Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Iron influence on dissolved color in lakes of the Upper Great Lakes States

  • Patrick L. Brezonik ,

    Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Department of Civil, Environmental, and Geo- Engineering, University of Minnesota, Minneapolis, MN, United States of America

  • Jacques C. Finlay,

    Roles Funding acquisition, Investigation, Project administration, Resources, Supervision, Writing – review & editing

    Affiliation Department of Ecology, Evolution, and Behavior, University of Minnesota, St. Paul, MN, United States of America

  • Claire G. Griffin,

    Roles Data curation, Investigation, Methodology, Supervision, Visualization, Writing – review & editing

    Affiliation Department of Ecology, Evolution, and Behavior, University of Minnesota, St. Paul, MN, United States of America

  • William A. Arnold,

    Roles Methodology, Resources, Supervision, Writing – review & editing

    Affiliation Department of Civil, Environmental, and Geo- Engineering, University of Minnesota, Minneapolis, MN, United States of America

  • Evelyn H. Boardman,

    Roles Investigation, Methodology, Writing – review & editing

    Current address: Fitzgerald Environmental Associates, LLC, Colchester, VT, United States of America.

    Affiliation Department of Ecology, Evolution, and Behavior, University of Minnesota, St. Paul, MN, United States of America

  • Noah Germolus,

    Roles Investigation, Methodology, Writing – review & editing

    Current address: Parsons Laboratory, MIT, Cambridge, MA, United States of America.

    Affiliation Department of Civil, Environmental, and Geo- Engineering, University of Minnesota, Minneapolis, MN, United States of America

  • Raymond M. Hozalski,

    Roles Funding acquisition, Investigation, Project administration, Supervision, Writing – review & editing

    Affiliation Department of Civil, Environmental, and Geo- Engineering, University of Minnesota, Minneapolis, MN, United States of America

  • Leif G. Olmanson

    Roles Investigation, Visualization, Writing – review & editing

    Affiliation Remote Sensing Laboratory, Department of Forest Resources, University of Minnesota, St. Paul, MN, United States of America

Iron influence on dissolved color in lakes of the Upper Great Lakes States

  • Patrick L. Brezonik, 
  • Jacques C. Finlay, 
  • Claire G. Griffin, 
  • William A. Arnold, 
  • Evelyn H. Boardman, 
  • Noah Germolus, 
  • Raymond M. Hozalski, 
  • Leif G. Olmanson


Colored dissolved organic matter (CDOM), a major component of the dissolved organic carbon (DOC) pool in many lakes, is an important controlling factor in lake ecosystem functioning. Absorption coefficients at 440 nm (a440, m-1), a common measure of CDOM, exhibited strong associations with dissolved iron (Fediss) and DOC in 280 lakes of the Upper Great Lakes States (UGLS: Minnesota, Wisconsin, and Michigan), as has been found in Scandinavia and elsewhere. Linear regressions between the three variables on UGLS lake data typically yielded R2 values of 0.6–0.9, suggesting that some underlying common processes influence organic matter and Fediss. Statistical and experimental evidence, however, supports only a minor role for iron contributions to a440 in UGLS lakes. Although both DOC and Fediss were significant variables in linear and log-log regressions on a440, DOC was the stronger predictor; adding Fediss to the linear a440-DOC model improved the R2 only from 0.90 to 0.93. Furthermore, experimental additions of FeIII to colored lake waters had only small effects on a440 (average increase of 0.242 m-1 per 100 μg/L of added FeIII). For 136 visibly stained waters (with a440 > 3.0 m-1), where allochthonous DOM predominates, DOM accounted for 92.3 ± 5.0% of the measured a440 values, and Fediss accounted for the remainder. In 75% of the lakes, Fediss accounted for < 10% of a440, but contributions of 15–30% were observed for 7 river-influenced lakes. Contributions of Fediss in UGLS lakes to specific UV absorbance at 254 nm (SUVA254) generally were also low. Although Fediss accounted for 5–10% of measured SUVA254 in a few samples, on average, 98.1% of the SUVA254 signal was attributable to DOM and only 1.9% to Fediss. DOC predictions from measured a440 were nearly identical to those from a440 corrected to remove Fediss contributions. Overall, variations in Fediss in most UGLS lakes have very small effects on CDOM optical properties, such as a440 and SUVA254, and negligible effects on the accuracy of DOC estimated from a440, data for which can be obtained at broad regional scales by remote sensing methods.


Research associating iron (Fe) concentrations and organic color (now called colored dissolved organic matter, or CDOM) in surface waters extends back to studies in Finland [1] and Sweden [2], but its nature and significance were poorly understood for many decades. CDOM plays a major role in the ecological functioning of lakes by affecting light penetration, temperature structure, metal bioavailability, and photochemical processes. Several recent studies, e.g., [3,4], have implicated Fe as a factor in the long-term increases observed in CDOM across Scandinavia [5, 6] and some other temperate regions–the so-called “browning” phenomenon [7]. Increasing total Fe (FeT: dissolved plus particulate Fe) in 27 of 30 Swedish rivers was estimated to account for an average of 25% of the variations in CDOM and up to 74% in northern Sweden [2]. Ekström et al. [8] proposed that long-term CDOM trends in Swedish rivers could be related to increasing Fe mobilization driven by increasing temperature and river discharge that increase the probability of anoxic conditions conducive to Fe solubilization.

Whether the Fe-CDOM relationship is actually causative or merely correlative may affect the use of CDOM, which can be retrieved on regional scales from satellite imagery, e.g., [9,10], to estimate concentrations of DOC, a major component in the aquatic carbon cycle. If Fe affects absorption coefficients (aλ) at the wavelength (λ) used to quantify CDOM, variations in dissolved Fe or in the fraction of aλ caused by Fe could affect the accuracy of DOC estimated from aλ. Here we address this issue for lakes in the U.S. Upper Great Lakes States (UGLS).

Most recent studies on the influence of Fe on CDOM have focused on Swedish lakes. Based on observations from multi-basin Lake Mälaren, Köhler et al. [11] found decreasing dissolved Fe (Fediss) as water flowed through the basins, with concurrent declines in CDOM and a shift from colored allochthonous material to less colored autochthonous DOM. Weyhenmeyer et al. [4] found a linear relationship between dissolved organic carbon (DOC) and CDOM (measured as absorption coefficients at 420 nm, a420) in a large dataset from Sweden and Canada, but the carbon-specific a420 (a420/DOC) increased nonlinearly, approaching an asymptotic value, with increasing FeT, which the authors considered to be all Fediss. Based on these findings, the authors inferred that Fediss affected apparent CDOM levels (i.e., absorption coefficients at 420 nm, a420) and concluded that Fediss, pH, water residence time, and colored DOC all may be important factors for regional changes in lake browning. Alternative explanations for the browning phenomenon, including climate change [12] and recovery from acidification by atmospheric acid deposition, e.g., [5,13], are not necessarily inconsistent with a role for Fe.

Effects of Fe on UV absorbance are well studied, but effects in the visible range are less well known. Weishaar et al. [14] found that absorbance at 254 nm (A254) increased with FeIII at the same rate in solutions with or without DOM. Poulin et al. [15] found A254 increased linearly with FeIII in DOM-containing solutions but found no effect for added FeII. They concluded that FeIII should be accounted for in measurements of specific UV absorbance at 254 nm (SUVA254; i.e., A254 normalized by DOC) and provided an equation to make such corrections. Maloney et al. [16] found a nonlinear increase in carbon-specific absorptivity, a320/DOC in the Fediss range of 1–4 mg/L in a humic-rich lake and reported that the spectral slope in the range 280–400 nm decreased as Fediss increased from 0.0 to 0.5 mg/L. They hypothesized that Fediss likely would affect light conditions in the visible range but made no measurements in this region. Kritzberg and Ekström [3] and Xiao et al. [17]) reported that adding FeIII to CDOM-containing waters linearly increased absorptivity at 410–420 nm. Adding FeIII (1600–3600 μg/L) to humic and fulvic acid reference materials also decreased spectral slopes in the UV range [17].

We have been studying characteristics of CDOM in UGLS lakes and mapping its distribution by field studies and satellite imagery [10,18,19]. The occurrence of major iron ore deposits in Minnesota led us to question whether Fe contributes to observed CDOM levels (measured as a440) and/or affects its other optical properties, and whether that could affect DOC values inferred from a440. This study had three primary objectives: (1) quantify the association between Fe and a440 (our measure of CDOM) in surface waters of three UGLS ecoregions; (2) experimentally determine whether the association is causal, and if so evaluate the extent of Fe contributions to a440 and other optical properties that characterize CDOM; and (3) quantify the influence of Fe variability on DOC estimated from a440 in natural waters.


Sampling sites

We collected 450 samples from 280 water bodies (mostly lakes) in northern and central Minnesota, Wisconsin, and Michigan over the period 2014–2018 (Fig 1). Sampling occurred during summer (June-September). Nearly all 2014 and 2015 samples were from two lake-rich ecoregions of northeastern and east-central Minnesota: Northern Lakes and Forest (NLF) and North-Central Hardwood Forest (NCHF), which together contain ~ 9800 of the state’s ~12,000 lakes. The NLF is ~ 50% forested, and nearly a third of its area is wetlands or lakes. Agriculture and urban land cover constitute only small portions of this ecoregion (7 and 4%, respectively). In contrast, the NCHF is ~ 48% agricultural land and 9% urban. Forest cover constitutes 25% and wetlands ~ 10% of the NCHF. In 2016, sampling was extended to NLF and NCHF areas in Wisconsin and Michigan and the Northern Minnesota Wetlands (NMW) ecoregion, which has < 200 lakes, only a few of which are road-accessible.

Fig 1. Map of study area showing ecoregions and sampling sites.

Red circles: sites with Fediss, a440 and DOC data; black circles: additional sites with only a440 and DOC data.

Water usually was collected by small boat or kayak in the open water area, but collections were made from the ends of docks on small lakes where boat access was not feasible. Sites were selected to include a diversity of lake types, CDOM levels, and catchment land cover. The vast majority of sites were lakes, but six large rivers and six impoundments on large rivers were included. NLF lakes were in forested catchments (mixed conifers and hardwoods with substantial wetlands) with little to no human development; supplemental sampling in 2018 focused on the NLF ecoregion, where the vast majority of CDOM-rich lakes occur in the study area. Some NCHF lakes were in minimally developed catchments, but most were in urban to exurban areas and a few had catchments with row-crop agriculture.

Sampling/Field procedures

Water samples were collected from ~ 0.25 m depth using acid-washed, triple-rinsed polycarbonate or high-density polyethylene bottles and stored on ice until processed, usually the same evening. Secchi depth (SD) was measured by standard limnological procedures. Samples were collected at various depths on a few NLF lakes in 2018 to determine effects of stratification on CDOM and Fediss. Raw water was filtered through 0.45 μm Geotech trace-metal-certified capsule filters or pre-combusted (4 h at 450 oC) 0.7 μm Whatman glass fiber filters. Filtered water for DOC and Fediss analyses was acidified using 0.1 mL of 2 M HCl per 50 mL of sample and refrigerated (DOC) or frozen (Fediss) in pre-cleaned glass or plastic bottles, respectively. Unfiltered water for FeT analyses was acidified with 1 mL of concentrated HNO3 per 50 mL sample and stored in the same manner as the Fediss samples. Un-acidified filtered water for CDOM analysis was refrigerated in 40 mL glass vials with no headspace. Filter blanks (DI water) showed no measurable DOC or CDOM. Chlorophyll-a (chl-a) was collected by vacuum filtration of water samples onto 0.22 μm cellulose nitrate filters that were then stored frozen until analysis.

Analytical methods

Absorbance was measured within a month of sample collection by scanning from 250 to 700 nm using a Shimadzu 1601UV-PC dual beam spectrophotometer with 1 cm or 5 cm quartz, depending on CDOM levels, and nanopure water in the reference cell. We tested whether length of storage affected a440 measurements on filtered, refrigerated samples from three colored lakes with a440 values of 5–30 m-1 and found no detectable decreases in a440 after one month of storage, which agrees with other studies [20]. Samples were allowed to warm to room temperature on the benchtop prior to measurements. Absorbance was converted to Napierian absorption coefficients using: (1) where: aλ is the absorption coefficient (m-1) and Aλ is absorbance, both at wavelength λ, and is cell path length (m). Absorbance was blank-corrected before conversion. CDOM is reported as absorption coefficient (m-1) at 440 nm, a440. SUVA254 (L mg-1 m-1) was calculated by dividing absorbance at 254 nm by DOC concentration (mg/L), after correcting for cell path length. Contributions of Fediss to SUVA254 were calculated using the equation of Poulin et al. [15]; subtraction of the Fediss contribution yielded DOM-based values, SUVA254,DOM. Spectral slopes (Sλ2-λ1) were calculated from absorbance data for three wavelength regions (275–295, 350–400, and 400–460 nm) by taking the natural logarithm (ln) of A and computing slopes in Excel or by nonlinear fit of absorptivity data to Eq (2): (2) where λref is a reference wavelength and S is the slope.

DOC was measured on a Shimadzu TOC L-CSN analyzer. Chl-a was measured by fluorometry after 90% acetone extraction of the chl-a filters. Fediss and FeT were analyzed in triplicate with 200 μg/L of yttrium added as an internal standard on a Thermo Scientific iCAP 6500 DUO ICP-OES or iCAP 7600 DUO ICP-OES instrument. Fediss was not analyzed on some low-CDOM waters sampled in 2016 that, based on 2014–2015 results, were expected to have low Fediss nor on some high-CDOM samples from lakes sampled multiple times in 2016. Based on analysis of the 2014–2016 data, we collected additional samples in 2018 from some rivers and lakes fed by rivers to measure FeT and Fediss and calculated particulate Fe (Fepart) by difference.

Fe addition experiment.

The effect of adding FeIII on a440 was measured for surface water samples from six northern Minnesota lakes with a range of ambient a440 and Fediss. The lake waters were circumneutral (pH 6.0–8.0). We used FeIII because Poulin et al. [15] found no effect of FeII on UV absorbance of CDOM-containing solutions. FeIII is the thermodynamically stable form in oxic water at circumneutral pH, which suggests that FeIII-humic complexes predominate in surface waters. A 500 mL aliquot of filtered lake water (0.7 μm glass fiber filters) was placed in a 1.0 L beaker on a magnetic stirrer, and five 0.6 mL increments of a solution containing 77.1 mg/L of FeIII were added sequentially. The additions were designed to yield measurable increases in FeIII (total of 460 μg/L over the five increments) but not over-saturate the DOM. The ratio FeIII/DOC was < 1 μmol/mg for the highest additions, lower than reported iron-binding capacities for humic materials, e.g., [21,22]. We also added similar amounts of FeIII to deionized water to determine whether a440 increased from uncomplexed FeIII and to 0.01 M EDTA to determine whether a440 increased when FeIII was added to a colorless chelating agent. The FeIII solution was prepared in 0.1 M HNO3 from reagent-grade Fe2(SO4)3·nH2O, and the resulting FeIII concentration was determined by triplicate ICP-OES analysis. After each FeIII increment, sample pH was adjusted to within 0.1 of its ambient value by dropwise addition of 1 M NaOH. A preliminary experiment showed that the acidic FeIII solution decreased the pH enough to affect the measured a440. The effect of pH on CDOM absorbance is well known [23]. After pH stabilization, 5 mL aliquots were stored in the dark at 4 oC with no further filtration until absorbance was measured, ~ 24 h later.

Data analysis.

All observations (site-date combinations) were treated as separate data points; i.e., multiple samples from a lake across or within years were not averaged. Statistical analyses were done in JMP Pro 13.1 except for some simple regressions done in Excel 2016. Initial data inspection showed that distributions for a440, DOC, and Fediss were skewed to low values but otherwise well distributed over the range of observed values (S1 Fig). Natural log transforms yielded more Gaussian-looking distributions but still did not satisfy the Shapiro-Wilks test for normality. Unless stated otherwise, statistical results are reported for untransformed data. In addition to simple and multiple regression analyses on subsets of the untransformed and log-transformed data, we analyzed relationships between a440 and “de-trended” values of DOC and Fediss. The de-trended DOC analysis regressed the residuals from a regression of DOC vs. Fediss (i.e., the variance in DOC not explained by Fediss) against a440. The de-trended Fediss analysis similarly used the residuals from a regression of Fediss vs. DOC (i.e., the variance in Fediss not explained by DOC) in a regression vs. a440.

Results and discussion

Overview of water quality variables in study lakes

Broad ranges of a440, DOC, and Fediss and two basic limnological variables, SD and chl-a, were measured in the study, and large differences were found between the two major ecoregions (NLF and NCHF; Table 1). Median, mean and maximum values of a440, Fediss, and DOC were substantially higher for NLF lakes (dominated by forests) than NCHF lakes (dominated by agriculture). The median chl-a in NLF lakes was 4.1 μg/L (range 0–25 μg/L), and the median in NCHF lakes was 7.6 μg/L (range 1–98 μg/L). Lakes with obvious color (defined here as a440 > 3.0 m-1) also had low chl-a, nearly all < 20 μg/L [19]. The SD range was more limited (0.4–5.5 m) in NCHF lakes than NLF lakes (0.3–19.5 m), where high SD values were associated with deep, ultra-oligotrophic mine pit lakes. NLF lakes with a440 > 3 m-1 generally had SD < 3 m, and CDOM levels were the controlling factor for SD in highly colored lakes [24], some of which had SD values as low as 0.3 m. Higher mean than median values for the five variables (especially for a440 and Fediss) in both ecoregions are indicative of non-normal (skewed) distributions.

Table 1. Summary statistics for a440, DOC, and Fediss and two basic limnological variables in the NLF and NCHF ecoregions.

A principal components analysis to examine relationships among the above five variables showed that 90% of the variance was explained by the first two principal components (PCs) (Fig 2). DOC, a440, and Fediss were clustered together with high positive loadings on PC1, which accounted for 67.2% of the variance. SD also had a high PC1 loading but in a negative direction. PC2, which accounted for 22.5% of the variance, was driven by a high negative loading of chl-a and smaller positive loadings of SD and Fediss. Overall, the results support the idea that a440, Fediss, and DOC behave similarly as variables but behave differently from chl-a and SD.

Fig 2. Plot of eigenvalues for the first two principal components of a principal components analysis of five variables (a440, DOC, Fediss, chl-a, and SD) for the whole data set.

Our 2014–2016 measurements were on near-surface samples because most CDOM effects of interest are near-surface phenomena. Other recent studies on Fe-CDOM interactions, e.g., [3,4], also focused on surface water samples. Data from summer 2018 for lakes with a range of near-surface a440 indicate that a440 and Fediss may vary with depth, with a440 decreases and Fediss increases in near-bottom waters of highly colored lakes (S1 Table). These trends could be caused by seasonal variations in CDOM and in-lake cycling processes for Fe and CDOM, a topic beyond the scope of this paper.

All 2014–2016 samples analyzed for Fe were filtered (0.7 μm filters) prior to analysis, and the results are defined operationally as Fediss, which comprises Fe in true solution, including that complexed by DOM, and colloidal Fe associated with macromolecular DOM and hydrous Fe oxide particles too small to be retained on filters. We excluded particulate Fe (Fepart) associated with filterable particles from analysis because we considered it inappropriate to include Fe in plankton or mineral particles. Three lines of evidence support this decision.

First, analyses of FeT and Fediss on samples collected in 2018 from moderate- to high-CDOM lakes showed that Fepart was only a small fraction of FeT (average of 10.3%, range 0–23%; S2 Table), and most of FeT was Fediss. These samples were from lakes with moderate-to-high Fediss concentrations, and most of the lakes had river inflows. Samples from the associated rivers had a higher fraction of Fepart−average of ~ 27%, or ~ 20% when one sample from a high runoff event was excluded (S2 Table). These results agree with the findings of Weyhenmeyer et al. [4], who reported that Fepart was not an important component of FeT in Swedish lakes. Kritzberg and Ekström [3] found that Fepart was an important fraction of FeT in Swedish rivers. Our more limited sampling found that Fepart was more important in rivers than lakes, but Fediss was still dominant in rivers.

Second, concentrations of total suspended matter (TSM) were generally < 10 mg/L in lakes with a440 > 3.0 m-1; the average TSM for 53 lakes sampled in 2016 with a440 > 3.0 m-1 was only 3.8 mg/L. Third, CDOM-rich UGLS lakes occur in highly vegetated, forested catchments, where soil erosion is low, similar in terrain and ecological conditions to the Swedish and Canadian lakes where Fepart was found not to be important [4].

Fediss is linearly correlated a440 and DOC

Strong correlations were found between Fediss and a440, our measure of CDOM, as well as for Fediss and DOC, for each year and for the complete data set (Fig 3, Table 2). Similar correlations were obtained for log-transformed data (S3 Table). Values of Fediss and a440 generally were higher and more scattered in 2016 than in the two previous years, probably for two reasons. First, unusually high precipitation across Minnesota in 2016 broke many daily and monthly records at individual locations, likely resulted in higher export of Fe and DOM from catchments to lakes, and thus led to higher concentrations. Second, we sampled three times as many sites in 2016 than in 2014 or 2015. These sites covered a larger geographic range and had a greater proportion of catchments in agricultural, urban, or mixed-use landscapes, resulting in a greater diversity of geochemical conditions among sites than for previous years, thus accounting for the greater scatter. Because of the inter-annual differences, R2 for the total data set for Fediss vs. a440 (0.67) was lower than for the individual years. Overall, however, the results are consistent with century-old [1] and more recent studies [4] associating CDOM and Fe concentrations in lakes.

Fig 3.

(a) a440 and (b) DOC vs. Fediss for sampling years 2104–2016; triangles = 2014, squares = 2015, circles = 2016; dashed lines are regression fits for all years (see Table 1 for regression statistics).

DOC is a stronger a440 predictor than Fediss

Although many catchment and water quality conditions affect lake CDOM levels, we are most interested here in the relative effects of DOC and Fediss on a440 because a440 is used to quantify CDOM and often used to predict DOC, e.g., [19]. As shown below, both DOC and Fediss are strong predictors of a440, but DOC is stronger. We performed simple and multiple regression analyses with a440 as predicted variable and DOC and Fediss as predictor variables (Table 3). Regressions were performed using the entire a440 range and just for sites with a440 > 3.0 m-1 because related work [19] showed a break in the DOC-a440 relationship around a440 = 3.0 m-1. A tight fit between the two variables was found above this value, but much more scatter and a higher slope were found below. Griffin et al. [19] interpreted this finding to indicate that low-color DOM from autochthonous and anthropogenic sources was an important, but variable DOC contributor in waters with a440 < 3.0 m-1, and these sources were less important in high-CDOM waters dominated by allochthonous (humic-like) DOM.

Table 3. Simple and multiple regression relationships for a440 vs. DOC and Fediss.

DOC exhibited stronger relationships with a440 than did Fediss, but both variables were significant in multiple regressions (Table 3). Addition of Fediss as a second variable increased R2 by only 0.03 for the entire a440 data range and 0.01 for a440 > 3.0 m-1. Similar results were found using log-transformed data (S4 Table), except that R2 increased more when adding Fediss as a second variable (0.09 for all data, 0.03 for a440 > 3.0 m-1). De-trending to remove the influence of Fediss on the a440–DOC relationship yielded an R2 of 0.32. Removal of the influence of DOC on the a440-Fediss relationship yielded an even lower R2 of 0.12.

Weyhenmeyer et al. [14] similarly found that a least squares model using ln DOC and ln Fediss explained 86% of the variance in ln a420. Linear de-trending of their data showed that DOC explained 38% of the variance when the Fe signal had been removed, and Fe explained 25% of the variance when the DOC signal was removed. Comparable de-trended values for ln-ln relationships of our data are 25% for DOC with the Fediss signal removed and 12% for Fediss when the DOC signal was removed. Although numerical values of Weyhenmeyer et al.’s original and de-trended R2 results differ from ours, the overall outcomes of the analyses are similar: de-trending caused a large decrease in fit for aλ-DOC relationships and even a larger decrease for aλ-Fediss relationships. Together, these findings indicate that DOC is the more important explanatory variable statistically, but Fediss does explain some variance in a440 beyond that produced by the correlation between DOC and Fediss.

As noted above, DOC and Fediss, the two main chemical determinants of a440, are themselves moderately correlated for the complete data set (Fig 3B) and within each year. In each case, R2 for the Fediss-DOC relationship was lower than that for the corresponding Fediss-a440 relationship (Table 2), and 2016 values were more scattered than those for the previous years; R2 for the total data set was only 0.58. Regression equations between Fediss and a440 (Table 2, Fig 3A) had x-intercepts of a440 < 1 m-1. In contrast, best-fit lines for linear regressions of Fediss vs. DOC had x-intercepts of 5–7 mg/L DOC (Table 2, Fig 3B). Together, these findings suggest that (i) Fediss is associated with the colored component of DOM and (ii) on average across all sites ~ 6 mg/L of DOC is not associated with Fediss. This likely represents low-color DOM with a low abundance of Fe-binding ligand groups, probably of autochthonous or anthropogenic origin. Photo-degradation of CDOM also could contribute to the low-color DOM pool, but it is uncertain whether CDOM photo-degradation reduces Fe binding capacity.

Weyhenmeyer et al. [14] found a curvilinear relationship (R2 = 0.49) between the ratio aλ/DOC and Fediss that might be interpreted as a measure of the effect of Fediss on the fraction of DOC that is colored. We found a similar relationship (Fig 4A) for our data; R2 = 0.64 for a440/DOC vs. ln Fediss. As discussed above, however, the nature of DOM in low-CDOM waters (a440 < 3.0 m-1) likely differs from that in high-CDOM waters. The latter consists primarily of allochthonous, humic-like DOM; the former derives from various sources with generally lower color intensity and probably fewer binding sites for Fediss. Consequently, trends in a440/DOC vs. Fediss may simply reflect changes in the nature of DOM as a440/DOC increases. A plot of the relationship for sites dominated by allochthonous DOM (those with a440 > 3.0 m-1; Fig 4B) yielded an R2 of only 0.46, and there was little trend in the ratio for Fediss > 300 μg/L. Overall, the close fit between a440 and DOC for waters with a440 > 3.0 m-1 (Table 3) suggests that the DOM for these sites was dominated by humic-colored DOM.

Fig 4.

(a) a440/DOC vs. Fediss for all data, R2 = 0.64; (b) a440/DOC vs. Fediss for sites with a440 > 3.0 m-1, R2 = 0.46. (R2 are for fit of a440/DOC to ln(Fediss)).

Fediss had minor effects on other CDOM optical properties

The above results show that Fediss should be considered when evaluating a440. Thus, it is worthwhile to assess whether other common optical measurements also are affected by Fediss. Results similar to those for a440/DOC were obtained for SUVA254, a more common DOC-normalized optical measure. For the whole data set, a moderate fit was found for both uncorrected SUVA254 vs. ln Fediss (R2 = 0.67) and for SUVA254,DOM vs. ln Fediss (R2 = 0.64). For samples dominated by allochthonous DOM (a440 > 3.0 m-1), both relationships had lower R2 (0.50 and 0.45, respectively, for uncorrected and Fe-corrected SUVA254), with little trend above Fediss = 300 μg/L (Fig 5A).

Fig 5.

(a) SUVA254,DOM vs. Fediss for sites with a440 > 3.0 m-1, R2 = 0.45 (R2 is for fit to ln(Fediss)); (b) fraction of SUVA254 for all sites caused by DOM (i.e., corrected for FeIII contribution) vs. Fediss.

Fediss contributions to SUVA254, calculated according to [15], were small (mean = 1.9%, std. dev. = 1.9%, n = 271); on average, across all UGLS samples with Fediss and SUVA254 data, more than 98% of the SUVA254 signal thus could be attributed to DOM. A small number of samples, however, had larger Fediss contributions to SUVA254 (Fig 5B); 18 had Fediss contributions > 5%, and two had contributions > 10%. The lake with the largest contribution (11.9%), Crystal Lake (WI), is an ultra-clear oligotrophic seepage lake with low DOC (2.5 mg/L) and a SUVA254 of only 0.64 L mg-1 m-1; it is an outlier relative to most lakes in the region. More relevant here are samples with higher Fediss and DOC. Seven samples with Fediss of 1000–1500 μg/L, had Fediss contributions to SUVA254 of 4.4–6.7%, and Fediss contributions for three samples with Fediss > 1500 μg/L were 6.9–10.3%.

SUVA254 values corrected for Fediss were slightly lower than uncorrected values, but of the 15 samples with original SUVA254 > 5.0 L mg-1 m-1, 10 still had values > 5.0 after correction. The common upper limit for DOM-caused SUVA in natural waters is 5.0 L mg-1 m-1 [15]. Average SUVA254 values for the 15 samples before and after correction (S5 Table) were 5.33 and 5.15 L mg-1 m-1, respectively. A third of these samples were from Johnson Lake, Minnesota (Itasca County), a small bog lake that generally had the highest CDOM and SUVA254 levels in our studies. The average SUVA254 before Fe-correction for Johnson Lake of 5.41 L mg-1 m-1 decreased to 5.23 L mg-1 m-1 after correction. Although high nitrate/nitrite concentrations (tens of mg/L range) may affect levels of SUVA254 [25], concentrations of these ions were very low (few μg/L) in Johnson Lake and the other lakes we studied.

Spectral slopes, a measure of DOM composition, are also influenced by Fediss [16,17]. Plots of the spectral slopes S350-400 and S400-460 versus the ratio Fediss/a440 showed no trends, but S275-295 had a trend of smaller slopes with increasing Fediss/a440, albeit with considerable scatter. S275-295 values > 0.020 generally were from sites with a440 < 3.0 m-1, where low-colored autochthonous and anthropogenic DOM was dominant. Sites dominated by allochthonous DOM (a440 > 3.0 m-1) had lower scatter, but the trend explained little variance in S275-295 (R2 = 0.13). The trend in S275-295 generally agrees with findings of others [16,17], who reported that Fediss decreased spectral slopes. The lack of trends in S350-400 and S400-460, however, reinforces the conclusion that Fediss at levels found in UGLS lakes does not strongly influence absorbance in the UV-A and visible regions.

Addition of Fediss had minor effects on a440

To measure effects of Fediss on a440 directly, we added known amounts of an acidified FeIII solution to six lake waters with a range of a440, DOC, and Fediss (Table 4). Although a440 increased linearly with added FeIII after readjusting the pH to the original value (Fig 6), the rate was small. The average rate of increase, 0.242 m-1 per 100 μg/L of added FeIII, was within the range observed by others: 0.19 and 0.29 m-1 per 100 μg/L of added FeIII ([3,17], respectively). The weak response to FeIII additions indicates that changes in Fediss have only small effects on a440, and inspection of absorbance spectra over the range 250–500 nm showed no changes in shapes of the spectra. Addition of 3.0 mL of acidified stock FeIII solution to 500 mL of deionized water or to 500 mL of 0.01 M EDTA at pH 6.5 yielded no measurable increases in a440.

Fig 6. a440 vs. Fediss for six waters in iron addition experiment; slopes and statistical information on best-fit lines in Table 4.

Table 4. Chemical characteristics of lakes and results for iron addition experiment.

If the relationships in Fig 6 apply across the entire FeIII range, a substantial fraction of the ambient a440 remains at Fediss = 0; extrapolating the best-fit lines in Fig 6 to the ordinate yielded a440 values in the range 90.3–99.7% (mean of 95.5%) of the ambient a440. On average for the six lakes, ~ 95% of measured a440 thus can be attributed to DOM and only ~ 5% to enhanced absorptivity from Fediss or Fediss-DOM complexes. Xiao et al. [17] reported that Fediss (or Fe-DOM complexes) was responsible for up to 56% of a410 in 13 natural waters. Their highest value, however, was from a Finnish groundwater spring with very low a410 (0.2 m-1) and Fediss = 42 μg/L. DOC was not reported but likely very low. It was an outlier among the waters, and Fe contributions to a410 for the 12 other samples were 0.6–8.7% (mean = 2.8%).

The fact that a440 did not increase when FeIII was added to DI water or to an EDTA solution but increased by small amounts when added to CDOM-rich waters indicates that the increase is caused by interaction of FeIII with DOM molecules and not FeIII absorbance itself. The chemical nature of Fe interactions with DOM is complicated [26], and how they may affect absorption of visible light (e.g., at 440 nm) still is not well understood. FeIII-complexes with carboxylate groups in humic substances can undergo photochemical reduction to FeII [27,28], and the occurrence of FeII-humic complexes in oxic waters thus cannot be ruled out. There also is evidence that some Fe associated with aquatic humic substances is bound irreversibly, apparently not as conventional metal-ligand complexes [2931]. The literature has conflicting information on FeII stability in the presence of humic substances. Complexation by humic substances inhibited FeII autoxidation rates (oxidation by O2) [32], but fulvic acid accelerated FeII oxidation by hydrogen peroxide (an intermediate in O2 reduction to H2O) [33]. Nonetheless, several studies [34,35] reported that FeIII forms stronger complexes with DOM than FeII and probably is the predominant Fe-DOM form in oxic waters. Stability constants (Kf) for FeIII and FeII with 12 DOM sources [35] were 102−104 higher for FeIII than for FeII although stability constants varied widely among the DOM sources. Overall, our results indicate that FeIII complexation by DOM has very small effects on CDOM chromophoric groups.

Application of experimental results to field data

We applied the experimental results to our field data to further evaluate Fediss effects on a440, which is critical to know before attempting to use a440 to predict DOC. For example, the Fediss-a440 regression of the 2015 data (Fig 7) showed that some data points were far from the regression line. For the largest outliers (six high and eight low), we estimated the change in a440 that would occur if Fediss were adjusted to the “best fit” values of the regression relationship. The difference between measured and best-fit Fediss, multiplied by 0.242 m-1 per 100 μg/L of Fediss (average slope of the a440-FeIII relationship, Fig 6), provided estimates of the a440 change caused by the Fediss change. The results showed small a440 changes even for waters with large differences between measured and best-fit Fediss. For example, Blueberry Lake had the highest measured Fediss (1224 μg/L), and the best-fit Fediss for its measured a440 (19.6 m-1) is 690 μg/L. If the latter value represented the Fediss in this lake, a440 would be 18.3 m-1, a decrease of 1.3 m-1 (a 6.6% change). Similar changes were found for the other waters with large differences between measured and best-fit Fediss (S6 Table); the average a440 change for the six high outliers was– 3.7% (range −1.8 to − 6.6%), and the average for the eight low outliers was +2.2% (range 1.2 to 2.9%).

Fig 7. Fediss vs. a440 for 2015 data; best-fit regression line and statistics given in Table 2.

Circled data: outliers examined for effects of Fediss on a440 (S6 Table). Dotted line: best-fit linear relationship: Fediss = 36.8a440−30.7; R2 = 0.79. Dashed lines illustrate change in a440 for some outliers when Fediss was changed to “best-fit” value.

“Iron-corrected” a440 values for samples with Fediss data are estimates of the a440 attributable to DOM alone (a440,OM). These were obtained by multiplying measured Fediss values by 0.242 (average slope of the a440-FeIII relationships in Fig 6) and subtracting the result from measured a440. For 136 waters with a440 > 3.0 m-1, a440,OM was 92.3 ± 5.0% (range of 71.3–99.7%) of measured a440. Fediss accounted for < 10% of a440 in most lakes (102, or 75%), but in seven lakes it accounted for 15–30% of measured a440 (S7 Table). These lakes generally were river-influenced systems with relatively high Fediss concentrations and/or high Fediss/DOC ratios, and all were samples collected in the high rainfall year 2016. Of the six lakes that also had data for 2014 or 2015, only Big Sandy Lake and Big Sandy River Lake had Fediss contributions to a440 > 10% in those years. We conclude that high rainfall promotes Fe export to lakes, resulting in higher Fediss contributions to a440, and that lakes influenced by rivers with high CDOM and Fe are more likely to have relatively high Fediss contributions to a440.

DOC can be predicted from a440 without correcting for Fediss

It is now possible to assess whether correction for the presence of Fediss is needed to allow accurate prediction of DOC from a440. Regressions of measured DOC versus a440,OM and measured DOC versus measured a440 yielded very similar relationships for the same data set (a440 > 3.0 m-1; N = 134): (3) (4) The DOC predicted by Eq 3 from the estimated a440,OM for each site was compared to the DOC predicted from the measured a440-DOC relationship (Eq 4); the relationship was almost exactly 1:1 (slope = 0.997; R2 = 0.99) (S2 Fig). Consequently, we conclude that DOC predictions from measured a440 are just as accurate for our study sites as DOC predictions from a440 corrected to remove the influence of Fediss. Moreover, for waters where a440 > 3.0 m-1, a440 is a good predictor of DOC (Table 3).

Long-term trends in CDOM and the role of Fediss in UGLS lakes

Long-term data across UGLS surface waters for CDOM [18] and Fediss are scarce. The extent of regional increases in CDOM, or whether such increases could be attributed to increases in Fediss, thus is unknown. Smaller-scale analyses suggest, however, that regional CDOM trends are more complicated than observed in Scandinavia and likely driven by climatic and hydrologic variations. For example, substantial intra- and inter-annual variations but no monotonic trends were found in a440 and DOC for 20 small lakes in Upper Michigan over six years [36]. Climatic conditions that affected carbon loadings from upland forests and wetlands were considered the drivers of these variations. Similarly, Brezonik et al. [18] found large a440 variations in seven lakes of the northern Wisconsin LTER program (all within a radius of 10 km, [37]) for the period 1990–2012, but only colored Crystal Bog had increasing a440 over the whole period. In a study on optical properties of the LTER lakes, Jane et al. [37] found inconsistent trends in DOC since 1990, with increases in two (including Crystal Bog), decreases in four, and no trend in one. Optical properties related to DOM chemical characteristics varied more with climatic conditions than DOC concentrations.

Björnerås et al. [38] recently reported temporal increases in Fe on broad scales in European and North American freshwaters. Their data overlap our region at only one site (in north-central Wisconsin). Additional data from the Wisconsin LTER lakes, which were not included in the Björnerås et al. study, showed that Fe increased in Crystal Bog, decreased in Trout Bog, and had no trends in the other five lakes since the 1980s (Mann-Kendall test; N. Lottig, Univ. Wisconsin, pers. comm., 2017). As noted above, Crystal Bog is the only LTER lake with increasing CDOM during the same period. The intra- and inter-annual variability in Fe and CDOM was high in all the lakes. In Crystal Bog, Fediss averaged 200 μg/L for six pre-1990 measurements and 330 μg/L for eight post-2010 measurements; the average increase of 130 μg/L could account for an a440 increase of only ~ 0.3 m-1 (based on the average slope in Fig 6), but a440 in Crystal Bog actually increased by ~ 5.5 m-1 over this time [18]. Moreover, pH data for Crystal Bog showed no trends over the period of record (1981–2016) (S3 Fig). The increases in DOC [37] and a440 [18] in Crystal Bog thus cannot be explained by declining acidity, and the lack of similar trends in the other LTER lakes suggests that long-term climatic changes also are not responsible for the trends.


Our examination of the role of dissolved iron in optical properties of DOM in lakes supports three main conclusions. First, a440 and Fediss are well correlated in surface waters of the UGLS, with R2 values of 0.73–0.77 for individual years, as has been shown in studies elsewhere. Second, experimental data show that iron has small effects on CDOM measured as a440, but it is not the dominant factor for a440 or SUVA254 variations in UGLS lakes. The average increase in a440 with added Fediss (0.242 m-1 per 100 μg/L) means that increasing Fediss by 400 μg/L would increase a440 by only ~ 1.0 m-1. Even this level of Fediss variation leads to an a440 change less than the expected error in using a440 as a proxy for DOC (RMSE of 1.3–1.8 m-1, Table 3). Third, estimates of DOC based on measured a440 and a440,DOM (i.e., a440 corrected for Fediss) were essentially the same. Consequently, our data indicate that ambient levels of Fediss have only a minor influence on CDOM optical properties (a440 and SUVA254) and do not affect DOC estimates based on a440 in lakes of our study area.

Supporting information

S1 Fig. Histograms of data distributions for a440 (CDOM), DOC, Fediss, and SUVA254.

Upper plots: untransformed data; lower plots: log-transformed (ln) values.


S2 Fig. DOC predicted from Fe-corrected a440 (a440,OM) (Eq 3) vs. DOC predicted from measured a440.

(Eq 4). Best-fit line: a440,OM = 0.997a440 + 0.032; R2 = 0.99; RMSE = 0.685, slope SE = 0.0086, p < 0.0001.


S3 Fig. Time trend for pH in Crystal Bog, Vilas County, Wisconsin, 1981–2016; data from the North Temperate Lakes Long Term Ecological Research (LTER) program (


S1 Table. Vertical profile data for three NLF lakes with a wide range of surface CDOM (a440) values.


S2 Table. FeT, Fediss, and % Fediss for 2018 lake and associated river samples from the NLF ecoregion.


S3 Table. Fediss-a440 and Fediss-DOC relationships for log-transformed data.


S4 Table. Log-transformed regression relationships for a440 vs. DOC and Fediss.


S5 Table. SUVA254 values for samples with measured SUVA254 > 5.0 before and after Fediss correction.


S6 Table. Changes in a440 for 2015 waters that had large differences between measured and best-fit Fediss after Fediss was changed to the best-fit value.


S7 Table. Samples with measured a440 > 3.0 m-1 having 15–30% of a440 caused by Fediss and 70–85% caused by colored DOM.



We gratefully acknowledge support from the National Science Foundation, the Minnesota Environmental and Natural Resources Trust fund, as recommended by the Legislative-Citizen Commission on Minnesota Resources, and Univ. of Minnesota’s Office of the VP for Research and Retirees Association, U-Spatial Program, Sea Grant Program, and Agricultural Experiment Station. We thank numerous collaborators and student workers for assistance in sample collection and analysis. The senior author gratefully acknowledges mentoring early in his career by G.F. Lee, himself an iron researcher.


  1. 1. Aschan O. Soluble humus material of northern fresh waters. J Prakt Chemie 1908;77: 172.
  2. 2. Lohammar G. Wasserchemie und höhere Vegetation schwedischer Seen. Symb. Bot. Upsaliens 1938;3: 1–252.
  3. 3. Kritzberg ES, Ekström SM. Increasing iron concentrations in surface waters–a factor behind brownification? Biogeosciences 2012;9: 1465–1478.
  4. 4. Weyhenmeyer GA, Prairie YT, Tranvik LJ. Browning of boreal freshwaters coupled to carbon-iron interactions along the aquatic continuum. PLoS ONE 2014;9: (2) e88104. pmid:24505396
  5. 5. Monteith DT, Stoddard JL, Evans CD, de Wit HA, Forsius M, Hørgåsen T, et al. Dissolved organic carbon trends resulting from changes in atmospheric deposition chemistry. Nature 2007;450: 537–539. pmid:18033294
  6. 6. Haaland S, Hongve D, Laudon H, Riise G, Vogt RD. Quantifying the drivers of the increasing colored organic matter in Boreal surface waters. Environ. Sci. Technol. 2010;44: 2975–2980. pmid:20329770
  7. 7. Roulet N, Moore TR. Environmental chemistry—Browning the waters. Nature 2006;444: 283–284. pmid:17108948
  8. 8. Ekström SM, Regnell O, Reader HE, Nilsson PA, Löfgren S, Kritzberg ES. Increasing concentrations of iron in surface waters as a consequence of reducing conditions in the catchment area. J. Geophys. Res. Biogeosci. 2016;121: 479–493.
  9. 9. Kutser T, Pierson DC, Kallio KY, Reinart A, Sobek S. Mapping lake CDOM by satellite remote sensing. Remote Sens. Environ. 2005;94: 535–540.
  10. 10. Olmanson LG, Brezonik PL, Finlay JC, Bauer ME. Comparison of Landsat 8 and Landsat 7 for regional measurements of CDOM and water clarity in lakes. Remote Sens. Environ. 2016;185: 119–128.
  11. 11. Köhler SJ, Kothawala D, Futter MN, Liungman O, Tranvik LJ. In-lake processes offset increased terrestrial inputs of dissolved organic carbon and color to lakes. PLoS ONE 2013;8: 8 e70598. pmid:23976946
  12. 12. Tranvik LJ, Jansson M. Climate change (Communication arising): Terrestrial export of organic carbon. Nature 2002;415: 861–862.
  13. 13. Garmo OA, Skjelkvale BL, De Wit HA, Colombo L, Curtis C, Fölster J, et al. Trends in surface water chemistry in acidified areas in Europe and North America from 1990 to 2008. Water Air Soil Pollut. 2014;225: 1–14.
  14. 14. Weishaar JL, Aiken GR, Bergamaschi BA, Fram MS, Fujii R, Mopper K. Evaluation of specific ultraviolet absorbance as an indicator of the chemical composition and reactivity of dissolved organic carbon. Environ. Sci. Technol. 2003;37: 4702–4708. pmid:14594381
  15. 15. Poulin BA, Ryan JR, Aiken GR. Effects of iron on optical properties of dissolved organic matter. Environ. Sci. Technol. 2014;48, 10098–10106. pmid:25084347
  16. 16. Maloney KO, Morris DP, Moses CO, Osburn CL. The role of iron and dissolved organic carbon in the absorption of ultraviolet radiation in humic lake water. Biogeochemistry 2005;75: 393–407.
  17. 17. Xiao YH, Sara-Aho T, Hartikainen H, Vähätalo AV. Contribution of ferric iron to light absorption by chromophoric dissolved organic matter. Limnol. Oceanogr. 2013;58: 653–662.
  18. 18. Brezonik PL, Olmanson LG, Finlay JC, Bauer ME. Factors affecting the measurement of CDOM in optically complex inland waters. Remote Sens. Environ. 2015;157: 199–215.
  19. 19. Griffin CG, Finlay JC, Brezonik PL, Olmanson LG, Hozalski RM. Limitations on using CDOM as a proxy for DOC in temperate lakes. Water Res. 2018;144: 719–727. pmid:30099300
  20. 20. Peacock M, Freeman C, Gauci V, Lebron I, Evans CD. Investigations of freezing and cold storage for the analysis of peatland dissolved organic carbon (DOC) and absorbance properties. Env. Sci.: Processes Impacts 2015;17: 1290–1301.
  21. 21. Shapiro J. Effect of yellow organic acids on iron and other metals in water. J. Amer. Water Works Assoc. 1964;56: 1062–1082.
  22. 22. Garcia-Mina JM. Stability, solubility and maximum metal binding capacity in metal-humic complexes involving humic substances extracted from peat and organic compost. Org. Geochem. 2006;37:1960–1972.
  23. 23. Singley JE, Harris RH, Maulding JS. Correction of color measurements to standard conditions. J. Amer. Water Wks. Assoc. 1966;58: 455–457.
  24. 24. Brezonik PL, Bouchard RW Jr, Finlay JC, Griffin CG, Olmanson LG, Anderson JP, et al. Color, chlorophyll a and suspended solids effects on Secchi depth in lakes: implications for trophic state assessment. Ecol. Appl. 2019;29: (in press).
  25. 25. Dilling J, Kaiser K. Estimation of the hydrophobic fraction of dissolved organic matter in water samples using UV photometry. Water Res. 2002;36: 5037–5044. pmid:12448552
  26. 26. Brezonik PL, Arnold WA. Water chemistry. New York, Oxford: Oxford University Press; 2011.
  27. 27. Miles CJ, Brezonik PL. Oxygen consumption by a photochemical ferrous-ferric catalytic cycle. Environ. Sci. Technol. 1981;15: 1089–1095. pmid:22284115
  28. 28. Xie H, Zafiriou OC, Cai WJ, Zepp RG, Wang Y. Photooxidation and its effects on the carboxyl content of dissolved organic matter in two coastal rivers in the southeastern United States. Environ. Sci. Technol. 2004;38: 4113–4119. pmid:15352449
  29. 29. Hutchinson GE. A Treatise on limnology, vol. 1. New York: Wiley-Interscience; 1957.
  30. 30. Senesi N, Griffith SM, Schnitzer M, Townsend MG. Binding of Fe3+ by humic materials. Geochim. Cosmochim. Acta 1977;41: 969–976.
  31. 31. Sedlacek J, Gjessing E, Rambaek JP. Isotope exchange between inorganic iron and iron naturally complexed by aquatic humus. Sci. Tot. Env. 1987;62: 275–279.
  32. 32. Theis TL, Singer PC. Complexation of iron(II) by organic matter and its effect on iron(II) oxygenation. Environ. Sci. Technol. 1974;8: 569–573.
  33. 33. Voelker BM, Sulzberger B. Effects of fulvic acid on Fe(II) oxidation by hydrogen peroxide. Environ. Sci. Technol. 1996;30: 1106–1114.
  34. 34. Emmenegger L, Sigg L, Sulzberger B. Light-induced redox cycling of iron in circumneutral lakes. Limnol. Oceanogr. 2001;46: 49–61.
  35. 35. Rose AL, Waite TD. Kinetics of iron complexation by dissolved natural organic matter in coastal waters. Mar. Chem. 2003;84: 85–103.
  36. 36. Pace ML, Cole JJ. Synchronous variation of dissolved organic carbon and color in lakes. Limnol. Oceanogr. 2002;47: 333–342.
  37. 37. Jane SF, Winslow LA, Remucal CK, Rose KC. Long-term trends and synchrony in dissolved organic matter characteristics in Wisconsin, USA, lakes: Quality, not quantity, is highly sensitive to climate. J. Geophys. Res. Biogeosci. 2017;122: 546–561.
  38. 38. Björnerås C, Weyhenmeyer GA, Evans CD, Gessner MO, Grossart H-P, Kangur K, et al. Widespread increases in iron concentration in European and North American freshwaters. Global Biogeochem. Cycles 2017;31: 1488–1500.