American bison (
The American bison (
The ecological role of bison has become better understood as populations have recovered (Allred et al., 2001; Hansen, 1984; Knapp et al., 1999). Bison feed preferentially on grasses (Plumb and Dodd, 1993; Steuter and Hidinger, 1999) and often enhance forb diversity as a result (Collins and Steinauer, 1998; Hartnett et al., 1996; Towne et al., 2005). They tend to graze in preferred meadows during winter and search broadly for the most energy-dense forages during the growing season (Fortin et al., 2003; Geremia et al., 2019), often in areas which have recently burned (Allred et al., 2011; Coppedge and Shaw, 1998; Vinton et al., 1993). Combined, these observations suggest that bison select for forage quality rather than quantity, which likely impacts their efflux of methane – which all ruminants emit – because ruminant methane emission is related to feed quality (Hammond et al., 2016), including cellulose and hemicellulose intake (Moe and Tyrrell, 1979). It remains unclear how much methane results from the cellulose-rich grass-dominated diet of bison given their preference for fresh foliage and if management for bison may increase or diminish the greenhouse gas burden of ruminant-based agriculture.
Atmospheric methane concentrations have been rising at an accelerated rate
since 2016 for reasons that remain unclear (Nisbet et al., 2019), and there
is an urgent need to improve our understanding of its surface–atmosphere
flux. Between 30 % and 40 % of current anthropogenic methane emissions
are due to enteric fermentation in livestock (Kirschke et al., 2013), and the
greenhouse gas burden of cattle alone is some 5 Pg carbon dioxide equivalent yr
Bison in North America are thought to have been responsible for some 2.2 Tg yr
Here, we measure methane flux from a bison herd on winter pasture using the eddy covariance technique (Dengel et al., 2011; Felber et al., 2015; Prajapati and Santos, 2018; Sun et al., 2015). We use flux footprint analyses combined with bison locations determined using automated cameras to estimate methane flux on a per-animal basis and discuss observations in the context of eddy covariance methane flux measurements from other ruminants.
The study site is a 5.5 ha fenced pasture on the Flying D Ranch near
Gallatin Gateway, Montana, USA (45.557
The study site near Gallatin Gateway, MT (45.557
A 3 m tower was installed near the center of the study pasture during November 2017 (Fig. 1) and surrounded by electric fencing to avoid bison damage. Four game cameras (TimelapseCam, Wingscapes, EBSCO Industries, Inc., Birmingham, Alabama, USA) were mounted to the tower and pointed in cardinal directions. Two additional game cameras were mounted near the pasture edge facing the tower. Cameras captured images every 5 min, and an example of an individual image from the south-facing camera located on the northern edge of the study pasture is shown in Fig. 2. Bison locations at the 0.5 h time interval of the eddy covariance measurements were estimated by manually attributing bison locations to squares in a 20 m grid overlaid on the pasture area (Fig. 1). The 20 m grid size represents the grid that we felt that we were able to attribute bison locations given features of the field that could be identified by camera, and we treat these observations as an initial guess that is subject to uncertainty. We test the sensitivity of per-animal methane efflux estimates to bison location estimates as described in the section on spatial uncertainty below.
A sample image of bison as viewed from the south-facing time-lapse camera located to the north of the study area (Fig. 1). The eddy covariance installation is visible toward the center of the study site. Please note that the date format in this figure is month day year (mm dd, yy) and that the time format uses the 12 h clock (AM and PM).
Incident and outgoing shortwave and longwave radiation and thereby the net radiation were measured using an NR01 net radiometer (Hukseflux, Delft, the Netherlands) mounted 1.5 m a.g.l. (meters above ground level). An SR50 sonic distance sensor (Campbell Scientific Inc., Logan, UT, USA) was installed at 1.3 m to gauge snow depth, and air temperature and relative humidity were measured at 2.25 m using an HMP45C probe (Vaisala, Vantaa, Finland). Average 0–30 cm soil moisture and temperature were collected using CS650 probes (Campbell Scientific). Meteorological variables were measured once per minute, and 0.5 h averages were stored using a CR3000 datalogger (Campbell Scientific).
Three-dimensional wind velocity was measured using a CSAT3 sonic anemometer (Campbell Scientific) at 2.0 m a.g.l. (meters above ground level). Carbon dioxide mixing ratios were measured at 10 Hz using an LI-7200 closed-path infrared gas analyzer (LI-COR Biosciences, Inc.) with an inlet placed at the same height as the center of the sonic anemometer. Methane mixing ratios were measured at 10 Hz using a LI-7700 open-path infrared gas analyzer (LI-COR Biosciences, Inc., Lincoln, Nebraska, USA) with the center of the instrument likewise located at 2.0 m and a 22 cm horizontal offset from the sonic anemometer; open- and closed-path infrared gas analyzers for eddy covariance have similar performance in field settings (Detto et al., 2011; Deventer et al., 2019). We use the atmospheric convention in which flux from the biosphere to the atmosphere is positive. Measurements were made during winter daytime hours from 07:00 to 17:00 local time to avoid depleting the battery bank and to ensure sufficient light to estimate bison location using game cameras. Flux measurements began on 14 November 2017 and ended on 14 February 2018.
Bison are dangerous and will charge humans. Their presence complicated data retrieval and game camera upkeep; some high-frequency flux measurements were overwritten, and cameras were shut down during exceptionally cold periods, resulting in missing measurements. Simultaneous flux and photographic data were obtained for the 7 January to 13 February 2018 period excluding 10 January 2018 when instruments were obstructed by snowfall. Flux data without accompanying game camera footage were obtained for the periods from 14 to 29 November 2017 and 31 December 2017 to 6 January 2018.
Methane and carbon dioxide fluxes were calculated using an EddyPro (LI-COR
Biosciences, Lincoln, Nebraska, USA). Standard double rotation, block averaging
and covariance maximization with default processing options were applied.
Spike removal was performed as described by Vickers and Mahrt (1997), and
spikes were defined as more than 3.5 standard deviations from the mean
mixing ratio for carbon dioxide and more than 8 standard deviations from the
mean mixing ratio for methane given the expectation of intermittent methane
spikes from the bison herd. The default dropout, absolute limit and
discontinuity tests were applied using the default settings following
recommendations by Dumortier et al. (2019), and the Moncrieff et al. (1997)
and Moncrieff et al. (2004) low- and high-pass filters were applied. The
Webb–Pearman–Leuning correction (Webb et al., 1980) was applied to calculate
methane efflux using the open-path LI-7700 sensor. Estimates of storage flux
in the 2 m airspace below the infrared gas analyzers were assumed to be
minor and excluded from the flux calculation. Flux measurements for which
the quality control flag was greater than one following Mauder and Foken (2011) (see also Foken et al., 2004) were discarded, and the net effect of
all corrections when bison were present was a methane flux reduction of
14 %. Measurements that exceeded an absolute value of 1
The eddy covariance flux footprint was calculated using the approach of
Hsieh et al. (2000) extended to two dimensions following Detto and Katul (2006). Such analytical footprint models have been found to give minimally
biased estimates of point-source fluxes in field settings (Dumortier et al.,
2019). We performed the footprint analysis on a 1 m grid and aggregated
values to the 20
The momentum roughness height (
The calculation of
Given that mean methane emissions were not significantly different from zero
in the absence of bison – as detailed in “Results” – we assume that observed
methane emissions are due to bison in the flux footprint. The relative
contribution of bison to each 0.5 h eddy covariance measurement was
calculated by expanding the approach of Dumortier et al. (2019) (see also
Prajapati and Santos, 2019) for multiple point sources. From the definition
of the footprint function (e.g., Schmid, 1997), the measured density of a
scalar
We only adopt this approach for calculating average methane efflux per bison,
as measured carbon dioxide fluxes in the absence of bison were significantly
greater than zero. Methane efflux values less than
An eddy covariance flux footprint calculated following
Our observations are subject to multiple sources of uncertainty including uncertainty from eddy covariance measurements, footprint models and bison location estimates. Uncertainty of the eddy covariance methane flux measurements was determined by Deventer et al. (2019) to be between 6 % and 41 % for 0.5 h fluxes. We use an uncertainty of 41 %, as we are primarily concerned with providing a conservative uncertainty assessment and take the absolute value of the measurements multiplied by this percentage to calculate uncertainty due to eddy covariance measurements. Uncertainty due to the flux footprint was calculated as the mean percent difference in per-bison flux calculated using the Hsieh et al. (2000) and Kljun et al. (2015) footprint models.
Uncertainty due to bison location estimates is more difficult to calculate. The location of bison in the pasture was approximated visually by identifying the position of bison in relation to static cues in the study area using 5 min photographs. Observations were then aggregated to 0.5 h flux measurement periods. This approach results in spatial uncertainty in bison location, especially due to movements within 0.5 h periods and potential misallocation to nearby grid cells (Fig. 1). We acknowledge that uncertainty in the bison location estimate is likely using our approach and explored the sensitivity of per-bison methane flux estimates to bison location using stochastic simulations in order to arrive at a conservative uncertainty estimate.
The camera measurements resulted in many pixels where bison were not observed (e.g., Fig. S1), but there is a finite probability that this absence was in error. Pixels near populated pixels likely have a higher probability that bison were located within them because small movements within 0.5 h periods were common and because their locations may have been misallocated due to measurement uncertainty. We therefore sought an approach that simulates a spatial distribution of bison that is constrained by the camera measurements. To do so, we treated the camera measurements as an initial guess of their location that helped us define a likelihood surface. The likelihood surface was determined using two-dimensional Tikhonov regularization (Tikhonov and Arsenin, 1977), a classic mathematical technique to solve ill-posed problems, here the challenge of estimating the likelihood of bison location with intermittent and uncertain observations as described in detail in the Supplement. The probability of the 39 bison landing in a pixel is informed by this likelihood surface, and we used 100 simulations for both the Hsieh et al. (2000) footprint and the Kljun et al. (2015) footprint along with four different values of the spatial smoothness of the probability surface defined by the Lagrange multiplier (Eq. S1). An example of a likelihood surface generated for a single 0.5 h observation of bison locations and different values of the Lagrange multiplier is shown in Fig. S1. We explore the sensitivity of per-bison methane emissions to the Tikhonov regularization approach in the Supplement (Figs. S2 and S3).
We took the percent difference between the calculated per-bison methane emissions and values from the 200 stochastic simulations as the uncertainty due to bison location. Total uncertainty was then calculated by summing variances for the spatial uncertainty, footprint model uncertainty and eddy covariance uncertainty. We suggest strategies for reducing uncertainty in the “Discussion” section.
Air temperature averaged
A wind rose following Pereira (2020) for periods when eddy covariance measurements and bison location measurements were available. WS: wind speed.
Methane fluxes from 0.5 h periods averaged 0.048
The daily mean and standard error carbon dioxide and methane fluxes with standard error during daytime hours (07:00–17:00) from the study pasture near Gallatin Gateway, Montana, USA. The gray background denotes the interval during which bison were present on the study site.
The relationship between carbon dioxide and methane fluxes from the study pasture is shown for periods when bison were present (filled circles) and when bison were absent (open circles).
Methane flux was significantly and positively related to friction velocity
in the absence of bison at
Methane
Time-lapse camera footage yielded usable imagery for 444 0.5 h periods, of which 245 0.5 h periods had available eddy covariance observations and of which 177 had eddy covariance measurements that passed quality control criteria. Bison tended to aggregate in an area on the western side of the pasture near the location where supplemental hay was often provided (Fig. 9a). They intermittently visited the area north of the tower in mornings and afternoons and intermittently made sporadic mass movements to the southernmost edge of the field near its gate during midday periods (Fig. 9b–d).
Average proportional bison density for three periods of the day. Each colored pixel represents a 20 m grid square; red dots denote the location of the eddy covariance tower; and subplot titles refer to local time. Color denotes the average number of bison present in each grid cell for the 39-animal herd.
Bison were located within the 90 % flux footprint 40 % of the time
(Fig. 10). There were 158 0.5 h observations with bison in the flux
footprint when applying the Hsieh et al. (2000) footprint model, and 146 observations were available when applying the Kljun et al. (2015) footprint
model; an average of eight (seven) bison were within the 90 % flux footprint
of the Hsieh et al. (2000) (Kljun et al., 2015) models. When excluding
periods for which bison were absent from the flux footprint, this value
increased to 21 (20), respectively (Fig. 10). Per-bison methane emission
estimates when using the Hsieh et al. (2000) footprint model had an average (mean
Per-bison methane flux estimates from stochastic simulations of bison
location were sensitive to the smoothness of the likelihood surface (Fig. 12). Combining per-bison methane flux estimates from all 100 simulations
resulted in a standard deviation of 6.2
The probability (
The eddy covariance flux footprint analysis coupled to bison location
estimates from automated camera images resulted in a mean methane flux of 55
It is important to study methane emissions from other grazing systems to
place our observations into a broader context and, moving forward, to design
grazing systems that minimize greenhouse gas burdens. From this perspective,
our simple seasonal scaling exercise may underestimate or overestimate
methane emissions from bison grazing systems for multiple reasons that must
be kept in mind when interpreting results. Methane emissions from cattle
have been observed to be on the order of 10 %–17 % higher in summer than
winter (Todd et al., 2014; Prajapati and Santos, 2018, 2019) such that our wintertime methane flux observations may be lower than
what full annual measurements would yield. Our observations were similar to
wintertime measurements of beef cattle in a feedlot, on the order of 75 g CH
Kernel density estimates of the distribution (
The estimated mean per-bison CH
We did not observe significant differences in methane efflux over the course
of the day, noting that our observations were limited to daytime periods
because we had little basis to determine animal location at night. Other
studies have observed higher methane efflux from cattle during feeding times
(Gao et al., 2011), but bison also frequently graze at night, leaving it
unclear if they also exhibit daytime and nighttime differences in methane
flux with implications for scaling flux across time. Methane efflux was not
significantly higher during days when supplemental hay was provided (
Nutritional needs also impact methane efflux; dairying buffalo cows for example are estimated to have higher methane emissions than other buffalo (Cóndor et al., 2008). The study herd comprised numerous pregnant females (Table S1) that have higher metabolic requirements such that methane flux values may be higher than a herd with fewer pregnant animals. Taken as a whole, there is no evidence from our measurements that bison have more or less methane efflux than typical values reported for cattle. We note that it is critical to make full year-round methane flux measurements to understand the seasonal course of bison methane efflux to establish defensible annual sums.
Methane flux was not related to air or soil temperature but was related to
Insufficient evidence exists in our data record to attribute observed
methane efflux to the onset of freezing conditions in soil (Mastepanov et
al., 2008). We note that extensive snow trampling (e.g., Fig. 2) likely
resulted in a situation where snow depth (Fig. 4c) and its insulating
effect on soil temperature (Fig. 4a) varied across the field and therefore
differed from snow and soil measurements taken within the instrumentation
enclosure. Regardless, mean methane flux when bison were absent,
It is important to note that potential methane fluxes from bison manure may
have been dampened by freezing conditions but may be an important methane
source during warmer conditions if it enters anoxic conditions. Manure is
thought to contribute a nontrivial portion (10–14 Tg CH
Ruminant behavior is an important consideration when measuring field-scale efflux (Gourlez de la Motte et al., 2019). The spatial distribution of bison in the study pasture often varied from morning to midday and afternoon (Fig. 9). It is difficult to infer from the available data whether the study bison are more active during morning and evening hours in the pasture environment like cattle (Gregorini, 2012). Supplemental hay was made available to the bison approximately 50 m west of the tower, and increases in the frequency of bison appearance there are associated with the animals' preferred feeding times after dawn and before dusk, but observed methane flux did not vary as a function of time of day (e.g., Dengel et al., 2011) as noted above. Regardless, ruminant methane flux measurements are simpler to make when animals congregate (Coates et al., 2017; Tallec et al., 2012) as was often observed in our study (e.g., Figs. 2, 9 and 10). Aggregation behavior in our study bison herd was often upwind of the eddy covariance tower (Figs. 5 and 9) and resulted in more overlap between flux footprint and bison location than would have occurred if bison locations were randomly distributed throughout the study area, emphasizing the importance of tower placement in eddy covariance studies of grazing systems.
Despite the largely favorable location of the herd with reference to wind direction and the flux footprint, spatial uncertainties in bison location dominated the total uncertainty calculated here. More accurate location observations are a logical way to reduce this uncertainty. Uncertainties in flux footprint modeling for methane source attribution were also nontrivial on the order of 33 % of total uncertainty. Footprint models of the type used here have been found to accurately estimate point sources of trace gas flux (Heidbach et al., 2017; Dumortier et al., 2019), but it is important to note that footprint modeling techniques play a large role in the spatial attribution of observed fluxes of ruminant trace gas flux (Felber et al., 2015). Prajapati and Santos (2018), for instance, found that an analytical model (Kormann and Meixner, 2001) predicted flux footprint areas 5 to 6 times larger than did an approximation of a Lagrangian dispersion model (Kljun et al., 2002) did such that footprint model uncertainty is a major source of uncertainty for measuring methane flux from multiple point sources as we also find here. Regarding the footprint model it is also important to note that emitted gas is warmer than the surrounding environment in our case. It is unclear how well typical eddy covariance flux footprint models simulate the release location of heated parcels, but we note that heat is also transferred more efficiently than passive scalars like methane in the convective sublayer (Katul et al., 1995) such that methane transport should not be assumed to behave like heat. It is also unclear for our case if a point near the snow surface accurately represented the typical parcel release height. We were unable to track individual animals with different muzzle heights, noting that the animals were also frequently grazing with muzzle below the snow surface such that the true parcel release point represented a wide range of heights that we had little basis to simulate from available observations.
Methane efflux cannot be completely removed from ruminant grazing systems; some 4.6 %–6.2 % of gross energy intake is lost as methane in cattle, sheep and goats worldwide (Johnson and Ward, 1996) with cattle often falling on the higher end of the observed range (Lassey et al., 1997). But there are other aspects of bison ecology that merit consideration when designing greenhouse-gas-cognizant grazing systems. For example, cattle tend to graze close to water more frequently than bison do (Allred et al., 2011) with unclear consequences for riparian vegetation, water quality and potential methane efflux from wallows. Cattle also tend to graze for longer periods than bison (Plumb and Dodd, 1993), and it is unclear if there is an associated consequence for methane efflux. Future work should consider the large inter-animal variability in methane efflux (Lassey et al., 1997), possibly using advanced techniques for identifying individual animals through photographs (Merkle and Fortin, 2014) or tracking devices (Felber et al., 2015). Animal age and size are also important factors in ruminant methane efflux (Jiao et al., 2014), and individual tracking may improve our estimates of this variability in a field setting. That being said, it will be difficult to measure the methane contributions of different animals in species that tend to herd using eddy covariance.
Adding seasonal foraging behavior, estimating emissions from individual animals, and addressing seasonal and inter-annual variability and trends in forage nutrition are likely to further improve prediction of methane emissions from grazing systems (Moraes et al., 2014). Advanced eddy covariance algorithms are also likely to improve flux estimates on short timescales, noting that non-stationary bursts have not been found to create systematic bias in methane budgets measured over longer time periods using eddy covariance (Göckede et al., 2019). Of these, advanced footprint attribution techniques like environmental response functions designed to create improved maps of surface–atmosphere fluxes (Metzger et al., 2013; Xu et al., 2017) may be uniquely applicable to the challenging case presented by grazing systems with mobile point sources and intermittent biogeochemical hotspots created by animal waste. Going forward, increases in atmospheric carbon dioxide concentrations are likely to decrease forage quality (Jégo et al., 2013), resulting in higher ratios of leaf carbon to nitrogen and increasing ruminant methane emissions (Lee et al., 2017), all else being equal. Understanding greenhouse gas fluxes from ruminants is therefore likely to be even more important in the future. An ongoing interest in bison reintroduction and ungulate ecology coupled with established micrometeorological measurement techniques will help us understand the present and future role that bison and other alternative grazing systems play in the Earth system.
We measured methane efflux from a bison herd from an enclosed pasture using the eddy covariance method. Measurements were made during winter and background methane flux measurements in the absence of bison were not different from zero. Bison were free to graze and were also fed supplemental hay, which likely resulted in different methane efflux from that of a natural herd. Regardless of potential differences in greenhouse gas fluxes between wild and managed bison, bison are not domesticated, and it is difficult to make measurements of their greenhouse gas efflux using standard techniques like chambers or the sulfur hexafluoride method. Our results suggest that eddy covariance is a promising method for measuring trace gas fluxes from non-domesticated ruminants and that improved technologies for tracking animal movement is a logical way to reduce total uncertainties in observations. There is little evidence from our observations that methane efflux from the study herd differed from wintertime methane efflux from a cattle feedlot system, but full annual flux observations are necessary to understand if methane efflux from bison differs from the cattle management systems that originally replaced them and are now in turn being increasingly replaced by bison across much of their native range by a diverse group of Native American tribes and private landowners who share a common interest in bison reintroduction and conservation.
Code for the Hsieh et al. (2000) footprint is available at
The supplement related to this article is available online at:
PCS designed the study with AAC, JED and WK and wrote the paper with all coauthors. AAC collected data and analyzed them with PCS and TG. NK assisted with the footprint analysis.
The authors declare that they have no conflict of interest.
We wish to acknowledge the Indigenous nations on whose ancestral lands the study took place and recognize that the infrastructure used for this project was built on Indigenous land. We recognize multiple Indigenous nations as past, present and future caretakers of this land, whose stewardship of the region was interrupted through their physical removal by the 1830 Indian Removal Act and through US assimilation policies explicitly designed to eradicate Indigenous language and ways of being until the 1970s. Paul C. Stoy acknowledges support from the Graduate School at Montana State University and the University of Wisconsin–Madison. Funding for the LI-7700 methane analyzer used in this work was provided to JED by an NSF-EPSCoR award (no. EPS-1101342) and Montana State University. Daniel Salinas, Gabriel Bromley, Zheng Fu and James Irvine provided technical assistance, and Aaron Bird Bear and colleagues provided cultural knowledge. This work could not have been completed without permission of Turner Enterprises, Inc. and the assistance of Carter Kruse and Danny Johnson.
This research has been supported by the US National Science Foundation (grant nos. DEB-1552976 and OIA-1632810) and the USDA National Institute of Food and Agriculture (grant no. 228396).
This paper was edited by Lutz Merbold and reviewed by four anonymous referees.