Introduction

Nearly half of the Earth’s terrestrial surface is located 500 m above sea level, with 25% of the Earth’s terrestrial surface classified as having mountain relief (Barry 2008). Much of this mountain terrain can be considered “complex terrain”, with dramatic changes in elevation, slope, geology, topography, climate, soils, and vegetation over short horizontal distances. Areas of high-elevation complex terrain are known to influence regional climate (e.g. orographic precipitation, rain shadows, mountain waves, valley/mountain breezes, etc.), thus these regions play a major role in the supply of water resources to the surrounding areas. Due to the large influence of these regions on climate and resources, an understanding of the interaction between the surface and vegetation with the lower atmosphere is important. The exchange of energy, water, and carbon between the surface and atmosphere represents the lower boundary conditions which help govern the climate system. Both empirical model estimates and remote-sensing observations in complex terrain suffer from errors associated with poor spatial resolution that cannot capture the surface spatial heterogeneity, and thus lack an understanding of surface processes that are required to properly run and calibrate models.

Additionally, mountain regions have recently been appreciated as sensitive ecosystems, thus observations in these regions over the last 20–50 years may serve as early warning indicators of global environmental changes (Williams et al. 2002; Seastedt et al. 2004). For example, observations of vegetation dynamics at the alpine tree line have indicated changes in species composition (e.g. Elliot and Baker 2004; Keller et al. 2000; Pauli et al. 2001; Walther et al. 2005), increased growth rates (e.g. Paulsen et al. 2000; Motta and Nola 2001), and increased tree establishment (e.g. Körner 2003; Daly and Shankman 1985). Many mountain regions such as the European and Swiss Alps have experienced simultaneous changes in both land use and climate, thus it can be difficult to identify the specific causes of these vegetation changes (see discussions by Beniston et al. 1997 and Gehrig-Fasel et al. 2007).

In North America, the boreal tree line has received far more attention in terms of paleoclimatological and paleoecological research than the alpine tree line (e.g. Kerwin et al. 2004; MacDonald et al. 1993). In the Rocky Mountain region in north-central Colorado, however, significant changes in both climate and land use/land cover have and are occurring. Kittel et al. (2007) found that over the period 1953–2006, the mean maximum air temperature in the subalpine forest site near that used in this study increased for much of the year at a rate of +0.4°C per decade, yet there was no trend in precipitation changes. Near the alpine site used in this study, Kittel et al. (2007) found (over the same period) a decrease in the mean maximum air temperatures in winter (−0.4 to −0.6°C per decade), and an increase in October through April precipitation (+100 mm per decade). Natural fire occurrence (Sherriff et al. 2001), together with logging and mining have all, at times, affected forest cover (e.g. Baker et al. 2007), and the recent mountain pine beetle outbreak will drastically alter the landscape within the next decade (Negron and Popp 2004). In addition, the study area along the continental divide is adjacent to one of the faster growing population regions in the United States, thus the deposition of pollutants with the upslope easterly winds can impact ecosystem functions (e.g. Sievering et al. 2007).

There have been few observations of exchanges of energy, water and carbon in complex mountainous terrain, despite the significance of such regions as described above. Studies conducted at the subalpine forest site that is used in this study have shown that eddy-covariance can be a reliable method in complex terrain (Turnipseed et al. 2002), and that horizontal advection of CO2 at that site can be significant during periods of low turbulent mixing (Sun et al. 2007; Yi et al. 2008). In alpine tundra near the alpine study site, Cline (1997) showed that the aerodynamic profile method could be used to give reliable estimates of the surface energy balance, and other short-term eddy-covariance studies conducted in steep, complex terrain have shown that quality measurements can be made (e.g. Gu et al. 2008; Lewicki et al. 2008; Hammerle et al. 2007).

We present observations of water and carbon dioxide exchange above high-altitude alpine tundra near Niwot Ridge, Colorado, USA, and compare and contrast the alpine measurements to simultaneous measurements conducted over an adjacent subalpine forest, over a period spanning two summers and one winter. After first discussing the quality of the eddy-covariance based measurements, we then compare the general meteorological and surface flux measurements between sites. Despite large differences in above-ground biomass, surprising similarities and differences were found in the diurnal and seasonal water and carbon exchanges.

Materials and methods

The eddy covariance method was used to measure turbulent fluxes of water (latent heat flux, λE), heat (sensible heat flux, H), and CO2 (F C) over a period from June 9, 2007 through July 3, 2008. Pertinent details describing the alpine tundra site and subalpine forest site are presented below.

Alpine tundra site

The alpine study site is located above tree line at an elevation of 3,480 m above sea level (40°03′11′′N; 105°35′11′′W) on a flat ridge top (Niwot Ridge, Colorado). To our knowledge, this is the highest-elevation eddy covariance site in the world, exceeding the elevation of flux measurements made on the Tibetan Plateau (3,250 m; Gu et al. 2008). The site is considered an alpine fellfield, one of the most widespread Front Range alpine communities (Walker et al. 2001), and is characterized by high wind speeds, dry soils, low snow accumulation, and sparse vegetation cover. Vegetation is short (less than 5 cm tall) with a leaf area index, LAI, less than 0.5 m2 m−2, and is dominated by Carex rupestris (curly sedge), Kobresia myosuroides (Bellardi bog sedge) with a high root: shoot ratio (Humphries et al. 1996), and Paronychia pulvinata (Rocky Mountain nailwort) (Walker et al. 2001). Long-term (1981–2008) continuous climate data collected at the nearby Saddle site indicate a mean annual air temperature of −2.15°C and 966 mm total annual precipitation.

A pair of identical eddy covariance systems were located 50 m apart along an east-west axis (parallel to the prevailing wind direction: Fig. 1). A sonic anemometer (CSAT 3, Campbell Scientific) and open-path gas analyzer (LI-7500, LI-COR) were mounted 3 m above the ground facing north. Additional instrumentation mounted at 3 m included air temperature and humidity (shielded HMP 45C, Vaisala), and net radiation (R n; NR-Lite, Kipp and Zonen). The soil heat flux (G) was measured at each site using a pair of soil heat flux plates buried at a depth of 2 cm (HFT3, REBS). Half-hour statistics from data sampled at 10 Hz (sonic and gas analyzer) and 5 s (all other sensors) were calculated using a datalogger (CR3000, Campbell Scientific). All raw data were transmitted in real-time via radio (900 MHz Spread Spectrum, FreeWave) and archived by the Niwot Ridge LTER program at INSTAAR at the University of Colorado in Boulder. 120-V AC power was available at both sites. Post-processing of the eddy covariance data to calculate the turbulent fluxes included all of the standard corrections and adjustments (e.g. coordinate rotation, Webb adjustment, etc.) following Aubinet et al. (2000) and Lee et al. (2004), similar to those corrections for the subalpine site on Niwot Ridge.

Fig. 1
figure 1

Frequency of wind by direction and speed (ms−1) during the summer months (June-July-August) for the alpine (top panels) and forest (bottom panels) sites during daytime (left panels; 8 am–4 pm), and nighttime (right panels; 8 pm–4 am) periods

Full size image

In complex terrain that may be especially prone to katabatic cold-air drainage, the vertical turbulent flux of CO2 as measured using the eddy covariance method (F C; umol m−2 s−1) may not equal the net ecosystem exchange (NEE) during periods of low wind speeds due to changes in the time rate of change of CO2 concentration beneath the sensor height (F S; umol m−2 s−1), and the horizontal advection of CO2 (F H; umol m−2 s−1). Horizontal advection, in fact, has been reported in relatively flat sites during calm conditions associated with noctural drainage flows (e.g. Aubinet et al. 2003; Leuning et al. 2008). Therefore, we approximate the alpine NEE as:

$$ NEE = F_{\text{C}} + F_{\text{S}} + F_{\text{H}} = \frac{1}{{V_{\text{m}} }}\overline{{w^{\prime}c^{\prime}}} \left( {z_{r} } \right) + \int\limits_{0}^{{z_{r} }} {\left[ {\frac{1}{{V_{\text{m}} }}\frac{{\partial \overline{c} \left( {z_{r} } \right)}}{\partial t} + \frac{1}{{V_{\text{m}} }}\overline{u} \left( {z_{r} } \right)\frac{{\partial \overline{c} \left( {z_{r} } \right)}}{\partial x}} \right]} {\text{d}}z $$
(1)

where V m is the molar volume of dry air (m3 mol−1), w is the vertical wind component (m s−1), c is the molar fraction of CO2 (umol mol−1) measured at the reference height z r  = 3 m, primes note deviations from the 0.5 h means, t is time (seconds), and \( \overline{u} \left( {z_{r} } \right) \) is the component of the mean horizontal wind speed (m s−1) aligned with the horizontal (Δx = 50 m) CO2 concentration gradient. In practice, we used finite differences to approximate the spatial derivatives in Eq. 1, and used one-level measurements at z r to approximate the integrals due to the low measurement height, sparse canopy, and windy conditions, coupled with the uncertainty of evaluating vertical profiles as described next.

Even with technically demanding and expensive experiments specifically designed to measure horizontal and vertical advection, Leuning et al. (2008) has shown that the spatial and temporal variations in vertical wind velocities and horizontal concentration gradients make the measurements of the advection terms so difficult that they can not be calculated to the required accuracy. In this study, using sonic anemometers with a vertical wind speed resolution of 0.5 mm s−1, gas analyzers with a RMS of 0.0043 mmol m−3, and a horizontal separation distance of 50 m (similar to the separation used in other studies such as Feigenwinter et al. 2008; Leuning et al. 2008), the uncertainly in the estimate of F H at night was large (standard deviations of 0.0427 and 0.0736 umol m−2 s−1, summer and winter nights, respectively), relative to the means shown in Table 1. This implies that even when trying to account for flows not aligned with the two towers (Kutsch et al. 2008), there still were difficulties in measuring F H due to the sensitivity of F H to errors in measurements of wind vectors and horizontal concentration gradients.

Table 1 Mean values of alpine NEE, eddy-covariance turbulent flux (F C), horizontal advection flux (F H), and the storage flux (F S) measured during the daytime (8 am–4 pm MDT) and nighttime (8 pm–4 am MDT), summer (JJA) and winter (DJF)
Full size table

As described in the Results and Discussion section, use of the double coordinate rotation method (or planar fit method) forced vertical advection to zero, hence vertical advection was not explicitly included in Eq. 1 because vertical profiles of CO2 concentrations were not measured. Some studies, however, have reported that vertical advection is of similar magnitude but opposite sign to F H (e.g. Aubinet et al. 2003; Feigenwinter et al. 2004; Sun et al. 2007; Wang et al. 2005), and therefore recommend to include neither or both in CO2 budget studies. In this study, persistent high wind speeds and associated turbulent mixing coincided with a small F H relative to F C, so including only the one advective term could be erroneous, but nonetheless has been included to show the potential magnitude of the adjective terms.

At the alpine site, the wind direction was nearly exclusively from the west in close alignment with the orientation of the two towers. Wind roses (Figs. 1 and 2) show that upslope (easterly) winds were fairly common during the daytime in the summer at the forest site yet seldom reached the alpine site. Downslope (westerly) winds prevailed at both sites at night during the summer, and both day and night during the winter. Following the approach described in Kutsch et al. (2008), horizontal advection could also occur in mixed situations of wind and slope-aligned katabatic drainage flows, and therefore a wind vector proportional to the gradient-based alignment of the towers to account for winds not perfectly aligned with the two towers was calculated. Although multiple sampling locations, hence towers, are ideal to quantify both F S and F H (e.g. Fegenwinter et al. 2008; Sun et al. 2007), only measurements at two locations were available to estimate F S and F H as has been the case in several studies (e.g. Aubinet et al. 2003; Staebler and Fitzjarrald 2004; Sun and Mahrt 1994; Wang et al. 2005).

Fig. 2
figure 2

Frequency of wind by direction and speed (ms−1) during the winter months (December-January-February) for the alpine (top panels) and forest (bottom panels) sites during daytime (left panels; 8 am–4 pm), and nighttime (right panels; 8 pm–4 am) periods

Full size image

Accounting for the recently recognized potential effects of additional sensible heat exchange from the top, bottom, and spars of the open-path gas analyzer on the alpine F C measurements (Burba et al. 2008 using their “Method 4”), on average a 6% reduction in F C over the entire measurement period, similar to the magnitude of the horizontal advection term, was found. Although not particularly significant on a short-term (i.e. diurnal) basis, this new correction could be significant when looking at long-term (i.e. annual) carbon budgets, thus this correction warrants further investigation and the application of this correction should be treated with caution (Burba et al. 2008).

Subalpine forest site

The subalpine forest site was located roughly 4 km east of the alpine site, at an elevation of 3050 m above sea level (40°01′58′′N; 105°32′47′′W) on a ~5% east-sloping surface. The site is considered a low-productivity subalpine forest, and is dominated by Picea engelmannii (Engelmann spruce), Abies lasiocarpa (subalpine fir), Pinus flexilis (limber pine), and Pinus controrta (lodgepole pine) with a mean height of 11.5 m and an LAI of 4.2 m2 m−2 (Monson et al. 2002; Turnipseed et al. 2002, 2003). Conditions are warmer and drier than the alpine site, with a long-term (1952-present) mean annual air temperature of 1.5°C and 800 mm annual precipitation.

Instrumentation at this site is fully described in other papers (e.g. Turnipseed et al. 2002, 2003). Briefly, eddy covariance instruments were mounted 21.5 m above the ground from a 26 m tall scaffolding tower. Several instruments (e.g. sonic anemometers, open and closed-path gas analyzers, net radiometers) were located at the site, with intercomparisons presented in Turnipseed et al. (2002). Both local-scale and meso-scale effects that influence airflow and turbulent flux measurements are described in Turnipseed et al. (2003) and Turnipseed et al. (2004), respectively, and a detailed investigation of the role of advection in net ecosystem exchange is presented in Sun et al. (2007) and Yi et al. (2008). The influences of snowpack on soil respiration are described by Monson et al. (2002, 2006) and Bowling et al. (2009).

Results and discussion

Assessments of data quality

Although short-term turbulent flux measurements have been made near the alpine site in the past (e.g. Cline 1997), the focus of that study was on the snowmelt period, and CO2 fluxes were not measured. In addition, fluxes were calculated based on aerodynamic formulae (i.e. profiles) requiring empirical atmospheric stability corrections. Improvements in both our understanding of micrometeorology in complex terrain (e.g. Hammerle et al. 2007) and advances in instrumentation (e.g. fast-response sensors) have increased the quality of turbulent flux measurements in complex terrain. A key assessment of the quality of flux measurements is surface energy balance closure; the degree to which the non-turbulent terms (available energy: R n-G) balance the turbulent terms (λE and H). Figure 3 shows an example of surface energy balance closure at the alpine site during the summer (JJA) and winter (DJF) months. Based on the slope of the linear regression lines, λE + H was within 9% of R n-G indicating a closure of 91% during the summer months, and within 29% (closure of 71%) during the winter months with an expected decrease in winter-time closure due to a marked decrease in the magnitude of the turbulent fluxes (see Fig. 10). This degree of the summer-time surface energy balance closure is comparable to that observed at the forest site on Niwot Ridge (84%; Turnipseed et al. 2002), and overall similar or improved compared to other high-latitude alpine sites (e.g. 70%; Gu et al. 2008), and sites located in steep, complex terrain (e.g. 77%; Lewicki et al. 2008: 72%; Hammerle et al. 2007). Also, since the alpine site was located on a relatively flat yet high-altitude ridge, the rotation transform angle for the sonic anemometer in the horizontal-vertical plane varied only by roughly 5 degrees depending on wind direction regardless of season, with outliers associated with low wind speeds (Fig. 4). In addition to the double rotation method that was applied to the alpine flux data (the mean cross-wind and vertical-wind velocity components were rotated to zero for each 0.5 h; McMillen 1988) the planar fit method, thought to be preferable in complex terrain (Wilczak et al. 2001), was also applied. Using the planar fit method resulted in similar rotations angles (e.g. 3.6o in the horizontal-vertical plane), and therefore there was no significant affect of rotation method on the turbulent flux calculations.

Fig. 3
figure 3

Alpine 0.5 h mean available energy (net radiation minus the soil heat flux) versus the sum of the latent and sensible heat fluxes for summer (June-July-August) and winter (December-January-February) months. Equation of the linear regressions (lines) are summer: y = 0.91x + 16 W m−2 (r 2 = 0.83; n = 4,000), and winter: y = 0.71x−7 W m−2 (r 2 = 0.66; n = 4,318)

Full size image
Fig. 4
figure 4

The rotation transform angle for the sonic anemometer in the horizontal-vertical plane for summer (filled circles: June-July-August) and winter (open circles: December-January-February)

Full size image

The high quality of the turbulent flux measurements was also largely due to the relatively short flux footprint, calculated using the analytical solution to the diffusion equation as described by Schuepp et al. (1990). During all conditions (typical daytime, nighttime, and neutral atmospheric stability), the upwind distance that the turbulent fluxes measurements were most sensitive to occurred within 40–50 m of the tower (Fig. 5a). The cumulative footprint (Fig. 5b) shows that 80% of the turbulent fluxes originated from distances of 340, 381, and 368 m during typical daytime, nighttime, and neutral atmospheric stability conditions, respectively. The similarity in the footprint calculations for both daytime and nighttime conditions compared to neutral atmospheric stability conditions was indicative of the persistent windy, near-neutral stability conditions.

Fig. 5
figure 5

Turbulent flux footprints showing the upwind distance that the flux measurements were most sensitive to (a) and the cumulative footprint showing how much of the flux originated with upwind distance (b), for typical daytime (solid), nighttime (dotted) and neutral (dashed) atmospheric stability conditions

Full size image

The relatively short turbulent flux footprints relative to the scale of the spatial heterogeneity, ridge-top site location, and lack of strong sources or sinks of CO2 in the sparse tundra vegetation upwind of the alpine site coupled with high horizontal wind speed and friction velocity (u *) combined to result in a small F S and F H relative to F C. Several studies (all conducted in forests) have demonstrated that F S and F H tended to be large at night when u * was low, and hence many have used a u * filter (threshold) to replace F C values at low u * with some form of soil or air temperature-based estimate of respiration, thus avoiding the advection issue (e.g. Goulden et al. 1996). Figure 6a shows that the nighttime alpine F C showed only a weak dependence on u *, only below roughly 0.15 m s−1, slightly less than the u * threshold of 0.20 m s−1 determined at the forest site used in this study (Monson et al. 2002). The windy, turbulent conditions experienced at the alpine site resulted in relatively few low u * conditions even at night (Fig. 6d). Over the entire study period, u * was less than 0.15 m s−1 only 10% of the time, thus rejecting F C at u * < 0.15 m s−1 had no significant effect on our estimate of NEE (see Fig. 14).

Fig. 6
figure 6

A: Eddy-covariance measured turbulent flux (a: F C), horizontal advection flux (b: F H), and storage flux (c: F S) during the night (8 pm–4 am MDT) shown as means corresponding to friction velocity bin classes of 0.1 m s−1. Histogram (d) shows the distribution of the nocturnal friction velocity

Full size image

Both F H and F S showed similar behavior with u * to that reported in other studies at forested sites; F H decreased as u * increased (Fig. 6b), and F S increased until u * reached ~0.4 m s−1 and subsequently decreased (Fig. 6c). The increase in F S until ~0.4 m s−1 occurred because a minimum amount of turbulence was required to evacuate any CO2 accumulation beneath the sensor height. As u * increased, the horizontal advection of CO2 beneath the sensor height decreased as turbulent mixing displaced the cold, dense low-lying air, or decreased as turbulent mixing reduced the horizontal gradients of CO2.

An example of the diurnal behaviors of F C, F H, and F S during a 5 day summer period shows that F H tended to be large relative to F C at night, then decreased but remain positive during the daytime (Fig. 7). F S tended to be small relative to F C except for a few brief periods at sunrise. Averaged over summer and winter months, F H varied from less than 1% (winter) to a maximum 5–6% (summer) of NEE (Table 1). When averaged over daytime and nighttime periods, F S was small due to the high wind speeds and relatively low measurement height.

Fig. 7
figure 7

Diurnal example of the magnitudes of the eddy-covariance measured turbulent flux (F C; solid line), horizontal advection flux (F H; dotted line), and storage flux (F S; dashed line) at the alpine site

Full size image

Comparison of meteorological conditions

The seasonal trend in air temperature (T a) and relative humidity were similar between sites (Fig. 8a, b) indicative of the same synoptic-scale air mass influences. The mean T a at the alpine site, however, was roughly 2°C lower than the forest during the summer (Table 2), and nearly 3°C lower during the winter (Table 3). Following the dry adiabatic lapse rate of 1°C per 100 m, the decrease in T a due to the 430 m change in elevation was 4.30°C (or an estimated 2.15°C with a minimum moist adiabatic lapse rate of 0.5°C per 100 m), so differences in T a could be due to differences in vegetation cover.

Fig. 8
figure 8

24 h mean air temperature (a), relative humidity (b), and horizontal wind speed (c) measured at the alpine (thick line) and forest (thin line) sites

Full size image
Table 2 Summer (June–July–August) mean 0.5 h conditions observed at the alpine and forest sites, 24 h, daytime (8 am–4 pm), and nighttime (8 pm–4 am) periods
Full size table
Table 3 Winter (December–January–February) mean 0.5 h conditions observed at the alpine and forest sites, 24 h, daytime (8 am–4 pm), and nighttime (8 pm–4 am) periods
Full size table

Air temperature at the alpine site was measured 3 m above the ground, compared to 21 m above the ground, 9.6 m above the 11.4 m-tall canopy at the forest site. The effect of the forest vegetation canopy on T a was evident when comparing T a measured below the forest canopy, but at similar height above the ground (2 m) compared to the 3 m alpine measurement height (Fig. 9a). Compared to the alpine and above-canopy forest T a, the beneath-canopy T a values were higher during the daytime (downward sensible heat from the canopy) and lower at night (sub-canopy katabatic drainage flows; Fig. 1). While following similar annual trends (Fig. 8b), differences in relative humidity between sites were more pronounced in the winter than summer (Tables 2 and 3) due to the lower T a at the alpine site and non-linear saturation vapor pressure-temperature relationship.

Fig. 9
figure 9

Sample 0.5 h mean air temperature (a), relative humidity (b), and horizontal wind speed (c) measured at the alpine (thick lines), and above-forest (21 m: thin lines) and below-forest canopy (2 m: dashed lines) sites (June 24th and 25th)

Full size image

The most characteristic meteorological variable of the alpine/subalpine forest and most noticeable difference between sites was the high horizontal wind speed (U). At both sites, U increased in the winter periods, corresponding closely to the period when T a decreased below 0°C (Fig. 8c), and air was flowing downslope at both sites (Fig. 2). The average winter U exceeded 13 m s−1 at the alpine site during the night and day, roughly double that observed above the canopy at the forest site (Table 3), yet the average U during the summer was much lower and more similar between sites (Table 2). As described above, this persistent strong downslope wind at the alpine site helped to moderated horizontal advection of CO2 at the alpine site.

Comparison of the surface energy balance

The differences in vegetation (trees versus tundra) coupled with the cooler, slightly more humid, and windier conditions at the alpine compared to the forest site was apparent in all components of the surface energy balance. For example, R n was always greater at the forest compared to the alpine site, presumably due to a lower albedo (Figs. 10a and 11a). The mean daytime R n at the two sites were more similar in the winter months (128 W m−2; Table 3) compared to the summer months (147 W m−2; Table 2) indicating that vegetation, not seasonal changes in snow cover, played a major role in seasonal differences in R n between sites.

Fig. 10
figure 10

24 h mean surface energy balance components for the alpine (think lines) and forest (thin lines) sites: net radiation (a), sensible heat flux (b), latent heat flux (c), and soil heat flux (d), all in W m−2

Full size image
Fig. 11
figure 11

0.5 h mean surface energy balance components for the alpine (thick lines) and forest (thin lines) sites: net radiation (a), sensible heat flux (b), latent heat flux (c), and soil heat flux (d), all in W m−2, for June 24 and 25, 2007

Full size image

The sensible heat flux (H) was generally small with similar patterns at both sites for most of the measurement period except for early winter (Figs. 10b and 11b). In the early winter, H from the forest could exceed 100 W m−2 (24 h average) compared to an early-winter negative (downward) H at the alpine site. During the early-winter period, the forest canopy was relatively warm compared to the ambient air temperature, whereas the alpine tundra was relatively cold compared to the ambient air temperature. Seasonal differences in the latent heat flux (λE) between sites were also more apparent during the winter period (Fig. 10c) compared to the summer periods (e.g. Fig. 11c). Although the contribution to λE from transpiration can not be isolated, the summertime forest λE exceeded that from the alpine likely due to greater transpiration from the forest. During the winter, the larger λE from the forest site was likely due to the added contribution to λE from the sublimation of intercepted and beneath-canopy snow Molotch et al. (2007), since transpiration should be negligible at that time of year. The seasonal change in the partitioning of R n between H and λE (the evaporative fraction; λE/(H + λE)) was similar at the two sites, despite the large difference in above-ground biomass. The soil heat flux, however, was markedly different (Figs. 10d and 11d), with a much higher G at the alpine site due to the lack of vegetation cover and higher fraction of exposed bare ground, and lack of continuous snow cover in the winter months.

Comparison of net ecosystem exchange of CO2

Figure 12a shows that the 24 h mean NEE at both sites was most similar during the winter season. The forest site continued a net daytime CO2 uptake (negative NEE) much longer into the fall and switch to net CO2 uptake much earlier in the spring (Fig. 12b). Nighttime NEE (respiration) was larger from the forest site in the summer (Fig. 12c). At both sites, the switch from positive NEE (respiration) to negative NEE (assimilation) corresponded to changes in T a to +10°C (Table 4). An example of a typical diurnal pattern of NEE at the two sites (Fig. 13) further illustrates the larger uptake and release of CO2 by the forest compared to the alpine site. The differences, however, in the diurnal NEE patterns were not as large as initially expected given the difference in LAI at the two sites, since others have reported strong correlations between LAI and NEE (e.g. Jarlan et al. 2008; Shi et al. 2006).

Fig. 12
figure 12

Mean net ecosystem exchange of CO2 (umol m−2 s−1) at the alpine (thick lines) and forest (thin lines) sites for 24 h (a), daytime (b) and nighttime (c) periods

Full size image
Table 4 Summary of conditions for the periods with consecutive days of daily-averaged negative NEE (net CO2 uptake by the surface) and positive NEE (net loss of CO2 from the surface)
Full size table
Fig. 13
figure 13

0.5 h mean NEE at the alpine (thick line) and forest (thin line) sites (June 24th and 25th)

Full size image

One of the most valued benefits of using the eddy covariance method to continuously measure carbon and water cycling is the ability to estimate the annual carbon and water budgets. As has been discussed, however, NEE estimates based on eddy covariance measurements need to directly account for horizontal advection and storage of CO2 (e.g. Eq. 1), or alternatively, use a u * filter to replace F C measurement below a u * threshold with temperature-based estimates of nocturnal respiration. In Fig. 14, three estimates of the long-term NEE measured at the alpine site are shown; the vertical turbulent flux (F C), F C with F C values when u * < 0.15 m s−1 (Fig. 6a) were replaced with an estimate of respiration based on the empirical relationship of F C versus air temperature (r 2 = 0.70; not shown), and NEE calculated using Eq. 1. Figure 14 shows that the use of the u * filter tended to give consistently higher estimates of the cumulative C exchange compared to the other two methods. F C tended to be lowest of the three estimates, especially during periods of net uptake, and NEE calculated with F H and F S was somewhere between the other two estimates. At the end of the measurement period, however, all three methods were similar (153.7 163.6, 163.8 g C m−2, F C, F C with u * filter, and Eq. 1, respectively), and all indicated that the alpine site was a net source of C when integrated over two summers and one winter.

Fig. 14
figure 14

Cumulative carbon balance at the alpine site based on three estimates of NEE: F C alone (dotted line), F C with a u * filter (dashed line), and F C + F H + F S (solid line). Values at the end of the measurement period were 153.7, 163.6, and 163.8 g C m−2, F C, F C + F H + F S, and F C with u * filter, respectively

Full size image

Since the three estimates of the cumulative C balance at the alpine site were quite similar, Fig. 15 compares the alpine NEE using Eq. 1 to the forest’s NEE, and compares both to the cumulative atmospheric water exchange (E). Over the course of the entire measurement period, the alpine site had a net loss of both carbon and evaporative water, whereas the forest site had a net carbon gain coupled with an evaporative water loss greater than the alpine site.

Fig. 15
figure 15

Cumulative NEE expressed in g m−2 of carbon (a) and total evaporative water loss (b) for the alpine (thick lines) and forest (thin lines) sites over matching periods. Values at the end of the measurement period were 139 and −62 g C m−2, 408 and 688 mm H2O, alpine and forest, respectively

Full size image

Comparing NEE and E over the two complete periods with a net C gain to the surface (NEE > 0) and a net C loss from the surface (NEE < 0), interesting differences emerged (Table 4). The start of the alpine net C uptake period was delayed by 44 days compared to the forest site, and at both sites the start date coincided with the date when T a reached +10°C. The summer period of alpine net C uptake ended 64 days before the forest site (therefore the period of C uptake was 100 days for the alpine site compared to 208 days at the forest site), and the end of this period coincided with the date when T a decreased below −10°C. The ecosystem water use efficiency of the two sites, calculated as the ratio of the total C gain to the total H2O loss (and recognizing that this represents the surface as a whole since not 100% of H2O was transpired) was 0.25 g C/kg H2O at the forest compared to 0.08 g C/kg H2O the alpine site. Comparing the ecosystem water use efficiencies shows that the forest was more than three-times more efficient than the alpine tundra.

The switch to net C loss at both sites coincided with the date when T a decreased below −10°C (Table 4), which occurred earlier in the year at the alpine site. Net C loss ended when T a increased above 0°C at both sites, and since this occurred earlier in the year at the forest site, the overall length of the net C loss period was 96 days longer at the alpine site. The net C loss was more than three-times that from the forest, and coupled with similar H2O losses, the resulting ratio of C loss to water loss was nearly four-times greater at the alpine site.

Conclusions

The first results from, to the authors’ knowledge, the highest elevation alpine tundra eddy covariance site in the world, were presented and compared to simultaneous measurements conducted 4 km down- slope above a subalpine forest, over a period of two summers and one winter. The experiment showed that long-term (measurements are on-going), nearly continuous turbulent flux measurements, with real-time data transmission, can be made at the alpine site. Surface energy balance closure of between 71–91% was achieved, in-part due to the relatively flat ridge-top location, the short flux footprint, and windy conditions. The advection of CO2 was small, likely due the location on the ridge top away from strong upwind sources of CO2, persistent windy and turbulent conditions, and the influence of CO2 drainage flows down the adjacent valleys.

Despite a vertical and horizontal separation of only 430 m and 4 km, respectively, the alpine site had lower air temperatures and higher relative humidity than the forest site, but less than that explained by the dry adiabatic lapse rate. Both sites were windy, with higher mean wind speeds at the alpine site, and downslope winds prevailed at the alpine site compared to a diurnal switch to daytime upslope flows at the forest site.

The comparison between the surface energy balance components at each site showed the net radiation was higher over the forest (lower albedo), as was the sensible heat flux, especially in early in the winter. The latent heat flux at the alpine site was consistently less than that over the forest, yet the evaporative fraction was similar between the two sites, indicating similar partitioning of the available energy between latent and sensible heat. Although the magnitude of the soil heat flux was relatively small at both sites, it was greater at the alpine site, indicative of the low LAI and high fraction of exposed ground.

Calculating NEE at the alpine with or without accounting for horizontal advection either directly or by using a friction velocity-based filter did not have a significant effect on cumulative NEE estimates. The meteorological differences observed at the sites, namely windy, cooler conditions at the alpine site, resulted in a 108 day shorter negative (net uptake) NEE period, and a 96 day longer positive (net loss) NEE period, compared to the forest site. The ecosystem water use efficiency was more than three-times greater at the forest site compared to the alpine site, given the longer growing season, greater leaf area index, and presumably lower levels of climate-induced physiological stress.

The eddy covariance method of measuring the turbulent exchange of water and carbon dioxide has the advantage of continuously sampling relatively large areas with minimal disturbance. With suitable site selection and experimental design, this paper shows that this method can be used to make reliable measurements of surface-atmosphere interactions at a high elevation alpine tundra site situated in complex terrain. Over the bulk of the growing season, results show that the alpine tundra played a contrasting yet significant role in water and CO2 exchange despite its low leaf area index. It is hoped that these results will be used to calibrate and validate both modeling tools and remote sensing products for use in predicting the responses of alpine tundra to climate change.