Tree Physiology 20, 745–753
© 2000 Heron Publishing—Victoria, Canada
Estimating CO2 flux from snowpacks at three sites in the Rocky
Mountains
NATE G. MCDOWELL,1,4 JOHN D. MARSHALL,1 TOBY D. HOOKER2 and ROBERT
MUSSELMAN3
1
Department of Forest Resources, University of Idaho, Moscow, ID 83844-1133, USA
2
Department of Natural Resource Sciences, University of Rhode Island, Kingston, RI 02881, USA
3
USDA Forest Service, Rocky Mountain Research Station, Fort Collins, CO 80526-2098, USA
4
Present address: Department of Forest Science, Oregon State University, Corvallis, OR 97331, USA
Received June 3, 1999
Summary Soil surface CO2 flux (Fs) is the dominant respiratory flux in many temperate forest ecosystems. Snowpacks increase this dominance by insulating the soil against the low
temperature to which aboveground components are exposed.
However, measurement of Fs in winter may be impeded by
snow cover. Likewise, developing annual Fs models is complicated by seasonal variation in root and microbial metabolism.
We compared three methods of measuring sub-snow Fs: (1)
dynamic chamber measurements at the upper snowpack surface (Fsnow), (2) dynamic chamber measurements at the soil
surface via snowpits (Fsoil), and (3) static estimates based on
measured concentrations of carbon dioxide ([CO2]) and conductance properties of the snowpack (Fdiffusional). Methods were
compared at a mid-elevation forest in northeastern Washington, a mid-elevation forest in northern Idaho, and a high-elevation forest and neighboring meadow in Wyoming. The
methods that minimized snowpack disturbance, Fdiffusional and
Fsnow, yielded similar estimates of Fs. In contrast, Fsoil yielded
rates two to three times higher than Fsnow at the forested sites,
and seven times higher at the subalpine meadow. The ratio
Fsoil /Fsnow increased with increasing snow depth when compared across all sites. Snow removal appears to induce elevated
soil flux as a result of lateral CO2 diffusion into the pit. We
chose Fsnow as our preferred method and used it to estimate annual CO2 fluxes. The snowpack was present for 36% of the
year at this site, during which time 132 g C m –2, or 17% of the
annual flux, occurred. We conclude that snowpack CO2 flux is
quantitatively important in annual carbon budgets for these
forests and that the static and dynamic methods yield similar
and reasonable estimates of the flux, as long as snowpack disturbance is minimized.
Keywords: snow, soil-surface CO2 flux, winter CO2 flux.
Introduction
Soil-surface CO2 flux (Fs) during winter has recently been
identified as a large contributor to annual CO2 flux from ter-
restrial ecosystems (Sommerfeld et al. 1993, Goulden et al.
1996). Annual Fs estimates in seasonally snow-covered areas
must account for this winter metabolism to estimate annual
CO2 flux (Oechel et al. 1997, Fahnestock et al. 1998). Recent
research has shown Fs can exceed 50% of total ecosystem respiration (Law et al. 1999). This flux is sensitive to both temperature and water availability (Paul and Clark 1989). Soil
surface CO2 flux results from the diffusional movement of
CO2 from two belowground sources: root and microbial respiration (Paul and Clark 1989, Ryan et al. 1997).
Sub-snow Fs has been measured by several independent
methods, including eddy correlation (Goulden et al. 1996),
static boundary layer gas collection (Coyne and Kelley 1974,
Sommerfeld et al. 1993, Brooks et al. 1997) and dynamic
chamber methods (Winston et al. 1995, 1997, Oechel et al.
1997). The gas collection method (Sommerfeld et al. 1993)
utilizes Fick’s Law for gas diffusion through porous media,
and is traditionally the most common method for estimates of
CO2 flux through snow. The chamber methods vary primarily
in placement of the chambers—either at the soil or snow surface. Each method has unique assumptions (presented in the
Methods and Discussion sections).
Extensive research on winter CO2 flux is accumulating.
Most such research has been conducted in arctic, subalpine
and boreal ecosystems (e.g., Sommerfeld et al. 1996, Oechel
et al. 1997, Winston et al. 1997). Less research has been done
in the more productive, low-elevation temperate forests (but
see Winston et al. 1995, Goulden et al. 1996). These forests
may have higher rates of CO2 flux through snow than more
northerly or higher elevation ecosystems because of their high
productivity and the positive correlation between soil CO2
flux and productivity (Raich and Schlesinger 1992). Estimates
of winter CO2 flux from such forests will improve global carbon budget models (Raich and Potter 1995).
Our objectives were: (1) to compare the static gas collection
method and two dynamic chamber methods and (2) to determine the percentage of annual soil CO2 flux that occurs
through snow at a mid-elevation, temperate forest.
746
MCDOWELL, MARSHALL, HOOKER AND MUSSELMAN
Materials and methods
Site descriptions
The northern Idaho site was located within a 70-year-old
mixed-conifer forest. The site is at 1100 m elevation in the Palouse range on the University of Idaho’s Experimental Forest
(46°5′ N, 116°5′ W). The climate is characterized by cool,
mild winters and warm, dry summers. Annual precipitation is
840 mm, the majority falling in the period from November
through May. We made measurements on eight blocks across
a 0.4-hectare plot. Overstory density is 1522 stems ha –1, and
consists of Pinus ponderosa Dougl. ex Laws., Pseudotsuga
menziesii var. glauca [Beissn.] Franco, Larix occidentalis
Nutt., Abies grandis (D. Don ex Lamb.) Lindl. and Betula
papyrifera Marsh. var. occidentalis with an understory of
Thuja plicata J. Donn ex D. Don, Abies grandis and Acer
glabrum Torr. The forest floor has sparse herbaceous cover
and a litter depth of approximately 0.015 m. The soil is a moderately well-drained silt loam derived from volcanic ash over
loess. At the time of the methods comparison, mean snow
depth was 0.62 m, no new snow had fallen in the previous
week, and the snowpack was in the early phases of spring melt.
Method comparisons were made on March 20 and 21, 1997.
The northern Washington site was located within a 98year-old Rocky Mountain Douglas-fir (Pseudotsuga menziesii
var. glauca) stand. The site is at 912 m elevation in the Kettle
River range near Curlew, WA (48°85′ N, 118°67′ W). Climate
is similar to the northern Idaho site, but with less precipitation
(660 mm year –1, 260 mm year –1 as snow). The soil is formed
from volcanic ash and loess over glacial till. We used two
404-m 2 plots, of which one was treated with 230 kg ha –1 N and
196 kg ha –1 K in 1988 and the other left untreated. There were
nine measurements per location per date. Method comparison
measurements were conducted once per month in December
1996, February 1997 and March 1997, and consisted of nine
measurements per block per date. Snow depth was 0.23 m in
December, 0.36 m in February, and 0.31 m in March.
The third site was located in the Glacier Lakes Ecosystem
Experiments Site (GLEES) in the Rocky Mountains of south
central Wyoming (41°20′ N, 106°20′ W). This location is at
3200 m elevation, and is characterized by long, cold winters
with deep snowpack accumulation. A subalpine area used in
previous studies of winter CO2 flux was chosen (Sommerfeld
et al. 1993, 1996, Massman et al. 1997) and includes both
meadow and forest (Picea engelmannii Parry ex Engelm.).
The two ecosystems were considered blocks, and each had
three measurement locations per method. Mean snow depth
was 1.9 m for the forest and 1.8 m for the meadow. Further details on site climatic and edaphic characteristics are described
in Sommerfeld et al. (1996). Measurements were made on 3
days between March 27 and April 1, 1997.
Method descriptions and assumptions
Soil CO2 flux is often measured by monitoring CO2 change in
a closed system over time. The Fs chamber methods utilize
closed- or open-system infrared gas analyzers (Field et al.
1989, Norman et al. 1992). There is only one possible source
of CO2 release, the surface under the open chamber (soil or
snow), and hence the change in CO2 concentration ([CO2]) is
attributable to that surface. Fluxes are calculated from the
slopes of [CO2] versus time curves (Figure 1), system volume,
and surface area of the measured surface. The [CO2] is corrected for temperature, pressure, and any change in system
water vapor. The chamber headspace [CO2] must be near ambient during measurement to approximate the [CO2] gradient
between the surface and aboveground atmosphere; alteration
of this gradient directly influences the rate of CO2 flux from
the surface (R.L. Garcia, Li-Cor, Inc., Lincoln, NE, personal
communication; Hanson et al. 1993, Healy et al. 1996). Pressure differentials between the chamber and outside atmosphere are minimized to avoid pressure-pumping driven CO2
flux (Hanson et al. 1993, Massman et al. 1997). Water vapor is
maintained near constant during measurements to prevent absorption or release of CO2 that is not measured by the infrared
optical cell (Model 6000-09, Li-Cor Inc., Lincoln, NE). A seal
between the chamber and surface is critical to prevent leaks
between the system and the surrounding atmosphere; this is
usually obtained by insertion of a chamber-attachable collar
before measurement (Norman et al. 1992, Norman et al.
1997). Static chambers, such as those that use alkali or soda
lime as a CO2 absorbent, have been shown to alter soil CO2
flux by changing headspace [CO2], and were not tested here
(Nay et al. 1994, Healy et al. 1996, Jensen et al. 1996,
Pongracic et al. 1997).
Method 1: Soil surface (Fsoil) Apparent soil CO2 flux, Fsoil,
measures CO2 flux directly from the soil surface, and hence requires removal of the snowpack to access the forest floor. Collars are generally inserted between 0.02 and 0.05 m into the
ground. Soil CO2 flux measurements are made 24 h to 2 weeks
later to allow the disturbed soil to recover from collar insertion.
We used a closed system infrared gas analyzer (LI-6200,
Figure 1. Typical response of snow surface CO2 flux (µmol m –2 s –1)
to headspace chamber CO2 concentration (µmol µmol –1). The dashed
line represents the CO2 concentration outside the chamber, and the
corresponding CO2 flux. Data were collected in March 1997 at the
Pseudotsuga menziesii var. glauca site in northern Washington.
Snow depth was approximately 0.31 m.
TREE PHYSIOLOGY VOLUME 20, 2000
SUB-SNOW CO2 FLUX
Li-Cor, Inc.) with a chamber attachment with surface area
165 cm 2 (15.24-cm diameter) and internal volume of 1881
cm 3. The measurement of Fsoil assumes that CO2 flux from the
soil surface was not altered by collar insertion or snowpack removal.
Measurements of Fsoil were conducted in snow pits. We dug
snow pits for placement of PVC collars 24 h before measurements began. Pits were at least 5 m from a site of measurement
by the other methods. The area of exposed soil within each
snow pit was no larger than necessary to allow collar insertion
(approximately 200 cm 2) and thus minimized disturbance of
within-soil physical conditions. Collars were placed in the
ground such that the bottom edge of the collar was at least
0.015 m below the mean forest floor depth. We placed the
snow back in the pit after collar placement and between measurements to minimize changes in soil temperature, and carefully excavated it approximately 1 h before measurement.
This was not possible at GLEES because of deep snow, so plywood boards were placed over the pits and then covered with
snow. Soil temperature (Tsoil, °C) was measured at 0.10 m
depth at the time of each measurement of Fsoil. Snow pits appear to have had no effect on Tsoil. Within-site variation in Tsoil
was less than 0.3 °C. Furthermore, Tsoil did not differ between
soil pits and neighboring sub-snow soils at the mixed conifer
site (n = 5, P = 0.83).
We monitored changes in CO2 flux over time after chamber
placement. We found changes in flux on the scale of minutes
were dependent on headspace [CO2] (Figure 1). As expected
based on Fick’s law, CO2 fluxes decreased as chamber [CO2]
rose. Headspace [CO2] was maintained near ambient by scrubbing CO2 to approximately 10 µmol mol –1 below ambient, and
then allowing it to increase during measurement to approximately 10 µmol mol –1 greater than ambient (Figure 1). There
was no relationship between CO2 flux and time elapsed from
snow removal on the scale of 1 to 3 hours (11 days tested,
highest r 2 = 0.02, P-value = 0.60). We found no change of CO2
flux on the scale of days.
Method 2: Snow surface (Fsnow) Measurement of Fsnow (apparent snow CO2 flux) is similar to measurement of Fsoil except
that the chamber is placed directly on the snowpack surface
(Figure 2). A seal between chamber and snow is made by insertion of the bottom edge of the chamber 0.03–0.10 m into the
snow immediately before measurement. The Fsnow method assumes that snow surface CO2 flux was not altered by chamber
insertion, and that the snowpack CO2 pool is at steady state.
Disturbance of the CO2 concentration gradient by foot travel
must be avoided by standing as far away as possible from the
measurement location.
Measurements of Fsnow were conducted by placing the
chamber directly on the snow surface. Areas chosen for Fsnow
measurement had an undisturbed area of snow of at least 3 ×
3 m to avoid any preferential pathways for CO2 flux from the
snow. The chamber used for Fsoil measurements was adapted
for Fsnow measurements by attaching a “snowshoe” to its base.
The snowshoe was a square, 0.43 × 0.43 m reinforced window
screen that kept the chamber afloat on the snow surface (Fig-
747
Figure 2. A schematic representation of the chamber design utilized
for Fsnow measurements. The letters correspond to: (A) the snow surface, (B) hollow PVC ring forming the chamber bottom that is submerged below the snow surface, including a beveled lower edge, (C)
mesh platform made from a window screen, attached to the Plexiglas
chamber by latches and glued to (B), (D) Plexiglas chamber, (E) perforated tubing to allow even circulation of air incoming from the
IRGA, (F) sensor head containing humidity and temperature sensors,
and (G) tubing between the IRGA and chamber.
ure 2). A PVC ring was placed in the middle of the screen with
latches that attached it snugly to the chamber, preventing
leaks. The ring base was submerged 0.05 m into the snow.
This design minimized snow disturbance by balancing the
chamber while maintaining a known system volume for all
measurements (a prerequisite for closed system flux measurements). Any CO2 released by snow disturbance flows outside
the measured area because the submerged ring has an outward-facing bevel. Human disturbance of snow was
minimized by wearing snowshoes. Headspace [CO2] was
maintained near ambient as with Fsoil.
Method 3: Snowpack (Fdiffusional) Apparent diffusional CO2
flux, Fdiffusional (µmol m –2 s –1), is based on Fick’s law of diffusion:
Fdiffusional = −φ t D ( dc / dz).
(1)
Equation 1 requires measurements of [CO2] at the snow–soil
interface and the surface of the snowpack to calculate the
[CO2] gradient across the snowpack (dc/dz, µmol mol –1 m –1),
as well as estimates of snowpack porosity (φ) and tortuosity (t)
(Sommerfeld et al. 1993). The diffusion coefficient of CO2 in
air (D; m 2 s –1) requires correction for temperature (T) and
pressure (P) (Fuller et al. 1966):
D × 10 − 4 = DC ( P0 / P ) ( T / T0) 1.81,
(2)
where DC is the binary diffusion constant for CO2 in air (0.138
× 10 –4, m 2 s –1 at standard temperature (T0) and pressure (P0))
(Massman 1998). We assumed that the snowpack was isothermal based on spot measurements of snowpack temperature.
The measurement of Fdiffusional does not account for CO2 storage within the snowpack; it assumes that the snowpack CO2
TREE PHYSIOLOGY ON-LINE at http://www.heronpublishing.com
748
MCDOWELL, MARSHALL, HOOKER AND MUSSELMAN
pool is at steady state.
The CO2 collection system duplicated that of Sommerfeld et
al. (1993, 1996). Gas collectors were placed at the soil surface
and 1.6-mm teflon tubing was run along the soil surface a distance of approximately 1 m, then attached to a metal conduit
that ran from the soil to the snow surface. The average volume
of the tubing system from stopcock to the collection end was
19.7 ml. Gas samples (10 ml) were drawn from stopcocks at
the ends of the tubing with 20-ml nylon syringes and analyzed
for CO2 on a gas chromatograph within 4 h (Model 5730A,
Hewlett-Packard Co., Palo Alto, CA). Gas samples were collected after purging the system by removing a 20-ml sample
and discharging the gas into the atmosphere. This resulted in
the 10 ml sample originating from the soil–snow interface.
Leakage of CO2 from the syringes averaged < 20 µmol mol –1
based on trials with a 2000 µmol mol –1 standard gas.
Snowpack characteristics were quantified using the walls of
the eight Fsoil snowpits. Porosity estimates were obtained by
measuring the thickness and density of each snow layer within
each snowpit (Table 1). Snowpack porosity estimates at each
collection location were derived from the mean of the nearest
snow pit. Tortuosity (t = 0.7 to 0.9; unitless) for each snow pit
was derived from the correlation with porosity (Prieur du
Plessis and Masliyah 1991):
t = 1 − (1 − φ) 2/ 3.
(3)
Statistical analysis
We used a complete randomized block design with repeated
measures for comparison of measured soil CO2 flux by each
method, block and period. Blocks of homogenous stem density, snow depth, and ground topography were selected and
each method was randomly placed within 5 m of the block
center (Sommerfeld et al. 1996). Data were log transformed to
meet assumptions of normality. Tukey’s HSD was performed
after detection of significant differences for the Idaho data.
Analyses were done with the SYSTAT 5.03 statistical package
(Wilkinson 1992) with a level of significance (α) of 0.05.
we measured Fs, Tsoil and gravimetric water content to 0.15 m
depth on 10 dates, four during periods of snow cover. Water
content on each date was measured on replicate soil samples
collected with a bulk density corer. Soil samples were sealed
and weighed fresh within 24 h, then dried at 105 °C for approximately 72 h for measurement of dry mass. Both Fsnow and
Fsoil were measured on the four dates with snow cover. We
tested models including linear, polynomial and exponential
equations fitted to temperature, water, or both (Coble 1997,
Law et al. 1999). Our statistical criteria for model selection included fit (R 2), variability (MSE) and significance tests of coefficients. The simplest model was chosen if the statistical
parameters were otherwise similar.
An annual model of daily soil temperature was required to
estimate daily Fs. Mean daily Tsoil was related to air temperature averaged over several days before measurement. Based
on air temperature data averaged from two nearby (~10 km)
weather stations, we tested averaging periods ranging from
3 to 48 days. The 28-day mean provided the best correlation to
measured Tsoil (R 2 = 0.92; Figure 3). This equation was used to
predict daily mean Tsoil for the 12-month period. The diel amplitude of Tsoil was 0.4 °C at the northern Washington site in
June 1997. This small temperature amplitude was associated
with a small CO2 flux amplitude of 0.5 µmol m –2 s –1. Therefore, we did not include a diel temperature amplitude in the
model.
To examine seasonal variation in metabolic activity, we fit
the data to a standard exponential model:
Flux = β 1 exp(β 2 Tsoil ),
(4)
where β1 is the intercept and β2 is the temperature response coefficient. Mean respiration rates for each date were normalized to 10 °C using β2:
F10 = Ft soil exp(β 2(10 ° C − Tsoil )),
(5)
Winter and annual budget
We began by comparing models that predicted mean Fs from
Tsoil and gravimetric water content (g g –1). At the Idaho site,
Table 1. Idaho snowpack characteristics. Mean values, standard deviations and ranges for soil-surface [CO2] (µmol mol –1), porosity
(unitless), depth (m), density (g cm –3), and gsnow (µmol m –2 s –1). n = 8
for each.
[CO2] Day 1
[CO2] Day 2
Porosity
Depth
Density
gsnow
Mean
SD
Range
2479
3053
0.41
0.62
0.54
300
298
360.1
0.01
0.13
0.01
64.5
854.6
1190.5
0.03
0.42
0.02
198.3
Figure 3. Measured soil temperature at 10 cm (°C) versus the previous
28-day mean air temperature (°C) from two nearby weather stations
from winter 1997 to winter 1998 at the mixed conifer site in northern
Idaho. Regression equation: Tsoil = 0.027x 2 + 0.146x + 1.046 (R 2 =
0.92).
TREE PHYSIOLOGY VOLUME 20, 2000
SUB-SNOW CO2 FLUX
where F10 is the CO2 flux at 10 °C and Ft soil is the measured
flux at Tsoil.
The percentage of annual CO2 flux under snow was calculated first by estimating the period of snow cover from a regression of snow depth at the site and snow depth at a local
weather station (R 2 = 0.99). We then predicted the first and last
days of snow cover from this regression. These dates were accurate within ≤ 10 days based on our observations. Finally, the
sum of the modeled CO2 flux from this period was divided by
the summed annual flux and multiplied by 100.
Results
Methods comparison
Estimates of Fs during periods of snow cover showed significant variation among methods (Table 2). Apparent soil CO2
flux, Fsoil, was threefold higher than Fsnow or Fdiffusional in Idaho
(F2,14 = 25.49, P < 0.01); there was no difference between Fsnow
and Fdiffusional (Tukey’s HSD; P = 0.76). Similarly, Fsoil was
significantly higher than Fsnow during all 3 months in Washington (Table 2, F1,21 = 21.53, P < 0.01). The value of Fsoil was
also significantly higher than Fsnow in Wyoming (Table 2, F1,9
= 19.26, P < 0.01).
We analyzed the relationships between Fsnow and snowpack
depth and conductance (gsnow) to examine the extent to which
snowpack characteristics control short-term CO2 flux through
snow. Parameter gsnow (µmol m –2 s –1; Table 1) can be derived
from integration of Fick’s law (Equation 1) and conversion to
749
common conductance units (Pearcy et al. 1991, Campbell and
Norman 1997):
g snow = −[( φ t D)/ ∆z ] 0.446 ( P / P0 ) ( T0 / T )(10 6).
(6)
In Idaho, the only site with sufficiently detailed snowpack
analysis, we found no relationship between Fsnow and gsnow
(n = 16, r 2 = 0.02, P = 0.77), snow depth (n = 16, r 2 = 0.03,
P = 0.5) or soil surface [CO2] (n = 16, r 2 = 0.02, P = 0.56). We
conclude that neither snow pack depth nor gsnow directly controlled Fsnow at this site.
We analyzed spatial and temporal variation in Fs. Spatial
variation in Fs was not statistically significant among blocks in
Idaho (F7,14 = 2.16, P = 0.11) or in Washington (F1,21 = 1.30,
P = 0.27). The value of Fs was not significantly different between the meadow and forest in Wyoming (Table 2, F1,9 =
3.60, P = 0.09). Fluxes in Washington were significantly
greater in December 1996 than February or March 1997 (Table 2, F2,42 = 25.40, P < 0.01). This temporal variation in Fsoil
and Fsnow was linearly correlated with Tsoil (n = 6, r 2 = 0.96,
P < 0.01 and n = 6, r 2 = 0.55, P = 0.09, respectively). No temporal differences were observed in Wyoming (Table 1, F2,18 =
0.22, P = 0.81).
Winter and annual CO2 flux
We compared models for predicting daily Fs in winter and
throughout the year. We selected Fsnow as our preferred
method for winter estimates. Although both linear and exponential temperature relationships were similar, Fs was better
Table 2. Mean sub-snow CO2 fluxes (µmol m –2 s –1) for a mid-elevation mixed conifer site in Idaho, a mid-elevation Pseudotsuga forest in Washington, and a subalpine Picea forest and meadow in Wyoming. Results presented by site, method and period (day for mixed conifer and subalpine
sites, month for Pseudotsuga) along with site mean soil temperature at 0.10 m (Tsoil, °C). Numbers following in parenthesis are standard deviations (all blocks included, n = 8 for each method in Idaho, 9 in Washington and 3 in Wyoming). In the expression f1 /f 2 (right-hand column), f1 =
Fsnow and f2 = Fsoil. In the case of the mixed conifer stand, the first ratio (0.33) is the mean of Fsnow and Fdiffusional divided by Fsoil, and the second ratio (1.03) is Fsnow/Fdiffusional.
Period 1 flux
Period 2 flux
Mixed conifer forest
Fsoil
Fsnow
Fdiffusional
Tsoil
1.86 (1.25)
0.67 (0.38)
0.65 (0.19)
0.94 (0.11)
2.54 (2.28)
0.77 (0.39)
0.84 (0.26)
1.09 (0.15)
Pseudotsuga menziesii var. glauca forest
Fsoil
Fsnow
Tsoil
1.77 (0.58)
0.63 (0.14)
1.06 (0.28)
Subalpine meadow
Fsoil
Fsnow
Tsoil
Subalpine Picea engelmannii forest
Fsoil
Fsnow
Tsoil
Period 3 flux
f 1 /f 2
—
—
—
0.33
0.66 (0.44)
0.39 (0.14)
–1.40 (0.30)
0.69 (0.47)
0.47 (0.30)
–1.81 (0.14)
0.48
1.07 (0.63)
0.16 (0.08)
0.33 (0.13)
1.01 (0.74)
0.14 (0.12)
0.49 (0.11)
1.01 (0.16)
0.11 (0.03)
0.29 (0.08)
0.13
1.57 (0.95)
0.60 (0.54)
–0.13 (0.37)
1.67 (0.81)
0.30 (0.10)
–0.20 (0.50)
1.54 (0.22)
0.45 (0.04)
–0.76 (0.28)
0.28
TREE PHYSIOLOGY ON-LINE at http://www.heronpublishing.com
1.03
750
MCDOWELL, MARSHALL, HOOKER AND MUSSELMAN
not significantly different from zero and therefore was not included in the annual model. To compare annual estimates of
the two temperature-based models, we also fitted the CO2 flux
estimates to an exponential relationship with Tsoil:
Flux = 1113
.
exp( 0.092 Tsoil ),
(8)
( R2 = 0.77, MSE = 0.56, Q10 = 2.6).
Figure 4. Soil-surface CO2 flux versus soil temperature at 10 cm (°C)
at the mixed conifer site in northern Idaho. Data collected between
winter 1997 and winter 1998. Regression equation: Fs (µmol m –2 s –1)
= 0.63 + 0.25Tsoil (R 2 = 0.87, MSE = 0.31).
predicted by the linear model (Figure 4):
Flux = 0.63 + 0.25 Tsoil
( R2 = 0.87, MSE = 0.31). (7)
Daily variation in CO2 fluxes were first estimated with the linear model (Equation 7). Daily fluxes were consistently around
0.8 µmol m –2 s –1 during periods of snow cover (Figure 6a).
During periods without snow, fluxes ranged from 1.7 to
4.7 µmol m –2 s –1 (annual mean of 2.1 µmol m –2 s –1). The CO2
flux adjusted to Tsoil = 10 °C (F10) exhibited distinct seasonal
variation, with lowest rates in winter, highest rates in May and
October, and a late-summer depression (Figure 6b).
The fraction of annual Fs that occurred through snow was
estimated next. Snow was present at the Idaho site from the
start of the study (February 1, 1997) through April 1, 1997,
and from November 20, 1997 through the end of the annual
budget (January 31, 1998). Based on the linear temperature
model (Equation 7), winter and annual fluxes were 132 g C
Large seasonal variation was observed in soil water (Figure 5).
The soil was dry from mid-summer to early autumn, and relatively wet during spring snowmelt (the air-filled porosity during spring snowmelt equaled 0.23, assuming a bulk density of
0.7 g cm –3 and a particle density of 2.5 g cm –3; these are typical values of andic soils in this area (D. Page-Dumroese,
USDA Forest Service Rocky Mountain Research Station,
Moscow, ID, USA, personal communication). Despite the
large variation in soil water content, the water coefficient was
Figure 5. Seasonal variation in (a) measured soil temperature at 10 cm
(°C), and (b) soil water (g g –1) at the mixed conifer site in northern
Idaho, winter 1997 to winter 1998.
Figure 6. (a) Mean daily soil-surface CO2 flux at the mixed conifer
site in northern Idaho. Flux was modeled by the linear temperature response based on mean air temperature from the previous 28 days. Triangles refer to sub-snow flux. Open circles represent measured Fs. (b)
Soil surface CO2 flux adjusted to 10 °C (F10) using the exponential
temperature response equation. Open circles represent values during
snow cover, filled circles are snow-free values. Error bars are standard errors for each date.
TREE PHYSIOLOGY VOLUME 20, 2000
SUB-SNOW CO2 FLUX
m –2 and 764 g C m –2 year –1, respectively. Flux of CO2 through
the snowpack was equivalent to 17% of the annual total (Figure 6a). Based on the exponential temperature model (Equation 8), winter and annual fluxes were 116 g C m –2 and 745 g C
m –2 year –1, respectively. Flux through the snowpack was 16%
of the annual total.
Discussion
Three methods were used to estimate CO2 flux through snow.
Method Fsnow is a direct chamber measurement of CO2 flux
from the surface of the snowpack. Method Fsoil is a direct
chamber measurement of CO2 flux from the soil surface. In
winter, the soil surface is accessed by excavating the snow.
Method Fdiffusional estimates CO2 flux through snow with a linear model of CO2 diffusion. Similar estimates were obtained
by Fsnow and Fdiffusional, but Fsoil yielded values threefold higher.
Of the three methods, the Fsnow measurement causes the
least disturbance and makes the fewest assumptions. Measurement of Fsnow assumes that the snow CO2 pool is at steady state
and causes little disturbance to the snowpack. The assumption
of steady state in the snowpack CO2 pool is valid as long as no
snow has fallen recently (Coyne and Kelley 1974) and the
snowpack is not disturbed. To assess the rapidity with which
steady state is reestablished after disturbance, we constructed
a simple simulation model and used it to evaluate the effect of
a tripling of snow depth. The model was based on the same
equations as Fdiffusional and assumed constant CO2 production
by the soil. This procedure estimated that the system would return to steady state within 7 hours after a sudden tripling of
snow depth, with 90% of flux regained after 3.5 hours (data
not presented). Based on this exercise, we assume that disturbances, including snowfall, result in relatively brief excursions from steady state. Although there is no direct way of assessing the immediate effect of a disturbance like chamber insertion, we observed no change in CO2 flux over time other
than that associated with rising headspace [CO2] (see Methods
and Figure 1).
The measurement of Fdiffusional depends on more assumptions
than measurement of Fsnow, but shares the advantage of minimal disturbance. Like Fsnow, Fdiffusional assumes a steady state
[CO2] pool within the snowpack. Additionally, Fdiffusional assumes that snowpack porosity and tortuosity are accurately estimated. Sommerfeld et al. (1996) concluded that the greatest
error in their estimates of Fdiffusional lay in their porosity estimates. Error in porosity estimates could occur as a result of
heterogeneous snowpack structure such as ice lenses, depth
hoar and tree-wells (Winston et al. 1995). However, this error
may be reduced by thorough snowpack sampling. At the
mixed conifer site, scatter in porosity estimates was low (Table 1), possibly because of the large number of snowpack profiles sampled within a relatively small area. Tortuosity is difficult to measure and few estimates exist. Our estimate of 0.7 to
0.9 seems reasonable compared with measurements (0.75–
0.9) at the Wyoming site (Massman et al. 1997).
Of the three methods, Fsoil requires the most disturbance.
751
The snowpack represents a significant barrier to the diffusion
of CO2. Removal of the snowpack creates a localized breach in
the diffusion barrier. We hypothesize that lateral diffusion of
soil CO2 into this breach could maintain a steep concentration
gradient, artificially increasing the CO2 flux (Figure 7). The
threefold higher rates of Fsoil compared with Fsnow and Fdiffusional
support this hypothesis. Also, the ratio of Fsnow to Fsoil decreased along a gradient of increasing snow depth across sites
(Figure 8). We suspect that Fsoil overestimates soil CO2 flux.
The elevated Fsoil rates relative to Fsnow and Fdiffusional may alternatively be due to an unrecovered disturbance of steady
state CO2 boundary conditions at the soil surface. If this were
true, we would expect Fsoil to decline as time passed after removal of snow from the snow pit. However, regression analysis of Fsoil versus time from snow removal showed no
correlation. For all sites and days (11), the best fit was an r 2 of
0.02 and a P-value of 0.60. It appears that steady-state conditions within the snow pit had been re-established shortly after
snow removal. These conditions, however, were characterized
by a steeper [CO2] gradient than exists at the snow surface, resulting in higher CO2 fluxes from the soil surface than those
observed at the snow surface (Figure 7).
It must be noted that no clear relationship between Fsnow and
Fdiffusional yet exists. In contrast to our observation of a ratio of
Figure 7. A hypothetical model of the effects of snow removal on
snowpack and soil CO2 concentrations and CO2 diffusion patterns.
Snow removal results in lateral diffusion, increasing soil–atmosphere
CO2 flux.
Figure 8. The ratio of Fsnow /Fsoil versus snow depth. In order of increasing snow depth, the data points are for Washington, Idaho and
Wyoming.
TREE PHYSIOLOGY ON-LINE at http://www.heronpublishing.com
752
MCDOWELL, MARSHALL, HOOKER AND MUSSELMAN
Fsnow /Fdiffusional of 1.03, Winston et al. (1995) found that
Fsnow /Fdiffusional ranged from 0.2 to 5.5, and Mast et al. (1998)
found that Fsnow /Fdiffusional was 0.37. Winston et al. (1995) attributed the wide range to variation in snowpack characteristics, which suggests that greater sampling may be required to
characterize CO2 fluxes in spatially heterogeneous snowpacks. Conversely, Mast et al. (1998) suggested their observed
ratio was a result of inadequate seals between the snow and
Fsnow chamber. It is hard to assess whether leaks occurred;
however, Mast et al. (1998) used long measurement periods
(up to 30 min), which may lead to underestimates of flux because of elevated headspace [CO2] (Figure 1 and R.L. Garcia,
personal communication).
We selected Fsnow as our best estimate of winter CO2 flux.
This decision was based on the small number of required assumptions and the lack of snowpack disturbance. In addition,
the similarity to Fdiffusional increased our confidence in the Fsnow
estimates. Method Fsnow was therefore used for the annual budget.
We selected a linear model for predicting daily soil CO2 flux
from soil temperature. Daily CO2 flux was summed for the
snow-covered period and for the year. We expected an exponential temperature response (Raich and Schlesinger 1992).
The exponential response may have been damped by low soil
water (Figure 5, also see Davidson et al. 1998) and belowground phenology (Figure 6b, Ryan et al. 1997, Law et al.
1999). We note that the linear model presented here is an empirical fit and may not be applicable to other sites.
Our scaled CO2 flux estimates compare well to published
values for other mid-elevation, temperate forests. The annual
flux of 764 g C m –2 year –1 is similar to the mean value for temperate coniferous forests of 681 ± 95 g C m –2 year –1 (Raich
and Schlesinger 1992), and to other estimates from the Pacific
Northwest. Annual Fs of Pinus ponderosa in central Oregon
was 683 g C m –2 year –1 (Law et al. 1999) and was 890 and
750 g C m –2 year –1 in Pseudotsuga menziesii var. glauca and
Pinus ponderosa forests, respectively, in eastern Washington
(D. Zabowski, College of Forest Resources, University of
Washington and R.S. Sletten, Quaternary Research Center,
University of Washington, unpublished observations).
Total sub-snow CO2 flux of 132 g C m –2 from the mid-elevation, mixed conifer forest is between that reported from arctic tundra and subalpine forest ecosystems. Flux of CO2 was
5–70 g C m –2 in arctic tundra (Oechel et al. 1997, Fahnestock
et al. 1998), 40–55 g C m –2 in boreal forests (Winston et al.
1997) and 232 g C m –2 in the subalpine forest in Wyoming
(Sommerfeld et al. 1996). Our estimate of a 17% sub-snow
contribution to annual CO2 flux is similar to that estimated
from a Pinus ponderosa ecosystem in central Oregon (Law et
al. 1999). Comparing these estimates to the annual totals
above supports the contention that winter CO2 flux cannot be
ignored for annual budgets (Oechel et al. 1997, Fahnestock et
al. 1998).
We conclude that measurements of CO2 flux during periods
of snow cover should be made with a minimum of disturbance
to the snowpack. The Fsoil method appears to overestimate CO2
flux. We hypothesize that this overestimate results from
lateral diffusion of CO2 from beneath the adjacent snowpack.
Estimates from Fdiffusional agree well with measurements of
CO2 flux directly from the surface of the snowpack (Fsnow). Either Fdiffusional or Fsnow appear to give reasonable estimates of
CO2 flux and may be used interchangeably. This research
highlights the importance of lateral diffusion of CO2 within
soils and snowpacks, and its consequences for surface CO2
flux measurements. It also argues for the importance of measuring winter CO2 flux, which accounted for 17% of the annual total.
Acknowledgments
We thank Drs. Linda Joyce, Bill Massman and Dick Sommerfeld
(USDA Forest Service, Rocky Mountain Research Station, Fort Collins, CO), who provided support and technical assistance both during
our stay at GLEES and throughout this study. We thank Drs. Tom
Gower and John Norman for the use of their prototype measurement
chamber, and Nick Balster and Brian Austin for field assistance.
Thanks also to Dr. Jeff Smith (Washington State University) for the
generous use of his gas-chromatograph for [CO2] measurements.
Drs. Michael Ryan, Deborah Page-Dumroese (USFS) and Greg
Winston (USGS) provided helpful reviews of the manuscript. This
study was supported by a McIntire-Stennis Grant to John Marshall
and Jim Moore.
References
Brooks, P.D., S.K. Schmidt and M.W. Williams. 1997. Winter production of CO2 and N2O from alpine tundra: environmental controls and relationship to inter-system C and N fluxes. Oecologia
110:404–413.
Campbell, G.S. and J.M. Norman. 1997. An introduction to environmental biophysics. 2nd Edn. Springer-Verlag, New York, 286 p.
Coble, D.W. 1997. Trends in above- and below-ground production of
trees and non-tree vegetation on contrasting aspects in western
Montana. Ph.D. Thesis. University of Montana, Missoula, MT,
127 p.
Coyne, P.I. and J.J. Kelley. 1974. Variations in carbon dioxide across
an arctic snowpack during spring. J. Geophys. Res. 79:799–802.
Davidson, E.A., E. Belk and R.D. Boone. 1998. Soil water content
and temperature as independent or confounded factors controlling
soil respiration in a temperate mixed hardwood forest. Global
Change Biol. 4:217–227.
Fahnestock, J.T., M.H. Jones, P.D. Brooks, D.A. Walker and J.M.
Welker. 1998. Winter and early spring CO2 efflux from tundra
communities of northern Alaska. J. Geophys. Res. 103:
29023–29027.
Field, C.B., T.J. Ball and J.A. Berry. 1989. Photosynthesis: principles
and field techniques. In Plant Physiological Ecology. Eds. R.W.
Pearcy, J. Ehleringer, H.A. Mooney and P.W. Rundel. Chapman
and Hall, London, England, pp 209–253.
Fuller, E.N., P.D. Schettler and J.C. Geddings. 1966. A new method
for the prediction of binary gas-phase diffusion coefficients. Ind.
Eng. Chem. 58:19–27.
Goulden, M.L., J.W. Munger, S. Fan, B.C. Daube and S.C. Wofsy.
1996. Exchange of carbon dioxide by a deciduous forest: response
to interannual climate variability. Science 271:1576–1578.
Hanson, P.J., S.D. Wullschleger, S.A. Bohlman and D.E. Todd. 1993.
Seasonal and topographic patterns of forest floor CO2 efflux from
an upland oak forest. Tree Physiol. 13:1–15.
TREE PHYSIOLOGY VOLUME 20, 2000
SUB-SNOW CO2 FLUX
Healy, R.W., R.G. Striegl, T.F. Russell, G.L. Hutchinson and G.P.
Livingston. 1996. Numerical evaluation of static-chamber measurements of soil–atmosphere gas exchange: identification of
physical processes. Soil Sci. Soc. Am. J. 60:740–747.
Jensen, L.S., T. Mueller, K.R. Tate, D.J. Ross, J. Magid and N.E.
Nielsen. 1996. Soil surface CO2 flux as an index of soil respiration
in situ: a comparison of two chamber methods. Soil Biol. Biochem.
28:1297–1306.
Law, B.E., M.G. Ryan and P.M. Anthoni. 1999. Seasonal and annual
respiration of a ponderosa pine ecosystem. Global Change Biol.
5:169–182.
Mast, M.A., K.P. Wickland, R.T. Striegl and D.W. Clow. 1998. Winter fluxes of CO2 and CH2 from subalpine soils in Rocky Mountain
National Park, Colorado. Global Biogeochem. Cycles 12:
607–620.
Massman, W., R.A. Sommerfeld, A.R. Mosier, K.F. Zeller, T.J. Hehn
and S.G. Rochelle. 1997. A model investigation of turbulencedriven pressure-pumping effects on the rate of diffusion of CO2,
N2O and CH4 through layered snowpacks. J. Geophys. Res.
102:18851–18863.
Massman, W.J. 1998. A review of the molecular diffusivities of H2O,
CO2, CH4, CO, O3, SO2, NH3, NO2 and N2O in air, O2 and N2 near
STP. Atmos. Environ. 32:1111–1127.
Nay, S.M., K.G. Mattson and B.T. Bormann. 1994. Biases of chamber methods for measuring soil CO2 efflux demonstrated with a
laboratory apparatus. Ecology 75:2460–2463.
Norman, J.M., R. Garcia and S.B. Verma. 1992. Soil surface CO2
fluxes and the carbon budget of a grassland. J. Geophys. Res. 97:
18845–18853.
Norman, J.M., C.J. Kucharik, S.T. Gower, D.D. Baldocchi, P.M.
Crill, M. Rayment, K. Savage and R.G. Striegl. 1997. A comparison of six methods for measuring soil-surface carbon dioxide
fluxes. J. Geophys. Res. 102:28771–28777.
Oechel, W.C., G. Vourlitis and S.J. Hastings. 1997. Cold season CO2
emission from arctic soils. Global Biogeochem. Cycles. 11:
163–172.
753
Paul, E.A. and F.E. Clark. 1989. Soil microbiology and biochemistry.
1st Edn. Academic Press, New York, 275 p.
Pearcy, R.W., E.-D. Schulze and R. Zimmerman. 1991. Measurement
of transpiration and leaf conductance. In Plant Physiological Ecology. Eds. R.W. Pearcy, J. Ehleringer, H.A. Mooney and P.W.
Rundel. Chapman and Hall, London, pp 137–160.
Pongracic, S., M.U.F. Kirschbaum and R.J. Raison. 1997. Comparison of soda lime and infrared gas analysis techniques for in situ
measurement of forest soil respiration. Can. J. For. Res. 27:
1890–1895.
Prieur du Plessis, J. and J.H. Masliyah. 1991. Flow through isotropic
granular porous media. Transport Porous Med. 6:207–221.
Raich, J.W. and W.H. Schlesinger. 1992. The global carbon dioxide
flux in soil respiration and its relationship to vegetation and climate. Tellus 44B:81–99.
Raich, J.W. and C.S. Potter. 1995. Global patterns of carbon dioxide
emissions from soils. Global Biogeochem. Cycles 9:23–36.
Ryan, M.G., M.B. Lavigne and S.T. Gower. 1997. Annual carbon
cost of autotrophic respiration in boreal forest ecosystems in relation to species and climate. J. Geophys. Res. 102:28871–28884.
Sommerfeld, R.A., A.R. Mosier and R.C. Musselman. 1993. CO2,
CH4 and N2O flux through a Wyoming snowpack and implications
for global budgets. Nature 361:140–142.
Sommerfeld, R.A., W.J. Massman, R.C. Musselman and A.R.
Mosier. 1996. Diffusional flux of CO2 through snow: spatial and
temporal variability among alpine–subalpine sites. Global
Biogeochem. Cycles 10:473–482.
Wilkinson, L. 1992. SYSTAT for Windows: Version 5. Evanston, IL,
750 p.
Winston, G.C., B.B. Stephens, E.T. Sundquist, J.P. Hardy and R.E.
Davis. 1995. Seasonal variability in CO2 transport through snow in
a boreal forest. In Biogeochemistry of Seasonally Snow-Covered
Catchments. Eds. K.A. Tonnessen, M.W. Williams and M.
Tranter. IAHS Publ. 228:61–70.
Winston, G.C., E.T. Sundquist, B.B. Stephens and S.E. Trumbore.
1997. Winter CO2 fluxes in a boreal forest. J. Geophys. Res.
102:28795–28804.
TREE PHYSIOLOGY ON-LINE at http://www.heronpublishing.com