3.3.2 Predicted Fe and Mn Concentrations
Reservoir turnover had substantial impacts on Fe and Mn concentrations. At the beginning of the deployment (16 October 2020), 17 days prior to turnover, both Fe and Mn displayed large differences in concentration between the epilimnion and hypolimnion (Figures 4D-E). The average total Fe and total Mn concentrations across all hypolimnetic depths (6.2, 8.0, and 9.0 m) were 3.73 mg/L and 1.48 mg/L; across all epilimnetic depths (0.1, 1.6, and 3.8 m) they were 0.41 mg/L and 0.14 mg/L, respectively (Figures 4D-E). Substantial changes in epilimnetic concentrations were not observed until 24 hours prior to turnover. Within that 24 hour period, average epilimnetic total Fe and total Mn increased by 70% (0.61 to 1.04 mg/L) and 66% (0.29 to 0.48 mg/L), respectively.
In contrast to the epilimnion, we observed declining total Fe and Mn concentrations in the hypolimnion prior to turnover (Figures 4D-E). Between 16 October and 02 November 2020, hypolimnetic total Fe and total Mn concentrations declined at a rate of 0.13 and 0.11 mg/L/d, respectively. However, there were also periods of fluctuations in total Fe and total Mn concentrations by as much as 1 mg/L/d (Figure 4D-E). In the 24 hours prior to turnover, average hypolimnetic total Fe and total Mn decreased by 45% (2.09 to 1.14 mg/L) and 32% (0.82 to 0.55 mg/L), respectively.
A strong concentration gradient between the epilimnion and hypolimnion remained for total Fe and total Mn until full reservoir turnover on 02 November 2020. After turnover, water temperature and DO rapidly equalized across the full water column, coinciding with the rapid equalization of total Fe and Mn concentrations across the water column (Figures 4D-E). Total Fe and Mn concentrations were lower and less variable than during the pre-turnover period (Figures 4D-E). The reservoir remained well-mixed for 2 days, but then shifting thermal gradients led to a temporary re-stratification that began on 02 November 2020 and lasted until the end of the deployment on 09 November 2020 (Figures 4A-B). The re-stratification of the reservoir was also evident in total Fe and total Mn concentrations (Figures 4D-E).
3.4 Oxygen On Deployment
3.4.1 Water Temperature, Stratification, and DO
DO concentrations, water temperature, and Schmidt stability differed considerably between the two deployments (Figures 5A-C). At the start of the Oxygen On deployment (26 May 2021), 16 days prior to HOx activation, epilimnetic DO concentrations were high (5-15 mg/L) and exhibited a consistent decline throughout the deployment due to warm air temperatures (Figure S14). Metalimnetic and hypolimnetic DO concentrations were both approximately 0 mg/L throughout the deployment. The water temperature profile shows distinctly stratified layers in the reservoir prior to HOx operation, with a sharp temperature gradient throughout the epilimnion for the entire deployment and a slight temperature gradient in the hypolimnion (Figure 5B). Immediately following HOx activation on 11 June 2021, the water temperature profile equalized across layers below 6m depth, indicating mixing within the hypolimnion due to HOx activation (Figure 5B). The water temperature profile in the epilimnion was unaffected by HOx operation. Metalimnetic and hypolimnetic DO concentrations did not increase above 0 mg/L in the few days after activation of the HOx system. This is attributed to chemical oxygen demand in the hypolimnion resulting from high concentrations of reduced solutes (e.g., Fe(II) and Mn(II)).
3.4.2 Predicted Fe and Mn Concentrations
At the beginning of the deployment, the highest concentrations of total Fe and Mn were at the lowest depth (9m) and concentrations decreased upwards in the water column, with a sharp decrease between the hypolimnion and epilimnion (Figures 5D-E). In the first 24 hours of the deployment, total Fe and Mn concentrations averaged across all epilimnetic depths were 0.43 and 0.03 mg/L, respectively, while across the hypolimnetic depths they were 2.71 and 0.54 mg/L, respectively. Prior to HOx operation, both total Fe and Mn in the hypolimnion exhibited large, sub-daily fluctuations which resulted in concentration changes of up to 1.62 mg/L/hr and 0.19 mg/L/hr, respectively (Figures 5D-E). These sub-daily fluctuations were most pronounced at the lowest depth.
The spatial and temporal cycling dynamics of Fe and Mn were significantly affected by hypolimnetic oxygenation. Prior to activation of the HOx system on 11 June 2021, epilimnetic total Fe and Mn concentrations remained constant (sd = 0.07 mg/L and 0.004 mg/L, respectively) and low (maximum concentrations = 0.63 mg/L and 0.05 mg/L, respectively). Hypolimnetic total Fe and Mn concentrations during this period were much more variable (sd = 1.85 mg/L and 0.19 mg/L, respectively) and higher (maximum concentrations = 7.90 mg/L and 1.08 mg/L, respectively). Shortly after HOx activation, total Fe and Mn concentrations equalized contemporaneously with the equalization of water temperature across the hypolimnetic depths, indicating that this layer of the reservoir was well-mixed with respect to Fe and Mn (Figures 5B, 5D-E). In contrast, differences in total Fe and Mn concentrations across the epilimnetic depths increased slightly after activation of the HOx system.
Approximately 6 hours after turning on the HOx system, total Fe and Mn at 9m depth declined by approximately 2.5 mg/L and 0.25 mg/L, respectively (Figures 5D-E). Concentrations of total Fe and Mn at all hypolimnetic depths subsequently increased over the next 24 hours, before eventually stabilizing over the following 24 hours at concentrations of 1.5-3.5 mg/L and 0.5-0.75 mg/L, respectively. For the remainder of the deployment, total Fe and Mn concentrations remained equal across all hypolimnetic depths and exhibited less variability (Figures 5D-E).
3.5 Predicted Fe and Mn Soluble-to-Total Ratios
The ratio of predicted soluble to total Fe (SFe:TFe) and Mn (SMn:TMn) was calculated to assess redox transformations. We observed distinct changes in these ratios over the course of both deployments, most notably in the hypolimnion (Figure 6). During the Turnover Deployment, the hypolimnion was maintained at oxic conditions pre-turnover (due to HOx) and post-turnover (due to mixing). As expected, hypolimnetic SFe:TFe was approximately 0 during this entire deployment, indicating that all Fe in the hypolimnion was in the particulate fraction (soluble Fe + particulate Fe = total Fe). In contrast, hypolimnetic SMn:TMn was approximately 1 at the beginning of the deployment, indicating that all Mn was in the soluble fraction. However, in the week prior to turnover, hypolimnetic SMn:TMn oscillated between 0.5 and 1. Following turnover, SMn:TMn was greater than 0.75 and remained high until the end of the deployment.
At the beginning of the Oxygen On deployment, SFe:TFe differed greatly with depth in the hypolimnion, with ratios greater than 0.8 at 9m depth and ratios close to 0 at 6.2m and 8m depths (Figures 6C-D). Between the beginning of the deployment and HOx activation on 11 June 2021, the SFe:TFe at 6.2m and 8m increased continuously to approximately the same level as 9m (Figures 6C-D). Just before the initiation of HOx operation, the SFe:TFe at all hypolimnion depths was > 0.75, indicating that most of the Fe in the hypolimnion was in the soluble fraction. However, immediately after turning the HOx system on, the SFe:TFe in the hypolimnion decreased steadily. In the 48-hour period after HOx activation, the SFe:TFe in the hypolimnion declined to less than 0.25 and remained low until the end of the experimental period (Figure 6C-D), indicating oxidation processes. In contrast to Fe, SMn:TMn in the hypolimnion was > 0.90 for the entire deployment. We did not observe a significant effect of HOx operation on SMn:TMn (0.99 pre-HOx, 0.97 post-HOx).
4. Discussion
4.1 PLSR modeling of high frequency absorbance spectra can predict Fe and Mn concentrations
Using UV-visible absorbance spectra and PLSR modeling, we made hourly predictions of Fe and Mn concentrations at 6 depths in our study reservoir. Our results indicate that this method can successfully predict Fe and Mn concentrations based on their covariability with UV-vis absorbance spectra, despite the paucity of clearly-defined absorbance peaks for these elements. PLSR models were able to explain a high proportion of the variability in the sampling data (Table 1) and predictions agreed with expected Fe and Mn cycling dynamics. For example, the rapid decline in SFe:TFe following the onset of HOx operation (Figure 6C) matches expectations based on the rapid oxidation kinetics of Fe(II) in the presence of oxygen (Davison & Seed 1983); previous studies have also demonstrated substantial decreases in soluble Fe following short periods of HOx (Dent et al. 2014, Munger et al. 2016, Krueger et al. 2020). Based on model skill metrics (i.e., R2 and RMSEP) and visual inspection of the predicted time series, accurate predictions of Fe and Mn concentrations using this method are influenced by numerous factors, including: the range and variance of concentrations in the calibration dataset, the sample size used for calibration, the number of outliers in the calibration dataset, the number of components in the PLSR model, and the inherent predictability of each variable at a particular site (i.e., the strength of correlation with the UV-vis absorbance spectra).
Our results suggest that our methodology may be most appropriate for measuring elevated concentrations of Fe and Mn (> 0.1 mg/L). This result agrees with Vaughan et al. (2018), who suggested that the application of this method to predict riverine total phosphorus (TP) concentrations was best for sites with elevated TP (>0.1 mg/L) concentrations. In our study, PLSR models fit to data with lower concentrations of Fe and Mn (< 0.1 mg/L) generally did not perform well. For example, soluble Fe during the Turnover Deployment had median concentrations of 0.06 mg/L and 0.02 mg/L in the epilimnion and hypolimnion, respectively (Figure 3 and Table S1). Accordingly, the PLSR models for soluble Fe had the lowest R2 (epilimnion: 0.74; hypolimnion: 0.06) and highest RMSEP relative to median calibration concentration out of any model for the Turnover Deployment (Tables 1, S1).
Our PLSR models were also sensitive to the range and variance of sampling data used for calibration. Preliminary model testing revealed that PLSR models were hindered by the distinct water chemistry between epilimnetic and hypolimnetic depths (Fe and Mn mean differences >1.3 mg/L and 0.8 mg/L, respectively; see Figure 3) and therefore models were generated separately for each reservoir layer. This conforms with findings of previous studies using in situUV-vis spectrophotometers and PLSR in waterbodies, which all achieved higher accuracy with site-specific models (Avagyan, Runkle, & Kutzbach 2014, Vaughan et al. 2018, Etheridge et al. 2014). However, when comparing pairs of PLSR models (i.e., the same variable + depth combination) between the two deployments, the models fit to data with a higher standard deviation had higher R2 values, with the sole exception of hypolimnetic total Fe (Tables 1 and S1). These results suggest that there is a tradeoff between capturing the maximum variability in observed concentrations and the limitations imposed by the degree of covariability between the UV-vis absorbance spectra and the variable of interest (also observed by Avagyan, Runkle, & Kutzbach 2014 and Allen 2021). To achieve an accurate predictive model, grouping data based on the spatial and temporal context of measurement achieved a better fitting model while still maximizing the variability captured in the calibration data.
Birgand et al. (2016) used a similar approach for making predictions of soluble Fe concentrations in FCR after the activation of a HOx system. They obtained a slightly better model fit, indicated by an R2 value of 0.94, compared to our R2values of 0.79 and 0.75 (epilimnion and hypolimnion, respectively) for the Oxygen On Deployment. We used calibration sample sizes of 48 and 45 (epilimnion and hypolimnion, respectively) while Birgand et al. (2016) used 27. However, they used 5 components in their PLSR model, whereas we used 4 components. Thus, the higher R2 value for their model may be attributed to a higher ratio of components to sample size (18%) compared to our study (8-9%).
4.2 Fe and Mn Concentrations Change Gradually in Response to Weakening Stratification and Rapidly in Response to Full Turnover
Trends in predicted Fe and Mn concentrations shed light on the changes that occurred in the reservoir before and after turnover. Hypolimnetic concentrations of Fe and Mn began declining 17 and 9 days prior to turnover, respectively, and shorter periods of more rapid concentration fluctuations were superimposed upon these patterns of decline (Figures 4D-E). Combined, these results suggest that turnover, at least in our study reservoir, is not a discrete event, but rather a process occurring over an extended time period. McMahon (1969) measured a similar decrease in soluble Fe using daily samples for nine days across spring mixing in a dimictic lake; soluble Fe concentrations decreased by more than one order of magnitude 5 days prior to full circulation. McMahon (1969) did not offer any interpretation of this phenomenon, simply stating that the changes in soluble Fe were concurrent with vernal circulation. Similar trends have also been observed in other parameters of biogeochemical relevance. For example, Kankaala et al. (2007) found that the majority of CH4 in the hypolimnion of a lake was microbially oxidized at the oxycline boundary during a month-long period of weakening stratification before complete mixing occurred, resulting in lower effluxes of CH4 to the atmosphere during turnover.
Predicted Fe and Mn concentration data can be compared to other time series data to infer mechanisms behind the declining Fe and Mn concentrations prior to turnover. Based on trends in Schmidt stability and water temperature (Figures 4A-B), reservoir stratification was weakening for a 9-day period prior to full turnover, in response to daily and hourly shifts in meteorological conditions, including air temperature and wind speed (Figure S13). Mixing between the hypolimnion and metalimnion, as indicated by the homogenization of water temperature between these layers, occurred periodically throughout the deployment, with an increasing frequency as turnover approached (Figures 4A-B, S15). These ephemeral periods of mixing between the hypolimnion and metalimnion likely led to exchange of Fe and Mn between layers, which suggests that hydrodynamic processes occurring on hourly to daily time scales may have a substantial influence of Fe and Mn cycling. However, without Fe and Mn concentration data at a high spatiotemporal resolution, these patterns would not be observed.
The flexibility of using a multiplexor-spectrophotometer system with a customized prediction algorithm (e.g., site-specific PLSR models) allows for the quantification of high-resolution elemental stoichiometry by making predictions of both the soluble and total fractions of Fe and Mn. During the Turnover Deployment, Fe was predominantly composed of the total fraction, whereas Mn was largely composed of the soluble fraction until approximately one week before turnover, at which time the SMn:TMn ratio began to decline (Figure 6B). This coincided with the onset of declining total Mn concentrations that continued until turnover, excluding a 2-day period from 28 October to 30 October 2020 when total Mn concentrations temporarily increased (Figure 4E). The shift to declining SMn:TMn and total Mn concentrations also coincided with increased frequency of mixing between the metalimnion and hypolimnion and declining stratification intensity (Figures 4A-B and 6B). These trends suggest that declining total Mn concentrations in the pre-turnover period were the result of increased oxidation of Mn(II), perhaps due to the exposure of Mn(II) in the hypolimnion to Mn-oxidizing microbes that inhabit the metalimnion, as demonstrated by a previous study at FCR showing that the presence of Mn-oxidizing microorganisms can substantially increase Mn oxidation rates (Munger et al. 2016).
4.3 Hypolimnetic Oxygenation Causes Oxidation of Fe, but not Mn
The MUX-spectrophotometer system enabled us to observe Fe and Mn concentration changes in response to hypolimnetic oxygenation at an unprecedented spatiotemporal resolution. Fe and Mn concentrations in the hypolimnion both spiked in the 48 hours following oxygenation, then declined (Figures 5D-E). However, Fe concentrations decreased to levels lower than they were prior to oxygenation, especially at the lowest depth, whereas Mn concentrations declined to approximately the same levels prior to oxygenation (Figures 5D-E). These results indicate that the HOx system effectively physically mixed the hypolimnion with respect to both metals, as total Fe and total Mn concentrations quickly converged across hypolimnetic depths after turning on the HOx system (Figures 5D-E) The physical mixing induced by the HOx system appeared to affect Fe and Mn similarly, suggesting that the spike in total Fe and Mn immediately following HOx activation was a result of increased mixing and/or entrainment of particulates in the hypolimnion due to stirring of the bottom sediments. The convergence of Fe and Mn concentrations across hypolimnetic depths has previously been observed in response to HOx activation (Gerling et al. 2014), but results from this study reveal that this can occur in less than 24 hours, and may subsequently be followed by an ephemeral spike in total Fe and Mn concentrations.
Concentrations of total Fe and Mn displayed much greater short-term variability prior to HOx activation than they did post-activation. This was especially pronounced at the lowest depth (9 m) where concentrations fluctuated significantly over a period of less than 24 hours (Figures 5D-E). Given that the SFe:TFe ratio in the upper and middle hypolimnion (6.2m and 8m) steadily increased during the pre-HOx period (Figure 6C), likely due to diffusion of soluble Fe out of the lower hypolimnion, the rapid fluctuations in total Fe in the lower hypolimnion may have been attributed to shifting diffusion gradients. However, similar patterns in short-term variability were observed in Fe and Mn, despite the fact that Mn was predominantly in the soluble phase for the entire deployment, suggesting that diffusion of soluble Mn out of the lower hypolimnion was not responsible for the pre-HOx rapid fluctuations observed at 9 m.
The change in redox conditions caused by adding DO to the hypolimnion had a much more pronounced effect on Fe than Mn, as has been observed in other studies (e.g., Gantzer et al. 2009). The contrasting responses of Fe and Mn to oxygenation can be seen most clearly in the resulting changes in soluble:total ratios (Figure 6). The SFe:TFe ratio in the hypolimnion exhibited a nearly constant linear decline in the 48 hours post-oxygenation and remained below 0.25 for the remainder of the deployment. This indicates that soluble Fe in the water column was rapidly oxidized by the HOx system, even though there was no measurable increase in hypolimnetic DO. This is further supported by the fact that the mean hypolimnetic total Fe concentration was consistently lower after HOx operation began than it was previously. The observed trends in SFe:TFe ratios agree with previous research on the effects of HOx systems on Fe in lakes and reservoirs. For example, Dent et al. (2014) found that SFe:TFe declined to 0.58 after 8 hours of hypolimnetic oxygenation. In our study, it took approximately twice as long (16 hours) for SFe:TFe to reach 0.58. However, the Fe concentrations in Dent et al. (2014) were lower (0.17 - 2.88 mg/L) than those in our study (0.31 - 7.42 mg/L).
In contrast to Fe, the SMn:TMn ratio in the hypolimnion displayed only a very slight response (approximately 2% decrease) to HOx activation, demonstrating that hypolimnetic oxygenation did not result in significant oxidation of Mn. Our results agree with those from Dent et al. (2014), who found that Mn was still 100% in the soluble phase 8 hours after oxygenation. Furthermore, previous studies at FCR have also showed that soluble Mn does not respond significantly to oxygenation alone and that other factors, such as microbially-mediated oxidation, reservoir pH (range 6.4 - 7.1 observed in the hypolimnion during our study) and dilution from physical mixing, are more important variables impacting hypolimnetic soluble Mn than oxygenation (Munger et al. 2016, Krueger et al. 2020).
4.4 Study Limitations
The MUX pumping system enabled us to monitor multiple depths simultaneously, which is invaluable for investigating biogeochemical processes in spatially heterogeneous systems such as thermally-stratified reservoirs. However, there are several limitations to be improved upon in future research. In our reservoir, the cuvette fitted on the spectrophotometer experienced fouling, likely due to Fe and Mn in the hypolimnion that oxidized and precipitated on the cuvette walls upon exposure to oxygen. Despite our efforts to limit fouling (see Methods), there was still a fouling signal detected in several periods of our time series (Figures S4-5). PLSR models provided a remarkably good numerical correction for this fouling signal, indicating that the collection of additional calibration samples obtained at regular intervals between servicing dates may yield lower uncertainties in future deployments. We also found that truncating the UV-vis absorbance spectra used for calibration to only include wavelengths greater than 250 nm substantially improved the model skill and diminished spikes in the time series of predictions that corresponded to periods of heavy fouling (Figures S6-7).
Our results captured sub-weekly patterns in Fe and Mn dynamics in FCR, but the PLSR-predicted time series of Fe and Mn concentrations was not able to adequately capture some of the high-magnitude, sub-daily fluctuations that were observed in the sampling data (Figures 4 and 6). This is likely due to varying PLSR model skill, which is related to the sample size and distribution of data used for calibration, the number of PLSR model components, and the inherent predictability of each variable. Therefore, it follows that the strength of correlation between the UV-vis absorbance spectra and Fe/Mn concentrations plays a strong role in determining the limits to the temporal resolution. This relationship can be refined through the methodological suggestions outlined above, but ultimately depends upon the spectral properties of the study system.
5. Conclusions
Results from this study demonstrate that coupling a spectrophotometer with a pumping system enabled unprecedented high-frequency monitoring of Fe and Mn at multiple depths in our study reservoir, providing a unique ability to observe hour-resolution biogeochemical dynamics in a freshwater ecosystem. Our findings underscore the importance of implementing robust and consistent methodologies for obtaining calibration concentrations, choosing the number of components in PLSR models, and quantifying the uncertainty around predictions.
The high-spatio-temporal resolution predictions provide novel insights into Fe and Mn cycling dynamics that could improve aquatic monitoring programs and reservoir management practices. First, we demonstrated that Fe and Mn concentrations can fluctuate significantly on time scales much shorter than those employed by most traditional monitoring programs. For example, sub-daily fluctuations of total Fe and Mn during the Oxygen On Deployment resulted in concentration changes of up to 1.62 mg/L/hr and 0.19 mg/L/hr, respectively. Considering that the secondary drinking water standards for Fe and Mn are 0.3 and 0.05 mg/L, respectively, sub-daily concentration changes of this magnitude are critical for water quality management. Second, we observed an increase in total hypolimnetic Fe and Mn in response to the re-stratification of our study reservoir two days after turnover, which contradicts the common assumption that metals concentrations equalize and remain consistently low during the mixed period following turnover. Last, our results offer new insights on the rapid response of Fe to hypolimnetic oxygenation; within hours of activating the system, the soluble to total Fe ratio indicated oxidation of Fe, even though there was no measurable increase in DO. This study emphasizes the power of high spatiotemporal resolution data for improving our understanding of biogeochemical cycles by unveiling previously-unobserved processes altering Fe and Mn cycling.