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

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.


Introduction
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 (Fe T : 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 (a λ ) at the wavelength (λ) used to quantify CDOM, variations in dissolved Fe or in the fraction of a λ caused by Fe could affect the accuracy of DOC estimated from a λ . 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 (Fe diss ) 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, a 420 ) in a large dataset from Sweden and Canada, but the carbon-specific a 420 (a 420 /DOC) increased nonlinearly, approaching an asymptotic value, with increasing Fe T , which the authors considered to be all Fe diss . Based on these findings, the authors inferred that Fe diss affected apparent CDOM levels (i.e., absorption coefficients at 420 nm, a 420 ) and concluded that Fe diss , 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 (A 254 ) increased with Fe III at the same rate in solutions with or without DOM. Poulin et al. [15] found A 254 increased linearly with Fe III in DOM-containing solutions but found no effect for added Fe II . They concluded that Fe III should be accounted for in measurements of specific UV absorbance at 254 nm (SUVA 254 ; i.e., A 254 normalized by DOC) and provided an equation to make such corrections. Maloney et al. [16] found a nonlinear increase in carbon-specific absorptivity, a 320 /DOC in the Fe diss 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 Fe diss increased from 0.0 to 0.5 mg/L. They hypothesized that Fe diss 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 Fe III to CDOMcontaining waters linearly increased absorptivity at 410-420 nm. Adding Fe III (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 a 440 ) and/or affects its other optical properties, and whether that could affect DOC values inferred from a 440 . This study had three primary objectives: (1) quantify the association between Fe and a 440 (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 a 440 and other optical properties that characterize CDOM; and (3) quantify the influence of Fe variability on DOC estimated from a 440 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.
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 Fe diss . Raw water was filtered through 0.45 μm Geotech trace-metal-certified capsule filters or pre-combusted (4 h at 450 o C) 0.7 μm Whatman glass fiber filters. Filtered water for DOC and Fe diss analyses was acidified using 0.1 mL of 2 M HCl per 50 mL of sample and refrigerated (DOC) or frozen (Fe diss ) in pre-cleaned glass or plastic bottles, respectively. Unfiltered water for Fe T analyses was acidified with 1 mL of concentrated HNO 3 per 50 mL sample and stored in the same manner as the Fe diss 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 a 440 measurements on filtered, refrigerated samples from three colored lakes with a 440 values of 5-30 m -1 and found no detectable decreases in a 440 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: where: a λ is the absorption coefficient (m -1 ) and A λ is absorbance, both at wavelength λ, and ℓ is cell path length (m). Absorbance was blank-corrected before conversion. CDOM is reported as absorption coefficient (m -1 ) at 440 nm, a 440 . SUVA 254 (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 Fe diss to SUVA 254 were calculated using the equation of Poulin et al. [15]; subtraction of the Fe diss contribution yielded DOM-based values, SUVA 254,DOM . Spectral slopes (S λ2-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): where λ ref 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. Fe diss and Fe T 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. Fe diss was not analyzed on some low-CDOM waters sampled in 2016 that, based on 2014-2015 results, were expected to have low Fe diss 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 Fe T and Fe diss and calculated particulate Fe (Fe part ) by difference.
Fe addition experiment. The effect of adding Fe III on a 440 was measured for surface water samples from six northern Minnesota lakes with a range of ambient a 440 and Fe diss . The lake waters were circumneutral (pH 6.0-8.0). We used Fe III because Poulin et al. [15] found no effect of Fe II on UV absorbance of CDOM-containing solutions. Fe III is the thermodynamically stable form in oxic water at circumneutral pH, which suggests that Fe III -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 Fe III were added sequentially. The additions were designed to yield measurable increases in Fe III (total of 460 μg/L over the five increments) but not over-saturate the DOM. The ratio Fe III /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 Fe III to deionized water to determine whether a 440 increased from uncomplexed Fe III and to 0.01 M EDTA to determine whether a 440 increased when Fe III was added to a colorless chelating agent. The Fe III solution was prepared in 0.1 M HNO 3 from reagent-grade Fe 2 (SO 4 ) 3 �nH 2 O, and the resulting Fe III concentration was determined by triplicate ICP-OES analysis. After each Fe III 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 Fe III solution decreased the pH enough to affect the measured a 440 . The effect of pH on CDOM absorbance is well known [23]. After pH stabilization, 5 mL aliquots were stored in the dark at 4 o C 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 a 440 , DOC, and Fe diss 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 a 440 and "de-trended" values of DOC and Fe diss . The de-trended DOC analysis regressed the residuals from a regression of DOC vs. Fe diss (i.e., the variance in DOC not explained by Fe diss ) against a 440 . The detrended Fe diss analysis similarly used the residuals from a regression of Fe diss vs. DOC (i.e., the variance in Fe diss not explained by DOC) in a regression vs. a 440 .

Overview of water quality variables in study lakes
Broad ranges of a 440 , DOC, and Fe diss 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 a 440 , Fe diss , 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 a 440 > 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 a 440 > 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 a 440 and Fe diss ) in both ecoregions are indicative of non-normal (skewed) distributions.
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, a 440 , and Fe diss 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 Fe diss . Overall, the results support the idea that a 440 , Fe diss , and DOC behave similarly as variables but behave differently from chl-a and SD.
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 a 440 indicate that a 440 and Fe diss may vary with depth, with a 440 decreases and Fe diss 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 Fe diss , 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 (Fe part ) 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 Fe T and Fe diss on samples collected in 2018 from moderate-to high-CDOM lakes showed that Fe part was only a small fraction of Fe T (average of 10.3%, range 0-23%; S2 Table), and most of Fe T was Fe diss . These samples were from lakes with moderateto-high Fe diss concentrations, and most of the lakes had river inflows. Samples from the associated rivers had a higher fraction of Fe part −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 Fe part was not an important component of Fe T in Swedish lakes. Kritzberg and Ekström [3] found that Fe part was an important fraction of Fe T in Swedish rivers. Our more limited sampling found that Fe part was more important in rivers than lakes, but Fe diss was still dominant in rivers.
Second, concentrations of total suspended matter (TSM) were generally < 10 mg/L in lakes with a 440 > 3.0 m -1 ; the average TSM for 53 lakes sampled in 2016 with a 440 > 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 Fe part was found not to be important [4].

Fe diss is linearly correlated a 440 and DOC
Strong correlations were found between Fe diss and a 440 , our measure of CDOM, as well as for Fe diss 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 Fe diss and a 440 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, R 2 for the total data  Table 1  set for Fe diss vs. a 440 (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.

DOC is a stronger a 440 predictor than Fe diss
Although many catchment and water quality conditions affect lake CDOM levels, we are most interested here in the relative effects of DOC and Fe diss on a 440 because a 440 is used to quantify CDOM and often used to predict DOC, e.g., [19]. As shown below, both DOC and Fe diss are strong predictors of a 440 , but DOC is stronger. We performed simple and multiple regression analyses with a 440 as predicted variable and DOC and Fe diss as predictor variables (Table 3). Regressions were performed using the entire a 440 range and just for sites with a 440 > 3.0 m -1 because related work [19] showed a break in the DOC-a 440 relationship around a 440 = 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 lowcolor DOM from autochthonous and anthropogenic sources was an important, but variable DOC contributor in waters with a 440 < 3.0 m -1 , and these sources were less important in high-CDOM waters dominated by allochthonous (humic-like) DOM. DOC exhibited stronger relationships with a 440 than did Fe diss , but both variables were significant in multiple regressions ( Table 3). Addition of Fe diss as a second variable increased R 2 by only 0.03 for the entire a 440 data range and 0.01 for a 440 > 3.0 m -1 . Similar results were found using log-transformed data (S4 Table), except that R 2 increased more when adding Fe diss as a second variable (0.09 for all data, 0.03 for a 440 > 3.0 m -1 ). De-trending to remove the influence of Fe diss on the a 440 -DOC relationship yielded an R 2 of 0.32. Removal of the influence of DOC on the a 440 -Fe diss relationship yielded an even lower R 2 of 0.12.
Weyhenmeyer et al. [14] similarly found that a least squares model using ln DOC and ln Fe diss explained 86% of the variance in ln a 420 . 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 lnln relationships of our data are 25% for DOC with the Fe diss signal removed and 12% for Fe diss when the DOC signal was removed. Although numerical values of Weyhenmeyer et al.'s original and de-trended R 2 results differ from ours, the overall outcomes of the analyses are similar: de-trending caused a large decrease in fit for a λ -DOC relationships and even a larger decrease for a λ -Fe diss relationships. Together, these findings indicate that DOC is the more important explanatory variable statistically, but Fe diss does explain some variance in a 440 beyond that produced by the correlation between DOC and Fe diss .
As noted above, DOC and Fe diss , the two main chemical determinants of a 440 , are themselves moderately correlated for the complete data set ( Fig 3B) and within each year. In each case, R 2 for the Fe diss -DOC relationship was lower than that for the corresponding Fe diss -a 440 relationship (Table 2), and 2016 values were more scattered than those for the previous years; R 2 for the total data set was only 0.58. Regression equations between Fe diss and a 440 ( Table 2, Fig 3A) had x-intercepts of a 440 < 1 m -1 . In contrast, best-fit lines for linear regressions of Fe diss vs. DOC had x-intercepts of 5-7 mg/L DOC ( Table 2, Fig 3B). Together, these findings suggest that (i) Fe diss is associated with the colored component of DOM and (ii) on average across all sites~6 mg/L of DOC is not associated with Fe diss . 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 (R 2 = 0.49) between the ratio a λ / DOC and Fe diss that might be interpreted as a measure of the effect of Fe diss on the fraction of DOC that is colored. We found a similar relationship (Fig 4A) for our data; R 2 = 0.64 for a 440 / DOC vs. ln Fe diss . As discussed above, however, the nature of DOM in low-CDOM waters (a 440 < 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 Fe diss . Consequently, trends in a 440 / DOC vs. Fe diss may simply reflect changes in the nature of DOM as a 440 /DOC increases. A plot of the relationship for sites dominated by allochthonous DOM (those with a 440 > 3.0 m -1 ; Fig  4B) yielded an R 2 of only 0.46, and there was little trend in the ratio for Fe diss > 300 μg/L. Overall, the close fit between a 440 and DOC for waters with a 440 > 3.0 m -1 (Table 3) suggests that the DOM for these sites was dominated by humic-colored DOM.

Fe diss had minor effects on other CDOM optical properties
The above results show that Fe diss should be considered when evaluating a 440 . Thus, it is worthwhile to assess whether other common optical measurements also are affected by Fe diss. Results similar to those for a 440 /DOC were obtained for SUVA 254 , a more common DOC-normalized optical measure. For the whole data set, a moderate fit was found for both uncorrected SUVA 254 vs. ln Fe diss (R 2 = 0.67) and for SUVA 254,DOM vs. ln Fe diss (R 2 = 0.64). For samples dominated by allochthonous DOM (a 440 > 3.0 m -1 ), both relationships had lower R 2 (0.50 and 0.45, respectively, for uncorrected and Fe-corrected SUVA 254 ), with little trend above Fe diss = 300 μg/L (Fig 5A).
Fe diss contributions to SUVA 254 , calculated according to [15], were small (mean = 1.9%, std. dev. = 1.9%, n = 271); on average, across all UGLS samples with Fe diss and SUVA 254 data, more than 98% of the SUVA 254 signal thus could be attributed to DOM. A small number of samples, however, had larger Fe diss contributions to SUVA 254 ( Fig 5B); 18 had Fe diss 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 SUVA 254 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 Fe diss and DOC. Seven samples with Fe diss of 1000-1500 μg/L, had Fe diss contributions to SUVA 254 of 4.4-6.7%, and Fe diss contributions for three samples with Fe diss > 1500 μg/L were 6.9-10.3%.
SUVA 254 values corrected for Fe diss were slightly lower than uncorrected values, but of the 15 samples with original SUVA 254 > 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 SUVA 254 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 SUVA 254 levels in our studies. The average SUVA 254 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 SUVA 254 [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 Fe diss [16,17]. Plots of the spectral slopes S 350-400 and S 400-460 versus the ratio Fe diss /a 440 showed no trends, but S 275-295 had a trend of smaller slopes with increasing Fe diss /a 440 , albeit with considerable scatter. S 275-295 values > 0.020 generally were from sites with a 440 < 3.0 m -1 , where low-colored autochthonous and anthropogenic DOM was dominant. Sites dominated by allochthonous DOM (a 440 > 3.0 m -1 ) had lower scatter, but the trend explained little variance in S 275-295 (R 2 = 0.13). The trend in S 275-295 generally agrees with findings of others [16,17], who reported that Fe diss decreased spectral slopes. The lack of trends in S 350-400 and S 400-460 , however, reinforces the conclusion that Fe diss at levels found in UGLS lakes does not strongly influence absorbance in the UV-A and visible regions.

Addition of Fe diss had minor effects on a 440
To measure effects of Fe diss on a 440 directly, we added known amounts of an acidified Fe III solution to six lake waters with a range of a 440 , DOC, and Fe diss (Table 4). Although a 440 increased linearly with added Fe III 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 Fe III , was within the range observed by others: 0.19 and 0.29 m -1 per 100 μg/L of added Fe III ( [3,17], respectively). The weak response to Fe III additions indicates that changes in Fe diss have only small effects on a 440 , 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 Fe III 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 a 440 . If the relationships in Fig 6 apply across the entire Fe III range, a substantial fraction of the ambient a 440 remains at Fe diss = 0; extrapolating the best-fit lines in Fig 6 to the ordinate yielded a 440 values in the range 90.3-99.7% (mean of 95.5%) of the ambient a 440 . On average for the six lakes,~95% of measured a 440 thus can be attributed to DOM and only~5% to enhanced absorptivity from Fe diss or Fe diss -DOM complexes. Xiao et al. [17] reported that Fe diss (or Fe-DOM complexes) was responsible for up to 56% of a 410 in 13 natural waters. Their highest value, however, was from a Finnish groundwater spring with very low a 410 (0.2 m -1 ) and Fe diss = 42 μg/L. DOC was not reported but likely very low. It was an outlier among the waters, and Fe contributions to a 410 for the 12 other samples were 0.6-8.7% (mean = 2.8%).  Table 4. https://doi.org/10.1371/journal.pone.0211979.g006

Iron and CDOM in lakes
The fact that a 440 did not increase when Fe III 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 Fe III with DOM molecules and not Fe III 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. Fe III -complexes with carboxylate groups in humic substances can undergo photochemical reduction to Fe II [27,28], and the occurrence of Fe II -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 [29][30][31]. The literature has conflicting information on Fe II stability in the presence of humic substances. Complexation by humic substances inhibited Fe II autoxidation rates (oxidation by O 2 ) [32], but fulvic acid accelerated Fe II oxidation by hydrogen peroxide (an intermediate in O 2 reduction to H 2 O) [33]. Nonetheless, several studies [34,35] reported that Fe III forms stronger complexes with DOM than Fe II and probably is the predominant Fe-DOM form in oxic waters. Stability constants (K f ) for Fe III and Fe II with 12 DOM sources [35] were 10 2 −10 4 higher for Fe III than for Fe II although stability constants varied widely among the DOM sources. Overall, our results indicate that Fe III 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 Fe diss effects on a 440 , which is critical to know before attempting to use a 440 to predict DOC. For example, the Fe dissa 440 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 a 440 that would occur if Fe diss were adjusted to the "best fit" values of the regression relationship. The difference between measured and best-fit Fe diss , multiplied by 0.242 m -1 per 100 μg/L of Fe diss (average slope of the a 440 -Fe III relationship, Fig 6), provided estimates of the a 440 change caused by the Fe diss change. The results showed small a 440 changes even for waters with large differences between measured and best-fit Fe diss . For example, Blueberry Lake had the highest measured Fe diss (1224 μg/L), and the best-fit Fe diss for its measured a 440 (19.6 m -1 ) is 690 μg/L. If the latter value represented the Fe diss in this lake, a 440 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 Fe diss (S6 Table); the average a 440 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%).
"Iron-corrected" a 440 values for samples with Fe diss data are estimates of the a 440 attributable to DOM alone (a 440,OM ). These were obtained by multiplying measured Fe diss values by 0.242 (average slope of the a 440 -Fe III relationships in Fig 6) and subtracting the result from measured a 440 . For 136 waters with a 440 > 3.0 m -1 , a 440,OM was 92.3 ± 5.0% (range of 71.3-99.7%) of measured a 440 . Fe diss accounted for < 10% of a 440 in most lakes (102, or 75%), but in seven lakes it accounted for 15-30% of measured a 440 (S7 Table). These lakes generally were river-influenced systems with relatively high Fe diss concentrations and/or high Fe diss /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 Fe diss contributions to a 440 > 10% in those years. We conclude that high rainfall promotes Fe export to lakes, resulting in higher Fe diss contributions to a 440 , and that lakes influenced by rivers with high CDOM and Fe are more likely to have relatively high Fe diss contributions to a 440 .

DOC can be predicted from a 440 without correcting for Fe diss
It is now possible to assess whether correction for the presence of Fe diss is needed to allow accurate prediction of DOC from a 440 (Table 3).

Long-term trends in CDOM and the role of Fe diss in UGLS lakes
Long-term data across UGLS surface waters for CDOM [18] and Fe diss are scarce. The extent of regional increases in CDOM, or whether such increases could be attributed to increases in Fe diss , 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 a 440 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 a 440 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 a 440 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, Fe diss 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 a 440 increase of only~0.3 m -1 (based on the average slope in Fig 6), but a 440 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 a 440 [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.

Conclusions
Our examination of the role of dissolved iron in optical properties of DOM in lakes supports three main conclusions. First, a 440 and Fe diss are well correlated in surface waters of the UGLS, with R 2 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 a 440 , but it is not the dominant factor for a 440 or SUVA 254 variations in UGLS lakes. The average increase in a 440 with added Fe diss (0.242 m -1 per 100 μg/L) means that increasing Fe diss by 400 μg/L would increase a 440 by only~1.0 m -1 . Even this level of Fe diss variation leads to an a 440 change less than the expected error in using a 440 as a proxy for DOC (RMSE of 1.3-1.8 m -1 , Table 3). Third, estimates of DOC based on measured a 440 and a 440,DOM (i.e., a 440 corrected for Fe diss ) were essentially the same. Consequently, our data indicate that ambient levels of Fe diss have only a minor influence on CDOM optical properties (a 440 and SUVA 254 ) and do not affect DOC estimates based on a 440 in lakes of our study area.