quantum yields from dissolved organic matter photolysis using optical properties and antioxidant capacity. Environ. Sci. Technol. 51: 5404–5413. doi:10.1021/ acs.est.6b06372 McKay, G., J. A. Korak, P. R. Erickson, D. E. Latch, K. McNeill, and F. L. Rosario-Ortiz. 2018. The case against charge transfer interaction in dissolved organic matter photophysics. Environ. Sci. Technol. 52: 406–414. doi: 10.1021/acs.est.7b03589 McNally, A. M., E. C. Moody, and K. McNeill. 2005. Kinetics and mechanism of the sensitized photodegradation of lignin model compounds. Photochem. Photobiol. Sci. 4: 268–274. doi:10.1039/B416956E McNeill, K., and S. Canonica. 2016. Triplet state dissolved organic matter in aquatic photochemistry: Reaction mechanisms, substrate scope, and photophysical properties. Environ. Sci. Process. Impacts. 18: 1381–1399. doi: 10.1039/C6EM00408C Mopper, K., D. J. Kieber, and A. Stubbins. 2015. Chapter 8 Marine photochemistry of organic matter: Processes and impacts, p. 389–450. In D.A. Hansell and C.A. Carlson [eds.], Biogeochemistry of marine dissolved organic matter. Academic Press. MPCA. 2014. Development of a fish-based index of biological integrity for Minnesota’s Rivers and streams (wqbsm2–03). Minnesota Pollution Control Agency. MPCA. 2016. Guidance for calculating the MPCA’s human disturbance score (wq-bsm3–10). Minnesota Pollution Control Agency. € nster, U., and R. J. Chro st. 1990. Origin, composition, Mu and microbial utilization of dissolved organic matter, p. st [eds.], Aquatic micro8–46. In J. Overbeck and R.J. Chro bial ecology: Biochemical and molecular approaches. Springer. Murphy, K. R., C. A. Stedmon, D. Graeber, and R. Bro. 2013. Fluorescence spectroscopy and multi-way techniques. PARAFAC. Anal. Methods 5: 6557. doi:10.1039/c3ay41160e NatureServe. 2009. International Ecological Classification Standard: Terrestrial Ecological Classifications. NatureServe Central Databases. Neath, A. A., and J. E. Cavanaugh. 2012. The Bayesian information criterion: Background, derivation, and applications. Wiley Interdiscip. Rev. Comput. Stat. 4: 199–203. doi:10.1002/wics.199

matter in lakes driven by climate and hydrology. Nat. Commun. 5: 1–8. doi:10.1038/ncomms4804 Kohn, T., and K. L. Nelson. 2007. Sunlight-mediated inactivation of MS2 coliphage via exogenous singlet oxygen produced by sensitizers in natural waters. Environ. Sci. Technol. 41: 192–197. doi:10.1021/es061716i Kujawinski, E. B., R. Del Vecchio, N. V. Blough, G. C. Klein, and A. G. Marshall. 2004. Probing molecular-level transformations of dissolved organic matter: Insights on photochemical degradation and protozoan modification of DOM from electrospray ionization Fourier transform ion cyclotron resonance mass spectrometry. Mar. Chem. 92: 23–37. doi:10.1016/j.marchem.2004.06.038 Larson, J. H., P. C. Frost, M. A. Xenopoulos, C. J. Williams, A. M. Morales-Williams, J. M. Vallazza, J. C. Nelson, and W. B. Richardson. 2014. Relationships between land cover and dissolved organic matter change along the river to lake transition. Ecosystems. 17: 1413–1425. doi:10.1007/ s10021-014-9804-2 Laszakovits, J. R., S. M. Berg, B. G. Anderson, J. E. O’Brien, K. H. Wammer, and C. M. Sharpless. 2017. p-nitroanisole/ pyridine and p-nitroacetophenone/pyridine actinometers revisited: Quantum yield in comparison to ferrioxalate. Environ. Sci. Technol. Lett. 4: 11–14. doi:10.1021/acs. estlett.6b00422 Lehmann, J., and M. Kleber. 2015. The contentious nature of soil organic matter. Nature 528: 60–68. doi:10.1038/ nature16069 Leifer, A. 1988. The kinetics of environmental aquatic photochemistry: Theory and practice. American Chemical Society Maizel, A. C., and C. K. Remucal. 2017a. The effect of probe choice and solution conditions on the apparent photoreactivity of dissolved organic matter. Environ. Sci. Process. Impacts 19: 1040–1050. doi:10.1039/C7EM00235A Maizel, A. C., and C. K. Remucal. 2017b. Molecular composition and photochemical reactivity of size-fractionated dissolved organic matter. Environ. Sci. Technol. 51: 2113– 2123. doi:10.1021/acs.est.6b05140 MathWorks. 2014. MATLAB v. R2014a. €ike. 2005. Export of Mattsson, T., P. Kortelainen, and A. Ra DOM from boreal catchments: Impacts of land use cover and climate. Biogeochemistry 76: 373–394. doi:10.1007/ s10533-005-6897-x McCabe, A. J., and W. A. Arnold. 2016. Seasonal and spatial variabilities in the water chemistry of prairie pothole wetlands influence the photoproduction of reactive intermediates. Chemosphere 155: 640–647. doi:10.1016/ j.chemosphere.2016.04.078 McCabe, A. J., and W. A. Arnold. 2017. Reactivity of triplet excited states of dissolved natural organic matter in stormflow from mixed-use watersheds. Environ. Sci. Technol. 51: 9718–9728. doi:10.1021/acs.est.7b01914 McGuire, K. J., J. J. McDonnell, M. Weiler, C. Kendall, B. L. McGlynn, J. M. Welker, and J. Seibert. 2005. The role of 21

McCabe and Arnold


correlation of environmentally relevant properties and fate in an evaluative environment. Chemosphere 29: 1155–1224. doi:10.1016/0045-6535(94)90252-6 Sobek, S., L. J. Tranvik, Y. T. Prairie, P. Kortelainen, and J. J. Cole. 2007. Patterns and regulation of dissolved organic carbon: An analysis of 7,500 widely distributed lakes. Limnol. Oceanogr. 52: 1208–1219. doi:10.4319/lo.2007.52.3. 1208 Spencer, R. G. M., P. J. Hernes, R. Ruf, A. Baker, R. Y. Dyda, A. Stubbins, and J. Six. 2010. Temporal controls on dissolved organic matter and lignin biogeochemistry in a pristine tropical river, Democratic Republic of Congo. J. Geophys. Res. 115: G03013. doi:10.1029/2009JG001180 Stenson, A. C., A. G. Marshall, and W. T. Cooper. 2003. Exact masses and chemical formulas of individual Suwannee River Fulvic Acids from ultrahigh resolution electrospray ionization fourier transform ion cyclotron resonance mass spectra. Anal. Chem. 75: 1275–1284. doi:10.1021/ac026106p Stubbins, A., and others. 2010. Illuminated darkness: Molecular signatures of Congo River dissolved organic matter and its photochemical alteration as revealed by ultrahigh precision mass spectrometry. Limnol. Oceanogr. 55: 1467–1477. doi:10.4319/lo.2010.55.4.1467 Tiner, R. W. 2014. Dichotomous keys and mapping codes for wetland landscape position, landform, water flow path, and waterbody type descriptors: Version 3.0. U.S. Fish and Wildlife Service, National Wetlands Inventory Program. Tiwari, T., F. Lidman, H. Laudon, W. Lidberg, and A. M. ˚ gren. 2017. GIS-based prediction of stream chemistry A using landscape composition, wet areas, and hydrological flow pathways. J. Geophys. Res. Biogeosci. 122: 65–79. doi:10.1002/2016JG003399 Turro, N. J., V. Ramamurthy, and J. C. Scaiano. 2010. Modern molecular photochemistry of organic molecules. University Science Books. USFWS. 2017. United States Fish and Wildlife Service: National Wetland Inventory Wetland Mapper. Waggoner, D. C., A. S. Wozniak, R. M. Cory, and P. G. Hatcher. 2017. The role of reactive oxygen species in the degradation of lignin derived dissolved organic matter. Geochim. Cosmochim. Acta 208: 171–184. doi:10.1016/ j.gca.2017.03.036 Watanabe, A., K. Moroi, H. Sato, K. Tsutsuki, N. Maie, L. Melling, and R. Jaffe. 2012. Contributions of humic substances to the dissolved organic carbon pool in wetlands from different climates. Chemosphere 88: 1265–1268. doi:10.1016/j.chemosphere.2012.04.005 Weishaar, J. L., G. R. Aiken, B. A. Bergamaschi, M. S. Fram, R. Fujii, and K. Mopper. 2003. Evaluation of specific ultraviolet absorbance as an indicator of the chemical composition and reactivity of dissolved organic carbon. Environ. Sci. Technol. 37: 4702–4708. doi:10.1021/es030360x Wenk, J., S. N. Eustis, K. McNeill, and S. Canonica. 2013. Quenching of excited triplet states by dissolved natural

Nebbioso, A., and A. Piccolo. 2013. Molecular characterization of dissolved organic matter (DOM): a critical review. Anal. Bioanal. Chem. 405: 109–124. doi:10.1007/s00216012-6363-2 Ogawa, H. 2001. Production of refractory dissolved organic matter by bacteria. Science (80-.). 292: 917–920. doi: 10.1126/science.1057627 Ohno, T. 2002. Fluorescence inner-filtering correction for determining the humification index of dissolved organic matter. Environ. Sci. Technol. 36: 742–746. doi:10.1021/ es0155276 Parker, C. A., and W. J. Barnes. 1957. Some experiments with spectrofluorimeters and filter fluorimeters. Analyst 82: 606–618. doi:10.1039/AN9578200606 Parker, K. M., J. J. Pignatello, and W. A. Mitch. 2013. Influence of ionic strength on triplet-state natural organic matter loss by energy transfer and electron transfer pathways. Environ. Sci. Technol. 47: 10987–10994. doi:10.1021/es401900j Parker, K. M., and W. A. Mitch. 2016. Halogen radicals contribute to photooxidation in coastal and estuarine waters. Proc. Natl. Acad. Sci. USA 113: 5868–5873. doi:10.1073/ pnas.1602595113 Rampi, L. P., J. F. Knight, and M. Bauer. 2016. Minnesota Land Cover Classification and Impervious Surface Area by Landsat and Lidar: 2013 Update. Raymond, P. A., and R. G. M. Spencer. 2015. Chapter 11 Riverine DOM, p. 509–533. In D.A. Hansell and C.A. Carlson [eds.], Biogeochemistry of marine dissolved organic matter. Academic Press. Remucal, C. K. 2014. The role of indirect photochemical degradation in the environmental fate of pesticides: A review. Environ. Sci. Process. Impacts 16: 628–653. doi: 10.1039/c3em00549f Running, S. 2015. MOD17A3H MODIS/Terra Net Primary Production Yearly L4 Global 500m SIN Grid V006. Sayre, R., P. Comer, H. Warner, and J. Cress. 2009. A new map of standardized terrestrial ecosystems of the conterminous United States: U.S. Geological Survey Professional Paper 1768. Schmidt, M. W. I., and others. 2011. Persistence of soil organic matter as an ecosystem property. Nature 478: 49– 56. doi:10.1038/nature10386 Sharpless, C. M., M. Aeschbacher, S. E. Page, J. Wenk, M. Sander, and K. McNeill. 2014. Photooxidation-induced changes in optical, electrochemical, and photochemical properties of humic substances. Environ. Sci. Technol. 48: 2688–2696. doi:10.1021/es403925g Sharpless, C. M., and N. V. Blough. 2014. The importance of charge-transfer interactions in determining chromophoric dissolved organic matter (CDOM) optical and photochemical properties. Environ. Sci. Process. Impacts 16: 654– 671. doi:10.1039/c3em00573a , and D. Mackay. Shiu, W.-Y., K.-C. Ma, D. Varhanıcˇkova 1994. Chlorophenols and alkylphenols: A review and 22

McCabe and Arnold


organic matter. Environ. Sci. Technol. 47: 12802–12810. doi:10.1021/es402668h Wilson, M. A. 1987. NMR techniques and applications in geochemistry and soil science. Pergamon Press. Wilson, H. F., and M. A. Xenopoulos. 2009. Effects of agricultural land use on the composition of fluvial dissolved organic matter. Nat. Geosci. 2: 37–41. doi:10.1038/ngeo391 Yu, X., J. Hawley-Howard, A. L. Pitt, J.-J. Wang, R. F. Baldwin, and A. T. Chow. 2015. Water quality of small seasonal wetlands in the Piedmont ecoregion, South Carolina, USA: Effects of land use and hydrological connectivity. Water Res. 73: 98–108. doi:10.1016/j.watres.2015.01.007 Zepp, R. G., N. L. Wolfe, G. L. Baughman, and R. C. Hollis. 1977. Singlet oxygen in natural waters. Nature. 267: 421– 423. doi:10.1038/267421a0 Zepp, R. G., P. F. Schlotzhauer, and R. M. Sink. 1985. Photosensitized transformations involving electronic energy transfer in natural waters: Role of humic substances. Environ. Sci. Technol. 19: 74–81. doi:10.1021/es00131a008

Acknowledgments We thank Molly Martin (Minnesota Department of Natural Resources) for help collecting surface water samples. We thank Mark Gernes (Minnesota Pollution Control Agency) for help collecting surface water samples, for calculating the Human Disturbance Scores, and providing valuable insights on this manuscript. We thank Dean Bock, Jeanne Moua, Conner Dunteman, and Joshua Kirk for help processing samples and collecting water chemistry data. Funding was provided by the Minnesota Natural Resources Trust Fund as recommended by the Legislative and Citizen Commission on Minnesota Resources and by a Doctoral Dissertation Fellowship from the University of Minnesota. We declare no competing financial interests.

Conflict of Interest None declared. Submitted 14 October 2017 Revised 05 March 2018 Accepted 31 March 2018 Associate editor: Peter Hernes

23