Sediment organic carbon (C) burial and CO2 fluxes in inland waters are quantitatively important in regional and global carbon budgets. Estimates of C fluxes from inland waters are typically based on limited temporal resolution despite potential large variations with season and weather events. Further, most freshwater C budget studies have focused on natural soft-water lakes, while reservoirs and hard-water systems are globally numerous. Our study quantifies C fluxes in two hard-water, human constructed reservoirs (Ohio, USA) of contrasting watershed land use (agriculture vs. forest) using high-resolution mass balance budgets. We show that during a dry summer, C retention and export via the dam were reduced compared to a wet summer. Both reservoirs were net CO2 sources during a wet summer, but CO2 sinks during a dry summer. Despite weather-related summer differences, annual C fluxes within each reservoir were similar between years. Both reservoirs appear to be net autotrophic despite often being CO2 sources based on budgets. This is likely because CO2 fluxes in our hard-water reservoirs were more strongly associated with DIC than DOC. Using our C fluxes and statewide watershed land use, we determined the regional importance of Ohio reservoirs in OC burial and CO2 emissions. We estimate that Ohio reservoirs bury up to 4 times more OC, but emit <25% of CO2, than predicted based on their area and recent global mean estimates in lentic ecosystems. Our results provide evidence that moderately old (~50 years), temperate hard-water reservoirs are important OC sinks but contribute little to CO2 emissions.
 A desire to balance the global carbon budget was the main impetus for quantifying carbon (C) transport, storage, and emissions in inland waters (lakes, reservoirs, and ponds) and streams in the 1980s and 1990s [Kling et al., 1991; Mulholland and Elwood, 1982; Schlesinger and Melack, 1981]. Historically, many believed that inland waters would play an insignificant role in regional and global C cycling, because they comprise only about 3% of the earth's continental surface [Downing et al., 2006]. Despite their small global extent, inland waters receive, process, sequester, and release carbon at high rates, rendering them relevant to regional and global carbon budgets [Cole et al., 2007; Tranvik et al., 2009]. For example, inland waters receive large quantities of terrestrial organic carbon (generated by terrestrial primary production), and some of this carbon is permanently buried in their sediments and represents a significant global C sink [Cole et al., 2007; Dean and Gorham, 1998; Mulholland and Elwood, 1982]. Global annual burial of organic carbon (OC) in the sediments of lakes (~50 Tg yr-1) and reservoirs (~180 Tg yr-1) exceeds that buried in ocean sediments (120 Tg yr-1) [Cole et al., 2007; Dean and Gorham, 1998; Sarmiento and Sundquist, 1992; Tranvik et al., 2009].
 Many inland waters are also considered net heterotrophic, i.e., they are sources of CO2 to the atmosphere [Cole et al., 1994]. CO2 emissions by inland waters and streams are estimated to be 1.4 Pg C yr-1, which is a relevant value in terms of global carbon cycling [Tranvik et al., 2009]. Carbon inputs to the majority of north temperate and boreal soft-water lakes are dominated by dissolved organic carbon (DOC) [Dillon and Molot, 1997; Rantakari and Kortelainen, 2008]. Terrestrial DOC inputs increase in-lake microbial production and respiration above what is possible without this addition and allows lakes to be supersaturated with CO2 [Duarte and Prairie, 2005]. Because terrestrial DOC inputs fuel lake CO2 fluxes, it could be argued that aquatic CO2 emissions are part of the terrestrial carbon cycle, but traditional measures of terrestrial net ecosystem exchange would not quantify these emissions via inland waters [Buffam et al., 2011]. Importantly, inclusion of aquatic CO2 fluxes to regional carbon budgets can turn a region from being a sink of atmospheric carbon to being equilibrated with the atmosphere [Richey et al., 2002] and more adequately describe the overall regional C budget [Buffam et al., 2011].
 Another consideration with CO2 fluxes in lakes, is that most C flux studies focus on soft-water and boreal lakes, and recent work suggests that they process carbon very differently than saline and hard-water lakes [Duarte et al., 2008; Finlay et al., 2010; Lopez et al., 2011; Tranvik et al., 2009]. Hard-water lakes are globally abundant [Wetzel, 2001]; and in these systems, terrestrial C inputs are often dominated by inorganic C. The regulation of CO2 fluxes in hard-water systems appears to be driven by hydrological inputs of inorganic C rather than by in-lake metabolism [Duarte et al., 2008; Finlay et al., 2009; Finlay et al., 2010; Lopez et al., 2011; Stets et al., 2009]. For example, hydrological inputs of inorganic carbon can lead to in-lake efflux of CO2 through either direct atmospheric exchange or as the evolution of CO2 from the formation of CaCO3 [Finlay et al., 2010; Kling et al., 1992; Stets et al., 2009; Striegl and Michmerhuizen, 1998]. These losses are not related to microbial activity and can exceed CO2 efflux from biological metabolism [Finlay et al., 2010; Stets et al., 2009].
 Given that different inland water ecosystems may have contrasting dominance of OC and IC inputs and subsequently divergent in-lake C processing, it is not surprising that inland waters from varying landscapes and latitudes with disparate productivities, display a large range of CO2 fluxes, carbon burial, and downstream export of carbon [Tranvik et al., 2009]. Thus the magnitude and direction of CO2 flux and the magnitude of carbon burial will likely depend on a suite of abiotic and biotic factors including: productivity, geology, land use, lake and watershed morphometry, and water body type (e.g., natural lake, human constructed reservoir, wetland). Reservoirs are globally numerous and continue to be built for irrigation, water supply, hydroelectric power, and recreation [Downing et al., 2006; Smith et al., 2002]. Reservoir carbon fluxes likely differ from natural lakes, primarily because reservoirs often receive elevated carbon and nutrient inputs via streams and because they generally have large watershed area to reservoir area ratios. Both of these landscape attributes can result in the atmospheric drawdown of CO2 by phytoplankton production, elevated burial of terrestrially derived particulate and dissolved carbon, and elevated burial of carbon produced within the reservoir [Cole et al., 2007; Dean and Gorham, 1998; Downing et al., 2008; Hanson et al., 2004; Mulholland and Elwood, 1982; Stallard, 1998; Tranvik et al., 2009]. Work has been conducted in reservoirs on carbon burial in the sediments [Dean and Gorham, 1998; Downing et al., 2008; Mulholland and Elwood, 1982; Ritchie, 1989] and CO2 fluxes in reservoirs [Abril et al., 2005; Demarty et al., 2009; St Louis et al., 2000], but few have simultaneously quantified C inputs, outputs, CO2 fluxes, and burial [Finlay et al., 2010; Lopez et al., 2011]. Additionally, there is a lack of research in hypereutrophic reservoirs, which are expected to have the highest rates of C burial and CO2 influxes [Cole et al., 2007].
 Currently, accurate estimates of regional and global C fluxes in reservoirs are compromised by a lack of well-constrained C budgets for these systems. To accurately determine whether reservoirs are a net C source or sink, simultaneous quantification of burial and CO2 fluxes is needed, but few studies have examined both. Moreover, it is rare to have C flux measurements with enough resolution to determine the importance of temporal variation at both annual and weather-event scales. Past studies have primarily used a multi-site “snapshot” approach, comparing numerous water bodies with limited sampling frequency (often once per water body). This approach allows for comparisons among many systems over a large geographical range, but it does not account for variation in C fluxes among years, seasons or weather events. Given that climate change is already altering precipitation patterns [Karl et al., 2008], we need to understand how C fluxes and net C balance in reservoirs vary over different time scales. Intensive studies are needed to complement the multi-site approach. Infrequent sampling also limits confidence in extrapolated annual fluxes, i.e., the error associated with an annual flux rate is either not estimated or can be relatively large [Lehrter and Cebrian, 2010].
 We constructed carbon budgets to: 1) examine whether two hard-water reservoirs of contrasting watershed land use (agricultural versus forested) are net C sources or sinks, 2) evaluate whether organic or inorganic carbon appear to drive CO2 fluxes in our hard-water reservoirs, and 3) determine the significance of temporal/weather variability on C fluxes. We predicted that the agricultural reservoir would be a large CO2 sink and large net C sink, while the forested reservoir would be a small CO2 source and moderate net C sink. Fluxes of all carbon forms (DOC; POC, particulate organic carbon; DIC, dissolved inorganic carbon; PIC, particulate inorganic carbon) were measured over two complete water years (WY) using a high-frequency, flow-dependent sampling regime at stream inlets and dam outlets (Figure 1). CO2 atmospheric exchange was measured at a shallow and deep site within each reservoir (Figure 1). High-resolution sampling allowed us to estimate uncertainty in annual fluxes, facilitating comparisons between years and lakes. Using data on the distribution and sizes of reservoirs in the state of Ohio, we then were able to confidently extrapolate our estimates to the regional scale.
2.1 Study Sites
 Acton Lake is a hypereutrophic reservoir, located in southwestern Ohio, USA (39°34′N, 84°44.5′W), with watershed land use dominated by intensive row-crop agriculture (Table 1; Figure 1; Knoll et al., 2003). Acton was built in 1957 and the quantity of agricultural land use in the watershed has not changed drastically over the lifetime of the reservoir. However, row-crops have increased from about 50% of agricultural land in the 1950s to approximately 95% today, with a consequent decrease in pasture [Medley et al., 1995; Renwick et al., 2008]. Row-crop agricultural practices have changed considerably in the watershed with a pronounced increase in conservation tillage in the 1990s [Renwick et al., 2008]. Soils in the Acton watershed are high-lime, glacial till capped with very productive silt loess [Medley et al., 1995]. We quantified carbon fluxes in the three main streams draining into Acton (Figure 1). These streams, Four Mile Creek, Little Four Mile Creek, and Marshall's Branch collectively represent 86% of the watershed drainage [Renwick et al., 2008; Vanni et al., 2001].
Table 1. General Characteristics of the Two Study Reservoirs. Concentrations and pH Represent Simple Means of All Dates Sampled in the Two-Year Time Period (n = 61 for Acton, n = 29 for Burr Oak)
(% of watershed area)
(% of watershed area)
Surface area (km2)
Watershed area (km2)
Mean depth (m)
Residence time (yr)
Chl – a (µg L-1)
Total P (µg L-1)
DOC (mg L-1)
DIC (mg L-1)
POC (mg L-1)
 Burr Oak Lake, in southeastern Ohio (39°31.7′N, 82°2.6′W), is moderately productive and has land use dominated by forests (Table 1; Figure 1; Knoll et al., 2003). Burr Oak was built in 1950; and since this time, the percentage of forest in the watershed has increased from approximately 40% to 81% due to the establishment of the Wayne National Forest in 1934 and the subsequent regrowth of forest from previously cleared land [Birch and Wharton, 1982; Vanni et al., 2011]. Burr Oak's watershed is unglaciated, with soils mainly deriving from sandstone, shale, and siltstone. Carbon fluxes were measured at the East Branch of Sunday Creek, the main stream flowing into the reservoir, which represents 70% of the watershed drainage [Vanni et al., 2011].
2.2 Stream and Dam Carbon/Water Fluxes
 Streams flowing into the reservoirs (three on Acton and one on Burr Oak) were sampled intensively during two water years (1 October 2006 – 30 September 2007 and 1 October 2007 – 30 September 2008) for dissolved and particulate constituents of carbon (DOC, DIC, POC, PIC) and suspended solids (SS). We employed a high-resolution, flow-dependent sampling regime (i.e., more frequent sampling during storm events) utilizing ISCO automated water samplers. During baseflow periods, carbon samples were taken from streams three times per week, and suspended solids (SS) samples were taken once per day using ISCO samplers [Vanni et al., 2001]. During storm events, both carbon and SS samples were taken every 8 hours throughout the entire event. Sampling frequency during storms events was intensified, because during storms the concentrations of suspended sediments and nutrient forms increase significantly [Vanni et al., 2001]. ISCO samplers were also used to collect water exiting the reservoirs via their dam outflow. During low flow periods, outflow samples were taken three times a week; and during high flow periods, samples were taken daily. Samples were collected at a lower frequency at the dam outflows, because we expected concentrations in the outflow to change during storms but not as rapidly as in the inflow streams. Samples from the streams and outflow were collected from ISCO samplers every seven days, and carbon concentrations were not affected by this length of storage time. We processed many samples from each stream inflow or dam outflow for each C form (mean from stream inflow: DIC = 507, DOC = 469, PC = 162, SS = 1892; mean from dam outflow: DIC = 299, DOC = 276, PC = 292).
 DIC was measured on samples passed through pre-combusted Gelman AE glass fiber filters (1.0 µm nominal pore size) using a gas chromatograph (Shimadzu GC-8A) following the syringe gas-stripping method [Stainton, 1973]. Samples for DOC were filtered through a pre-combusted Gelman GFF glass fiber filter (0.7 µm nominal pore size) and subsequently analyzed on a Total Organic Carbon analyzer (Shimadzu TOC-V). Particulate carbon samples were analyzed on a CHN elemental analyzer (Perkin-Elmer Series 2400 CHN Analyzer, Waltham, MA, USA). Prior to analysis, particulate carbon samples were filtered onto pre-combusted Gelman AE glass fiber filters (1.0 µm nominal pore size). Two subsamples were taken in order to determine POC and PIC. To measure PIC, one subsample was ashed at 550 °C for four hours. The other subsample, used to estimate total particulate carbon, was not ashed. POC concentrations were then determined as the difference between the non-ashed and ashed samples. To minimize costs, particulate carbon (PC, i.e., POC plus PIC) was analyzed only on a sub-set of the samples, because previous work in Acton and Burr Oak streams has shown a strong relationship between PC and suspended solids [Renwick et al., 2008; Vanni et al., 2001], which is much less expensive to measure. Therefore, we used stream-specific SS-PC regressions to estimate PC concentrations on samples for which PC was not measured. All regressions using untransformed data between SS and PC had an r2 greater than 0.8. Predicted PC concentrations were then split into PIC and POC constituents based on regressions from the analyzed samples. Since PIC concentrations were often zero or negligible and the percentage of PIC (relative to total PC) did not vary significantly with flow, we assumed PIC was a fixed percentage of PC (Four Mile mean = 5.7%, n = 84; Little Four Mile mean = 5.1 %, n = 69, Marshall's Branch mean = 3.7%, n =78; East Branch mean = 6.1%, n =54). Samples for SS were filtered onto pre-weighed, pre-combusted Gelman AE glass fiber filters (1.0 µm nominal pore size). SS was then determined as the difference between dry mass on pre-combusted AE filters before and after filtering by weighing samples on a Mettler UMT ultra-microbalance (Mettler-Toledo, Columbus, OH, USA).
 For Acton streams, stage was recorded every ten minutes using pressure transducers and dataloggers. Hourly discharge (QSTREAMS) was then calculated using standard rating-curve techniques [Renwick et al., 2008; Vanni et al., 2001]. We were not able to directly record stage on the Burr Oak stream. Hourly QSTREAMS for this stream was therefore calculated using the following method:
where QDAM equals the hourly discharge from the dam outflow. These data were obtained from the US Army Corps of Engineers, Huntington District, which operates the Tom Jenkins Dam at Burr Oak. Hourly precipitation data and temperature data for potential evapotranspiration (PET) were obtained for the DCP114 site (Deer Creek) from the EPA CASTNET (Clean Air Status and Trends Network) database (~ 90 km east of Burr Oak). Change in lake volume was calculated using hourly lake level data (also obtained from the US Army Corps of Engineers) and lake bathymetry.
 Carbon and SS loading from the streams were calculated similarly to previous studies [Renwick et al., 2008; Vanni et al., 2001]. These loading estimates follow a commonly used method that has been used for decades for studies in which samples are collected frequently [Porterfield, 1972]. Hourly loading was calculated using the following equation:
where, Lh is hourly loading, Ch is the C concentration for hour h, and QSTREAMSh is the mean discharge for hour h. When there was a sample taken during a particular hour, the concentration from that sample was assumed for the entire hour. For hours when no samples were taken, we used a discharge (Q) proportionate interpolation method for DIC, DOC, and SS [Vanni et al., 2001]. Q-proportionate interpolation can be used when a strong relationship between discharge and concentration is found, because it adjusts for variations in concentration due to variations in discharge and changes in discharge that occur in between sample points. For this interpolation method, the slope of logQSTREAMSh - logCh regression is used. Residuals from the regression are linearly interpolated though time and applied to the calculation of concentrations:
where B0 and B1 are the intercept and slope of the logQSTREAMSh - logCh regression, and Rh is the interpolated residual from that regression [Vanni et al., 2001]. Once hourly loadings were obtained, they were then summed to obtain daily, monthly and annual loadings.
 Discharge via dams (QDAMh) was calculated differently for Acton and Burr Oak due to differences in data availability. For Acton, hourly QDAMh was calculated by taking into account water inputs and outputs using the following equation:
where QSTREAMSh equals the sum of discharge from the three gaged streams divided by 0.86 to scale up to entire watershed discharge. Hourly precipitation and temperature for potential evapotranspiration (PET) were obtained from the EPA CASTNET (Clean Air Status and Trends Network) database at the OXF122 site (Oxford) located ~5 km from Acton. Change in lake volume was calculated using hourly lake level data (continuously recorded via a lake level gauge) and lake bathymetry. For Burr Oak, direct values for QDAMh were available, because these data are recorded hourly by the US Army Corps of Engineers, Huntington District.
 Carbon exports via the dams were generally calculated the same as stream loadings. However, only simple interpolation was used to interpolate missing hourly carbon concentrations, since relationships between discharge and C concentrations are not particularly strong at dam outflows. Similarly, we did not find strong relationships between SS and PC, so we measured PIC and POC on all samples using methods described above for streams. Hourly export was calculated as hourly concentration (Ch) multiplied by hourly discharge (QDAMh). Hourly exports were then summed to get daily, monthly or annual exports.
2.3 Carbon Dioxide Fluxes
 In order to calculate atmospheric flux of CO2, atmospheric and lake partial pressure CO2 (pCO2, µatm) were determined. Lake pCO2 was quantified using two methods; it was either directly quantified from reservoir water, or using an established method, it was estimated from surface water DIC and pH measurements, correcting for temperature and ionic strength [Cole et al., 1994; Kling et al., 1992]. Samples for direct pCO2 measurements were collected at two sites within the reservoirs two to four times per month, when boat sampling was possible. One site was located near stream inlets at a shallow lake site, and the other was near dam outlets at a deep lake site (Figure 1). At each site, pH, temperature, and conductivity were taken with calibrated hand-held meters (YSI Model 60, YSI Model 30, respectively), and a sample for DIC was collected at 0.1m. Directly measured pCO2 samples were collected in duplicate in gas-tight, glass syringes. Surface water pCO2 was quantified using the headspace equilibration method at a depth of 0.1m [Cole et al., 1994; Raymond et al., 1997]. Atmospheric pCO2 in the ambient air, 1m above the reservoir surface, was collected at the same time as surface water pCO2. Surface water and atmospheric pCO2 samples were measured in the laboratory using a gas chromatograph fitted with a thermal conductivity detector (Shimadzu GC-8A). To verify that calculated pCO2 estimates were comparable to those from the headspace equilibrium method, we quantified pCO2 using both methods from Burr Oak and Acton for numerous dates and found a strong relationship (both lakes: n = 120, r2 = 0.72).
 Chemically enhanced CO2 flux calculations (mmol m-2 day-1) for each sampling date were calculated using the following equation:
where FCO2 is atmospheric CO2 flux, CO2(H2O) is the measured surface-water CO2 concentration, and CO2(eq) is the equilibrium CO2 concentration adjusting for Henry's constant for CO2 [Plummer and Busenberg, 1982]. Piston velocity (cm hr-1), k, can be calculated for a lake with an established equation [Cole and Caraco, 1998]; however, k values must be adjusted if pH is > ~8.0 for substantial periods of time, because chemical reactions at elevated pH can affect the diffusion rate [Bade and Cole, 2006; Wanninkhof and Knox, 1996]. Acton and Burr Oak reservoirs often have pH above 8.0; thus, piston velocity must be corrected for chemically enhanced diffusion. The chemical enhancement factor, α, was quantified using established equations based on gas piston velocity, temperature, and pH [Bade and Cole, 2006; Stets et al., 2009; Wanninkhof and Knox, 1996]. Piston velocity, corrected for chemically enhanced diffusion, kenh, was calculated as kenh = αk.
 Because it was not feasible to sample the reservoirs during winter months, we were unable to measure CO2 fluxes year-round. For Acton, we could sample CO2 fluxes during nine months of the year. For two of the remaining months, the reservoir was generally covered by ice so there would be no CO2 movement between the reservoir and atmosphere. For the remaining month (when ice cover was intermittent and unstable), an average of all the measured fluxes was used. CO2 fluxes were directly measured from Burr Oak seven to eight months of the year. For the remaining ice-free months, we estimated pCO2 using temperature and pH data available from the Burr Oak Regional Water District and using DIC concentrations collected from the dam outflow using automated water samplers (see above).
2.4 Carbon Mass Balance Budgets
 To create total carbon budgets, a mass-balance approach was applied using the following equation:
where CRETENTION is the net retention of carbon in each reservoir, CSTREAMS is the loading of carbon into the reservoirs via stream inlets, CO2FLUX is positive if there is an efflux and negative if there is an influx, and CDAM is the export of carbon from the reservoir via dam outlets.
 Carbon retention in our budgets was calculated by difference of C inputs and C outputs, because sediment re-suspension and the age of the reservoirs (< 60 years old) make it difficult to obtain accurate C burial rates on a yearly time-scale. We feel this approach is appropriate because our rates are similar to long-term burial rates [Vanni et al., 2011], and the annual change in C mass within the water columns in either water year represented only 0.2-3% of C stream inputs minus C dam outputs. To determine the relative contribution OC and IC to retention, we applied the percentage of OC or IC found in the sediments from long-term sediment coring data in these reservoirs (55% OC in Acton, 99% OC in Burr Oak [Vanni et al., 2011]) to the total C retention. To estimate the mass of incoming IC converted to OC or consumed as alkalinity, we subtracted IC inputs (stream loading and influxes of CO2) from IC outputs (dam export and effluxes of CO2) and the mass of IC retained in the sediments. We are unable to directly identify these processes or quantify their rates. However, we expect them to be considerable, particularly alkalinity consumption in Burr Oak where DIC stream input concentrations are high (20 mg L-1), and the percentage of IC making up total carbon content in lake sediments is low (1%). We did not include a value depicting conversion of OC to IC, because we think this conversion is not quantitatively important in our hard-water systems. While there is good evidence for photobleaching in Acton and Burr Oak [Overholt et al., 2008], the conversion of OC to IC is not likely to be quantitatively important in the overall C budget. For example, even if all of the DOC loading from the streams was converted to DIC, this would alter DIC by only 6% in Acton and 20% in Burr Oak. Even in soft-water systems with low DIC, photobleaching serves primarily to convert high molecular weight DOC to lower molecular weight compounds with little change in DOC concentration [Osburn et al., 2001].
 As a check on our C budgets, we also calculated alkalinity balances. If alkalinity loss in our reservoirs is due to CaCO3 precipitation, we would expect to see a balance between the alkalinity brought in from stream inlets with the sum of alkalinity lost via dam outlets and buried IC (this value is derived from our C budgets). We would also expect to see a balance between the difference of alkalinity stream inputs and alkalinity dam outputs (as buried CaCO3) with IC burial from our C budgets. Our data do not allow us to do a complete alkalinity balance, because we lack pH and temperature data in the streams and at the dam outlet station. However, based on our data availability, we created alkalinity balances for Acton and Burr Oak during the summer months only. We estimated stream and dam outlet alkalinity by using within lake pH, lake temperature, and DIC concentration from the stream inlet and dam outlet. The within lake data were obtained from our shallow site (near stream inflows) and our deep site (near dam outflows). This is not ideal as lake pH and lake temperature may not match inlets and outlets. We then calculated alkalinity stream load and dam export using similar methods as our C load and export calculations. Alkalinity flux in eq day-1 was then converted to Mg day-1 assuming 1 mol equals 1 equivalent. This seems a reasonable assumption, because HCO3- makes up 96% of alkalinity in both Acton and Burr Oak (based on pH).
 Carbon budgets were constructed for the 2007 and 2008 WYs as well as the summers (i.e. May – October) of 2007 and 2008. We calculated summer budgets for two reasons: to compare to other studies in which fluxes were measured only in summer and to account for high variation in precipitation between summers. This allowed us to examine how weather mediates C budgets.
2.5 Bootstrapping Analyses
 Bootstrapping was used to provide error estimates on total C retention (Mg year-1) using total C inputs via streams and total C outputs via dams (stream inputs – dam outputs). All CO2 fluxes were included as retention in these estimates because at annual scales, there was net CO2 efflux from both lakes in both years. This approach is similar to a recent nitrogen (N) retention study in which N removal from lakes and reservoirs was equal to N inputs minus N outputs via surface water outlets and denitrification losses [Harrison et al., 2009]. Separate error estimates on CO2 fluxes were not made, because our CO2 measurements were collected at a coarser temporal resolution. Bootstrapping was conducted by breaking each water year into wet and dry time periods, which resulted in two time periods for WY 2007 and three for WY 2008. The data were split because of large fluctuations based on seasonal variability. These splits helped to ensure that bootstrapping subsamples would come from a similar sampling distribution. Each bootstrap simulation randomly chose a retention value (stream inputs – dam outputs) from the appropriate time period. To obtain each annual estimate, a stratified bootstrap was run over the two 2007 time periods and over the three 2008 time periods, i.e., the proportion of samples drawn from each period was equal to the proportion of the annual dates that period represented. Bootstrap simulations were performed 99,999 times and were run for mean and 95% confidence intervals. All bootstrap simulations were run using the boot package in R [Canty and Ripley, 2010].
3.1 C Stream Inputs and Dam Outputs
 Over the two water years, DIC was the dominant carbon form contributing to total carbon load for both reservoirs, while PIC was the least dominant form (Figure 2). Of the remaining carbon constituents, POC represented a greater fraction of TOC loading than DOC for Acton's inlets (DOC/TOC = 37% in 2007, 34% in 2008), while DOC was greater than POC in Burr Oak's inlet (DOC/TOC = 68% in 2007, 69% in 2008; Figure 2). The percentage of total export from the dam outlets by C form followed the same pattern for both reservoirs in that DIC > DOC > POC > PIC (Figure 2) with DOC comprising a larger percentage of TOC export than POC in both Acton (DOC/TOC = 61% in 2007, 58% in 2008) and Burr Oak (DOC/TOC = 83% in 2007, 86% in 2008).
 Acton had greater monthly carbon fluxes (Mg C month-1) than Burr Oak for all C forms via the inlets or outlets, most notably for DIC and POC (Figure 3). Monthly retention efficiency (i.e., retention/stream inputs) for each carbon form for each month varied between the reservoirs, with Burr Oak often retaining a larger percentage than Acton (Figure 3). DIC retention efficiency in Acton was moderate, except for a high retention period during the late summer of the 2007 water year and 2008 water year (Figure 3), potentially owing to low DIC inputs and high algal production (i.e., DIC uptake) during this period. Furthermore, during the late summer of 2007, we also found DIC concentration to be higher in the inlets than outlets for Acton (45, 30 mg L-1, respectively) and Burr Oak (25, 12 mg L-1, respectively) indicating retention of DIC within the lakes. It should be noted that DIC retention efficiency calculations do not incorporate CO2 fluxes, and thus we cannot distinguish retention as those retained in the sediments and those that are lost to the atmosphere as CO2. DOC retention efficiency in Acton was often negative, indicating net DOC export (outputs exceed inputs), except during periods of low-flow (Figure 3). DOC retention efficiency in Burr Oak was always positive, and during most of the 2007 summer, DOC retention was nearly 100% (Figure 3). In agreement with these results, we found DOC concentrations to be higher at the inlet than the outlet (3.8, 3.1 mg L-1, respectively). Additionally, flow from the outlet was low during these periods. Particulate carbon tended to be retained in Acton in moderate quantities, but during time periods with low-flow, export via the outlet far exceeded inlet loading, probably due to export of phytoplankton-derived C. This resulted in exceptional net POC export during a few months (Figure 3). Like other C forms, POC retention efficiency in Burr Oak was always positive and was less variable than Acton, and as in Acton, retention efficiency was high during low-flow and moderate the rest of the year (Figure 3).
3.2 CO2 Fluxes
 CO2 fluxes (Mg C month-1) with the atmosphere were often near zero in Acton, with an influx noted in the summer of 2007 and a moderate efflux found at the end of the 2007 water year and beginning of 2008 water year (Figure 4). CO2 fluxes in Burr Oak were more temporally variable, displaying effluxes during the fall and winter seasons, an influx in the summer of 2007, and fluxes near zero during the 2008 summer (Figure 4). In both lakes, there was a trend toward lower effluxes and sometimes influxes in midsummer time periods.
 In Acton and Burr Oak, pCO2 (µatm) was correlated with temperature, pH, and DIC concentration, while pCO2 was correlated with primary production, TP, and chlorophyll-a only in Acton (Table 2). pCO2 was unrelated to DOC in either reservoir (Table 2). In both reservoirs, pH was considerably better at explaining pCO2 variation than any of the other predictor variables. Since some of the pCO2 values were estimated using pH, pCO2 – pH regressions were also generated using dates only in which pCO2 was directly measured in the field. Comparisons between the pH regressions with either direct or estimated pCO2 revealed little difference and show that pH remains the best predictor variable of pCO2 in the study reservoirs (Table 2).
Table 2. Regression Relationships Between pCO2 (the Dependent Variable) and Physical, Chemical, and Biological Parameters. The pH Was Used to Estimate pCO2 When Direct Measurements Were Not Available. The pH Regressions Were Generated Using All pCO2 Data and With Dates Where Only Direct Measurements Were Taken
Burr Oak (forested)
1128 – 26 * temp
1553 – 43 * temp
DIC (mg L-1)
−416 + 72*DIC
PPr (mg C m-2 day-1)
861 – 0.135*PPr
DOC (mg L-1)
TP (µg L-1)
Chl – a (µg L-1)
766 – 3.35*chl
pH all (n = 159, 72)
7453 – 811*pH
5744 – 629*pH
pH direct only (n = 89, 42)
7054 – 767*pH
5737 – 628*pH
3.3 Growing Season Carbon Budgets
 Acton inlet TC loading by mass during the growing season (Mg summer-1) was much higher in 2008 than 2007 for the May – October budget corresponding to higher stream discharge in 2008 during these months (Figures 5 and 6). Similarly, TC outputs via the outlet were also higher in 2008 (Figure 5). CO2 fluxes were in opposite directions between years (i.e. net influx in 2007, net efflux in 2008), but in both years the fluxes were relatively low in magnitude and comprised a small percentage of TC fluxes (Figure 5). The majority of IC was lost from the reservoir water column via outlet fluxes in 2007 and 2008, while OC losses via the outlet and retention were similar (Figure 5).
 Inlet loading of TC by mass into Burr Oak was quite different between the May – October 2007 and 2008 budgets with higher loading in 2008 (Figure 5). As for Acton, precipitation and discharge were higher in these months in 2008 as were loading of all C forms during this time period (Figures 5 and 6). During the growing season, TC fluxes via the outlet were more than 20X greater in 2008 than 2007, while retention was similar between the years with OC dominating the C pool (Figure 5). Growing season CO2 fluxes were a moderately high influx in 2007 and an efflux in 2008 (Figure 5).
3.4 Annual Carbon Budgets
 Annual Acton and Burr Oak budgets revealed that TC loading was strikingly similar in 2007 and 2008 despite large differences in their growing season budgets (Figure 5). On an annual basis, 2008 was slightly wetter (discharge was 1.1-1.4X higher) than 2007 (Figure 6); in addition, based on precipitation, 2008 was wetter and 2007 was drier than an average year [Vanni et al., 2001]. Seasonal patterns of precipitation and stream discharge showed that both were higher during late fall and winter in WY 2007 than 2008 (even though 2007 was drier on an annual basis), particularly for the agricultural reservoir (Figure 6). High discharge during these months in 2007 corresponded with high DIC and POC loading, likely compensating for low TC loads during summer 2007 (Figures 5 and 6).
 Both reservoirs were net C sinks but of different magnitudes. Annually, total C retention rate was 3-4X greater in Acton, while CO2 flux to the atmosphere was 2-39X greater in Burr Oak (Figure 5). Annually, both reservoirs were a small source of CO2 to the atmosphere with the mesotrophic Burr Oak having higher CO2 evasion than the hypereutrophic Acton (Figure 5). Carbon retention in Burr Oak was dominated by OC, while OC and IC were approximately equivalent contributors to Acton C retention (Figure 5).
 We generated bootstrapped estimates of confidence intervals on annual retention rates (Figure 5). Generally, confidence intervals were <30% for total C (± the bootstrapped mean), imparting a relatively high level of confidence in our estimates.
 Reservoir carbon budgets revealed that, on an annual basis, both were a small source of CO2 to the atmosphere with the forested, mesotrophic Burr Oak having higher CO2 evasion (78 – 102 Mg yr-1) than the agricultural, hypereutrophic Acton (2 – 56 Mg yr -1). These results are surprising, since many researchers have suggested that inland waters with high primary production rates are likely to be large sinks of atmospheric CO2 [Cole et al., 2007; Downing et al., 2008; Hanson et al., 2004; Tranvik et al., 2009]. However, parameters associated with lake metabolism were either unrelated or weakly related to pCO2 in Acton and Burr Oak. This deviates from studies in low-productivity, low-pH lakes where DOC is often positively correlated with pCO2 [Jonsson et al., 2003; Sobek et al., 2003] but is consistent with results from other hard-water lakes [Finlay et al., 2009; Finlay et al., 2010; Lopez et al., 2011; Tranvik et al., 2009]. Our reservoirs are also generally considered net autotrophic based on whole lake ecosystem metabolism measurements using diel dissolved oxygen dynamics; a recent study using metabolism measurements showed that Acton Lake is net autotrophic in the summer (Solomonet al., 2013). Thus, it appears that our study reservoirs are net CO2 sources, based on C budgets, but that these emissions are not as strongly linked to lake metabolism as they are in soft-water lakes. Further, our reservoirs are generally considered net autotrophic when using whole lake ecosystem metabolism estimates. These are important distinctions and given that productive, hard-water lakes and reservoirs are globally abundant in area and volume [Wetzel, 2001], estimation of the role of lentic ecosystems in global C budgets will require more thorough knowledge of these systems.
 Whole-lake mass balance budgets for six hard-water lakes in Canada revealed that CO2 fluxes accounted for ~2% of total C fluxes, and there was large CO2 inter-annual variation [Finlay et al., 2010]. CO2 flux rates were somewhat lower in Acton (average = +3.5, range = −15.0 to +21.1 mmol C m-2 day-1) and Burr Oak (average = +11.6, range = −28.0 to +46.4 mmol C m-2 day-1) than in the six mesotrophic – eutrophic, Canadian lakes (range = −100 to +200 mmol C m-2 day-1[Finlay et al., 2010] but were more comparable to two low-productivity, hard-water lakes in Minnesota [Stets et al., 2009], large (100 – 3000 km2) eutrophic reservoirs in Canada that were > 30 years old [Demarty et al., 2009], as well as boreal and temperate lakes in general [Cole and Caraco, 1998; Del Giorgio et al., 1999; Rantakari and Kortelainen, 2005]. In addition, both reservoirs often had pH above 8, particularly in late summer, which resulted in considerable chemical enhancement factors (Acton mean = 3.0, range 1.0 – 8.4; Burr Oak mean = 2.1, range 1.0 – 10.5). Chemical enhancement factors in Acton and Burr Oak are comparable to other hard-water lakes where CO2 fluxes were also found to contribute relatively small fluxes total C fluxes [Finlay et al., 2009; Stets et al., 2009].
 OC retention per unit reservoir area was 274–340 g m-2 yr-1 in Acton and 126–133 g m-2 yr-1 in Burr Oak. These rates are higher than those found in most natural lakes, 6–94 g m-2 yr-1 [Mulholland and Elwood, 1982] but generally lower than the recently reported median rate in small, eutrophic agricultural reservoirs (2122 g m-2 yr-1) in Iowa [Downing et al., 2008]. Differences between Iowa rates and those in Acton, a hypereutrophic reservoir dominated by row-crop agricultural land use, may be due to improved land management in Acton's watershed. Since the 1990s, conservation tillage has become the dominant land management practice in Acton's watershed resulting in reduced nutrient and sediment loads into Acton via streams [Renwick et al., 2008]. We also note that our OC retention rates are similar with those based on sediment cores in both Acton and Burr Oak [Vanni et al., 2011]. IC retention rates in Acton were 224–279 g m-2 yr-1 while only 1.27-1.34 g m-2 yr-1 in Burr Oak. Few studies have reported IC burial rates in lentic waters, but rates in Burr Oak are lower than previous reports in hard-water lakes, while Acton rates are up to 5X greater [Finlay et al., 2010; Stets et al., 2009], perhaps because Acton water is supersaturated with CaCO3 [Green et al., 1985].
 Our results also show the temporal scale dependence of carbon budgets. Within each reservoir, variability for any given C flux was much greater between summers than between water years. Within-reservoir differences were especially pronounced in early summer (May-July), probably because of variable precipitation and stream discharge. Summer 2008 discharge was 4.8-6.4X higher than summer 2007, resulting in greater stream C inputs than in 2007. Within each reservoir, C retention efficiency (retention/stream inputs) was higher in the dry summer, probably because of increased water residence time given that water flow from the outlet during late summer was either non-existent or negligible. We also found that, in 2007 and 2008, water residence time was greater in Burr Oak than Acton (Table 1), which likely played a role in Burr Oak retaining a higher percentage of C. Also, both reservoirs were CO2 sinks in the dry summer but CO2 sources during the wet summer. Previous work suggests that precipitation may be positively related to the magnitude of CO2 evasion in boreal lakes [Einola et al., 2011], where CO2 fluxes are associated with terrestrial DOC inputs [Sobek et al., 2003]. We observed a similar relationship with precipitation in our hard-water reservoirs even though CO2 fluxes did not appear to be tightly coupled with DOC. DIC inputs would also be expected to increase during precipitation events [Raymond and Oh, 2007]. Nonetheless, given the similar trends in these disparate water bodies, we may be able to assume that dry summers will generally have elevated C retention efficiency, coupled with either reduced CO2 evasion or a CO2 influx, in contrast to wet summers. Identifying patterns such as these is central to understanding the consequences of climate change and altered hydrological regimes on C budgets.
 We also highlight the role of variable stream inputs, because the largest inputs often occurred during time periods not included in typical summer studies. Thus, 67-79% of annual POC loads via streams occurred collectively during only 10% of days (i.e., dates on which daily discharge was > in the 90th percentile), and 62-89% of these dates occurred outside of summer [May-October; Knoll, 2011]. Our high resolution stream data also allow us to differentiate between the relative contribution of POC and DOC to OC loading. In Acton, POC can represent 17-78% of the OC load (mean = 44%) and 12-50% in Burr Oak (mean = 25%), reflecting the large variation that would not be captured without storm based sampling regimes. Further, C inputs via streams are only rarely measured directly in carbon budgets. Therefore, better estimates of these fluxes could significantly alter inferences drawn from lake budgets [Finlay et al., 2010; Sobek et al., 2006; Stets et al., 2009], particularly if C loading is underestimated during storm events. In general, variation in non-summer precipitation will likely generate large differences in C runoff and loads in temperate areas, compared to variation in summer when terrestrial evapotranspiration is greater. In addition, CO2 emissions from lakes and reservoirs are relatively well described, but less is known about CO2 sources because until very recently, complete carbon budgets were rare.
 Mass-balance carbon budgets are an ideal way to elucidate carbon fluxes and dynamics in inland waters [Andersson and Sobek, 2006]; however, they are often constrained by the logistics and expense of accurately measuring all possible fluxes. While the current study was an improvement on many prior studies due to our high resolution estimates of DIC, DOC, POC, and PIC, we were unable to estimate some fluxes in detail. Ice-out release of CO2 can be a significant flux, particularly in systems with high DOC loads such as boreal lakes. Limited work indicates that ice-out CO2 flux in some hard-water lakes may represent a small contribution to total annual CO2 flux [Finlay et al., 2010]. Our estimates of CO2 fluxes for Acton during the winter and ice-out were constrained by our ability to safely sample the lake. Thus, our winter CO2 efflux estimates for Acton may be underestimated and represent a source of uncertainty in our budget. We had year-round pH, DIC, and temperature data for Burr Oak, so were able to estimate CO2 fluxes during these periods. Estimating k, piston velocity, from wind speed may also be an additional source of uncertainty in our CO2 flux calculations. However, empirical relationships between wind speed and k are more problematic in small (< 0.5 km2), wind sheltered lakes [Cole et al., 2010] than in reservoirs like Acton and Burr Oak. We also did not directly measure CaCO3 precipitation rates in our reservoirs. This process is important because it removes alkalinity from the water column and also increases CO2 evolution. Furthermore, if alkalinity is conserved in the reservoirs, we should be able to take the difference between alkalinity inputs and outputs to/from the water column and attribute this loss to CaCO3 precipitation and thus IC burial. Using limited data, we found agreement with our budgets and estimated alkalinity balances in Acton (Table 3). For Burr Oak, our alkalinity balance overestimated IC burial, and we attribute this to alkalinity consumption and CaCO3 dissolution (Table 3). Burr Oak may have increased alkalinity consumption via nitrification during fall turnover, because this reservoir has high ammonium concentrations in the hypolimnion (M.J. Vanni, unpublished data) and a larger volume of anoxic waters than Acton (1.5 x 106, 8.2 x 105 m3, respectively). In a reservoir with similar chemical conditions as Burr Oak, permanent IC burial was 88% lower than deposited carbonate due to dissolution [Wang et al., 2012]. Uncertainty in the loss of IC from Burr Oak is a source of potential error in our budgets.
Table 3. Comparison of Alkalinity Balances With C Budgets. Note All Time Periods Are From May – October Except for Acton 2008 (May – September).
Stream alkalinity load (Mg summer-1)
Dam alkalinity export (Mg summer-1)
IC buried from C budgets (Mg summer-1)
Alkalinity export + IC buried (Mg summer-1)
Alkalinity load – alkalinity export as buried CaCO3 (Mg summer-1)
 We did not measure direct atmospheric deposition of DOC, DIC, or POC onto reservoir surfaces. In seven unproductive lakes in Ontario, atmospheric inputs of DIC comprised 1 to 8% of total DIC inputs while atmospheric DOC inputs ranged from 2 to 13% of total DOC inputs [Dillon and Molot, 1997]. Stets et al.  showed that organic carbon inputs via precipitation were only important in a closed-basin lake while they represented a minor influx in a lake with high surface water input. Given that Acton and Burr Oak have large watersheds, and hence water inputs via surface runoff, we suspect that atmospheric inputs of C would represent a small and insignificant influx into these reservoirs. In addition, based on nutrient budgets for Acton Lake, atmospheric inputs of C are likely to be small compared to stream inputs [Vanni et al., 2011]. Groundwater inputs into some lakes are extremely rich in DIC and CO2 and can thus contribute greatly to these inputs [Stets et al., 2009; Striegl and Michmerhuizen, 1998]. We did not estimate groundwater inputs of C, but for both reservoirs, groundwater contributes a small amount of the hydrological inputs (W.H. Renwick, unpublished data).
 Because we used a high resolution sampling regime, and recognizing the uncertainties discussed above, we felt confident in our retention and CO2 emission rates. Thus, we scaled up C fluxes regionally, specifically for the state of Ohio, by estimating OC burial and atmospheric CO2 exchange rates in Ohio tributary reservoirs >0.5 km2 (n = 105 reservoirs statewide). We used watershed land use data [Hagenbuch, 2010] to classify reservoirs as either dominated by agriculture, forest, or of mixed land use (i.e., agriculture and forest). We then applied mean CO2 fluxes and OC burial from our budgets to three reservoir classification types (agricultural, forested, or mixed), using the average of agricultural and forested for mixed land use (Table 4). We assessed only OC burial to facilitate comparison with past global studies [Cole et al., 2007; Tranvik et al., 2009]. We estimate that Ohio reservoirs bury 105 Gg OC yr-1 and emit 8.5 Gg of C as CO2 yr-1 (Table 4). Reservoirs in agricultural landscapes buried 53% of statewide OC, while forested reservoirs buried only 8% (39% was buried by mixed land use). Agricultural and forested reservoirs each emitted ~25% of the CO2 emissions by Ohio reservoirs (those in mixed land use emitted 50%).
Table 4. CO2 Emissions And OC Burial In Ohio Reservoirs and Global Lentic Water Bodies.
Total surface area of reservoirs, statewide or global (km2)
% of global area of lentic ecosystems (lakes plus reservoirs)
Mean reservoir CO2 emissions (g m-2 yr-1)
Statewide or global CO2 emissions (Gg yr-1)
% of global CO2 emissions occurring from reservoirs, statewide
Mean reservoir OC burial (g m-2 yr-1)
Statewide or global OC burial (Gg yr-1)
% of global OC burial occurring from reservoirs, statewide
based on low and high global lake area estimates from references Meybeck  and Downing et al. ,
based on low and high estimated global C fluxes from Cole et al.  and Tranvik et al. .
 Using recently published estimates of global OC burial and CO2 emission rates of lentic ecosystems (lakes and reservoirs [Cole et al., 2007; Tranvik et al., 2009]), we calculated the percentage of global lentic fluxes attributable to Ohio reservoirs. We also examined whether these fluxes are proportional to the lentic area they occupy using published global estimates of lake and reservoir area [Downing et al., 2006; Meybeck, 1995] and C fluxes [Cole et al., 2007; Tranvik et al., 2009]. We used high [Downing et al., 2006; Tranvik et al., 2009] and low [Cole et al., 2007; Meybeck, 1995] estimates of global C fluxes and global lentic area to account for uncertainty. We find that Ohio reservoirs account for 0.02-0.05% of total global OC burial but only 0.001-0.002% of global CO2 emissions in lentic ecosystems (Table 4). Ohio reservoirs bury between 0.8-4.4X of the C mass that would be predicted based only on water body area, i.e., (Ohio lentic burial/global lentic burial) / (Ohio lentic area/global lentic area). On the other hand, CO2 emissions by Ohio reservoirs represent only 5-21% of emissions expected based on their area. We initially expected that productive Ohio reservoirs would be CO2 sinks, but our results indicate that they are generally small sources. Even when considering additional eutrophic Ohio reservoirs dominated by agricultural land, we find that these systems are sources of CO2 (Figure 7). These additional reservoirs may behave as Acton in that they receive large quantities of inorganic carbon from their watersheds, emit a portion of inorganic carbon as CO2, and are net CO2 sources despite being autotrophic. Thus, Ohio reservoirs appear to retain proportionally more C, but emit proportionally less CO2, than the global average. This is likely a consequence of their relatively large watershed areas and carbonate bedrock, which results in large quantities of C, particularly POC and DIC, being delivered to these reservoirs (Figure 3).
 Our budget results suggest that moderately to highly productive waters are not necessarily large CO2 sinks as previously expected [Cole et al., 2007; Hanson et al., 2004], and that watershed land use and hydrology (specifically, precipitation variability) modulate C fluxes. Regional estimates also suggest that Midwestern US reservoirs are burying significant amounts of OC but that the magnitude of CO2 flux is unexpectedly low and in the opposite direction as predicted for productive systems. The role of inland waters in regulating carbon will vary with changing climate, and the nature of these shifts will depend upon watershed and lake characteristics. To gain insights into how climate change may modify these fluxes in diverse systems, we can use high resolution baseline data and compare it to periods of extreme climatic events such as droughts or extreme storm events.
 We thank Robert Moeller, Stephen Glaholt, Burr Oak Regional Water District, and Beth Mette for providing assistance in the field and laboratory. We also thank Lyz Hagenbuch and the Ohio Division of Wildlife for land use and reservoir area data. Jon Cole and Edward Stets provided valuable assistance with CO2 flux calculations. This research was supported by a Miami University Research Enrichment Grant to M.J.V. and W.H.R., Miami University Field Workshop funds to L.B.K., an NSF REU Site grant (DBI 0353915), and two NSF LTREB grants (DEB 0235755 and 0743192) to M.J.V. and W.H.R. The research was also funded in part by the United States Environmental Protection Agency (EPA) under the Science to Achieve Results (STAR) Graduate Fellowship Program to L.B.K. EPA has not officially endorsed this publication and the views expressed herein may not reflect the views of the EPA.