Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Determining gas transfer velocities and CO₂ evasion fluxes from streams using carbon dioxide as a tracer McDowell, Mollie Jean 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_november_mcdowell_mollie.pdf [ 1.81MB ]
Metadata
JSON: 24-1.0355223.json
JSON-LD: 24-1.0355223-ld.json
RDF/XML (Pretty): 24-1.0355223-rdf.xml
RDF/JSON: 24-1.0355223-rdf.json
Turtle: 24-1.0355223-turtle.txt
N-Triples: 24-1.0355223-rdf-ntriples.txt
Original Record: 24-1.0355223-source.json
Full Text
24-1.0355223-fulltext.txt
Citation
24-1.0355223.ris

Full Text

 Determining gas transfer velocities and CO2 evasion fluxes from streams using carbon dioxide as a tracer by Mollie Jean McDowell B.A. (Cum Laude), Amherst College, 2014  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Geological Sciences)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) August 2017 © Mollie Jean McDowell, 2017  ii Abstract Evasion of carbon dioxide (CO2) from headwater streams is a dominant process controlling the fate of terrestrially-derived carbon (C) in inland waters. However, methodological limitations associated with determining the gas transfer velocity of carbon dioxide (kCO2) in headwater streams inhibit efforts to accurately quantify CO2 emissions. In this thesis, I present a proof of concept for a tracer gas method that mitigates common issues associated with conventional methods for determining kCO2. In this method, a datalogger controls in situ stream sensors that measure the partial pressure of CO2 (pCO2) and other stream parameters as well as a solenoid valve connected to a compressed CO2 cylinder. Automated injections of CO2 were made via an aquatic diffuser located on the stream bed. Infrared gas-analyzing (IRGA) CO2-type sensors enclosed in waterproof, gas-permeable membranes located downstream from the diffuser continuously measured aqueous pCO2 and equilibrate to elevated values during CO2 injections. The difference between upstream and downstream pCO2 values during CO2 injection relative to pre-injection concentrations permitted calculation of both the CO2 flux from the reach and kCO2. This method improves upon conventional methods due to its automation, in situ measurement, and use of CO2 as a tracer rather than another gas, thereby reducing analytical error and increasing the frequency and timing with which measurements can be made relative to conventional methods. I tested this method in a headwater stream in southwestern British Columbia. I calculated kCO2 and continuous CO2 emissions from the reach and compared both datasets to hydrogeomorphic parameters as well as values in the literature. Values of kCO2 were generally above the average values reported in the literature, but they corresponded well to values reported for steep, turbulent headwater streams. Values of kCO2 varied in relation to discharge, flow velocity, and stream temperature. CO2 emissions from the stream were highest  iii during high flow events. Headwater streams, which have been shown to be "hotspots" for CO2 emissions, can also be considered as exhibiting "hot moments" of CO2 evasion.    iv Lay summary Characterizing the global carbon cycle is important for many aspects of earth science, including those associated with climate change. Inland waters, such as streams, rivers, lakes, and estuaries, play a role in the carbon cycle by carrying carbon from the land into the ocean. Researchers have recently discovered that inland waters, and headwater streams in particular, do not just carry carbon but also process it, such that much of the carbon that enters an inland water body is either stored in sediments or evades into the atmosphere as carbon dioxide (CO2), a potent greenhouse gas. However, limitations of current methods for measuring CO2 evasion from streams have resulted in analytical errors and a dearth of data. In this thesis, I outline a new method that mitigates many of the errors associated with conventional methods for determining CO2 evasion from streams.    v Preface This thesis was completed under the supervision of Dr. Mark S. Johnson, who provided the project idea, grant funding, guidance, suggestions, and editing assistance. Dr. R. Dan Moore also provided support as a supervisory committee member, particularly regarding hydrological measurements and statistical analyses. It is the original and unpublished work of the author, Mollie J. McDowell.   vi Table of Contents Abstract ..................................................................................................................................... ii Lay summary ........................................................................................................................... iv Preface....................................................................................................................................... v Table of contents ...................................................................................................................... vi List of figures ............................................................................................................................ x List of symbols and acronyms ................................................................................................ xii Acknowledgements ................................................................................................................ xiv Dedication ............................................................................................................................... xv 1 INTRODUCTION ................................................................................................................. 1 2 METHODS ............................................................................................................................ 6 2.1 Site description................................................................................................................. 6 2.2 Experimental setup........................................................................................................... 9 2.3 CO2 injection and gas transfer velocity calculation ....................................................... 12 2.4 Field measurements, sensor calibration, and sensor maintenance ................................. 15 2.5 Gas transfer velocity data analysis and method validation ............................................ 17 2.6 Continuous CO2 emissions estimates............................................................................. 18 3 RESULTS ............................................................................................................................ 19 3.1 Continuous monitoring of stream and environmental parameters ................................. 19 3.1.1 Weather .............................................................................................................................. 19 3.1.2 CO2 variability and stream chemistry................................................................................. 19  vii 3.2 Discharge determinations from salt slug injections ....................................................... 23 3.2.1 Hydrology ........................................................................................................................... 23 3.3 Gas transfer velocity estimates ...................................................................................... 25 3.3.1 Relationships between gas transfer velocities and stream parameters ............................... 28 3.3.2 Linear regression models of gas transfer velocities and stream parameters ...................... 32 3.3.3 Model validation of k600 ..................................................................................................... 36 3.4 Continuous CO2 emissions estimates............................................................................. 38 4 DISCUSSION ...................................................................................................................... 42 4.1 Gas transfer velocity estimates ...................................................................................... 42 4.2 Relationships between gas transfer velocities and stream parameters ........................... 46 4.3 CO2 evasion estimates.................................................................................................... 48 4.4 Advantages and drawbacks of automated CO2 injections as a tracer for kCO2 and FCO2 50 4.5 Considerations................................................................................................................ 52 5 CONCLUSIONS.................................................................................................................. 54 REFERENCES ....................................................................................................................... 56 Appendix 1 .............................................................................................................................. 64 Appendix 2 .............................................................................................................................. 71 Appendix 3 .............................................................................................................................. 77 Appendix 4 .............................................................................................................................. 78     viii List of tables Table 1. Descriptions, manufacturers, measured parameter(s) (for this experiment), and accuracy of sensors used in experiment. ......................................................................... 10 Table 2. Variables used in calculation of kCO2. ....................................................................... 14 Table 3. Equations used in calculation of kCO2. ...................................................................... 15 Table 4. Seven models for predicting k600 (m d-1) based flow velocity (V, in m s-1), slope (S, unitless), stream depth (D, in m), discharge (Q, in m3 s-1), and the Froude number (Fr = V/(gD)0.5), described in Raymond et al. (2012). ............................................................. 18 Table 5. Monthly rainfall totals (mm) and average (ºC) temperatures during the study period and 30-year means. ......................................................................................................... 22 Table 6. Mean and median stream pCO2 values during each month of the study period. ...... 22 Table 7. Equations, coefficients of determination, and p-values of regression models that predict discharge (Q), flow velocity (V), and slug travel time (T) from stage (D), based on calculations from salt slug injections (n = 19). .......................................................... 23 Table 8. Values calculated from 38 CO2 injections: kCO2 (m d-1), k600 (m d-1), mean stream discharge (Q, in L s-1), modal stream velocity (V, in m s-1), and mean stream temperature (T, in ºC). .................................................................................................... 26 Table 9. Equations, coefficients of determination, residual standard errors (RSE), and p-values of linear regression models that predict log(kCO2) and log(k600) from mean stream discharge (Q), modal stream velocity (V), and mean stream temperature (T) as single regressor terms. ............................................................................................................... 35 Table 10. Equations, coefficients of determination, residual standard errors (RSE), and p-values of multiple linear regression models that predict log(kCO2) and log(k600) from  ix mean discharge (Q), modal stream velocity (V), and mean stream temperature (T) as regressor terms. ............................................................................................................... 36 Table 11. Equation, coefficients of determination, residual standard error (RSE), and p-value of a linear regression model that predicts modeled k600 from measured k600. ................. 38 Table 12. Equations, coefficients of determination, residual standard errors (RSE), and p-values of linear regression models that predict log(FCO2) from discharge (Q) and flow velocity (V) as single regressor terms. ............................................................................ 42     x List of figures  Figure 1. Location of the UBC Malcolm Knapp Research Forest within southwestern British Columbia, Canada. (Google Earth) ................................................................................... 7 Figure 2. (a) Location of study reach in experimental stream G-H (white box) within MKRF (designated by roads and streams), and (b) study reach in experimental stream G-H. Photo taken on March 18, 2017; perspective is facing downstream. (Google Earth) ...... 9 Figure 3. Schematic diagram of field setup. In situ stream sensors include IRGA CO2 probes, pH probes, CTD, and GS3 (upstream and downstream sensor bundles are identical). The weir hut contains the CR1000 datalogger and battery. ................................................... 11 Figure 4. Time series of 6-hour time-averaged (a) air temperature, (b) pCO2, (c) stream temperature, and (d) electrical conductivity observations during the study period. ....... 20 Figure 5. (a) Time series of 6-hour time-averaged wind speed observations during the study period and (b) wind rose of frequencies of 30-minute time-averaged wind speed observations by direction during the study period. ......................................................... 21 Figure 6. Time series of 6-hour time-averaged (a) stream discharge and (b) velocity during the study period. .............................................................................................................. 25 Figure 7. Boxplots of (a) kCO2 and (b) k600 at low discharge (Q < 100 L s-1) and high discharge (Q > 100 L s-1). ............................................................................................... 29 Figure 8. Boxplots of (a) kCO2 and (b) k600 at low flow velocity (V < 0.13 m s-1) and high flow velocity (V > 0.13 m s-1). ................................................................................................ 30 Figure 9. Boxplot of (a) kCO2 and (b) k600 at low stream temperature (T < 10 ºC) and high stream temperature (T > 10 ºC)....................................................................................... 31 Figure 10. Boxplots of (a) kCO2 and (b) k600 during daytime and nighttime. .......................... 32  xi Figure 11. Bivariate plots of (a) log(kCO2) and stream discharge, (b) log(k600) and stream discharge, (c) log(kCO2) and stream velocity, (d) log(k600) and stream velocity, (e) log(kCO2) and stream temperature, and (f) log(k600) and stream temperature (n = 38). Blue lines are linear regression models with equations and coefficients of determination (r2) given in panel (all p-values < 0.001). Shaded areas are 95% confidence envelopes. Error bars reflect minimum and maximum pCO2 measurements considering ±2% IRGA uncertainty (some within the margins of points). Color scale for values and error bars in (a) – (d) indicate stream temperature. ............................................................................. 34 Figure 12. Bivariate plot of measured k600 and mean of the modeled k600 calculated from seven models described in Raymond et al. (2012) (n = 38). The blue line is a linear regression model with equation:   k600 modeled = 0.761k600 measured + 33.541 (r2 = 0.66, p < 0.001). The shaded area is the 95% confidence envelope. The red line is the 1:1 line. Error bars reflect minimum and maximum k600 calculations considering ±2% IRGA uncertainty (some within the margins of points). ........................................................... 37 Figure 13. Time series of 6-hour time-averaged (a) kCO2, (b) k600, and (c) CO2 emissions (FCO2) during the study period. ....................................................................................... 40 Figure 14. Bivariate plots of (a) continuous discharge (Q) and log(FCO2) and (b) continuous flow velocity and log(FCO2) (n = 194). The blue lines are (a) a linear regression model with equation: log(FCO2) = 0.01Q + 2.58 (r2 = 0.98; p < 0.001) and (b) a loess regression model (RSE = 0.11). The shaded areas are the 95% confidence envelopes. .................. 41    xii List of symbols and acronyms Symbol/Acronym Units (constant) Definition Fr N/A Froude number adj. r2 N/A coefficient of determination adjusted for number of external regressors RSE m d-1 residual standard error r2 N/A coefficient of determination 𝑘𝐻⊝ atm M-1 (0.035) Henry's Law constant for solubility in water at standard temperature and pressure T⊝ K (298.15) temperature at standard temperature and pressure C g carbon CO2 N/A carbon dioxide CO2(air) atm atm-1 atmospheric pCO2 CO2(aq) atm atm-1 stream pCO2 d m depth/stage F g L-1 d-1 flux FCO2 g C L-1 d-1 flux of carbon dioxide IRGA N/A infrared gas analyzer k m d-1 gas transfer velocity k600 m d-1 gas transfer velocity with a Schmidt number = 600 kCO2 m d-1 gas transfer velocity of carbon dioxide kgas m d-1 gas transfer velocity of a gas other than carbon dioxide kH N/A Henry's Law constant for solubility in water at stream temperature pCO2 atm atm-1 partial pressure of carbon dioxide  xiii pH mol L-1 potential of hydrogen Q L s-1 or m3 s-1 discharge RMSE m d-1 root mean square error S degrees or percentage slope ScCO2 N/A Schmidt coefficient of carbon dioxide Scgas N/A Schmidt coefficient of a gas other than carbon dioxide T ºC or K temperature t s time V m s-1 velocity CO2 atm atm-1 excess CO2 in solution    xiv Acknowledgements I am incredibly grateful for the guidance and support I have received from mentors, colleagues, friends, and family throughout the process of completing this project. I thank my supervisor, Mark Johnson, and my supervisory committee member, Dan Moore, for their guidance, support, and patience in helping me develop as a researcher, regarding this project in particular but also through their respective courses. Additionally, this work was supported by an NSERC Discovery Grant to Mark Johnson.  I am very grateful for Iain Hawthorne's invaluable technical and field assistance and for the entire Ecohydrology Lab's support. I also thank Teddy Eyster, Kristi Collins, Michael Oh, Robin Arnold, Emma Luker, and Alida O'Connor for their field assistance. Further, I thank Zoran Nesic and Hughie Jones for their technical assistance, Brenda D'Acunha for her lab assistance, Ionut Aron for providing support at Malcolm Knapp Research Forest (and for saving me from getting stuck in the field twice), and Sally MacIntyre and Ted Tedford for their insightful conversations and imparting their expertise. Finally, I thank Rhy McMillan, whose editorial assistance and endless support have been a gift, the Miller-Vedam family, who have extended their home and love to me and therefore have supported me finishing this project immensely, and my family, who continues to support me from all over the continent, as they always have. This research was conducted at the University of British Columbia, which is on the unceded traditional territory of the Musqueam First Nation. The field experiment was conducted at the Malcolm Knapp Research Forest, which is on the unceded traditional territory of the Katzie First Nation.     i Dedication For my sisters.  1 1 INTRODUCTION Thorough characterization of the global carbon (C) cycle relies on accurate quantification of carbon fluxes into and out of all ecosystems. Among these fluxes, carbon dioxide (CO2) evasion from surface waters has received increasing attention due to the active roles that streams, rivers, lakes, and estuaries play in transforming terrestrially-derived carbon (Aufdenkampe et al., 2011; Battin et al., 2009). In this way, inland waters significantly reduce the amount of terrestrial C that rivers ultimately deliver to the ocean (Cole et al., 2007).  Terrestrially-derived imports of C to a freshwater ecosystem are subsequently partitioned into C lost to the atmosphere through evasion, C stored in sediments, and C exported from the ecosystem through drainage (Cole et al., 2007).  The majority of C inputs to streams and rivers are highly spatiotemporally variable (Wallin et al., 2011); they originate in the terrestrial environment, are transmitted via soils, and enter surface water systems through a range of hydrological flowpaths including groundwater-derived baseflow, shallow subsurface stormflow, and surficial runoff (Aufdenkampe et al., 2011). Headwater streams are particularly active sites of C cycling, not only because of their strong interactions with benthic substrates and the atmosphere (Benstead and Leigh, 2012), but also because they receive the majority of landscape drainage and are therefore closely coupled to terrestrial biogeochemical processes (Gomi et al., 2002).  A recent global estimate of 2.58 Pg C y-1 (Sawakuchi et al., 2017) suggests that almost half of terrestrially-derived C that enters streams and rivers is lost to the atmosphere through gaseous evasion, 36-64% of which evades from headwater streams (Marx et al., 2017; Raymond et al., 2013; Sawakuchi et al., 2017). Globally, pCO2 values in headwater streams and effluxes of CO2 from headwater streams are large but poorly constrained, resulting in  2 significant uncertainties in quantifying global C budgets (Butman and Raymond, 2011; Raymond et al., 2013; Marx et al., 2017). Due to the considerably higher partial pressures of CO2 (pCO2) in headwater streams relative to pCO2 in rivers (Butman and Raymond, 2011) and their proximity to terrestrially-derived C sources (Jones and Mulholland, 1998), many studies suggest that headwater streams are locations of substantial CO2 evasion (Schelker et al., 2016; Wallin et al., 2013). However, current methodological limitations compromise our ability to thoroughly evaluate the significance of CO2 fluxes between headwater streams and the atmosphere (Marx et al., 2017; Raymond et al., 2013). In particular, determination of the gas transfer velocity (k), a parameter that describes the rate of gas exchange across an air-water interface, presents a number of methodological challenges in headwater stream settings as it is spatiotemporally variable and dependent upon complex hydrogeomorphic stream properties (Raymond et al., 2012; Wallin et al., 2011).  In this study, I present an automated and field-deployable method for determining gas transfer velocities of CO2 (kCO2) in headwater streams. Accurate CO2 evasion flux estimates from streams rely on frequent and accurate determinations of kCO2 that can be scaled with hydrogeomorphic and flow parameters for a given reach; difficulties in determining and scaling kCO2 therefore result in large uncertainties in calculating evasion fluxes. I also present relationships between kCO2 values of a headwater stream reach determined using this method and stream parameters measured including discharge, velocity, and temperature, as well as estimates of continuous CO2 evasion from the reach. Thus, this method will allow researchers to determine kCO2 frequently and accurately in tandem with streamflow parameters, create site-specific scaling relationships, and calculate continuous CO2 evasion fluxes from a reach.  3  The flux of CO2 across an air-water interface (FCO2, in g C L-1 d-1) can be described in terms of the partial pressure gradient of CO2 between air (pCO2 air, in atm atm-1) and water (pCO2 aq, in atm atm-1) and kCO2 (in m d-1), such that: FCO2 = kCO2 x (pCO2 aq – pCO2 air) (1) Reorganizing equation (1), kCO2 can be calculated from FCO2, pCO2 aq, and pCO2 air, such that: kCO2 = (pCO2 aq – pCO2 air) / FCO2 (2) Specifically, kCO2 describes the height of a column of water that equilibrates with the atmosphere per unit of time (Frankignoulle et al., 1998; Wanninkhof et al., 2009).  In headwater streams, spatiotemporal variability in kCO2, driven primarily by surface water turbulence, is the largest determinant of CO2 effluxes (Hope et al., 2001; Tsivoglou and Neal, 1976; Zappa et al., 2007). Complex stream morphologies, including variable gradients and widths, streambed roughness, and tortuous flowpaths enhance the generation of surface water turbulence (MacIntyre et al., 1995; Wallin et al., 2011), with turbulence generally increasing with stream discharge and flow velocities (Billett and Harvey, 2013). Thus, the development of relationships between kCO2 and hydrogeomorphic and hydraulic parameters in headwater streams, calculated via accurate determinations of FCO2, may allow for more accurate prediction of kCO2 and its scaling among different catchments and stream orders (Raymond et al., 2012).  Multiple methods exist for both experimentally determining and modeling kCO2. Floating chambers can contain or be coupled to non-dispersive infrared (NDIR) sensors that directly and continuously monitor CO2 in the chamber headspace and provide time-weighted mean values based on water-air headspace equilibration, from which the rate of CO2 accumulation is measured (Alin et al., 2011; Campeau et al., 2014; Vachon et al., 2010).  4 However, researchers criticize this method due to inherent chamber effects on surface water turbulence and modified conditions inside the chamber (Gafalk et al., 2013; Crawford et al., 2014; Raymond and Cole, 2001). Additionally, some studies use the eddy covariance technique to determine kCO2 and FCO2 in large rivers (Huotari et al., 2013) and lakes (Jonsson et al., 2008), which allows for accurate and direct measurement of CO2 fluxes at the ecosystem scale (Huotari et al., 2013). However, this method is not appropriate for small streams, because it requires a large fetch area around the measurement location (Wallin et al., 2011).  Due to the limitations for floating chamber use, particularly for highly turbulent streams, tracer gas injections have become the most widely-used and robust method for determining kCO2 in headwater streams (Natchimuthu et al., 2017; Öquist et al., 2009; Tobias et al., 2009). This method involves injecting an inert volatile gas tracer (e.g., sulfur hexafluoride (SF6), propane (C3H8), or methyl chloride (CH3Cl)) and measuring its loss over a specified stream reach (MacIntyre et al., 1995; Wanninkhof et al., 1990). After adequate mixing has occurred, sampling of the stream in two locations downstream of the gas injection site allows for determination of the tracer concentration via headspace analysis on a gas chromatograph. The resulting ktracer is then converted to kCO2 using Schmidt dependences (e.g., Clark et al., 1995; Kokic et al., 2015; Looman et al., 2016). The conversion requires empirically-derived coefficients for both gases, as expressed in the following equation: ktracer / kCO2 = (Sctracer / ScCO2)-n (3) where Sc is the Schmidt number, defined as the ratio of the kinematic viscosity of water and the diffusion coefficient of the gas (Raymond et al., 2012), and n is the Schmidt number exponent, which is assigned a value of 1/2 to 2/3 depending on the surface state of the water (Jähne et al., 1987). Frequently, researchers couple tracer gas injection with a non-volatile  5 solute tracer injection (e.g., NaCl) to determine discharge and the travel time of the reach length (Genereux and Hemond, 1992; Marzolf et al., 1994; Shaw et al., 2010).  Tracer gas injections have been the preferred method for experimentally determining kCO2 in headwater streams as they do not affect the air-water interface, while providing an integrated measure of the exchangeability of CO2 in a stream reach at a specific point in time (Marx et al., 2017; Wallin et al., 2011). However, due to the manual nature of gas injection, stream sampling, and laboratory analysis, most tracer gas experiments do not readily allow for frequent sampling or continuous monitoring (Marx et al., 2017). In a metadata analysis of 563 tracer gas experiments, Raymond et al. (2012) determined relationships between k and stream hydraulic and slope parameters and reported seven regression equations (mean r2 = 0.63). Although these equations perform well over large spatial scales, the authors cautioned that direct measurements of k are still necessary in small-scale studies. In particular, systems with high slopes and velocities, such as headwater catchments, require direct measurements to accurately determine k (Raymond et al., 2012; Wallin et al., 2011). Additionally, measurements of k made at high spatial and temporal frequencies, especially those that capture diurnal, seasonal, hydrologic, and climatic variability, will aid in accurately determining flux measurements (Marx et al., 2017). To address these issues, I developed an automated method to determine kCO2 using CO2 as a tracer. I tested this approach under a range of flow conditions for a first-order stream of a headwater catchment in southwestern British Columbia, Canada. I used two in situ infrared gas analyzing (IRGA) CO2-type sensors enclosed in waterproof, gas-permeable membranes (Johnson et al., 2010) to continuously measure pCO2 downstream from a CO2 gas diffuser that delivered periodic injections of CO2 to the stream environment. Using CO2 as a tracer presents  6 advantages of being automatable and field-deployable, and it does not require supplemental gas chromatography or conversions via Schmidt dependences, as is the case for most tracer gas experiments. Thus, it can be used to make measurements at different temporal resolutions, including at night, or in response to events of scientific interest, such as storms. The automation and programmability of this method will allow researchers not only to augment current datasets of kCO2, but also to readily scale kCO2 with hydrogeomorphic parameters.  In the present study, I deployed this autonomous system to enable frequent and accurate determinations of kCO2 across a range of flow conditions. I hypothesized that a high stream slope and turbulent sections within the reach would result in high kCO2 values overall, and that kCO2 would correspond positively with flow parameters (e.g., discharge and stream velocity). Multiple determinations of kCO2 obtained during varying flow conditions allowed me to interpolate kCO2 values for the entire study period and calculate continuous CO2 emissions from the stream. As headwater systems are known to be “hotspots” for CO2 evasion, the proof of concept presented here suggests a route forward for better elucidating the role that CO2 fluxes from headwater streams play in the global C cycle.  2 METHODS 2.1 Site description I conducted the research reported on in this thesis at the University of British Columbia Malcolm Knapp Research Forest (MKRF) located in Maple Ridge, British Columbia, approximately 60 km east of Vancouver (49°16’N 122°34’W) (Figure 1). MKRF is located in the Fraser Valley of British Columbia and the Coastal Western Hemlock Biogeoclimatic Zone. It typically experiences mild, wet winters and warm, dry summers. At upper East Creek in  7 MKRF, mean annual precipitation is 2353 mm yr-1, mean annual temperature is 9°C, and mean monthly temperature ranges from 1.4-16.8°C (Richardson and Moore, 2010).   MKRF primarily contains a mixture of Douglas fir, western hemlock, and western red cedar, as well as big leaf maple, black cottonwood, and red alder (Turk et al., 1998). The understory vegetation primarily consists of vine maple, western sword fern, salal, and trailing blackberry (Turk et al., 1998). The soil is a Gleyed Brunisol with dominant textures of sandy loam and loamy sand (Tashe 1998). Parent material consists primarily of colluvium and till (Klinka 1976).    Figure 1. Location of the UBC Malcolm Knapp Research Forest within southwestern British Columbia, Canada. (Google Earth)  8  This work was carried out in experimental stream “G-H”, located in the southeastern part of MKRF (Figure 2). The 64-m reach of G-H used in the present study flows southwest below a debris jam and a weir at the outlet. The mean width of the reach is 4 m and the mean slope is 0.236 (13.3°). The reach is heavily shaded by vegetation, with a steep southeastern riparian area and a shallower northwestern riparian area intersected by a logging road. Riffle-run sequences dominate the upper part of the reach, generating turbulence, with variable depth pools more common in the lower part of the reach. Many streams in MKRF, G-H included, partially or entirely dry up in the summer months. The catchment for G-H is 0.54 km2 and ranges in elevation from 235–330 m.   9  Figure 2. (a) Location of study reach in experimental stream G-H (white box) within MKRF (designated by roads and streams), and (b) study reach in experimental stream G-H. Photo taken on March 18, 2017; perspective is facing downstream. (Google Earth)  2.2 Experimental setup A weir hut located on the stream bank at the outlet of the reach provided protection for a power supply and data acquisition system. A 12-V battery supplied power to a Campbell Scientific CR1000 datalogger and AM16/32B multiplexer, two IRGA CO2-type gas  10 analyzing sensors, six water quality sensors, a sonic anemometer, a wireless modem, and a solenoid valve connected to a compressed gas cylinder (Table 1).   Table 1. Descriptions, manufacturers, measured parameter(s) (for this experiment), and accuracy of sensors used in experiment. Sensor Manufacturer Parameter(s) measured Accuracy GMP221 Vaisala pCO2 ±1% of range + 2% of reading CSIM11-L Campbell Scientific pH ±0.1% over full range GS3 Decagon electrical conductivity stream temperature ±10% (EC) ±1 ºC (stream temperature) CTD-10 Decagon depth 0.05% of full scale at 20 ºC 81000 ultrasonic anemometer RM Young air temperature wind speed wind direction ±2 ºC (air temperature) ±0.05 m s-1 (wind speed) ±2º (wind direction)  Sensors were distributed between two groups, each of which was fixed on the stream bed: one group at the reach halfway point (32 m from diffuser) and one at the end of the reach (63 m from diffuser). Each group contained the following sensors: a Vaisala GMM220 transmitter module with a GMP221 CO2 probe (Vantaa, Finland), a Campbell Scientific CSIM11-L pH probe (Logan, Utah, USA), a Decagon GS3 ruggedized soil moisture, temperature, and electrical conductivity probe (Pullman, Washington, USA), and a Decagon CTD-10 electrical conductivity, temperature, and depth probe. Figure 3 provides a schematic diagram of field setup. The stream parameter values discussed below reflect measurements from the upstream  11 group of sensors, which were located mid-reach and were shaded by forest cover. The downstream portion of the reach was less fully shaded due to proximity to the weir and weir hut. An RM Young Model 81000 ultrasonic anemometer (Traverse City, Michigan, USA) was situated 2 m above the ground on the stream bank between the upstream and downstream sensors.   Figure 3. Schematic diagram of field setup. In situ stream sensors include IRGA CO2 probes, pH probes, CTD, and GS3 (upstream and downstream sensor bundles are identical). The weir hut contains the CR1000 datalogger and battery.   The datalogger program provided commands for the sensors and solenoid valve through the datalogger and multiplexer, and it compiled sensor outputs into two data tables (see Appendices 1 and 2 for CR1000 programs). The main program collected continuous data from most sensors at a 5-second scan interval, averaging data into 30-minute averages. The IRGA sensors, with 4W demand each, were programmed to turn on for the last five minutes of each half-hour time block. This 30-minute data table collected data continuously and generated 48 distinct values for each measured component per day. During gas injection Diffuser Gas cylinder and solenoid valve Upstream sensors Downstream sensors Sonic anemometer Weir hut  12 periods, described in detail below, which occurred twice daily for an hour per injection (5-6 am and 5-6 pm local time), the datalogger collected data from all sensors at 5-second intervals, storing this higher frequency data in a separate table. A cellular modem was turned on daily between 8-9 am local time, and the datalogger transmitted both data tables to a UBC laboratory computer. A data gap from December 25, 2016, to January 22, 2017, resulted from a period when access roads at the site were impassable due to snow and power to the system was lost due to an interruption in battery swaps. Other data gaps reflect tracer addition periods (CO2 injections or salt additions used for discharge determination), as well as periods removed due to sensor issues.   2.3 CO2 injection and gas transfer velocity calculation During gas injection periods, a solenoid valve was used to switch the CO2 injection gas flow on for one hour, allowing the compressed gas cylinder to continuously deliver industrial-grade CO2 at a streamflow-dependent flow rate of 5 L min-1 – 10 L min-1. CO2 flowed from the cylinder via U.S. Plastic Corporation Bev-A-Line tubing (Lima, Ohio) to a 1-meter long aquatic diffuser situated on the stream bed at the inlet of the reach, just below the debris dam. Upstream and downstream IRGAs recorded increasing pCO2 values as the injected CO2 dissolved and mixed into the reach, eventually equilibrating at elevated concentrations during the injection. The relative differences between baseline and elevated values for each sensor were used in calculation of kCO2. Henry's Law was used to determine the mass equivalence of dissolved CO2 (here expressed as in carbon equivalent terms as mg CO2-C L-1) as a function of stream temperature  13 and pCO2 in solution (Plummer and Busenberg, 1982). The difference between upstream and downstream mass equivalences of CO2-C allowed me to calculate the evasion flux of CO2 along the stream reach. This information, in combination with excess CO2 in solution, permitted determination of kCO2 using the variables listed in Table 2 via the equations given in Table 3. For the purpose of compatibility with existing literature, I report both kCO2 and k normalized to a Schmidt number of 600 (k600), which corresponds to a temperature of 20ºC for CO2 (Jähne et al., 1987). kCO2 can be converted to k600 using the following equation: k600 = kCO2 x (600 / ScCO2)-0.5 (4) I calculated kCO2 and k600 from 49 CO2 injections, discarded 6 negative and 5 very large outliers, and used the remaining 38 discrete measurements of kCO2 and k600 for assessing relationships with other variables. Negative kCO2 and k600 values are not physically meaningful in this stream due to continual CO2 supersaturation in streamwater and high turbulence. Issues regarding the gas cylinder setup and harsh winter weather presented challenges through February 2017, so the kCO2 and k600 values reported here were determined during the last four months of the experiment.    14 Table 2. Variables used in calculation of kCO2. Symbol Description (units) [constant] 𝑘𝐻 Henry's Law constant for solubility in water at stream temperature  𝑇 Temperature (K) 𝑘𝐻⊝ Henry's Law constant for solubility in water at standard temperature and pressure (STP) [0.035] 𝐷 Temperature dependence constant (K) [2400] 𝑇⊝ Temperature at STP (K) [298.15] 𝐶 Mass equivalence of dissolved CO2 (mg CO2-C L-1) 𝐶𝑂2 pCO2 resulting from injection relative to baseline (atm) 𝛥𝐶𝑂2 Excess CO2 in solution (atm) 𝐶𝑂2(𝑎𝑞) Mean pCO2 in stream (atm) 𝐶𝑂2(𝑎𝑖𝑟) pCO2 in atmosphere (atm) 𝐹 CO2 evasion flux along reach (g C L-1 d-1) d Depth (m) t Travel time between salt slug centroids (s) 𝑘𝐶𝑂2 Gas transfer velocity (m d-1)    15 Table 3. Equations used in calculation of kCO2.  Equation 𝑘H(1) = 𝑘H⊝ ×  exp(𝐷 (1𝑇(1)−1𝑇⊝)) 𝑘H(2) = 𝑘H⊝ ×  exp(𝐷 (1𝑇(2)−1𝑇⊝)) 𝑘H(𝑚𝑒𝑎𝑛) = 𝑘H⊝ ×  exp (𝐷 (1𝑇(𝑚𝑒𝑎𝑛)−1𝑇⊝)) 𝐶(1) = 𝑘H(1) ×  CO2(1) ×  12 𝐶(2) = 𝑘H(2) ×  CO2(2) ×  12 ΔCO2 = 𝑘H(𝑚𝑒𝑎𝑛) ×  (CO2(𝑎𝑞) - CO2(𝑎𝑖𝑟))×  12 𝐹 =  𝐶(1) −  𝐶(2) 𝑡 𝑘CO2 =𝛥𝐶𝑂2 ×  𝑑𝐹   2.4 Field measurements, sensor calibration, and sensor maintenance Many in situ sensors require both pre-deployment and routine calibration and maintenance to ensure proper functioning and data validity. Before sensor installation, I wrapped all sensors except the two CTDs in a protective perforated PVC wrap. Sensors were individually calibrated according to manufacturer calibration instructions. While the CTD sensors provide information on stream water EC, the EC output from the GS3 sensors has  16 higher resolution, enabling the development of stage-discharge relationships described below. I performed sensor maintenance on weekly field visits, except when logging roads in MKRF were impassable due to snow (much of January and February 2017). During field visits and where applicable, I cleaned all sensors with water or soapy water and a cloth, flushing any sediment or debris that accumulated in the protective PVC wrap. I used a Hanna Low Range pH/Conductivity/TDS PPM Tester HI98129 (Woonsocket, Rhode Island, USA) to obtain weekly pH values against which the continuously recorded pH values could be adjusted based on standard methods for in situ water quality sensors (Gibs et al., 2007).  Prior to installation, I calibrated the IRGA sensors using the Vaisala CARBOCAP Carbon Dioxide Calibrator (GMK220) according to the two-point calibration procedure described in the manual (Vaisala Oyj, 2006). During field visits, I checked the calibration of the IRGA sensors by deploying a third IRGA sensor next to the in situ sensors for a few minutes and ensured that the outputs were similar. This check ensured that both deployed sensors continued to perform in a similar manner and without drift throughout the study. Deployment of a third IRGA also allowed me to survey the stream reach for areas of potential groundwater inputs which could locally elevate pCO2, as well as to assess changes in pCO2 along the reach during several gas injections.  The functionality of an IRGA sensor in an aqueous environment requires that the light-source, detector, and gas bench are enclosed within a waterproof, gas-permeable membrane (Johnson et al., 2010). Both IRGA sensors were encased in polytetrafluoroethylene (PTFE) sleeves and sealed with liquid electrical tape according to the method described in Johnson et al. (2010). Diffusivity measurements of the sleeve indicate that it has a negligible effect on CO2 diffusion and sensor response time (Johnson et al., 2010).  17 I measured stream discharge using salt additions as described in Moore (2005) and Richardson et al. (2017) to develop a rating curve for the stream, as the V-notch on the weir was greater than 90º. Using this method, I added 100 g of salt (NaCl) dissolved in 1 L streamwater to the top of the reach and recorded EC with the upstream and downstream GS3 sensors as the salt addition passed. I calculated discharge by integrating the downstream salt pulse, using a calibration constant of 0.486 mg·NaCl·cm·µS-1·L-1 (Richardson et al., 2017), taking the mean values based on two repeated salt additions.  The travel time between centroid of the salt pulse as it passed each EC sensor was used to determine modal stream velocity (V) (Waldon 2004). I subsequently evaluated relationships between stage, discharge, and velocity using regression models. These relationships were used to calculate continuous discharge and velocity for the entire dataset, as well as to determine the CO2 evasion dynamics and gas travel time between IRGA sensors during CO2 injections.  2.5 Gas transfer velocity data analysis and method validation Data analysis was conducted in R version 3.3.3 (R Core Team, 2017). I determined logarithmic linear regression relationships between kCO2 and k600 and hydraulic stream properties, including stream discharge, velocity, and temperature. I compared the k600 values determined using this method to outputs from seven models of kCO2 (Table 4) based on hydrogeomorphic variables described in Raymond et al. (2012). Continuous long-term data are presented as 6-hourly averages to reduce overplotting.    18 Table 4. Seven models for predicting k600 (m d-1) based flow velocity (V, in m s-1), slope (S, unitless), stream depth (D, in m), discharge (Q, in m3 s-1), and the Froude number (Fr = V/(gD)0.5), described in Raymond et al. (2012). Model equation 𝑘600 = (𝑉𝑆)0.89±0.020 × 𝐷0.54±0.030 × 5037± 604 𝑘600 = 5937 ± 606 × (1 − 2.54± 0.223 × 𝐹𝑟2) × (𝑉𝑆)0.89±0.017 × 𝐷0.58±0.027 𝑘600 = 1162 ± 192 × 𝑆0.77±0.028 × 𝑉0.85±0.045 𝑘600 = (𝑉𝑆)0.76±0.027 × 951.5 ± 144 𝑘600 = 𝑉𝑆 × 2841± 107 + 2.02 ± 0.209 𝑘600 = 929± 141 × (𝑉𝑆)0.75±0.027 × 𝑄0.011±0.016 𝑘600 = 4725 ± 445 × (𝑉𝑆)0.86±0.016 × 𝑄−0.14±0.012 × 𝐷0.66±0.029  2.6 Continuous CO2 emissions estimates  I determined continuous kCO2 and k600 for the entire study period using the linear regression relationships between log(kCO2), log(k600), and discharge. Back-transforming logarithmic regression models requires a bias correction and uncertainty calculation (Baskerville, 1972). A regression model of the form: log(y) = x = b0 + b1x + e (5) can be solved using the original terms and the sample residual variance of the logarithmic equation (), in which (Baskerville, 1972): y = e(x + /2) (6) The estimated variance in y (A) can be calculated via (Baskerville, 1972):  19 A = e(2 + 2x) - e( + 2x) (7) I determined continuous CO2 emissions from the stream using Equation (2), with continuous kCO2, continuous dissolved pCO2, and the average atmospheric pCO2 above the reach (427 atm atm-1), with the latter determined via headspace analysis on a gas chromatograph (Kling et al., 1991) (See Appendix 3 for description of analysis). I also determined regression relationships between continuous CO2 emissions and continuous stream discharge and velocity.   3 RESULTS 3.1 Continuous monitoring of stream and environmental parameters 3.1.1 Weather Precipitation at MKRF totaled 1387 mm over the study period (November 2016 – June 2017). Mean air temperature during the study period was 4 ºC and ranged from -11 to +21 ºC (Figure 4). Mean wind velocity measured within the forest and adjacent to the stream was 0.3 m s-1 and ranged from 0.004 – 0.94 m s-1, with about half of all observations originating from the north and northeast (Figure 5). Table 5 provides monthly rainfall totals and average temperatures for the study period as well as 30-year means. 3.1.2 CO2 variability and stream chemistry Baseline pCO2 values ranged from 1133 – 2110 atm atm-1 over the study period, with mean and median values of 1434 and 1404 atm atm-1 (Figure 4). Streamwater pCO2 decreased slightly between November and February, and diurnal variation remained relatively stable and limited between November and April. Both pCO2 and diurnal variation increased during May  20 and June. See Table 6 for mean and median values and ranges of pCO2 during each month of the study period. Streamwater pH ranged from 6.5 – 7.2, with an average value of 6.9. Stream temperature ranged from 0.3 – 15.2 ºC during the study period, with a mean value of 7 ºC (Figure 4). Electrical conductivity remained relatively low, with mean and median values of 12 and 11 s cm-1 and a range of 10 – 19 s cm-1 (Figure 4).     Figure 4. Time series of 6-hour time-averaged (a) air temperature, (b) pCO2, (c) stream temperature, and (d) electrical conductivity observations during the study period.      21  Figure 5. (a) Time series of 6-hour time-averaged wind speed observations during the study period and (b) wind rose of frequencies of 30-minute time-averaged wind speed observations by direction during the study period. (a)(b)0.20.40.60.82016−11−012016−12−012017−01−012017−02−012017−03−012017−04−012017−05−012017−06−012017−07−01Wind speed (m s-1) 22 Table 5. Monthly rainfall totals (mm) and average (ºC) temperatures during the study period and 30-year means. Month Rainfall (mm) Temperature (ºC)  Study period 30-year mean Study period 30-year mean November 310 334 5.3 5.3 December 254 289 -2.5 2.4 January 137 259 0.4 2.2 February 167 230 -1.6 4 March 326 209 2.7 6.2 April 176 161 5.5 9.1 May 131 136 9.3 12.4 June 21 107 11.5 15.1   Table 6. Mean and median stream pCO2 values during each month of the study period. Month mean pCO2 (atm atm-1) median pCO2 (atm atm-1) November 1531 1533 December 1442 1444 January 1349 1355 February 1327 1330 March 1365 1372 April 1347 1350 May 1458 1449 June 1637 1623    23 3.2 Discharge determinations from salt slug injections I calculated discharge, velocity, and travel time between salt pulse centroids based on data recorded by two electrical conductivity sensors from 24 salt additions between January and May, 2017. Of these calculations, 19 were used to model regression relationships between hydrologic variables and stage, while 5 were discarded as statistical outliers due to a poor relationship with discharge. Non-linear regression models that predicted stream discharge, stream velocity, and travel time using stage as an external regressor performed well, with r2 values of 0.95 (p < 0.001), 0.90 (p < 0.001), and 0.86 (p < 0.001), respectively (Table 7). Calculated stream discharge during salt additions ranged from 56.8 L s-1 – 438 L s-1, and calculated velocities during salt additions ranged from 0.11 – 0.46 m s-1. Travel times for the 31 m between groups of sensors ranged from 70 s at higher flows to 295 s at lower flows.   Table 7. Equations, coefficients of determination, and p-values of regression models that predict discharge (Q), flow velocity (V), and slug travel time (T) from stage (D), based on calculations from salt slug injections (n = 19). Equation r2 p-value 𝑄 = 0.005 × 𝐷1.856 0.95 < 0.001 𝑉 = 0.001 × 𝐷 − 0.036 0.90 < 0.001 𝑇 = 72944 × 𝐷−1.105 0.86 < 0.001  3.2.1 Hydrology The hydrograph over the study period (Figure 6) indicated rapid responses to frequent storm events, with discharge surpassing 400 L s-1 during six events throughout the study  24 period. Overall, discharge was higher during the rainy months of November and December and lower in late January and throughout February, which were uncharacteristically cold and snowy compared to the long-term record. Higher discharge resumed with spring rains in March. Mean and median discharge during the study period was 117.2 and 92.6 L s-1, respectively, and ranged from 9.2 – 601.5 L s-1. Mean flow velocity over the study period (Figure 6) ranged from 0.02 – 0.45 m s-1, with mean and median values of 0.15 and 0.14 m s-1, respectively.   25  Figure 6. Time series of 6-hour time-averaged (a) stream discharge and (b) velocity during the study period.    3.3 Gas transfer velocity estimates Mean and median kCO2 values were 48.9 and 31.4 m d-1, respectively, and ranged from 18.32 to 186.16 m d-1. Mean and median k600 values were 66.0 and 42.3 m d-1, respectively,  26 and ranged from 20.4 to 261 m d-1. Values for kCO2 and k600 as well as discharge, stream velocity, and stream temperature values during each CO2 injection are given in Table 8.  Table 8. Values calculated from 38 CO2 injections: gas transfer velocities (kCO2), gas transfer velocities normalized to a Schmidt number of 600 (k600), stream discharge (Q), stream velocity (V), and stream temperature (T). kCO2 (m d-1) k600 (m d-1) Q (L s-1) V (m s-1) T (ºC) 82.46 118.34 264.97 0.28 6.51 116.60 162.36 230.53 0.25 7.65 105.26 148.76 194.54 0.23 7.09 82.44 115.65 199.32 0.23 7.37 77.33 108.80 182.23 0.22 7.26 186.16 260.98 266.37 0.28 7.39 136.99 194.78 247.48 0.27 6.86 74.22 103.92 220.25 0.25 7.44 55.85 79.10 201.42 0.23 7.01 65.07 88.89 100.73 0.15 8.37 29.26 40.14 66.64 0.11 8.22 32.33 44.38 80.05 0.13 8.20 47.07 64.24 70.52 0.12 8.42 22.17 29.01 50.51 0.09 10.02 21.23 27.26 44.53 0.08 10.73 55.61 73.41 122.56 0.17 9.68  27 35.25 45.09 115.31 0.16 10.90 26.55 34.31 101.32 0.15 10.50 72.04 93.73 153.64 0.20 10.24 53.31 71.52 141.49 0.19 9.07 55.60 73.55 137.98 0.18 9.60 38.27 52.04 114.32 0.16 8.55 34.04 44.78 98.16 0.15 9.82 26.41 35.68 84.74 0.13 8.81 25.37 33.28 73.12 0.12 9.93 26.95 36.07 64.95 0.11 9.16 22.84 29.75 57.19 0.10 10.20 27.31 33.87 20.47 0.04 12.08 29.08 35.08 19.36 0.04 13.17 30.44 37.55 18.67 0.04 12.30 26.25 30.94 16.86 0.04 14.08 27.30 32.73 15.69 0.03 13.41 19.16 21.85 13.80 0.03 15.39 19.03 21.97 12.96 0.03 14.91 18.83 20.96 11.16 0.02 16.39 18.38 20.81 10.65 0.02 15.71 18.32 20.43 9.94 0.02 16.30 18.92 22.20 9.59 0.02 14.26   28 3.3.1 Relationships between gas transfer velocities and stream parameters Overall, I found that kCO2 and k600 were positively associated with stream discharge, with low values at low discharge (Q < 100 L s-1) and high values at high discharge (Q > 100 L s-1) (Figure 7). Mean values of kCO2 and k600 at low discharge were 25.8 and 32.5 m d-1, and mean values of kCO2 and k600 at high discharge were 77.6 and 107 m d-1. Similarly, kCO2 and k600 were positively associated with stream velocity, with low values at low velocity (V < 0.13 m s-1) and high values at high velocity (V > 0.13 m s-1) (Figure 8). Mean values of kCO2 and k600 at low flow velocity were 25.3 and 31.7 m d-1, and mean values of kCO2 and k600 at high velocity were 72.6 and 100 m d-1. kCO2 and k600 were inversely related to stream temperature, with low values at high temperature (T > 10 ºC) and high values at low temperature (T < 10 ºC) (Figure 9). Mean values of kCO2 and k600 at low stream temperature were 67.0 and 92.9 m d-1, and mean values of kCO2 and k600 at high stream temperature were 26.7 and 32.8 m d-1 Of the 38 CO2 injections, 21 were conducted during the day, and 17 were conducted at night. Mean values of kCO2 and k600 during daytime and nighttime were 51.4 and 69.0 m d-1 and 45.9 and 62.3 m d-1, respectively. Although daytime values were slightly higher, interquartile ranges of daytime and nighttime kCO2 and k600 overlapped significantly (Figure 10).   29  Figure 7. Boxplots of (a) kCO2 and (b) k600 at low discharge (Q < 100 L s-1) and high discharge (Q > 100 L s-1).     30   Figure 8. Boxplots of (a) kCO2 and (b) k600 at low flow velocity (V < 0.13 m s-1) and high flow velocity (V > 0.13 m s-1).    31  Figure 9. Boxplot of (a) kCO2 and (b) k600 at low stream temperature (T < 10 ºC) and high stream temperature (T > 10 ºC).    32  Figure 10. Boxplots of (a) kCO2 and (b) k600 during daytime and nighttime.   3.3.2 Linear regression models of gas transfer velocities and stream parameters  Linear regression models that predicted log(kCO2) and log(k600) from stream discharge, stream velocity, and stream temperature as single regressor terms performed well, with  33 discharge having the highest predictive power (r2 = 0.86 and 0.88; p < 0.001 and 0.001), velocity having the second highest (r2 = 0.82 and 0.85; p < 0.001 and 0.001), followed by temperature (r2 = 0.58 and 0.65; p < 0.001 and 0.001). Regression relationships corroborated overall relationships described above, with log(kCO2) and log(k600) positively associated with discharge and velocity and negatively associated with temperature.  Figure 11 provides bivariate plots of log(kCO2) and log(k600) and stream discharge, velocity, and temperature, as well as linear regression fits and 95% confidence intervals. Table 9 provides linear regression coefficients and statistics of the six models.   34  Figure 11. Bivariate plots of (a) log(kCO2) and stream discharge, (b) log(k600) and stream discharge, (c) log(kCO2) and stream velocity, (d) log(k600) and stream velocity, (e) log(kCO2) and stream temperature, and (f) log(k600) and stream temperature (n = 38). Blue lines are linear regression models with equations and coefficients of determination (r2) given in panel (all p-values < 0.001). Shaded areas are 95% confidence envelopes. Error bars reflect minimum and (a)(e)(d)(b)(c)(f)log(kCO2) = 0.007Q + 2.941 r2 = 0.86log(k600) = 0.008Q + 3.125 r2 = 0.88log(kCO2) = 6.790V + 2.768 r2 = 0.82log(k600) = 7.588V + 2.923 r2 = 0.85log(kCO2) = -0.165T + 5.379 r2 = 0.58log(k600) = -0.191T + 5.907 r2 = 0.653450 100 200Discharge (L s-1)log(kCO2)  (m d-1)34560 100 200Discharge (L s-1)log(k600) (m d-1)3450.1 0.2Stream velocity (m s-1)log(kCO2) (m d-1)34560.1 0.2Stream velocity (m s-1)log(k600) (m d-1)3458 10 12 14 16Stream temperature (ºC)log(kCO2) (m d-1)34568 10 12 14 16Stream temperature (ºC)log(k600) (m d-1) 35 maximum pCO2 measurements considering ±2% IRGA uncertainty (some within the margins of points). Color scale for values and error bars in (a) – (d) indicate stream temperature.  Table 9. Equations, coefficients of determination, residual standard errors (RSE), and p-values of linear regression models that predict log(kCO2) and log(k600) from stream discharge (Q), stream velocity (V), and stream temperature (T) as single regressor terms. Variable Equation r2 adj. r2 RSE p-value Q log(𝑘𝐶𝑂2) = 0.01𝑄 + 2.94 0.86 0.85 0.24 < 0.001 V log(𝑘𝐶𝑂2) = 6.79𝑉 + 2.77 0.82 0.81 0.28 < 0.001 T log(𝑘𝐶𝑂2) = −0.16𝑇 + 5.38 0.58 0.57 0.41 < 0.001 Q log(𝑘600) = 0.01𝑄 + 3.12 0.88 0.87 0.25 < 0.001 V log(𝑘600) = 7.59𝑉 + 2.92 0.85 0.85 0.27 < 0.001 T log(𝑘600) = −0.19𝑇 + 5.91 0.65 0.64 0.41 < 0.001  Multiple linear regression models that predicted log(kCO2) and log(k600) from either stream discharge and temperature or velocity and temperature did not outperform single linear regression models that predicted log(kCO2) and log(k600) from either stream discharge or velocity, with no enhancement of the predictive power of the models by adding a stream temperature term. Stream discharge and temperature had a higher predictive power (r2 = 0.86 and 0.88; p < 0.001 and 0.001) than stream velocity and temperature (r2 = 0.82 and 0.85; p < 0.001 and 0.001). Table 10 provides linear regression coefficients and statistics of the four models.   36 Table 10. Equations, coefficients of determination, residual standard errors (RSE), and p-values of multiple linear regression models that predict log(kCO2) and log(k600) from stream discharge (Q), stream velocity (V), and stream temperature (T) as regressor terms. Variables Equation r2 adj. r2 RSE p-value Q + T log⁡(𝑘𝐶𝑂2) = 0.01𝑄 − 0.003𝑇 + 2.98 0.86 0.85 0.25 < 0.001 V + T log⁡(𝑘𝐶𝑂2) = 7.92𝑉 + 0.04𝑇 + 2.24 0.82 0.81 0.27 < 0.001 Q + T log⁡(𝑘600) = 0.001𝑄 − 0.03𝑇 + 3.50 0.88 0.87 0.25 < 0.001 V + T log⁡(𝑘600) = 7.93𝑉 + 0.01𝑇 + 2.76 0.85 0.84 0.27 < 0.001  3.3.3 Model validation of k600 Using empirically determined data for discharge, flow velocity, and stage during CO2 injections as well as stream slope, I calculated k600 from the seven models provided in Raymond et al. (2012). Calculations of k600 for each CO2 injection period estimated by the seven models described by Raymond et al. (2012) yielded mean and median values of 83.8 and 85.0 m d-1, respectively, and range from 12.3 – 172 m d-1. The models overestimated k600 compared to measured k600 values by 17.8 m d-1 on average. Figure 12 illustrates the relationship between measured k600 and modeled k600, as well as the linear regression model fit and a 1:1 line. The root mean square error (RMSE) of the model was 29.0 m d-1. Table 11 provides linear regression coefficients and statistics of the model.   37  Figure 12. Bivariate plot of measured k600 and mean of the modeled k600 calculated from seven models described in Raymond et al. (2012) (n = 38). The blue line is a linear regression model with equation:   k600 modeled = 0.761k600 measured + 33.541 (r2 = 0.66, p < 0.001). The shaded area is the 95% confidence envelope. The red line is the 1:1 line. Error bars reflect minimum and maximum k600 calculations considering ±2% IRGA uncertainty (some within the margins of points).         0501001502000 50 100 150 200Measured k600 (m d−1)Modeled k600 (m d−1) 38 Table 11. Equation, coefficients of determination, residual standard error (RSE), and p-value of a linear regression model that predicts modeled k600 from measured k600. Equation r2 adj. r2 RSE p-value 𝑘600⁡𝑚𝑜𝑑𝑒𝑙𝑒𝑑 = 0.76𝑘600⁡𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑 + 33.54 0.66 0.65 29.74 < 0.001   3.4 Continuous CO2 emissions estimates  I computed kCO2 and k600 on a continuous basis for the study period using the regression relationships between log(kCO2) and log(k600) and discharge (Table 9) and Equations (6) and (7). I also calculated continuous CO2 emissions (FCO2) from the stream using continuous kCO2 and Equation (1). Continuous kCO2 and k600 reflected variability in continuous discharge, with kCO2 and k600 exceeding 500 and 900 m d-1 on two occasions, during high-flow (Q > 440 L s-1) events (Figure 13). Mean and median kCO2 and k600 values during the entire study period were 64.6 and 95.1 m d-1 and 36.9 and 47.7 m d-1, respectively, and ranged from 20.2 and 24.5 m d-1 to 1440 and 2800 m d-1, respectively.  Similarly, continuous CO2 emissions reflected variability in continuous kCO2 and k600 and discharge, with the highest fluxes occurring during high flow events with the highest kCO2 and k600 values discussed above (Figure 13). There was not significant diurnal cycling in kCO2 and k600 or CO2 emissions during the study period. Estimated variance in continuous kCO2 and k600 was very high, with mean and median values of 10700 and 43300 m d-1 and 999 and 2270 m d-1, respectively.  A linear regression model that predicted log(FCO2) from continuous stream discharge performed well (r2 = 0.98, p < 0.001). Similarly, a weighted loess regression model that predicted log(FCO2) from continuous stream velocity performed well (RSE = 0.11). One outlier  39 (FCO2 = 883 g C L-1 d-1) was removed to reduce heteroscedasticity and normalize residuals of the loess fit. I found that log(FCO2) was positively associated with flow parameters. Figure 14 provides bivariate plots of log(FCO2) and discharge and flow velocity, as well as regression fits and 95% confidence intervals. Table 12 provides linear regression coefficients and statistics of the linear model. Appendix 4 provides predicted FCO2 values of the weighted loess model.    40  Figure 13. Time series of 6-hour time-averaged (a) kCO2, (b) k600, and (c) CO2 emissions (FCO2) during the study period.  41  Figure 14. Bivariate plots of (a) continuous discharge (Q) and log(FCO2) and (b) continuous flow velocity and log(FCO2) (n = 194). The blue lines are (a) a linear regression model with equation: log(FCO2) = 0.01Q + 2.58 (r2 = 0.98; p < 0.001) and (b) a loess regression model (RSE = 0.11). The shaded areas are the 95% confidence envelopes. (a)(b)345670.0 0.1 0.2 0.3 0.4Stream velocity (m s-1)log(FCO2) (g C L-1 d-1)345670 200 400Discharge (L s-1)log(FCO2) (g C L-1 d-1) 42 Table 12. Equation, coefficient of determination, residual standard error (RSE), and p-value of a linear regression model that predicts log(FCO2) from discharge (Q) as a single regressor term. Variables Equation r2 adj. r2 RSE p-value Q log(𝐹𝐶𝑂2) = 0.01𝑄 + 9.86  0.97 0.97 0.11 < 0.001   4 DISCUSSION 4.1 Gas transfer velocity estimates  Mean and median k600 values were 66.0 and 42.3 m d-1, respectively. These values were in the high range of those reported in the literature, but were in line with those reported for steep, turbulent headwater streams (e.g., Natchimuthu et al., 2017). The mean k600 value of the current study was most similar to the mean k600 value of 67 m d-1 reported in Natchimuthu et al. (2017), a study of CO2 emissions from the stream network of a forested, hemiboreal catchment in Sweden, in which k600 values were determined via propane injections. Streams with very steep reaches and steep waterfalls were included in calculation of this value; excluding the steepest reaches, the mean k600 was 6.5 m d-1 (Natchimuthu et al., 2017). Average k600 values of the present study were also similar to mean and median values of 41.0 and 16.1 m d-1 determined via propane injections in 12 peatland headwater streams in the United Kingdom (Billett and Harvey, 2013). A study of O2 exchange in the Colorado River, Grand Canyon, USA modeled mean k600 values of 1410, 37, and 113 m d-1 for the rapids, runs, and entire river, respectively (Hall et al., 2012).  The kCO2 and k600 values determined in the current study also corresponded well with those determined in other streams in the Pacific Northwest (Raymond et al., 2012; Stackpoole  43 et al., 2012; Stackpoole et al., 2017). A metadata analysis (Raymond et al., 2012) based on five datasets (Bott, 1996; Bott et al., 2006; Melching and Flores, 1999; Mulholland et al., 2001; Tsivoglou and Wallace, 1972) modeled k600 values for the Pacific Northwest region of the USA, which ranged from ~25 m d-1 to just under 40 m d-1. However, Raymond et al. (2012) cautioned that the models may overestimate k600 and are only applicable at large spatial scales.  In a study of aquatic carbon fluxes in five ecoregions of the Western USA, estimated kCO2 values for streams in the Western Cordillera, determined via hydraulic relationships, ranged from 10 to 80 m d-1 (Stackpoole et al., 2012), and included the models used in the present study (Melching and Flores, 1999; Raymond et al., 2012). In a recent study of riverine and lacustrine gas exchange in multiple regions of Alaska, mean river kCO2 values ranged from 9.2 m d-1 in the Northwest region to 27.5 m d-1 in the Southeast region, with a mean value for all regions of 14.6 m d-1 (Stackpoole et al., 2017).  The range of k600 values reported in the present study (20.4 – 261 m d-1) were also within the range of, and in some cases much lower than, the upper range of k600 values reported in the literature. A study of gas exchange in the Amazon and Mekong river systems reported k600 values determined using floating chambers for streams and small rivers (<100 m wide) as high as 71 m d-1 (Alin et al., 2011). Hall and Tank (2003) reported an upper range for k600 of 42 m d-1 determined via SF6 injections in small streams in Grand Teton National Park, Wyoming. Upper ranges for k600 in rapids, runs, and the entire river system in the Colorado River, Grand Canyon, USA were 7730, 143, and 113 m d-1, respectively (Hall et al., 2012). Natchimuthu et al. (2017) determined an upper range of 558.7 m d-1 in the steepest reaches of headwater streams in a Swedish hemiboreal catchment.  44  However, many previous studies have reported average k600 values for headwater streams and small rivers lower than the range of those determined in the present study. Butman and Raymond (2011) estimated a mean k600 value of 18 m d-1 based on a model using slope and flow velocity terms in the steep headwater streams of the American west. Hall and Tank (2003) determined a mean k600 value of 18.4 m d-1 in the small streams of Grand Teton National Park, Wyoming, USA. Wallin et al. (2011) determined mean and median k600 values of 13 and 7.4 m d-1, respectively, via propane injections in boreal headwater streams in Sweden. Mean and median k600 values for a first-order stream in Walker Branch, Tennessee, USA determined via propane injections were 5 and 7 m d-1, respectively (Wanninkhof et al., 1990), and ranged from 5.5 – 14.2 m d-1 (Roberts et al., 2007). Streams and small rivers (<100 m wide) in the Amazon and Mekong river systems yielded a mean k600 value of 5.6 m d-1 (Alin et al., 2011). Campeau et al. (2014) determined k600 values for a lowland boreal river system in Northern Quebec using suspended floating chambers and corroborated results with modeled values; the mean of all samples (stream orders 1 – 6) was under 1 m d-1, with lower orders having the lowest mean k600 values. Crawford et al. (2013) determined a mean value of 6.5 m d-1 for a boreal headwater stream network in Alaska using suspended floating chambers. The mean and range values of k600 for five small streams in the Northern Highlands Lake District of north central Wisconsin and Michigan, USA, determined via suspended floating chambers, were 3.9 m d-1 and 0.3 – 13.5 m d-1, respectively (Crawford et al., 2014). A mean k600 value for first-order streams in Sweden determined via equations with slope and depth parameters (O'Connor and Dobbins, 1958) was 6.5 m d-1 (Humborg et al., 2010). In a study of CO2 emissions from low-gradient (<4%) streams in a southwestern Alaskan watershed, Smits (2016) estimated a  45 mean k600 value of 5.9 m d-1 from an oxygen-based regression model, and 6 m d-1 from the models used in the present study.  Thus, k600 values determined in the present study were in the high range of, and in some cases much higher than, reported k600 values for streams and rivers in general, and headwater streams specifically. Billett and Harvey (2013) discussed potential reasons why their k600 estimates were higher than those previously published for headwater streams, noting that multiple measurements were conducted during high discharge events, and that the reaches they used were narrower and deeper than those used in other studies, with stream width to depth ratios ranging from 2 to 15. In the present study, mean discharge observed during CO2 injection periods ranged from 10 to 266 L s-1, with mean and median values of 101 and 82 L s-1, respectively. Mean and median discharge values observed during the entire study period were 117 and 93 L s-1, respectively, and ranged from 9 to 601 L s-1. Average k600 values therefore represent and likely slightly underestimate k600 values for the entire study period, with the range of k600 values excluding very high flow events. The mean stream width to depth ratio observed during CO2 injections was 19, which is higher than those observed in Billett and Harvey (2013), but lower than those observed in other studies (e.g., Hall and Tank, 2003).   There was good correspondence between modeled k600 values and measured k600 values, with a linear regression model using measured k600 as an external regressor explaining 66% of the variance in modeled values. This is slightly higher than the predictive power of the models in the original metadata analysis (mean r2 = 0.63) (Raymond et al., 2012). In the present study, models overestimated k600 by 17.8 m d-1 on average, which is generally consistent with conclusions of the original analysis (Raymond et al., 2012). The original analysis also found  46 that k600 values were generally highest in headwater streams in the Pacific Northwest region, compared to all other regions of the USA (Raymond et al., 2012).  In a study of gas exchange in a hemiboreal headwater stream network, Natchimuthu et al. (2017) found that the models for k600 performed well for values below 100 m d-1 but underestimated values above 100 m d-1 by as much as 80%, likely because the models were not developed in the context of reaches with steep slopes or high altitudes and correspondingly high k600 values (Natchimuthu et al., 2017; Raymond et al., 2012; Stackpoole et al., 2012). In the present study, models underestimated k600 for values above 139 m d-1. The slope of this reach (13.3º, equivalent to 23.6% slope) was slightly higher than the highest slope category (6 – 21%) studied in Natchimuthu et al. (2017), which the authors found to generate the highest k600 values. A high slope for this reach may therefore explain both generally high k600 values and model performance. Other studies have noted that slope is positively associated with k600 (Moog and Jirka, 1999; Raymond et al., 2012), although some studies have reported no correspondence between slope and k600 (e.g., Smits, 2016).  4.2 Relationships between gas transfer velocities and stream parameters  I found good correspondence between kCO2 and k600 and stream discharge, velocity, and temperature, with single linear regression models using these parameters as external regressors explaining 86 and 88%, 82 and 85%, and 58 and 65% of the variance in kCO2 and k600, respectively. Values for kCO2 and k600 were positively associated with discharge and flow velocity, and negatively associated with stream temperature. Boxplots of kCO2 and k600 at high and low values of stream discharge, velocity, and temperature illustrate these relationships (Figures 12 – 17).  47  Previous studies have found good correspondence between k600 and discharge (e.g., Natchimuthu et al., 2017), but the relationship is generally spatiotemporally variable (Billett and Harvey, 2013; Wallin et al., 2011) and depends on the relationship between discharge and turbulence in a given reach (Wallin et al., 2011). Gas exchange studies that have determined k600 for multiple reaches in a catchment under multiple flow conditions have found positive, negative, and zero correspondence between k600 and discharge in different reaches (Billett and Harvey, 2013; Natchimuthu et al., 2017; Wallin et al., 2011). Two studies that determined k600 for different reaches in the same stream found positive (Roberts et al., 2007) and zero (Genereux and Hemond, 1992) correspondence between k600 and discharge. Natchimuthu et al. (2017) found that the correspondence between k600 and discharge increased positively with increasing slope, suggesting that steeper reaches generate more turbulence with higher flow than flatter reaches, thereby increasing gas evasion (Wallin et al., 2011). This reach was relatively steep compared to previously studied reaches and generated turbulence in many areas under high flow conditions, which may explain the strong relationships between kCO2 and k600 and mean discharge. Correspondingly, many studies have found both a strong positive relationship (Billett and Harvey, 2013; Natchimuthu et al., 2017; Sand-Jensen and Staehr, 2012) and no relationship (Natchimuthu et al., 2017) between flow velocity and k600. Variable channel morphologies and bed characteristics in a given catchment therefore limit the potential for regional scaling of k600 based on discharge or flow velocity terms (Moog and Jirka, 1999; Wallin et al., 2011).  Additionally, previous work comparing studies from different latitudes has shown that k600 varies with stream temperature, although the predictor is not as strong as turbulence parameters (Aufdenkampe et al., 2011). Similarly, the present study shows that kCO2 and k600  48 were negatively associated with stream temperature, but the parameter explains less of the variance in kCO2 and k600 (58 and 65%) than discharge and velocity as single regressor terms. Water temperature may be a strong predictor for k600 in lakes, where convection at the air-water interface drives turbulence (Holgerson et al., 2017; MacIntyre et al., 2010). In streams, streamflow and morphological parameters drive turbulence, rather than convection (Wallin et al., 2011). Figure 11 indicates that stream temperature was strongly associated with discharge and velocity in the present study, suggesting that the relationship between stream temperature and kCO2 and k600 was due to the strong seasonal relationship between stream temperature and flow parameters. Further, kCO2 values calculated using mean pCO2, travel time, and stage parameters from CO2 injections showed no change with changing temperature, given that all other parameters stayed the same and temperature values from both sensors changed concurrently. Under these conditions, k600 values decreased with increasing stream temperature (e.g., by 4 m d-1 between 6 and 8ºC).  4.3 CO2 evasion estimates  CO2 evasion from headwater streams has been reported as the dominant process governing aqueous C fluxes in a boreal landscape (Wallin et al., 2013). Studies have reported high spatiotemporal variability in both stream pCO2 and CO2 emissions from headwater streams (Billett and Harvey, 2013; Sand-Jensen and Staehr, 2012; Wallin et al., 2013), with variability in gas transfer velocities controlling CO2 evasion (Wallin et al., 2011) and decreasing pCO2 with increasing gas transfer velocities (Dosch, 2014). Additionally, pCO2 decreases and CO2 evasion increases with increasing slope in headwater streams (Finlay, 2003; Natchimuthu et al., 2017; Wallin et al., 2011). Wallin et al. (2011) found that stream slope,  49 width, and depth parameters explained 83% of spatial variability in gas transfer velocities in a boreal stream network. In a hemiboreal headwater catchment, Natchimuthu et al. (2017) determined that the steepest slope category (6 – 21%) generated CO2 emissions four times greater than the mean emissions for all studied reaches, despite comprising 0.9% of the total stream surface area. The authors further hypothesized that high flow events in steep headwater streams may play a dominant role in annual CO2 emissions from a catchment, even if they occur over short periods of time (Natchimuthu et al., 2017).  In the present study, CO2 emissions were variable overall and highest during high flow events, with discharge explaining 97% of the variability in emissions and flow velocity explaining 89% of the variability in emissions during the study period. High flow events were likely "hot moments" of CO2 evasion (McClain et al., 2003); almost 80% of CO2 emissions occurred when discharge was greater than the median overall discharge of 92 L s -1. The high slope, high gas transfer velocities, and high degree of turbulence in this stream suggest that it may play an important role in CO2 emissions from the catchment, with future climate-driven hydrological regime changes potentially increasing current emissions estimates (Natchimuthu et al., 2017). Additionally, estimates of CO2 evasion calculated using k600 determined via the seven models used in the present study underestimated CO2 evasion by a mean of 1191 g C L-1 d-1, corroborating evidence that high flow events are significant determinants of CO2 evasion from headwater streams and are likely underrepresented by infrequent measurements and modeled calculations. Figure 15 provides a comparison of modeled CO2 evasion over the study period from the two methods; evasion calculated using k600 determined via the seven models used in the present study again underrepresented the highest modeled flux events. An important note is that CO2 evasion estimates were determined from back-transformed logarithmic  50 estimates of continuous kCO2, which had very high calculated uncertainties, in some cases higher than kCO2 estimates by multiple orders of magnitude. Thus, modeling kCO2 and k600 or FCO2 in this way may not be reliable, depending on the desired accuracy of the estimates.  4.4 Advantages and drawbacks of automated CO2 injections as a tracer for kCO2 and FCO2  Despite the widespread use of tracer gas injections for determining gas transfer velocities, shortcomings of current techniques result in analytical variability (Sand-Jensen and Staehr, 2012) and data scarcity (Marx et al., 2017). Sand-Jensen and Staehr (2012) provide a comprehensive discussion regarding limitations of using tracer gases (e.g. oxygen, propane, ethane, SF6) in aqueous CO2 emissions studies, including errors associated with Schmidt conversions (Cole and Caraco, 1998; Simonsen, 1974; Thyssen and Kelly, 1985), pH-dependent conversions of CO2 in aqueous environments (Ho et al., 1997), and differences in how different gases pass through an organic-rich air-water interface (Frew, 1997).  Conventional tracer gas studies also rely on manual sampling, which limits the frequency and timing with which samples can be collected. For example, there is a paucity of nighttime gas exchange data collected via tracer gas analysis (Marx et al., 2017), despite the likelihood that diurnal processes, such as in-stream metabolism (Crawford et al., 2013), affect gas transfer velocities and CO2 emissions (Schelker et al., 2017). In a study of CO2 evasion from a steep, high gradient stream network in the European Alps, evasion was highest at nighttime and lowest at daytime (Schelker et al., 2017). However, in a study of CO2 evasion from a large boreal river in Finland, Huotari et al. (2013) reported no significant diurnal cycling in CO2 fluxes. Similarly, in the present study, there were no significant diurnal trends in gas transfer velocities or CO2 fluxes. This finding is likely supported by the hypothesis that most  51 streamwater CO2 is terrestrially-derived (Dinsmore and Billett, 2008; Hope et al., 2004), and thus stream pCO2 was largely not governed by in-stream processes during the present study. However, year-round data collection as streamflow permits will allow for a better interpretation of the importance of in-stream processes for CO2 evasion dynamics. Continued research on diurnal trends in CO2 transfer velocities and emissions will elucidate processes and fill in data gaps. Additionally, it is worth noting that innovations in tracer gas techniques have incorporated diurnal sampling; for example, Tobias et al. (2009) measured gas transfer continuously for 32 hours by pumping dilute SF6-saturated water into a stream.   While methods that include direct measurements of CO2 fluxes, such as floating chamber techniques, can be deployed for extended periods of time and preclude analytical errors associated with conventional tracer gas studies, they have been criticized for underestimating fluxes by disrupting turbulence and creating artificial pressure and temperature conditions (Gafalk et al., 2013; Crawford et al., 2014; Raymond and Cole, 2001). Suspended chamber techniques have mitigated some of these effects and validated data with other methods (Crawford et al., 2013; Crawford et al., 2014).  Direct injection of CO2 for use as a tracer precludes analytical variability concerning Schmidt conversions, pH dependencies, and gas-dependent interactions at the air-water interface. In situ measurement of pCO2 also has advantages over ex situ measurement (e.g. systems that circulate air to a sensor outside of the stream), including enhanced precision and response time (Johnson et al., 2010). Further, automated injections allow for continuous monitoring at any desired time scale. Although trigger parameters for automated injections other than time of day were not tested in the present study, it is certainly possible to program a datalogger to collect tracer data in response to events of scientific interest, given that other  52 sensors connected to the datalogger are measuring ancillary variables. For example, a future direction of the present study may be to trigger CO2 injections during high-flow events by setting a threshold for stream depth.   Previous studies have tested the IRGA sensors used in the present study for output stability and long-term drift during continuous deployment in both streamwater (Johnson et al., 2010) and soil (Jassal et al., 2004) environments. Additionally, the sensors have exhibited output stability in a wide range of aqueous environments and have not required data cleaning (Johnson et al., 2010).  However, the experimental setup, including automated CO2 injections, has some disadvantages and may not be suitable for every field campaign. Researchers working in sites with public access should consider the possibility of equipment being tampered with, which can result in expensive losses, for example in the case of sensors being broken or stolen. Further, compressed gas cylinders may need to be replaced frequently and can also pose a safety hazard for inexperienced users and curious wildlife, such as bears. Gas cylinders should be covered properly, ideally in a locked container. Additionally, negative and positive outliers in the present study may have been due to inadequate gas mixing in the reach or sensor malfunction.  4.5 Considerations In general, I overestimated gas injection times that were needed for sensor equilibration, especially during winter months when stream discharge and flow velocity were high. In response, I manually adjusted gas flow rates as necessary during field visits in response to changing streamflow and stream temperature. Future experimental deployments using this method will benefit from adjusting gas flow and injection time based on the dynamics of  53 specific stream reaches. To further automate this method, researchers may wish to include streamflow and temperature thresholds in their datalogger programs that govern injection times and gas flow rates based on determined stream dynamics.  Another consideration is the potential influence of hyporheic exchange on evasion estimates in headwater stream systems. While hyporheic exchange likely does not significantly affect estimates during high flow conditions (Leach and Moore, 2014), it may play a larger role during lower flow conditions by impacting pCO2 and promoting evasion. CO2 generation in the hyporheic zone and efflux rates may be significant in streams (Schindler and Krabbenhoft, 1998); in the Pacific Northwest, hyporheic zone pCO2 has been shown to be highest during the summer and lowest during the winter, with winter storms quickly decreasing and subsequently increasing hyporheic zone pCO2 (Brandes, 2017). The presence of hyporheic exchange pathways can be determined by frequently monitoring pCO2 at many points in the study reach during various flow conditions, and particularly in response to high flow events. Nevertheless, its influence on evasion estimates should be mitigated through a long gas injection period during which sufficient time is provided for sensor equilibration.  Additionally, comparison of discharge estimates from upstream and downstream sensor locations during salt slug injections indicated that study reach was generally a gaining reach. Thus, larger-scale hydraulic forcing conditions may affect evasion estimates, perhaps more than hyporheic exchange, particularly as the losing or gaining flux increases (Fox et al., 2014).  It is also worth noting that significantly elevating pCO2 in an aqueous environment for long periods of time may have impacts on ecosystems by decreasing pH, resulting in litter and  54 algal quality decreases and food web effects (Ferreira and Chauvet, 2011; Hargrave et al., 2009). However, elevated pCO2 may also increase the density, biomass, and average individual size of benthic invertebrates (Hargrave et al., 2009). Further, although studies have focused on simulating potential climate change scenarios in streams by elevating pCO2 for long periods of time (e.g., 90 days), no study to my knowledge has looked at long-term effects of elevating aqueous pCO2 for short bursts (e.g., 1 hour twice per day). Additionally, in turbulent headwater streams particularly, all injected CO2 will likely evade within a confined reach, although there is a possibility that some is taken up by an ecosystem. More research may be needed to ensure that this method does not harm stream ecosystems, and site-specific evasion dynamics should be taken into consideration. One complementary path for this research may be isotopic analysis of CO2 during baseline conditions and CO2 injections, which could help discriminate among various CO2 pathways.  5 CONCLUSIONS  The limitations of current methods for determining gas transfer velocities of CO2 in headwater streams result in analytical variability and data scarcity. The method presented here mitigates common issues associated with gas transfer velocity estimations. The use of CO2 as a tracer precludes analytical variability associated with the use of alternative gases while maintaining natural conditions at the air-water interface, and automated injections allow researchers to determine gas transfer velocities at the desired temporal scale. In the present study, kCO2 and k600 showed good correspondence with stream discharge, velocity, and temperature, with correspondence decreasing in that order. Values of kCO2 and k600 associated positively with both discharge and flow velocity and negatively with stream temperature;  55 continuous kCO2 and k600 extrapolation indicated that the highest kCO2 and k600 values occurred during very high flow events, when turbulence was highest. Values of k600 were generally high but within the range of those reported in previous studies; high k600 values also corresponded well with stream morphological and turbulence parameters. Similar to other studies in steep headwater streams, k600 values determined via seven widely-used models performed well for values under 139 m d-1 but underestimated k600 above this threshold, suggesting that the models were not developed for steep headwater streams under high flow conditions. CO2 emissions estimates suggested that high flow conditions drove evasion during the study period and may become more important if climate-driven hydrological regime changes result in more frequent high flow events. Constraining estimates of CO2 evasion from headwater streams is a critical step in characterizing the global carbon cycle. The method presented in this thesis will allow researchers to increase the frequency and accuracy with which they can determine gas transfer velocities of CO2 in headwater streams, ultimately resulting in data-driven quantifications of CO2 emissions and fluxes in the global C cycle.    56 REFERENCES Alin, S. R., M. D. D. F. F. L. Rasera, C. I. Salimon, J. E. Richey, G. W. Holtgrieve, A. V. Krusche, and A. Snidvongs (2011), Physical controls on carbon dioxide transfer velocity and flux in low-gradient river systems and implications for regional carbon budgets, J. Geophys. Res., 116(G1), doi:10.1029/2010JG001398.  Aufdenkampe, A. K., E. Mayorga, P. A. Raymond, J. M. Melack, S. C. Doney, S. R. Alin, R. E. Aalto, and K. Yoo (2011), Riverine coupling of biogeochemical cycles between land, oceans, and atmosphere, Front. Ecol. Environ., 9(1), 53–60, doi:10.1890/100014. Baskerville, G. L. (1972), Use of Logarithmic Regression in the Estimation of Plant Biomass, Canadian Journal of Forestry 2(49), 49–53, doi:10.1139/x72-009. Battin, T. J., S. Luyssaert, L. A. Kaplan, A. K. Aufdenkampe, A. Richter, and L. J. Tranvik (2009), The boundless carbon cycle, Nat. Geosci., 2, 598–600, doi:10.1038/ngeo618 Benstead, J. P., and D. S. Leigh (2012), An expanded role for river networks, Nat. Geosci., 5, 678–679, doi:10.1038/ngeo1593.  Billett, M. F., and F. H. Harvey (2013), Measurements of CO2 and CH4 evasion from UK peatland headwater streams, Biogeochemistry, 114, 165–181 doi:10.1007/s10533-012-9798-9. Bott, T. L, (1996), Primary productivity and community respiration, Methods in Stream Ecology. Academic Press, San Diego, California, 533–556.  Bott, T. L., D. S. Montgomery, J. D. Newbold, D. B. Arscott, C. L. Dow, A. K. Aufdenkampe, J. K. Jackson, and L. A. Kaplan (2006), Ecosystem metabolism in streams of the Catskill Mountains (Delaware and Hudson River watersheds) and lower Hudson Valley, J. N. Am. Benthol. Soc., 25(4), 1018–1044, doi:10.1899 /0887-3593(2006)025[1018:EMISOT]2.0.CO;2.  Brandes, J. B. (2017), The vadose zone as a hyporheic carbon source: a look at temporal trends in hyporheic zone pCO2, M.Sc. thesis, Oregon State University, Corvallis. Butman, D., and P. A. Raymond (2011), Significant efflux of carbon dioxide from streams and rivers in the United States, Nat. Geosci., 4, 839–842, doi:10.1038/ngeo1294. Campeau, A., J.-F. Lapierre, D. Vachon, and P. A. del Giorgio (2014), Regional contribution of CO2 and CH4 fluxes from the fluvial network in a lowland boreal landscape of Québec, Global Biogeochem. Cycles, 28(1), 57–69, doi:10.1002/2013GB004685. Clark, J. F., P. Schlosser, H. J. Simpson, M. Stute, R. Wanninkhof, and D. T. Ho (1995), Relationship between gas transfer velocities and wind speeds in the tidal Hudson River determined by the dual tracer technique, Air-water gas transfer, 785-800.  57 Cole, J. J., et al. (2007), Plumbing the global carbon cycle: Integrating inland waters into the terrestrial carbon budget, Ecosystems, 10(1), 171–184, doi:10.1007/s10021-006-9013-8.
 Cole, J. J., and N. F. Caraco (1998), Atmospheric exchange of carbon dioxide in a low‐wind oligotrophic lake measured by the addition of SF6, Limnol. Oceanogr., 43(4), 647–656.  Crawford, J. T., N. R. Lottig, E. H. Stanley, J. F. Walker, P. C. Hanson, J. C. Finlay, and R. G. Striegl (2014), CO2 and CH4 emissions from streams in a lake-rich landscape: Patterns, controls, and regional significance, Global Biogeochem. Cycles, 28(3), 197–210, doi:10.1002/2013GB004661. 
 Crawford, J. T., R. G. Striegl, K. P. Wickland, M. M. Dornblaser, and E. H. Stanley (2013), Emissions of carbon dioxide and methane from a headwater stream network of interior Alaska, J. Geophys. Res., 118(2), 482–494, doi:10.1002/jgrg.20034. Dinsmore, K. J., and M. F. Billett (2008), Continuous measurement and modelling of CO2 losses from a peatland stream during stormflow events, Water Resour. Res., 44(12), W12417, doi:10.1029/2007WR007284.  Dosch, N. T. (2014), Spatiotemporal Dynamics and Drivers of Stream pCO2 in a Headwater Catchment in the Western Cascade Mountains, Oregon, M.Sc. thesis, Oregon State University, Corvallis. Ferreira, V., and E. Chauvet (2011), Future increase in temperature more than decrease in litter quality can affect microbial litter decomposition in streams, Oecologia, 167, 279–291, doi:10.1007/S00442-01 1-1976-2  Finlay, J. C. (2003), Controls of streamwater dissolved inorganic carbon dynamics in a forested watershed, Biogeochemistry, 62(3), 231-252.  Fox, A., F. Boano, and S. Arnon (2014), Impact of losing and gaining streamflow conditions on hyporheic exchange fluxes induced by dune-shaped bed forms, Water Resour. Res., 50(3), 1895–1907, doi:10.1002/2013WR014668 Frankignoulle, M., G. Abril, A. Borges, I. Bourge, C. Canon, B. DeLille, E. Libert, and J. M. Theate (1998), Carbon dioxide emission from European estuaries, Science, 282(5388), 434–436, doi:10.1126/science.282.5388.434  Frew, N.M. (1997), The role of organic films in air-sea gas exchange. The sea surface and global change, edited by P. S. Liss, and R. A. Duce, 121–163.  Gafalk, M., D. Bastviken, S. T. Fredriksson, and L. Arneborg (2013), Determination of the piston velocity for water-air interfaces using flux chambers, acoustic Doppler velocimetry, and IR imaging of the water surface, J. Geophys. Res.: Biogeosciences, 118(2), 770-782.  Genereux, D. P., and H. F. Hemond (1992), Determination of Gas-Exchange Rate Constants for a Small Stream on Walker Branch Watershed, Tennessee, Water Resour. Res., 28(9), 2365-2374.   58 Gibs, J., F. D. Wilde, and H. A. Heckathorn (2007), Use of Multiparameter Instruments for Routine Field Measurements, U.S. Geological Survey Techniques of Water-Resources Investigations (Book 9, Chapter A6, Section 6.8). Gomi, T., R. C. Sidle, and J. S. Richardson (2002), Understanding processes and downstream linkages of headwater systems, BioScience, 52(10), 905–916.
 Hall, R. O., T. A. Kennedy, and E. J. Rosi-Marshall, Air–water oxygen exchange in a large whitewater river (2012), Limnol. Oceanogr.: Fluids Environ. 2(1), 1–11, doi: 10.1215/21573689-1572535. Hall, R. O., and J. L. Tank (2003), Ecosystem metabolism controls nitrogen uptake in streams in Grand Teton National Park, Wyoming, Limnol. Oceanogr., 48(3), 1120–1128, doi: 10.4319/lo.2003.48.3.1120 Hargrave, C. W., K. P. Gary, and S. K. Rosado (2009), Potential effects of elevated atmospheric carbon dioxide on benthic autotrophs and consumers in stream ecosystems: a test using experimental stream mesocosms, Global Change Biology, 15(11), 2779–2790, doi:10.1111/j.1365-2486.2009.01897.x. Ho, D.T., L. F. Bliven, R. Wanninkhof, and P. Schlosser (1997), The effect of rain on air-water gas exchange, Tellus, 49(2), 149–158, doi: 10.1034/j.1600-0889.49.issue2.3.x  Holgerson, M. A., E. R. Farr, and
P. A. Raymond (2017), Gas transfer velocities in small forested ponds,
J. Geophys. Res. Biogeosci., 122, 1011–1021, doi:10.1002/2016JG003734. 
 Hope, D., S. M. Palmer, M. F. Billett, and J. J. C. Dawson (2001), Carbon dioxide and methane evasion from a temperate peatland stream, Limnol. Oceanogr., 46(4), 847–857, doi: 10.4319/lo.2001.46.4.0847 Hope, D., S. M. Palmer, M. F. Billett, and J. J. C. Dawson (2004), Variations in dissolved CO2 and CH4 in a first-order stream and catchment: An investigation of soil-stream linkages, Hydrol. Process., 18(17), 3255–3275, doi:10.1002/hyp.5657. Humborg, C., C. M. Mörth, M. Sundbom, H. Borg, T. Blenckner, R. Giesler, and V. Ittekkot (2010), CO2 supersaturation along the aquatic conduit in Swedish watersheds as constrained by terrestrial respiration, aquatic respiration and weathering, Global Change Biol., 16(7), 1966–1978, doi:10.1111/j.1365-2486.2009.02092.x. Huotari, J., S. Haapanala, J. Pumpanen, T. Vesala, and A. Ojala (2013), Efficient gas exchange between a boreal river and the atmosphere, Geophys. Res. Lett.,40(21), 5683–5686, doi:10.1002/2013GL057705. Jähne, B., G. Heinz, and W. Dietrich (1987), Measurement of the diffusion coefficients of sparingly soluble gases in water, J. Geophys. Res., 92(C10), 10767–10776, doi:10.1029/JC092iC10p10767.  59 Jassal, R. S., T. A. Black, G. B. Drewitt, M. D. Novak, D. Gaumont-Guay, and Z. Nesic (2004), A model of the production and transport of CO2 in soil: predicting soil CO2 concentrations and CO2 efflux from a forest floor, Agricultural and Forest Meteorology, 124(3-4), 219–236, doi:10.1016/j.agrformet.2004.01.013. Johnson, M. S., M. F. Billett, K. J. Dinsmore, M. Wallin, K. E. Dyson, and R. S. Jassal (2010), Direct and continuous measurement of dissolved carbon dioxide in freshwater aquatic systems—Methods and applications, Ecohydrology, 3(1), 68–78, doi:10.1002/eco.95.  Jones, J. B., and P. J. Mulholland (1998), Influence of drainage basin topography and elevation on carbon dioxide and methane supersaturation of stream water, Biogeochemistry, 40(1), 57–72, doi:10.1023/A:1005914121280.
 Jonsson, A., J. Aberg, A. Lindroth, and M. Jansson (2008), Gas transfer rate and CO2 flux between an unproductive lake and the atmosphere in northern Sweden, J. Geophys. Res. Biogeosci., 113, G4, doi:10.1029/2008JG000688 Kling, G. W., G. W. Kipphut, and M. C. Miller (1991), Arctic lakes and streams and gas conduits to the atmosphere: Implications for tundra carbon budgets, Science, 251(4991), 298–301, doi:10.1126/science.251.4991.298.  Kokic, J., M. B. Wallin, H. E. Chmiel,
B. A. Denfeld, and S. Sobek (2015), Carbon dioxide evasion from headwater systems strongly contributes to the total export of carbon from a small boreal lake catchment, J. Geophys. Res. Biogeosci., 120, 13–28, doi:10.1002/2014JG002706.  Leach, J. A., and Moore, R. D. (2014). Winter stream temperature in the rain-on-snow zone of the Pacific Northwest: influences of hillslope runoff and transient snow cover. Hydrology and Earth System Sciences, 18(2), 819–838, doi:10.5194/hess-18-819-2014 Looman, A., I. R. Santos, D. R. Tait, J. R. Webb, C. A. Sullivan, and D. T. Maher (2016), Carbon cycling and exports over diel and flood-recovery timescales in a subtropical rainforest headwater stream, Sci. Total Environ., 550, 645–657.
 MacIntyre, S., A. Jonsson, M. Jansson, J. Aberg, D. E. Turney, and S. D. Miller (2010), Buoyancy flux, turbulence, and the gas transfer coefficient in a stratified lake, Geophys. Res. Lett., 37(24), L24604, doi:10.1029/2010GL044164.
 MacIntyre, S., R. Wanninkhof, and J. P. Chanton (1995), Trace gas exchange across the air‐water interface in freshwater and coastal marine environments, in Biogenic Trace Gases: Measuring Emissions From Soil and Water, edited by P. A. Matson and R. C. Harriss, 52–97. Marx, A., et al, (2017), A review of CO2 and associated carbon dynamics in headwater streams: a global perspective, Reviews of Geophysics, doi:10.1002/2016RG000547  60 Marzolf, E. R., P. J. Mulholland, and A. D. Steinman, (1994), Improvements to the diurnal upstream-downstream dissolved oxygen change technique for determining whole-stream metabolism in small streams. Can. J. Fish. Aquat. Sci. 51(7), 1591–1599, doi:10.1139/f94-158.  McClain, M. E., et al. (2003), Biogeochemical hot spots and hot moments at the interface of terrestrial and aquatic ecosystems, Ecosystems, 6(4), 301–312, doi: 10.1007/s10021-003-0161-9  Melching, C. S., and H. E. Flores (1999), Reaeration equations derived from US Geological Survey database, J. Environ. Eng., 125, 407–414, doi:10.1061/(ASCE)0733-9372(1999)125:5(407).  Moog, D. B., and G. H. Jirka (1999), Stream reaeration in nonuniform flow: Macroroughness enhancement. J. Hydraul. Eng., 125(1), 11–16, doi:10.1061/(ASCE)0733-9429(1999)125:1(11). Moore, R. D. (2005), Slug injection using salt in solution, Streamline Watershed Management Bulletin, 8(2), 1-6. Mulholland, P. J., et al., (2001), Inter-biome comparison of factors controlling stream metabolism. Freshw. Biol. 46(11), 1503–1517, doi:10.1046/j.1365-2427.2001.00773.x.   Natchimuthu, S., M. B. Wallin, L. Klemedtsson, and D. Bastviken (2017), Spatio-temporal patterns of stream methane and carbon dioxide emissions in a hemiboreal catchment in Southwest Sweden, Sci. Rep., 7, 39729, doi:10.1038/srep39729. O’Connor, D., and W. Dobbins (1958), Mechanism of reaeration in natural streams, Trans. Am. Soc. Civ. Eng. 123(1), 641–666.  Öquist, M. G., M. Wallin, J. Seibert, K. Bishop, and H. Laudon (2009), Dissolved inorganic carbon export across the soil/stream interface and its fate in a boreal headwater stream, Environ. Sci. Technol., 43(19), 7364–7369, doi:10.1021/es900416h.  Plummer, L. N., and E. Busenberg (1982), The Solubilities of Calcite, Aragonite and Vaterite in CO2-H2O Solutions between 0 and 90°C, and an Evaluation of the Aqueous Model for the System CaCO3-CO2-H2O, Geochim. Cosmochim. Ac., 46(6), 1011-1040, doi:10.1016/0016-7037(82)90056-4 R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. Raymond, P. A., and J. J. Cole (2001), Gas exchange in rivers and estuaries: Choosing a gas transfer velocity, Estuaries, 24(2), 312–317, doi:10.2307/1352954.  61 Raymond, P. A., C. J. Zappa, D. Butman, T. L. Bott, J. Potter, P. Mulholland, A. E. Laursen, W. H. McDowell, and D. Newbold (2012), Scaling the gas transfer velocity and hydraulic geometry in streams and small rivers, Limnol. Oceanogr. Fluids Environ., 2(1), 41–53, doi:10.1215/21573689‐1597669.
 Raymond, P. A., et al. (2013), Global carbon dioxide emissions from inland waters, Nature, 503(7476), 355–359, doi:10.1038/nature12760.  Richardson, J. S., and R. D. Moore (2010), Malcolm Knapp Research Forest, Streamline Watershed Management Bulletin 14(1), 14–15. Richardson, M. E., G. Sentlinge, D. Moore, and A. Zimmermann (2017), Quantifying the Relation between Electrical Conductivity and Salt Concentration for Dilution Gauging Via Dry Salt Injection, Confluence: Journal of Watershed Science and Management, 1(2), 1–13, doi:10.22230/jwsm.2017v1n1a1. Roberts, B. J., P. J. Mulholland, and W. R. Hill (2007), Multiple scales of temporal variability in ecosystem metabolism rates: Results from 2 years of continuous monitoring in a forested headwater stream, Ecosystems 10(4), 588–606, doi:10.1007/s10021-007-9059-2. Sand-Jensen, K., and P. A. Staehr (2012), CO2 dynamics along Danish lowland streams: water-air gradients, piston velocities, and evasion rates, Biogeochemistry, 111(1-3), 615–628, doi: 10.1007/s10533-011-9696-6. Sawakuchi, H. O., et al. (2017), Carbon Dioxide Emissions along the Lower Amazon River, Front. Mar. Sci., 4, 76, doi:10.3389/fmars.2017.00076. Schelker, J., G. A. Singer, A. J. Ulseth, S. Hengsberger, T. J. Battin (2016), CO2 evasion from a steep, high gradient stream network: importance of seasonal and diurnal variation in aquatic pCO2 and gas transfer, Limnol. Oceanogr., 61(5), 1826–1838, doi: 10.1002/lno.10339. Schindler, J. E., and D. P. Krabbenhoft (1998), The hyporheic zone as a source of dissolved organic carbon and carbon gases to a temperate forested stream, Biogeochemistry 43(2), 157–174. Shaw, E. M., K. J. Beven, N. A. Chappell, and R. Lamb (2010), Hydrology in practice, 4th edn. CRC Press, Boca Raton. Simonsen, J.F. (1974), Oxygen fluctuations in streams, Ph.D. thesis, Danish Technical University, Copenhagen. Smits, A. P. (2016), Physical controls on land-water linkages: Carbon cycling and food webs in boreal watersheds, Ph.D. thesis, University of Washington, Seattle.  62 Stackpoole S, et al., (2012), Baseline carbon sequestration, transport, and emission from inland aquatic ecosystems in the western United States. Baseline and Projected Future Carbon Storage and Greenhouse-Gas Fluxes in Ecosystems of the Western United States, US Geological Survey Professional Paper, 1797, eds Z. Zhu, and B. C. Reed. Stackpoole, S. M., D. E. Butman, D. W. Clow, K. L. Verdin, B. V. Gaglioti, H. Genet, and R. G. Striegl (2017), Inland waters and their role in the carbon cycle of Alaska. Ecological Applications, 27(5), 1403–1420, doi: 10.1002/eap.1552. Tashe, N.C., (1998), The impact of vine maple on the biogeochemical nutrient cycle of conifer-dominated coastal forests in southwestern British Columbia, M.Sc. thesis, Simon Fraser University, Burnaby, British Columbia. Thyssen, N., and M. G. Kelly (1985), Water-air exchange of carbon dioxide and oxygen in a river: measurement and comparison of rates, Arch. Hydrobiol. 105(2), 219–228. Tobias, C. R., J. K. Böhlke, J. W. Harvey, and E. Busenberg (2009), A simple technique for continuous measurement of time-variable gas transfer in surface waters, Limnol. Oceanogr.: Methods, 7(2), 185–195, doi:10.4319/lom.2009.7.185. Tsivoglou, E. C., and R. J. Wallace (1972), Characterization of Stream Reaeration Capacity, Research Reporting Series, U.S. Environmental Protection Agency, 317.  Tsivoglou, E. C., and A. L. Neal (1976), Tracer measurement of reaeration, III: Predicting reaeration capacity of inland streams, J. Water Pollut. Control Fed., 48(12), 2669–2689.  Turk, T. D., M. G. Schmidt, and N. J. Roberts, (2008), The influence of bigleaf maple on forest floor and mineral soil properties in a coniferous forest in coastal British Columbia, Forest Ecology and Management, 255(5), 1874-1882. Vachon, D., Y. T. Prairie, and J. J. Cole (2010), The relationship between near‐surface turbulence and gas transfer velocity in freshwater systems and its implications for floating chamber measurements of gas exchange, Limnol. Oceanogr., 55(4), 1723–1732, doi:10.4319/lo.2010.55.4.1723.
 Waldon, M. G. (2004), Estimation of average stream velocity, J. Hydraul. Eng., 130(11), 1119–1122, doi:10.1061/(ASCE)0733-9429(2004)130.11(1119). Wallin, M. B., T. Grabs, I. Buffam, H. Laudon, A. Agren, M. G. Öquist, K. Bishop (2013), Evasion of CO2 from streams – The dominant component of the carbon export through the aquatic conduit in a boreal landscape, Global Change Biology, 19(3), 785–797, doi:10.1111/gcb.12083. Wallin, M. B., M. G. Öquist, I. Buffam, M. F. Billett, J. Nisell, and K. H. Bishop (2011), Spatiotemporal variability of the gas transfer coefficient K(CO2) in boreal streams: Implications for large scale estimates of CO2 evasion, Global Biogeochem. Cycles, 25(14), Gb3025, doi:10.1029/ 2010gb003975.   63 Wanninkhof, R., Mulholland, P. J., and Elwood, J. W. (1990), Gas exchange rates for a first‐order stream determined with deliberate and natural tracers, Water Resources Research, 26(7), 1621–1630, http://doi.org/10.1029/WR026i007p01621. Wanninkhof, R., W. E. Asher, D. T. Ho, C. Sweeney, and W. R. McGillis (2009), Advances in Quantifying Air-Sea Gas Exchange and Environmental Forcing, Annu. Rev. Mar. Sci., 1, 213-244, doi:10.1146/annurev.marine.010908.163742. Zappa, C. J., W. R. McGillis, P. A. Raymond, J. B. Edson, E. J. Hintsa, H. J. Zemmelink, J. W. H. Dacey, and D. T. Ho (2007), Environmental turbulent mixing controls on air‐water gas exchange in marine and aquatic systems, Geophys. Res. Lett., 34(10), L10601, doi:10.1029/2006GL028790.     64 Appendix 1  CR1000 program used for autonomous and unaccompanied measurement 'Program for WQ and Vaisala sensors 'date edited: October 14, 2016 'author: mollie  '====== VARIABLES =================================  'general  Public PTemp, batt_volt Units PTemp = Deg C Units batt_volt = Volts  Public lowPowerMode Public MinBattVolt Public powerPhone Public phoneManualON Public firstDataWindow, secondDataWindow  'time  Public real_time(9) Alias real_time(1) = year Alias real_time(2) = month Alias real_time(3) = DayX Alias real_time(4) = hours Alias real_time(5) = minutes Alias real_time(6) = seconds Alias real_time(7) = microseconds Alias real_time(8) = weekday Alias real_time(9) = jday 'Public mins_current_day  'constants Const port_phone_power = 6 Const port_AM32B_CLK = 1 Const port_AM32B_RES = 2 Const port_SV_power = 7 Const port_CO2_power = 8  'Solenoid valve  Public powerSV  'Vaisala CO2 variables  65 Public CO2_1_mV, CO2_1_ppm, CO2_2_mV, CO2_2_ppm, measureCO2, powerCO2, measureCO2_30, powerCO2_30  'CTD variables Public CTD_1(3), CTD_2(3) Alias CTD_1(1) = Depth_CTD_1 Alias CTD_1(2) = Temp_CTD_1 Alias CTD_1(3) = EC_CTD_1 Alias CTD_2(1) = Depth_CTD_2 Alias CTD_2(2) = Temp_CTD_2 Alias CTD_2(3) = EC_CTD_2  Units Depth_CTD_1 = mm Units Temp_CTD_1 = Deg C Units EC_CTD_1 = uS/m Units Depth_CTD_2 = mm Units Temp_CTD_2 = Deg C Units EC_CTD_2 = uS/m  'pH/ORP variables Public pH_1, pH_2, ORP_1 Dim pHMult_1, pHMult_2  Units pH_1 = pH Units pH_2 = pH Units ORP_1 = mV  'DO variables Public Tw_LDO, LDO  'GS3 variables Public GS3_1(3), GS3_2(3) Alias GS3_1(1) = Moisture_GS3_1 Alias GS3_1(2) = Temp_GS3_1 Alias GS3_1(3) = EC_GS3_1 Alias GS3_2(1) = Moisture_GS3_2 Alias GS3_2(2) = Temp_GS3_2 Alias GS3_2(3) = EC_GS3_2  'anemometer variables Public u, v, w, SonicT Units u = ms-1 Units v = ms-1 Units w = ms-1 Units SonicT = DegC   66 '======== Data tables ============================  '30 minute table DataTable(dt,true,-1)   DataInterval(0,30,Min,10)   Average(1,batt_volt,FP2,0)   Average(1,PTemp,FP2,0)   Average(1,CO2_1_ppm,IEEE4,measureCO2_30-1) 'this makes it track the data only if "measureCO2_30" = true   Average(1,CO2_2_ppm,IEEE4,measureCO2_30-1)   Average(3,CTD_1(),IEEE4,0)   Average(3,CTD_2(),IEEE4,0)   Average(1,ORP_1,IEEE4,0)   Average(1,pH_1,IEEE4,0)   Average(1,pH_2,IEEE4,0)   Average(1,LDO, IEEE4,0)   Average(1,Tw_LDO,IEEE4,0)   Average(3,GS3_1(),IEEE4,0)   Average(3,GS3_2(),IEEE4,0)   Average(1,u,IEEE4,0)   Average(1,v,IEEE4,0)   Average(1,w,IEEE4,0)   Average(1,SonicT,IEEE4,0) EndTable  '5 second table - only activated when measureCO2 = true DataTable(dt_co2,true,-1)   DataInterval(0,5,Sec,10)     Average(1,CO2_1_ppm,IEEE4,measureCO2-1) 'this makes it track the data only if "measureCO2" = true   Average(1,CO2_2_ppm,IEEE4,measureCO2-1)   Average(3,CTD_1(),IEEE4,0)   Average(3,CTD_2(),IEEE4,0)   Average(1,ORP_1,IEEE4,0)   Average(1,pH_1,IEEE4,0)    Average(1,pH_2,IEEE4,0)    Average(1,LDO, IEEE4,0)   Average(1,Tw_LDO,IEEE4,0)   Average(3,GS3_1(),IEEE4,0)   Average(3,GS3_2(),IEEE4,0)   Average(1,u,IEEE4,0)   Average(1,v,IEEE4,0)   Average(1,w,IEEE4,0)   Average(1,SonicT,IEEE4,0) EndTable  'Main Program  67 BeginProg   firstDataWindow  =  15   secondDataWindow =  16   phoneManualON = 0   MinBattVolt = 11.4    Scan (5,Sec,0,0)     '======== Panel Temp/Battery ============================     PanelTemp (PTemp,250)   Battery (batt_volt)       RealTime real_time      '====== Low power mode =====================     If batt_volt < MinBattVolt Then       lowPowerMode = 1     Else       lowPowerMode = 0     EndIf  '======== SDI: CTD/GS3 ============================    'CTD measurements     Delay (1,200,mSec)     SDI12Recorder (CTD_1(),3,1,"M!",1.0,0)     Delay (1,200,mSec)     SDI12Recorder (CTD_2(),3,6,"M!",1.0,0)        'GS3 measurements     Delay (1,200,mSec)     SDI12Recorder (GS3_1(),3,3,"M!",1.0,0)     Delay (1,200,mSec)     SDI12Recorder (GS3_2(),5,4,"M!",1.0,0)  '======== Solenoid and CO2 ============================    'SV switch - change to reflect how long to inject CO2     If (hours = 10 OR hours = 22) AND NOT lowPowerMode Then 'AND (minutes >= 0 AND minutes <= 59)        powerSV = 1     Else       powerSV = 0     EndIf     PortSet(port_SV_power,powerSV)       'CO2 ppm measurements - on or off       If (hours = 10 OR hours = 22) AND NOT lowPowerMode Then ' AND (minutes <= 59))   68       powerCO2 = 1     Else       powerCO2 = 0     EndIf  ' Put CO2 on 30 min output table  'CO2 ppm measurements - on or off     If ((minutes >= 25) AND (minutes <= 30)) OR ((minutes >= 55) AND (minutes <= 60)) AND NOT lowPowerMode Then       powerCO2_30 = 1     Else       powerCO2_30 = powerCO2     EndIf      If powerCO2_30 = 1 Then       powerCO2 = 1    EndIf         PortSet(port_CO2_power,powerCO2)           If powerCO2 AND (hours = 10 OR hours = 22) Then 'AND (minutes <= 59))        measureCO2 = 1     Else       measureCO2 = 0     EndIf           If powerCO2_30 AND (((minutes >= 28) AND (minutes <= 30)) OR ((minutes >= 58) AND (minutes <= 60))) Then       measureCO2_30 = 1     Else       measureCO2_30 = 0     EndIf      '======== PHONE ===========================   ' Turn the phone on for one hour during the first time window   ' (uncomment 'OR' and remove the 1st 'Then' to turn on for 30 mins during each of two time windows   If (hours = firstDataWindow)  AND NOT lowPowerMode Then 'OR hours = secondDataWindow) AND minutes < 30 Then     powerPhone = 1   Else     powerPhone = 0   EndIf   ' Use the phoneManualON flag to keep the phone on forever   ' the default value for phoneManualON will be 0 (OFF)  69   If phoneManualON = 1  AND NOT lowPowerMode Then     powerPhone = 1   EndIf   PortSet(port_phone_power,powerPhone)   '-------------------------------------------  '======== Multiplexer: pH, ORP, DO, CO2 ============================    'multiplexer ON     PortSet(port_AM32B_RES,1)      'pH measurements - see WQ program for cal slope lines   'change temp variable to GS3 if that's better     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B     pHMult_1=-1/(((Temp_CTD_1+273)/298)*59)     VoltDiff(pH_1,1,mV2500,1,True,0,_60Hz,pHMult_1,7)          PulsePort (port_AM32B_CLK,10000)     pHMult_2=-1/(((Temp_CTD_2+273)/298)*59)     VoltDiff(pH_2,1,mV2500,2,True,0,_60Hz,pHMult_2,7)        'ORP measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B       VoltDiff(ORP_1,1,mV2500,1,True,0,_60Hz,1,0)     'DO measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B     VoltDiff (LDO,1,mV2500,1,True,0,_60Hz,0.02,0)     VoltDiff (Tw_LDO,1,mV2500,2,True,0,_60Hz,0.05,0)     'vaisala CO2 measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B   VoltDiff (CO2_1_mV,1,mV2500,1,True,0,250,1.0,0)   CO2_1_ppm = CO2_1_mV * 4   VoltDiff (CO2_2_mV,1,mV2500,2,True,0,250,1.0,0)   CO2_2_ppm = CO2_2_mV * 4    'end multiplexer     PortSet(port_AM32B_RES,0)  '======== Diff Channels on CR1000: Sonic ============================     'sonic measurements     VoltDiff (u,1,mV5000,5,True ,0,250,1.0,0)     u = ((25*2/5000)*u)-25     VoltDiff (v,1,mV5000,6,True ,0,250,1.0,0)     v = ((25*2/5000)*v)-25  70     VoltDiff (w,1,mV5000,7,True ,0,250,1.0,0)     w = ((25*2/5000)*w)-25     VoltDiff (SonicT,1,mV5000,8,True ,0,250,1.0,0)     SonicT = ((100/5000)*SonicT)+220-273.15  '======== Call data tables ============================    CallTable(dt)   If measureCO2 = 1 Then     CallTable(dt_co2)    EndIf     NextScan EndProg     71 Appendix 2   CR1000 program used for autonomous and accompanied measurement 'Program for WQ and Vaisala sensors 'date edited: October 14, 2016 'author: mollie  '====== VARIABLES =================================  'general  Public PTemp, batt_volt Units PTemp = Deg C Units batt_volt = Volts  Public lowPowerMode Public MinBattVolt Public powerPhone Public phoneManualON Public firstDataWindow, secondDataWindow  'time  Public real_time(9) Alias real_time(1) = year Alias real_time(2) = month Alias real_time(3) = DayX Alias real_time(4) = hours Alias real_time(5) = minutes Alias real_time(6) = seconds Alias real_time(7) = microseconds Alias real_time(8) = weekday Alias real_time(9) = jday 'Public mins_current_day  'constants Const port_phone_power = 6 Const port_AM32B_CLK = 1 Const port_AM32B_RES = 2 Const port_SV_power = 7 Const port_CO2_power = 8  'Solenoid valve  'Public powerSV  'Vaisala CO2 variables Public CO2_1_mV, CO2_1_ppm, CO2_2_mV, CO2_2_ppm, measureCO2, powerCO2  72  'CTD variables Public CTD_1(3), CTD_2(3) Alias CTD_1(1) = Depth_CTD_1 Alias CTD_1(2) = Temp_CTD_1 Alias CTD_1(3) = EC_CTD_1 Alias CTD_2(1) = Depth_CTD_2 Alias CTD_2(2) = Temp_CTD_2 Alias CTD_2(3) = EC_CTD_2  Units Depth_CTD_1 = mm Units Temp_CTD_1 = Deg C Units EC_CTD_1 = uS/m Units Depth_CTD_2 = mm Units Temp_CTD_2 = Deg C Units EC_CTD_2 = uS/m  'pH/ORP variables Public pH_1, pH_2, ORP_1 Dim pHMult_1, pHMult_2  Units pH_1 = pH Units pH_2 = pH Units ORP_1 = mV  'DO variables Public Tw_LDO, LDO  'GS3 variables Public GS3_1(3), GS3_2(3) Alias GS3_1(1) = Moisture_GS3_1 Alias GS3_1(2) = Temp_GS3_1 Alias GS3_1(3) = EC_GS3_1 Alias GS3_2(1) = Moisture_GS3_2 Alias GS3_2(2) = Temp_GS3_2 Alias GS3_2(3) = EC_GS3_2  'anemometer variables Public u, v, w, SonicT Units u = ms-1 Units v = ms-1 Units w = ms-1 Units SonicT = DegC  '======== Data tables ============================  '30 minute table  73 DataTable(dt_co2on,true,-1)   DataInterval(0,30,Min,10)   Average(1,batt_volt,FP2,0)   Average(1,PTemp,FP2,0)   Average(1,CO2_1_ppm,IEEE4,measureCO2-1) 'this makes it track the data only if "measureCO2" = true   Average(1,CO2_2_ppm,IEEE4,measureCO2-1)   Average(3,CTD_1(),IEEE4,0)   Average(3,CTD_2(),IEEE4,0)   Average(1,ORP_1,IEEE4,0)   Average(1,pH_1,IEEE4,0)   Average(1,pH_2,IEEE4,0)   Average(1,LDO, IEEE4,0)   Average(1,Tw_LDO,IEEE4,0)   Average(3,GS3_1(),IEEE4,0)   Average(3,GS3_2(),IEEE4,0)   Average(1,u,IEEE4,0)   Average(1,v,IEEE4,0)   Average(1,w,IEEE4,0)   Average(1,SonicT,IEEE4,0) EndTable  '5 second table - only activated when measureCO2 = true DataTable(dt_co2_co2on,true,-1)   DataInterval(0,5,Sec,10)     Average(1,CO2_1_ppm,IEEE4,measureCO2-1) 'this makes it track the data only if "measureCO2" = true   Average(1,CO2_2_ppm,IEEE4,measureCO2-1)   Average(3,CTD_1(),IEEE4,0)   Average(3,CTD_2(),IEEE4,0)   Average(1,ORP_1,IEEE4,0)   Average(1,pH_1,IEEE4,0)    Average(1,pH_2,IEEE4,0)    Average(1,LDO, IEEE4,0)   Average(1,Tw_LDO,IEEE4,0)   Average(3,GS3_1(),IEEE4,0)   Average(3,GS3_2(),IEEE4,0)   Average(1,u,IEEE4,0)   Average(1,v,IEEE4,0)   Average(1,w,IEEE4,0)   Average(1,SonicT,IEEE4,0) EndTable  'Main Program BeginProg     PortsConfig (&B11111111,&B11111111)  74   firstDataWindow  =  15   secondDataWindow =  16   phoneManualON = 0   MinBattVolt = 11.4    Scan (5,Sec,0,0)     '======== Panel Temp/Battery ============================     PanelTemp (PTemp,250)   Battery (batt_volt)          RealTime real_time       '====== Low power mode =====================     If batt_volt < MinBattVolt Then       lowPowerMode = 1     Else       lowPowerMode = 0     EndIf   '======== SDI: CTD/GS3 ============================    'CTD measurements     Delay (1,200,mSec)     SDI12Recorder (CTD_1(),3,1,"M!",1.0,0)     Delay (1,200,mSec)     SDI12Recorder (CTD_2(),3,6,"M!",1.0,0)        'GS3 measurements     Delay (1,200,mSec)     SDI12Recorder (GS3_1(),3,3,"M!",1.0,0)     Delay (1,200,mSec)     SDI12Recorder (GS3_2(),5,4,"M!",1.0,0)  '======== Solenoid and CO2 ============================    'SV switch - change to reflect how long to inject CO2    ' If ((minutes >= 25 AND minutes < 30) OR (minutes >= 55)) AND NOT lowPowerMode Then     'powerSV = 1     'Else      ' powerSV = 0     'EndIf     'PortSet(port_SV_power,powerSV)       'CO2 ppm measurements - on or off   75    '  If ((minutes >= 23 AND minutes < 30) OR (minutes >= 53)) AND NOT lowPowerMode Then       powerCO2 = 1   '  Else     '  powerCO2 = 0    ' EndIf          PortSet(port_CO2_power,powerCO2)        '  If powerCO2 AND ((minutes >= 25 AND minutes < 30) OR (minutes >= 55)) Then       measureCO2 = 1   '  Else    '   measureCO2 = 0   '  EndIf      '======== PHONE ===========================   ' Turn the phone on for one hour during the first time window   ' (uncomment 'OR' and remove the 1st 'Then' to turn on for 30 mins during each of two time windows   If (hours = firstDataWindow)  AND NOT lowPowerMode Then 'OR hours = secondDataWindow) AND minutes < 30 Then     powerPhone = 1   Else     powerPhone = 0   EndIf   ' Use the phoneManualON flag to keep the phone on forever   ' the default value for phoneManualON will be 0 (OFF)   'If phoneManualON = 1  AND NOT lowPowerMode Then     'powerPhone = 1   'EndIf   PortSet(port_phone_power,powerPhone)   '-------------------------------------------  '======== Multiplexer: pH, ORP, DO, CO2 ============================    'multiplexer ON     PortSet(port_AM32B_RES,1)      'pH measurements - see WQ program for cal slope lines   'change temp variable to GS3 if that's better     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B     pHMult_1=-1/(((Temp_CTD_1+273)/298)*59)     VoltDiff(pH_1,1,mV2500,1,True,0,_60Hz,pHMult_1,7)          PulsePort (port_AM32B_CLK,10000)     pHMult_2=-1/(((Temp_CTD_2+273)/298)*59)     VoltDiff(pH_2,1,mV2500,2,True,0,_60Hz,pHMult_2,7)  76        'ORP measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B       VoltDiff(ORP_1,1,mV2500,1,True,0,_60Hz,1,0)     'DO measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B     VoltDiff (LDO,1,mV2500,1,True,0,_60Hz,0.02,0)     VoltDiff (Tw_LDO,1,mV2500,2,True,0,_60Hz,0.05,0)     'vaisala CO2 measurements     PulsePort (port_AM32B_CLK,10000) ' move to the next channel of AM32B   VoltDiff (CO2_1_mV,1,mV2500,1,True,0,250,1.0,0)   CO2_1_ppm = CO2_1_mV * 4   VoltDiff (CO2_2_mV,1,mV2500,2,True,0,250,1.0,0)   CO2_2_ppm = CO2_2_mV * 4    'end multiplexer     PortSet(port_AM32B_RES,0)  '======== Diff Channels on CR1000: Sonic ============================     'sonic measurements     VoltDiff (u,1,mV5000,5,True ,0,250,1.0,0)     u = ((25*2/5000)*u)-25     VoltDiff (v,1,mV5000,6,True ,0,250,1.0,0)     v = ((25*2/5000)*v)-25     VoltDiff (w,1,mV5000,7,True ,0,250,1.0,0)     w = ((25*2/5000)*w)-25     VoltDiff (SonicT,1,mV5000,8,True ,0,250,1.0,0)     SonicT = ((100/5000)*SonicT)+220-273.15  '======== Call data tables ============================    CallTable(dt_co2on)   CallTable(dt_co2_co2on)    NextScan EndProg    77 Appendix 3  Description of headspace analysis via gas chromatography I collected air samples from above the stream reach in duplicate for analysis on a gas chromatograph (Agilent 7890A GC system) and calculation of atmospheric CO2. In this method, I collected 30 mL of air in a syringe, attach a filter with a needle to the syringe, and inject the air into a sealed vial for analysis. I repeated this on multiple occasions during the study period to determine the average atmospheric CO2 concentration at the study site. This parameter is used in calculations of kCO2, k600, and CO2 emissions.    78 Appendix 4  Input stream velocity (V) (m s-1) and the logarithm of predicted CO2 emissions estimates (log(FCO2)) (g C L-1 d-1) of the loess fit V log(FCO2) 0.22 4.11 0.27 4.74 0.36 6.01 0.25 4.53 0.21 4.08 0.16 3.59 0.25 4.49 0.23 4.22 0.23 4.28 0.30 5.06 0.26 4.60 0.21 3.99 0.18 3.73 0.16 3.60 0.15 3.50 0.22 4.14 0.30 5.13 0.34 5.66 0.25 4.52 0.23 4.20  79 0.24 4.32 0.23 4.24 0.33 5.54 0.29 4.99 0.24 4.31 0.20 3.93 0.17 3.68 0.15 3.53 0.14 3.43 0.14 3.44 0.21 4.08 0.26 4.55 0.21 4.04 0.18 3.76 0.16 3.60 0.14 3.47 0.13 3.38 0.12 3.32 0.13 3.38 0.15 3.52 0.15 3.54 0.15 3.51 0.17 3.71 0.18 3.74 0.11 3.25  80 0.11 3.22 0.09 3.12 0.07 3.06 0.06 3.02 0.05 3.01 0.05 3.01 0.08 3.07 0.07 3.04 0.05 3.01 0.05 2.99 0.04 2.98 0.08 3.09 0.07 3.07 0.06 3.02 0.05 3.00 0.04 2.98 0.13 3.38 0.27 4.77 0.20 3.93 0.14 3.47 0.12 3.29 0.10 3.20 0.21 4.03 0.32 5.39 0.11 3.26  81 0.10 3.21 0.09 3.14 0.08 3.11 0.07 3.07 0.07 3.06 0.07 3.04 0.06 3.02 0.06 3.03 0.08 3.10 0.22 4.19 0.31 5.25 0.20 3.95 0.16 3.61 0.14 3.45 0.15 3.48 0.19 3.85 0.29 5.03 0.30 5.14 0.35 5.76 0.32 5.34 0.26 4.60 0.31 5.26 0.23 4.30 0.19 3.84 0.23 4.22  82 0.22 4.17 0.17 3.66 0.14 3.42 0.12 3.31 0.11 3.23 0.12 3.28 0.13 3.39 0.12 3.31 0.14 3.47 0.23 4.30 0.31 5.17 0.20 3.94 0.17 3.69 0.17 3.68 0.14 3.44 0.12 3.29 0.12 3.33 0.15 3.52 0.15 3.52 0.16 3.56 0.22 4.11 0.20 3.94 0.27 4.73 0.23 4.28 0.23 4.27  83 0.25 4.52 0.23 4.31 0.22 4.11 0.19 3.83 0.18 3.74 0.14 3.43 0.12 3.32 0.10 3.22 0.20 3.95 0.30 5.12 0.23 4.21 0.18 3.74 0.16 3.56 0.14 3.46 0.12 3.29 0.12 3.34 0.11 3.27 0.12 3.28 0.13 3.36 0.18 3.77 0.16 3.61 0.19 3.87 0.16 3.61 0.13 3.39 0.11 3.25  84 0.09 3.15 0.10 3.18 0.18 3.73 0.21 4.00 0.17 3.67 0.15 3.49 0.23 4.22 0.22 4.12 0.17 3.65 0.14 3.42 0.12 3.28 0.10 3.17 0.08 3.10 0.08 3.08 0.06 3.02 0.06 3.01 0.05 3.00 0.04 2.98 0.03 2.97 0.03 2.96 0.03 2.96 0.04 2.97 0.06 3.01 0.06 3.03 0.05 3.00  85 0.04 2.98 0.03 2.97 0.03 2.96 0.07 3.05 0.06 3.02 0.09 3.14 0.06 3.02 0.05 2.99 0.04 2.98 0.04 2.98 0.04 2.97 0.04 2.97 0.12 3.30 0.09 3.12 0.07 3.07 0.07 3.06 0.06 3.03 0.05 3.00 0.05 2.99 0.04 2.98 0.03 2.97 0.03 2.96 0.02 2.96 0.02 2.96  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0355223/manifest

Comment

Related Items