Authorship Statement
NWH, FB, and MES co-conceived the design of the study. NWH led field sampling, and MES, BB, and ABP assisted in field sampling. FB, CCC, and BB conducted the initial field testing for the MUX and development of the PLSR analysis workflow. ABP led the collection of auxiliary sensor data. FB, CCC, and MES contributed to conceptual development, which substantially improved the quality of the manuscript.
Abstract
The biogeochemical cycles of iron (Fe) and manganese (Mn) in lakes and reservoirs have predictable seasonal trends, largely governed by stratification dynamics and redox conditions in the hypolimnion. However, short-term (i.e., sub-weekly) trends in Fe and Mn cycling are less well-understood, as most monitoring efforts focus on longer-term (i.e., monthly to yearly) time scales. The potential for elevated Fe and Mn to degrade water quality and impact ecosystem functioning, coupled with increasing evidence for high spatiotemporal variability in other biogeochemical cycles, necessitates a closer evaluation of the short-term Fe and Mn cycling dynamics in lakes and reservoirs. We adapted a UV-visible spectrophotometer coupled with a multiplexor pumping system and PLSR modeling to generate high spatiotemporal resolution predictions of Fe and Mn concentrations in a drinking water reservoir (Falling Creek Reservoir, Vinton, VA, USA) equipped with a hypolimnetic oxygenation (HOx) system. We quantified hourly Fe and Mn concentrations during two distinct transitional periods: reservoir turnover (Fall 2020) and initiation of the HOx system (Summer 2021). Our sensor system was able to successfully predict mean Fe and Mn concentrations as well as capture sub-weekly variability, ground-truthed by traditional grab sampling and laboratory analysis. During fall turnover, hypolimnetic Fe and Mn concentrations began to decrease more than two weeks before complete mixing of the reservoir occurred, with rapid equalization of epilimnetic and hypolimnetic Fe and Mn concentrations in less than 48 hours after full water column mixing. During the initiation of hypolimnetic oxygenation in Summer 2021, we observed that Fe and Mn were similarly affected by physical mixing in the hypolimnion, but displayed distinctly different responses to oxygenation, as indicated by the rapid oxidation of soluble Fe but not soluble Mn. This study demonstrates that Fe and Mn concentrations are highly sensitive to shifting DO and stratification and that their dynamics can substantially change on hourly to daily time scales in response to these transitions.
Keywords: Hypolimnetic Oxygenation, Iron, Manganese, Spatiotemporal resolution, Spectrophotometer, Turnover
Highlights:
1. Introduction
Elevated levels of iron (Fe) and manganese (Mn) in lakes and reservoirs have negative consequences for ecosystem health and water quality. Increasing Fe concentrations have been linked to the long-term browning of lakes, which has numerous, significant ecological consequences (Kritzberg et al. 2020). Mn contamination of drinking water can pose serious risks to human health, especially in children (Wasserman et al. 2006). Furthermore, high concentrations of both metals negatively affect the taste, odor, and appearance of water and can damage water supply infrastructure through corrosion and deposition (World Health Organization 2017). As a result, the U.S. Environmental Protection Agency (EPA) has established secondary standards for Fe and Mn concentrations in drinking water of 0.3 and 0.05 mg/L, respectively (EPA 2021).
As Fe and Mn are redox-sensitive elements, their abundance in aquatic systems is largely influenced by dissolved oxygen (DO) concentrations (Hem 1972, Davison 1993). The oxidation state of Fe and Mn in natural waters is dominated by two forms: insoluble, oxidized Fe(III) and Mn(IV), and soluble, reduced Fe(II) and Mn(II) (Davison 1993). In most aquatic systems under circumneutral pH, this oxidation state is determined by the redox conditions at a given point in space and time. Under oxic conditions, Fe and Mn are generally present as insoluble Fe(III) and Mn(IV) solids in rocks and sediments. However, thermal stratification in lakes and reservoirs can create anoxic conditions in the hypolimnion and bottom sediments, promoting the microbial reduction of Fe and Mn in sediments and the subsequent release of soluble, reduced Fe and Mn into the water column (Lovely 1991). In such settings, soluble Fe and Mn can accumulate in hypolimnetic waters throughout the stratified period (McMahon 1969, Davison 1993, Beutel et al. 2008, Munger et al. 2016, Krueger et al. 2020).
An increasingly used in situ approach for mitigating high Fe and Mn in drinking water reservoirs is hypolimnetic oxygenation (HOx), which creates oxic conditions in previously anoxic waters and creates a thicker aerobic zone in bottom sediments (e.g., Beutel and Horne 1999, Bryant et al. 2011, Dent et al. 2014, Gantzer et al. 2009, Gerling et al. 2014). By increasing oxygen availability in the hypolimnion, HOx operation hinders the release of soluble Fe and Mn into sediment pore waters, slows upward diffusion into the water column, and promotes Fe and Mn oxidation and precipitation in the hypolimnion (Preece et al. 2019). HOx systems have been shown to effectively reduce soluble Fe and Mn in the hypolimnion of drinking water reservoirs (Gantzer et al. 2009, Bryant et al. 2011). However, removing soluble Mn from the water column requires more sustained oxygen inputs, due to its slower oxidation reaction kinetics (Bryant et al. 2011, Munger et al. 2016). To optimize water treatment using HOx systems, it is essential for drinking water managers to understand both the short-term (sub-weekly) and long-term (monthly to yearly) dynamics of Fe and Mn cycling in supply reservoirs.
Although Fe and Mn cycling in temperate lakes and reservoirs has predictable seasonal trends dictated by thermal stratification, there is a lack of research on short-term Fe and Mn dynamics. Quantifying short-term trends requires high-frequency data, which we define as having a temporal resolution of daily or shorter. To our knowledge, there is no standard definition for classifying data as ‘high-frequency’ or trends as ‘short-term.’ Thus, we developed operational definitions based on the contrast with traditional monitoring frequencies, which are typically weekly or longer (e.g., Marcé et al. 2016). The paucity of previous research on Fe and Mn cycling at sub-weekly scales represents a key knowledge gap, given that biogeochemical process rates can fluctuate rapidly over hourly to daily time scales (McClain et al. 2003). Studies have identified diel signals in the cycles of numerous biogeochemical variables, including Fe and Mn, and many biological and chemical processes in aquatic environments operate on hourly to daily scales, often with significant impacts on nutrient cycling and ecosystem productivity (Istvánovics, Osztoics, & Honti 2004, Nimick, Gammons, & Parker 2011, Kurz et al. 2013). Additionally, episodic hydrologic events, which may be missed by traditional sampling methods, can have pronounced effects on biogeochemical cycling dynamics (e.g., Marcé et al. 2016, Coraggio et al. 2022).
Studies analyzing the efficacy of HOx systems have observed substantial changes in Fe and Mn concentrations in response to changes in DO concentrations (Dent et al. 2014, Munger et al. 2019). For example, Dent et al. (2014) found that total Fe and Mn concentrations decreased by 71% and 73%, respectively, after 8 hours of oxygenation of a previously-anoxic reservoir hypolimnion. Conversely, Munger et al. (2019) found that Fe and Mn sediment fluxes into the water column were 1.4 and 4.5 times higher, respectively, two weeks after the onset of hypolimnetic anoxia in a reservoir. The dynamic behavior of Fe and Mn concentrations in response to both management and natural processes (e.g., seasonal thermal stratification) underscores the importance of quantifying these complex cycling dynamics, which could have substantial implications for drinking water management and water quality monitoring. To date, monitoring programs have been hindered by the coarse temporal frequency of months to seasons necessitated by traditional manual sampling and laboratory analysis techniques.
Recent developments in sensor technology have enabled high-frequency collection of some water quality variables in situ , without the need for manual sampling and laboratory analysis (Porter et al. 2009, Rode et al. 2016, Kruse 2018). However, most high-frequency sensors are only capable of measuring a single variable at a time and typically have a low spatial resolution. Moreover, numerous water quality variables, including Fe and Mn, lack instrumentation capable of unattended, reagent-less, high-frequency measurement.
To circumvent the limitations of current sensor technology, spectrophotometers have been designed to measure water quality variablesin situ at a high frequency using multi-wavelength absorbance patterns in the ultraviolet-visible (UV-vis) spectrum. These sensors do not require chemical reagents and are capable of measuring multiple variables simultaneously. To date, UV-vis spectrophotometers have been successfully used to measure chemical variables that have a strong correlation with known peaks in their absorbance spectra, such as nitrate and dissolved organic carbon (DOC) (Etheridge et al. 2014, Sakamoto, Johnson, & Coletti 2009). Additionally, several studies have had success using them to measure concentrations of other biogeochemical variables without well-defined spectral peaks, such as Fe, total phosphorus (TP), soluble reactive phosphorus (SRP), and dissolved silica (Si) (Birgand et al. 2016, Etheridge et al. 2014, Vaughan et al. 2018). Although Fe and Mn are not known to have well-defined spectral peaks, they absorb and scatter light at wavelengths across the UV-vis spectrum and they can affect the absorbance of a water sample through complexation with organic molecules (Poulin et al. 2014, Weishaar et al. 2003, Xiao et al. 2013). Therefore, the covariance between the variable of interest (e.g., Fe or Mn) and the overall “color matrix” of the water (the combination of multiple light-sensitive proxies) can be detected in the UV-vis absorbance spectra and used to predict concentrations of the variable of interest with statistical algorithms (Birgand et al. 2016). Laboratory-measured concentrations from manually collected samples are then subsequently used to develop predictive models that correlate known concentrations with absorbance spectra.
Numerous algorithms exist for calibrating UV-vis absorbance spectra to observed concentrations, but the most commonly-employed method is partial least squares regression (PLSR) (DiFoggio 2000, Birgand et al. 2016, Vaughan et al. 2018). PLSR is well-suited for modeling relationships within data that have a large number of highly correlated explanatory variables and relatively few observations, such as multi-wavelength spectral measurements (Wold et al. 2001). Previous studies have used in situ spectrophotometers and PLSR models to predict water quality variables in a variety of environments, including streams, lakes, estuaries, and oceans, with varying levels of predictive accuracy (Sakamoto, Johnson, & Coletti 2009, Avagyan, Runkle, & Kutzbach 2014, Etheridge et al. 2014, Birgand et al. 2016, Vaughan et al. 2018). However, to the best of our knowledge, only one study (Birgand et al. 2016) has evaluated the potential of this method to observe the high-frequency dynamics of metals in lakes and reservoirs.
Because the application of this method is relatively new, only a few studies have attempted to quantify the uncertainty of water chemistry predictions made using PLSR and spectrophotometric data (Bieroza & Heathwaite 2016, Vaughan et al. 2018). Uncertainty quantification is crucial for determining the accuracy and feasibility of these methods, especially in natural waters that have a complex chemical composition with an unknown relationship to the measured spectrophotometric color matrix (Bieroza & Heathwaite 2016; Rieger, Langergraber, & Siegrist 2006; Vaughan et al. 2018). Furthermore, the ability of a PLSR model to make accurate predictions is contingent upon how well the training data capture the variability in predicted concentrations, which in turn influences the generalizability of the model (Wold et al. 2001, DiFoggio 2000). In addition to these factors, photometric noise (i.e., random differences in spectral measurements) and spectral artifacts (e.g., instrument drift and fouling) can introduce error into model predictions (DiFoggio 2000).
Thus, in situ spectrophotometer data coupled to PLSR modeling potentially offer useful insight on rapidly changing metals concentrations in reservoirs and lakes. However, because of strong thermal gradients with depth in lakes, a single spectrophotometer cannot capture metals concentrations that may also rapidly change with depth. Additionally, the cost of in situ spectrophotometers ($8000-25000 USD as of 2022) prohibits the acquisition of many needed to characterize spatial dynamics as well. For this reason, Birgand et al. (2016) designed a multiplexed sequential sampling system to pump water from different depths at one site to one spectrophotometer used as a portable lab above the water’s surface. This system has proven to be able to characterize variable reservoir biogeochemical concentrations over both depth and time (Birgand et al., 2016).