Methods

Dissolved gas samples were collected at 0.5, 3, 5 and 7 m using the
headspace method. From January to March 2020, water at each discrete depth
was pumped directly into the bottom of a 1-L Nalgene bottle and flushed with
at least three times the volume before being capped with a rubber stopper.
60 mL of ambient air was added while 60 mL of sample water was removed from
the bottle and equilibrated by shaking for 90 seconds. From May 2020
onwards, water was pumped into a closed bottle system, and using syringe,
105mL of water was extracted and 35mL of ambient air was added. The
headspace was then equilibrated for 2 minutes by shaking and 10 mL of
equilibrated gas sample was then removed from the bottle and injected into a
5.9 mL Labco Exetainer vial that had been previously vacuumed. While in the
field, samples were stored in pouches within a survival suit to prevent
extreme temperature change. We analyzed the gas samples for CO2 and CH4 with
a gas chromatograph (GC-2014; Shimadzu Scientific Instruments) equipped with
a methanizer and flame ionization detector. Greenhouse gas concentrations
were calculated according to Henry’s law and corrected by measured ambient
air.<br/>

Methods

We determined stream CH4 concentrations at a large spatial extent (10
km) and fine grain (35,000 total measurements) by using a
biogeochemical mapping platform on a small boat [Crawford et al.,
2015]. Water was pumped on board where gases were stripped from the
water by using a sprayer-type equilibrator and analyzed with a Los
Gatos Research ultraportable greenhouse gas analyzer (using a
cavity-enhanced absorption technique). Survey speeds were very slow
(less than 3 kph) to enable the detection of small-scale changes in
CH4 concentrations over short distances. CH4 concentrations were
corrected for hydraulic and equilibrator lags by using first-order
step-change experiments detailed in Crawford et al. [2015] following
the outline provided in Fozdar et al. [1985]. Lag-corrected CH4 values
were georeferenced by using concurrent Global Positioning System
readings with the Wide Area Augmentation System capability enabled.
The highresolution transect was sampled on 24 and 25 July 2014 (two
morning to afternoon segments were combined into one data set). Both
days were similar in terms of weather and in-stream conditions.
Maximum daily air temperatures were 22.4degC and 20.9degC. Mean daily
air temperatures were 18.5deg and 19.1degC. At the middle site, daily
mean discharges were 3.3 and 3.7 × 10 2 m3 (Julys mean Q is 3.7 × 10 2
m3). Mean water temperatures were 18.1degC and 17.8degC (Julys mean
temperature is 19.6).
Using the high-resolution spatial CH4 data sets, we assessed the
degree of spatial autocorrelation by using semivariograms (spherical
model, using the function autofitVariogram in the R package automap).
We focused on the semivariogram range parameter which describes the
average scale of autocorrelation (i.e., the average patch size). We
also assessed the structure of spatial autocorrelation by using the
global Morans I statistic. The Morans I statistic evaluates whether a
series of geospatial observations are randomly distributed in space
(the null model), clustered, or dispersed. Statistically significant
positive values indicate spatial clustering, whereas negative values
indicate dispersed patterns. We used the Anselin Local Morans I
statistic for spatial cluster analysis of high-resolution CH4 data
[Anselin, 1995]. Statistically significant values of Local Morans I
identify regions (clusters) of high or low values relative to the
global data set, in addition to outliers (e.g., low outliers
surrounded by high values). The analysis was executed by using the
Spatial Statistics toolbox in ArcMap 10.2.
Groundwater Methane Sources
We analyzed groundwater CH4 from a series of wells near the middle and
lower sites and from wells at the head of the drainage near the spring
ponds (upper site) during the time frame of this study by using a
headspace equilibration method [Striegl et al., 2001]. Wells were
developed by using a peristaltic pump, and a minimum of two well
volumes were purged before sample collection. The goal was to evaluate
additional (external) lateral and vertical sources of CH4 beyond the
hyporheic zone. Despite a relatively homogenous sand aquifer,
groundwater flow paths and residence times are complex in this
catchment [Pint et al., 2003; Walker et al., 2003]. A combination of
flow paths including deep groundwater derived from meteoric recharge,
deep groundwater derived from lakes, and meteoric riparian water all
contribute to surface flow in the catchment. These water sources and
flow paths have been studied for over a decade as part of the USGS
WEBB program [Pint et al., 2003; Walker et al., 2003]. Differences in
substrate, organic matter availability, oxygen conditions, and other
metrics of redox state were previously shown to relate to the
concentrations of dissolved gases in wells at the middle site based on
historical data [Crawford et al., 2014b]. Here we expand the survey to
correspond to the timing of surface water mapping and to determine
whether patterns previously observed at the middle site held for the
catchment in general.
Sediment Methanogen Distribution and Abundance
CH4 production potential within stream sediments was first determined
by extracting DNA from stream sediment cores and quantifying the
abundance of methanogenic Archaea. We collected 14 cores approximately
22 cm long in sand and organic-rich wetland locations near the middle
site (locations correspond to odd numbered transect locations in
Figure 1; also corresponding to CH4 bubble trap locations described in
Crawford et al. [2014b]). Sediment cores were collected by using a 2.5
cm diameter, 30 cm length, stainless steel corer with an internal
polycarbonate tube attached to a one-way flow valve and a PVC
extension. Intact cores were transported to the laboratory within 2 h
and immediately frozen. Sediment cores were split into 2 cm segments
followed by DNA extraction by using a PowerSoil DNA isolation kit
(MoBio Laboratories Inc., Carlsbad, CA). We used quantitative
polymerase chain reaction (qPCR) targeting the gene encoding the alpha
subunit of methyl coenzyme-M reductase (mcrA) to quantify both
longitudinal and vertical distributions of methanogens. The mcrA gene
encodes a component of the terminal enzyme complex in the methane
generation pathway and is thought to be unique to methanogens and well
conserved [Thauer, 1998]. Many previous studies have used mcrA as a
genetic marker to determine methanogen abundance and community
composition [Luton et al., 2002; Earl et al., 2003; Freitag et al.,
2010; West et al., 2012]. Each extracted sample containing mcrA was
amplified in a 20 uL qPCR reaction in an ep gradient s realplex2
master cycler (Eppendorf), using SYBR Green as the reporter dye. Each
reaction contained 1 uL of 1/10 diluted sample DNA template, 1× iQ
SYBR Green Supermix (Biorad), and 0.25 uM of each primer targeting
mcrA: mcrAqF (50-AYGGTATGGARCAGTACGA-30) and mcrAqR
(50-TGVAGRTCGTABCCGWAGAA-30) [West et al., 2012]. Thermocycling
conditions for the mcrA qPCR were as follows: an initial denaturation
at 94degC for 1 min, followed by 40 cycles of 94degC denaturation for
40 s, 54degC annealing for 30 s, 72degC elongation for 30 s, and a
fluorescent detection at 85degC for 20 s. Melting curves were run to
ensure absence of nonspecific amplification. Amplification,
fluorescence data collection, and initial data analysis were all
performed by using the Eppendorf realplex2 software (Eppendorf,
Hauppauge, NY, USA).
Despite collecting greater than 20 cores, we were not able to perform
cluster analysis similar to that for CH4 concentrations on the genetic
data because the sample size was too low. Instead, we elected to
compare organicrich versus sand sediment mcrA gene abundance by using
a t test. To determine if methanogen abundance was correlated with CH4
production, we fit a linear model (log transformed) of mcrA abundance
versus average CH4 ebullition documented in the same year [see
Crawford et al., 2014b] with the R statistical programming language [R
Core Team, 2014]. We contend that the comparison between microbial
communities and integrated CH4 bubble flux over time is a stronger
comparison than that of point measurements of CH4 concentration.
Sediment Methane Production Potential
We collected surface sediments from Allequash Creek and placed them in
sealed jars the morning of the start of laboratory experiments. These
sediments were presumed to be mostly anoxic per the oxygen profile
study by Crawford et al. [2014b]. In the lab, about 75 mL of water
saturated surface sediments was transferred into a 150 mL glass
container, flushed with N2, sealed with a gas-tight lid equipped with
a butyl rubber septum for headspace gas sampling, and placed on a
shaker table in the dark at room temperature (about 22degC). Gas
samples were collected after 24 h for CH4 determination by using a
Shimadzu GC-2014 gas chromatograph. Headspace volume of each sample
was determined after gas sampling, then sediments were transferred to
a preweighted aluminum pan for drying (72 h at 50degC) and ashing (4 h
at 500degC). CH4 production potential was determined as headspace CH4
accumulation per gram of dry sediment and per gram of ash-free dry
mass (AFDM) per hour. Production rates were based on two-point
measurements of CH4 concentration and are thus presumed to be linear
over time. Because gas production rates could not be transformed to
meet assumptions of normality, significant differences among
treatments were assessed by using a Kruskal-Wallis test followed by
Wilcoxon rank tests for pairwise comparisons with Bonferroni-adjusted
P value using R.

Methods

This study used the R packages “GLMr” (https://github.com/GLEON/GLMr) and “glmtools” (https://github.com/USGS-R/glmtools) to interact with GLM-AED from the R statistical environment.

Hourly meteorological data were obtained for 2010-2016 from the North American Land Data Assimilation System (NLDAS) and included air temperature, relative humidity, short wave radiation, long wave radiation, wind speed, rain accumulation, and snow accumulation.

Average daily stream discharge and nitrogen and phosphorus loads were obtained from the United States Geological Survey (USGS) National Water Information System (NWIS) water quality database. All streams included in this study are monitored by the USGS. This study used 2014 nutrient loading data over the simulation time period since 2016 nutrient data were not yet available. Only the Yahara River and Pheasant Branch Creek have data on daily nutrient loads; thus, Pheasant Branch nutrient concentrations were also applied to both Dorn Creek and Sixmile Creek, which contribute near equivalent volume to Lake Mendota. Since GLM-AED requires nutrient concentration rather than load, water column concentrations were back calculated using load and discharge.

Daily OC loads (POC and DOC, individually) were estimated for each individual inflow (n=4) from observed weekly measurements of OC concentration (doi: 10.6073/pasta/86094298f5a9085869518372934bf9d7)
using stream-specific linear models (Lathrop et al. 1998). The linear relationship between known log-scaled OC loads and log-scaled discharge for each individual stream was used to predict the OC load for each day in the model simulation time period. Again, water column OC concentrations were back calculated using the load and discharge.

A manual calibration routine was conducted to optimize the goodness-of-fit of various state variables in GLM-AED, especially those known to influence OC and GHG dynamics. We used visual comparison of predictions and observations, as well as a quantitative approach, to assess goodness-of-fit. We calibrated water balance and water temperature to effectively simulate lake level and stratification. We also calibrated DO, Secchi depth, TN, TP, DOC, and CH4. Finally, we calibrated phytoplankton dynamics according to four phytoplankton functional groups, recognizing that phytoplankton contribute substantially to the OC pool.

Methods

In two consecutive years, we measured lake-wide spatial patterning of cyanobacteria using the FLAMe platform (Crawford et al. 2015). To evaluate early warning indicators of a critical transition, in the first year we induced a cyanobacteria bloom through nutrient addition in an experimental lake while using a nearby unmanipulated lake as a reference ecosystem (Pace et al. 2017). During the second year, both lakes were left unmanipulated. Proposed detection methods for early warning indicators were compared between the manipulated and reference lakes to test for their ability to accurately detect statistical signals before the cyanobacteria bloom developed.
Peter and Paul Lakes are small, oligotrophic lakes (Peter: 2.5 ha, 6 m, 19.6 m and Paul: 1.7 ha, 3.9 m, 15 m, for surface area, mean, and max depth respectively) located in the Northern Highlands Lake District in the Upper Peninsula of Michigan, USA (89°32’ W, 46°13’ N). These lakes have similar physical and chemical properties and are connected via a culvert with Paul Lake being upstream. Both lakes stratify soon after ice-off and remain stratified usually into November (for extensive lake descriptions, see Carpenter and Kitchell, 1993).
In the first year, Peter Lake was fertilized daily starting on 1 June 2015 (DOY 152) with a nutrient addition of 20 mg N m-2 d-1 and 3 mg P m-2 d-1 (molar N:P of 15:1) through the addition of H3PO4 and NH4NO3 until 29 June (day of year, DOY 180). The decision to stop nutrient additions required meeting four predefined criteria based on temporal changes in phycocyanin and chlorophyll concentrations indicative of early warning behavior of a critical transition to a persistent cyanobacteria bloom state. (Pace et al. 2017). Nutrients uniformly mix within 1-2 days after fertilization based on prior studies (Cole and Pace 1998). No nutrient additions were made to Paul Lake. In the second year (2016), neither lake received nutrient additions.
We mapped the surface water characteristics of both experimental lakes to identify changes in the spatial dynamics of cyanobacteria. In 2015, mapping occurred weekly from 4 June to 15 August (11 sample weeks). In 2016, when neither lake was fertilized, the lakes were mapped three times in early to mid-summer. In both years, mapping occurred between the hours of 07:00 to 12:00 (before the daily nutrient addition). We rotated the order that we sampled the lakes to avoid potential biases due to differences in time of day. Each individual lake sampling event was completed in approximately one hour.
The FLAMe platform maps the spatial pattern of water characteristics. A boat-mounted sampling system continuously pumps surface water from the lake to a series of sensors while geo-referencing each measurement (complete description of the FLAMe platform in Crawford et al. 2015). For this study, the FLAMe was mounted on a small flat-bottomed boat propelled by an electric motor and was outfitted with a YSI EXO2™ multi-parameter sonde (YSI, Yellow Springs, OH, USA). We focused for this study on measures of phycocyanin (a pigment unique to cyanobacteria) and temperature. Phycocyanin florescence was measured using the optical EXO™ Total Algae PC Smart Sensor. The Total Algae PC Smart Sensor was calibrated with a rhodamine solution based on the manufacturer’s recommendations. Phycocyanin concentrations are reported as ug/L; however, these concentrations should be considered as relative because we did not calibrate the sensor to actual phycocyanin nor blue-green algae concentrations. Geographic positions were measured using a Garmin echoMAP™ 50s. Sensor- data were collected continuously at 1 Hz and linked via timestamp to create spatially explicit data for each lake. Each sampling produced approximately 3500 measurements in the manipulated lake and 2000 in the reference lake. The measurements were distributed by following a gridded pattern across the entire lake surface to characterize spatial patterns over the extent of the lake.

Methods

lake_weekly_carbon_ghg.csv
Carbon Sample Analysis: Weekly observational data were collected on Lake Mendota between 15 April 2016 and 14 November 2016. All lake samples collected from the deep hole were taken at five discrete depths (3, 10, 12, 14, and 20 m), intended to span the seasonal thermocline. All lakes samples collected in littoral zones (Point, Ubay, and Yahara) were taken at two discrete depths (0.1 and 2 m). Two liters of water were collected at each sampling location and depth using a Van Dorn sampler for measurement of particulate organic carbon (POC), dissolved organic carbon (DOC), and dissolved inorganic carbon (DIC). Between 1200 mL and 1800 mL of water was passed through a ProWeigh 47 mm filter (Environmental Express, Charleston, SC, USA) depending on how quickly water became impassable. POC was estimated by performing loss on ignition on the filter. The difference in mass before and after combustion at 500°C was multiplied by 0.484 to account for the OC fraction of organic matter (Thomas et al. 2005). Filtrate was analyzed for DOC and DIC on a Shimadzu TOC-V-csh Total Organic Carbon Analyzer (Shimadzu Scientific Instruments, Kyoto, Japan), where organic carbon is measured by combustion and inorganic carbon after phosphoric acid digestion.
Dissolved Gas Analysis: Water samples for dissolved methane (CH4) and carbon dioxide (CO2) were collected at each depth in the lake using a Van Dorn sampler and stored in 30-mL serum vials. Serum vials were overfilled and capped in the field with a rubber septa and aluminum cap. Care was taken to ensure that no bubbles were present in the sample. Serum vials were then stored on ice until they could be placed in the refrigerator. Within 24 hours of collection, samples received a 3 mL N2 gas headspace, were shaken vigorously, and left to equilibrate at room temperature. Headspace CH4 and CO2 was analyzed on a Varian 3800 gas chromatograph, and headspace-water CH4 and CO2 partitioning was accounted for using Henrys Law.

Methods

The Fast Limnology Automated Measurement (FLAMe) platform is a novel flow-through system designed to sample inland waters at both low- (0 to appr. 10 km hr-1) and high-speeds (10 to greater than 45 km hr-1) described in Crawford et al. (2015). The FLAMe consists of three components: an intake manifold that attaches to the stern of a boat; a sensor and control box that contains hoses, valves, a circulation pump and sensor cradles; and a battery bank to power the electrical components. The boat-mounted intake manifold serves multiple purposes. First, sensors are mounted inside the boat, protecting them from potential damage. Second, the intake system creates a constant, bubble-free water flow, thus preventing any issues for optical sensors due to cavitation. Finally, to analyze dissolved gases, a constant water source is needed on board. Water flow via both the slow- and high-speed intakes is regulated by the onboard impeller pump, allowing for seamless switching between slow- and high-speed operations. Any number of sensors could be integrated into the platform with simple modifications, and can be combined with common limnological instruments such as acoustic depth-finders. In our example applications we used a YSI EXO2 multiparameter sonde (EXO2; Yellow Springs, OH, USA), and a Satlantic SUNA V2 optical nitrate (NO3) sensor (Halifax, NS, Canada), both integrated into the control box plumbing with flow-through cells available from the manufacturer. Additionally, a Los Gatos Research ultraportable greenhouse gas analyzer (UGGA) (cavity enhanced absorption spectrometer; Mountain View, CA, USA) was used to measure dry mole fraction of carbon dioxide (CO2) and methane (CH4) dissolved in surface water by equilibrating water with a small headspace using a sprayer-type equilibration system that has previously been shown to have fast response times relative to other designs16 (Figure S1). Both the EXO2 and the UGGA are capable of logging data at 1 Hz. Because the SUNA was operated out of the water and on a boat during warm periods, data were collected less frequently (appr. 0.1 Hz) to minimize lamp-on time and avoid the lamp temperature cutoff of 35° C. The EXO2 sonde uses a combination of electrical and optical sensors for: specific conductivity, water temperature, pH, dissolved oxygen, turbidity, fluorescent dissolved organic matter (fDOM), chlorophyll-a fluorescenece, and phycocyanin fluorescence. The SUNA instrument measures NO3 using in situ ultraviolet spectroscopy between 190-370 nm, has a detection range of 0.3-3000 microM NO3, and a precision of 2 microM NO3. The UGGA has a reported precision of 1 ppb (by volume). In order to translate time-series data from the instruments into spatial data, we also logged latitude and longitude at 1 Hz with a global positioning system (GPS) with the Wide Area Augmentation System (WAAS) functionality enabled allowing for less than 3 m accuracy for 95percent of measured coordinates. Synchronized time-stamps from the EXO2, UGGA, SUNA, and GPS were used to combine data streams into a single spatially-referenced dataset.
We ran a simple set of experiments to determine the residence time of the system and the overall response time of the EXO2 and UGGA sensors integrated into the platform. After determining first-order response characteristics of each sensor, we applied an ordinary differential equation method to correct the raw data for significant changes in water input resulting in higher accuracy spatial data (see Crawford et al. 2015).
Sensor response experiments
We conducted a series of sensor response experiments on Lake Mendota on August 1, 2014. The goal was to understand the potential lags and minimum response times for the EXO2 and UGGA sensors integrated into the FLAMe platform. These data were then used to develop correction procedures for higher accuracy spatial datasets. To test sensor responses to step-changes in water chemistry, we mixed a 40 L tracer solution into a plastic carboy that was connected to the reservoir port on the FLAMe. The reservoir was mixed with 50 mL of rhodamine WT to test the phytoplankton fluorescence sensors, 6 mL of quinine sulfate solution in acid buffer (100 QSE) to test the fDOM sensor, 14 g of KCl to test the conductivity sensor, and appr. 2 kg of ice to reduce the temperature of the solution relative to lake water. The mixture volume was increased to 40 L using tap water. We did not modify the CO2 concentration or pH in the carboy as we found the municipal water source to have greater than ambient lake CO2 (4300 vs. 290 microatm, respectively) and lower pH (7.5 vs 8.3, respectively). At the beginning of the experiments, we allowed lake water to circulate through the system for appr. 10 minutes. We then switched to the tracer solution for a period of five minutes, followed by five minutes of lake water, then back to the tracer solution for an additional five minutes.
Using the step-change experiment data, we determined each sensors hydraulic time constant (Hr) and parameter time constant (taus). The sensor-specific Hr is a function of system water residence time and sensor position/shielding within the system. Taus is the time required for a 63 percent response to a step-change input. Hr was calculated based on the plateau experiments and was indicated by the first observation with a non-zero rate of change. The CO2 and CH4 sensors had a much greater Hr than the EXO2 sensors because water must travel further through the system before equilibrating with the gas solution being pumped to the UGGA. Using these Hr values, we offset response variables thus removing the hydraulic lag. This correction does not account for sensor-specific response patterns (tau s). The EXO2 sensors have manufacturer-reported taus values between 2-5 s, but these values are not appropriate to apply to the FLAMe system because they do not include system hydraulic lag and mixing. In order to match sensor readings with spatial information, we first applied Hr values from each sensor output according to equation 2. This step aligns the time at which each sensor begins responding to the changing water, and accounts for the physical distance the water must travel before being sensed
In order to match individual sensor response characteristics and to obtain more accurate spatial data, we then applied sensor-specific corrections using Equation 3 (Fofonoff et al., 1974).
We first smoothed the raw data using a running mean of 3 observations in order to reduce inherent noise of the 1 Hz data. We then calculated dX/dt using a 3-point moving window around Xc. Equation 3 should ideally lead to a step response to a step-change input. We note that this is the same strategy used to correct oceanographic conductivity and temperature instruments (see Fozdar et al., 1985). Overall, the taus-corrected data show good responses to step-change inputs and indicate that this is a useful technique for generating higher accuracy spatial data. We include three types of data for each variable including: raw (e.g., TempC), the hydraulic lag corrected (e.g., TempC_hydro) and the taus-corrected data (e.g., TempC_tau). Note that not all sensors were used in each survey and not all sensors have each type of correction. This data was from our preliminary FLAMe sampling campaigns and future studies will include additional sensor outputs and corrections.
We used the FLAMe throughout the summer of 2014 on four distinct aquatic ecosystems including: a small dystrophic lake, a stream/lake complex, a medium-sized eutrophic lake, and a managed reach of the Upper Mississippi River. Each of these applications demonstrates the spatial variability of surface water chemistry and the flexibility of FLAMe for limnological research.
References
Crawford JT, Loken LC, Casson NJ, Smith C, Stone AG, and Winslow LA (2015) High-speed limnology: Using advanced sensors to investigate spatial variability in biogeochemistry and hydrology. Environmental Science and Technology 49:442-450.
Fozdar FM, Parker GJ, and Imberger J (1985) Matching temperature and conductivity sensor response characteristics. Journal of Physical Oceanography 15:1557-1569.

Methods

CH4 data from streams and rivers are widely scattered, as values are often included as end-member in studies focused on other processes or types of ecosystems. Thus, while we sought to be as complete as possible in compiling existing data, some sources have undoubtedly been overlooked. Sources included journal articles, book chapters, dissertations, USGS open file reports, meeting proceedings, and unpublished results provided by individual investigators. Data incorporated into MethDB were strictly limited to surface waters of rivers and streams; values for groundwater, porewater, saturated soils, lakes, reservoirs, wetlands, estuaries, and floodplains were not included. Some papers were excluded because essential supporting information was missing (e.g., units), or extracting data from complex graphs was considered to be unwise. Data sources are listed in the Notes and Comments section below.

Methods

Rate of bubble releaseWe deployed 30 inverted funnel-style bubble traps (Molongoski and Klug, 1980; Baulch et al., 2011) on Allequash Creek on 31 May 2013 to measure volumetric bubble release rates. Fifteen traps were placed in two sandy sediment sections and 15 were placed in muck sediments in the wetland portion of the creek (number 8&ndash;22), which sits in-between the two sandy sections. Site 1 was the most downstream sampling site (water flows from East to West). Traps were sampled approximately every other day after 1 June 2013 until 31 October 2013 (we omitted the first samples collected 24 h following trap installation; total of 65 sample events per trap). Our sampling design allowed us to assess both the spatial and temporal variability in ebullition along Allequash Creek and how ebullition related to potential controlling factors such as sediment composition, atmospheric pressure, groundwater CH4, and organic matter content (discussed further below). To characterize our ebullition time series from Allequash Creek in the larger context of the NHLD, we installed an additional 12 traps on three additional creeks (Mann Creek, Stevenson Creek, and North Creek in the Trout Lake drainage; three per site in an even mix of sand and muck sediments) and the headwater spring ponds that drain into Allequash Creek on 23 June 2013 which we sampled approximately every week for the remainder of the study. Bubble traps had a bottom surface area of appr. 503 cm2 which narrowed at the top into a graduated (1 mL resolution) syringe and 3-way stopcock. Traps were attached to steel poles that were pounded into the substrate. Traps were almost completely submerged and contained no headspace at deployment. Water depth below traps averaged 55.7 cm, but we were unable to place traps in locations where water depth was shallower than 15 cm. Water velocity during baseflow at the traps averaged 0.06 m s -1 (range = 0.003&ndash;0.23 m s -1). We sampled traps by carefully approaching them either by boat (muck sites) or by wading (sandy sites) to avoid induced ebullition. Volume of accumulated gas in the trap was based on the graduated syringe, and volumes less than 1 mL were recorded as zero. Traps were reset between sampling events by refilling them completely with water to eliminate all headspace. To assess the hypothesis that declines in atmospheric pressure are related to increased bubble release (Mattson and Likens, 1990; Comas et al., 2011), we compared a 15 min resolution atmospheric pressure time series recorded using a Vaisala BAROCAP barometer deployed near trap number 7 with a subset of the bubble release time series.

Methods

Sampling DesignSampling of gas partial pressures, and gas transfer velocities was performed weekly at five stream sites (Mann Creek, Allequash Lower Creek, Allequash Middle, Stevenson Creek, North Creek) that drain into Trout Lake (Figure 1), one of the regions larger lakes. Sampling began in May 2012 and continued through September 2012. These data were used to establish variability in gas transfer rates for the basin, and to investigate spatiotemporal patterns. To test for the effects of upstream lakes, sampling at 30 additional longitudinal transect sites along 6 streams (5 sites per stream; Figure 1) was conducted approximately every 3 weeks beginning in May 2012. Transect streams were chosen based on the presence or absence of lakes in the upstream watershed. Streams with lakes (Lost, White Sand, Aurora) were sampled starting at the approximate lake outlet (site selection based on aerial photographs), and along a 2000 m transect (0m, 250m, 500m, 1000m, 2000m). Upstream lake chemistry (epilimnion) was also sampled during late July or early August 2012 at the lake center to allow for a direct comparison with streams. Streams without upstream lakes (Stella, Mud, and North) were sampled at an arbitrary upstream location (0m) and followed the same sampling progression as streams with lakes. We analyzed stream chemistry and stream morphology data from a regional stream survey (Lottig and others 2011) which we use to scale fluxes to the NHLD (Figure 1). We also studied groundwater CO2 and CH4 patterns along a hillslope transect at Allequash Creek during 2001. In 2002, we monitored hourly CO2 and O2 dynamics at the four WEBB sites to assess the role of ecosystem metabolism.

Methods

Site SelectionBecause lakes are a dominant feature of the region and stream characteristics could potentially differ based on their hydrologic connections to lakes, we classified streams into three categories as a function of their hydrologic connections to lakes. The first category was streams that had no lakes within the drainage network upstream of the sampling location. The second category was streams that originated from headwater lakes (i.e., no stream inlet but a stream outlet) and the headwater lake was the only lake in the drainage network above the sampling location. The final category had at least a single drainage lake (i.e., a lake with both stream inlet(s) and outlet) in the drainage network above the sampling location. We then used these categories to select sampling sites using a stratified random design for a variety of chemical and physical characteristics.All streams identified on 1:24,000 7.5 inch USGS topographical maps that crossed access points were selected as potential sampling locations and assigned to one of the three stream types. A stream could be classified by more than a single category depending on the sampling location within the drainage network. However, a single drainage network was never sampled more than once to ensure sample independence. Of the 500 possible sampling locations, 52 sites were selected and sampled.SamplingAll streams were sampled 7-10 channel widths upstream of an access point to minimize any influences caused by culverts and other features. Water samples were collected from the center of the channel using a peristaltic pump. Stream discharge was measured after Gore (2007) using cross sectional area and water velocity.Chemical AnalysesAll samples for both studies were collected and processed following the North Temperate Long Term Ecological Research (NTL-LTER) protocols (http://lter.limnology.wisc.edu). Filtering was done in the field using an in-line 0.45 μm membrane filter. All samples were stored on ice and returned to the laboratory where they were preserved according to NTL-LTER protocols. Acid neutralizing capacity (ANC) was determined by Gran titration (APHA 2005). DOC was measured on a Shimadzu TOC-V carbon analyzer. Total nitrogen and phosphorus (unfiltered, TN and TP; filtered, TDN and TDP), nitrate+nitrite (NO3-N), and ammonium (NH4-N) were quantified with an Astoria-Pacific segmented flow auto-analyzer. Soluble reactive phosphorus (SRP) in streams was measured colormetrically on a Beckman DU-800 spectrophotometer (APHA 2005). Anions (Cl- and SO4 2-) were measured using a Dionix DX-500 ion chromatograph and cations (Ca, Mg, Na, K, Fe, K, and Mn) on a Perkin Elmer ICP mass spectrometerDissolved inorganic carbon (DIC) and pH were quantified differently in the lakes and stream data sets. For the lakes data, DIC was determined with a Shimadzu TOC-V carbon analyzer, whereas DIC for the streams dataset was determined by headspace equilibration of acidified water samples in the field and direct measurement of carbon dioxide (CO2) gas on a Shimadzu gas chromatograph (Cole et al. 1994). pH measurements for the lakes dataset were quantified on non-air equilibrated samples in the lab with a Accumet 950 pH meter while direct measurements were taken in the field for the streams dataset using a hand-held Orion model 266 pH meter that was allowed to equilibrated about 20 min in the center for the stream channel.Several variables presented in this study were determined from calculations based on measured values. In streams, dissolved organic nitrogen and phosphorus (DON and DOP, respectively) were determined by the difference between inorganic nutrients and total dissolved nutrients (e.g., DOP = TDP-SRP). We were unable to determine DON in lakes due to the lack of inorganic nitrogen data. It was assumed that DOP approximately equals TDP in lakes because dissolved inorganic phosphorus concentrations in the region are typically below detection limits in the epilimnion during the summer months and consequently not quantified (NTL-LTER unpublished data).