Exchange of Carbon Dioxide by a Deciduous Forest: Response to Interannual Climate Variability

The annual net uptake of CO2 by a deciduous forest in New England varied from 1.4 to 2.8 metric tons of carbon per hectare between 1991 and 1995. Carbon sequestration was higher than average in 1991 because of increased photosynthesis and in 1995 because of decreased respiration. Interannual shifts in photosynthesis were associated with the timing of leaf expansion and senescence. Shifts in annual respiration were associated with anomalies in soil temperature, deep snow in winter, and drought in summer. If this ecosystem is typical of northern biomes, interannual climate variations on seasonal time scales may modify annual CO2 exchange in the Northern Hemisphere by 1 gigaton of carbon or more each year.

The annual net uptake of Co2 by a deciduous forest in New England varied from 1.4 to 2.8 metric tons of carbon per hectare between 1991 and 1995. Carbon sequestration was higher than average in 1991 because of increased photosynthesis and in 1995 because of decreased respiration. Interannual shifts in photosynthesis were associated with the timing of leaf expansion and senescence. Shifts in annual respiration were associated with anomalies in soil temperature, deep snow in winter, and drought in summer. If this ecosystem is typical of northern biomes, interannual climate variations on seasonal time scales may modify annual CO2 exchange in the Northern Hemisphere by 1 gigaton of carbon or more each year.
Observations of atmospheric CO2 indicate that the carbon balance [net ecosystem production (NEP)] of the Earth's terrestrial biosphere varies by 1 gigaton of carbon per year (Gt C year-') ( (1). Many ecosystem processes are sensitive to weather (2), and the fluctuations in global NEP are probably a consequence of interannual climate variability. However, direct observations of the effects of climate variability on the CO2 exchange of whole ecosystems are required before the causes of global NEP variation can be assessed reliably.
We used a 5-year record of the turbulent exchange of CO2 between the atmosphere and a deciduous forest in New England [net ecosystem exchange (NEE)] (3)(4)(5) to evaluate the magnitude and causes of interannual variations in net production (NEP), respiration (R) (6), and gross ecosystem exchange (GEE) (7). The eddy-covariance technique (8,9) was used to measure hourly NEE from 28 October 1990 to 27 October 1995 (10) at Harvard Forest in central Massachusetts. Ecosystem respiration was measured directly during dark periods and estimated as a function of soil temperature during light periods (8). Hourly GEE was inferred by subtracting R from NEE.
The forest gained 30 to 60 kg C ha-1 day-1 in the growing seasons and lost 10 to 20 kg C ha-' day-' in the dormant periods ( Fig. 1). Annual net CO2 uptake ranged from 1.4 to 2.8 metric tons C ha-1 (Table  1) (12, 13), with above-average uptake in 1990-1991 and 1994-1995 (14). The rise in sequestration during 1990-1991 was caused by increased annual gross production, and the rise during 1994-1995 was caused by decreased annual R. Annual GEE and R varied as the result of 1to 2-monthlong episodes of anomalous activity (Fig. 2). For example, lower than average annual net production in 1993-1994 (Table 1) was a consequence of both higher than average respiration (Fig. 2) and lower than average gross production (Figs. 2 and 3) in spring. Large changes in annual GEE were associated with modest changes in the length of the growing season (Figs. 2 and 3). The leaves emerged 6 to 10 days later  Table 1. Values of NEE and R summed from day of year (DOY) 301 to DOY 300, and GEE summed from DOY 100 to DOY 300. Exchange from the forest is positive. Numbers in parentheses are 5 and 95 percentile confidence intervals (8,14). Net exchange in 1994-1995 was significantly more negative than in 1991-1992 and 1993-1994, which in turn were more negative than in 1992-1993. Net exchange in 1990-1991 was significantly more negative than in 1992-1993 and 1993-1994. in 1992, 1994, and 1995 than in 1991 and 1993, with similar delays in the uptake of carbon. Leaf expansion was correlated with air temperature, starting around 300 degree-days (15) and ending around 650 degree-days. Large shifts in annual GEE therefore resulted from brief anomalies in temperature during April and May. Canopy senescence began after the onset of cool nights (below 5°to 10"C). This occurred relatively late in 1992 and 1993, allowing photosynthesis to continue for 5 to 10 days longer than in 1994 and 1995 and increasing gross production by around 500 kg C ha-'.
Prolonged periods of cloudiness during mid-July 1992, mid-August 1992, and August 1994 each reduced gross production by around 400 kg C ha- (Fig. 2). The response of forest photosynthesis to the physical environment (light, temperature, and evaporative demand) varied little from summer to summer. We observed small enhancements (5 to 10%) in the instantaneous rates of photosynthesis at a given light level in 1993 and 1994 compared with 1991 and 1992, but these were only slightly larger than the long-term precision of the measurements (8). A modest (10%) reduction in photosynthesis at a given light level was associated with severe drought in August and September 1995 (Fig. 1).
The most striking period of anomalous respiration occurred in winter 1992-1993, when intermittent increases in efflux released a total of 1.6 to 2.0 tons C ha-I (Figs. 1 and 2) (16,17). These episodes coincided with high winds (8), a pattem we attribute to aspiration of CO2 accumulated in soil pores. The increases were observed only when the flux footprint was northwest of the tower, a poorly drained area that includes a bog. Periods of extreme efflux were not observed from this sector in other years and never from the southwest, an upland area of oaks and maples. The episodes started in December 1992, after a heavy snow on unfrozen soil (Fig. 4). This snowpack and subsequent rains possibly compressed regions of the bog, altering soil aeration and causing increased pore-space CO2 for several months as a result of accelerated decomposition. The cumulative uptake in 1993 after removal of winter periods with northwest winds was 3.3 tons C ha-1, significantly greater than in 1992, 1994, or 1995, and the cumulative R in 1993 was 9.6 tons C ha-l.
Smaller enhancements in fall, winter, and spring respiration (Fig. 2) were correlated with unusually warm soil temperatures (Fig. 4). An increase in respiration of 200 kg C ha-1 during fall 1993 coincided with a 2°C increase in soil temperature. Respiration rates and soil temperatures were higher than normal in winter 1994, despite colder than normal air temperatures, reflecting thermal insulation by deep snow. The sensitivity of spring and winter respiration to soil temperature was often greater than expected for a direct affect of temperature on metabolism. Soil temperatures averaged 2.7°C from 15 March to 30 April 1992, with freezing periods through most of April, compared with 4.50C during the other years, when freezing ended in late March or early April. The corresponding decline in respiration during 1992, around 500 kg C ha-' or 40%, exceeded the 13% reduction expected for a respiration coefficient Q1o of 2.0 (8,18). The rate of microbial decomposition near 0°C may be limited by freezing (19), potentially amplifying the response of ecosystem respiration to weather anomalies that affect soil frost, such as late arrival of spring or deep snow.
Respiration rates in summer were extremely consistent from 1991 to 1994 (Figs. 1 and 2) despite a range of mean air temperatures. A decline in respiration of nearly 1000 kg C ha-' during late summer 1995 (Figs. 1 and 2) coincided with a severe drought when only 10% of normal precipitation was recorded. Remarkably, this decrease in respiration (30%) more than offset 0 la. the simultaneous decrease in photosynthesis (10%). The depletion of water near the soil surface apparently reduced soil respiration, while water remaining deep in the soil column supported photosynthesis, resulting in above-average carbon storage during 1994-1995 (Table 1).
Annual CO2 exchange was particularly sensitive to four aspects of climate: (i) the length of the growing season, regulated by air temperature in spring and early fall, (ii) cloud cover in summer, (iii) snow depth and other factors affecting soil temperature in the dormant season, and (iv) drought in summer. Photosynthesis and respiration were relatively insensitive to other aspects of climate, including growing-season temperature. Shifts in annual CO2 exchange resulted from weather anomalies during periods when the forest was particularly sensitive, rather than from changes in annual mean conditions. Most predictions of the response of terrestrial ecosystems to climatic warming focus on a shift in annual mean temperature, ignoring the possibility that CO2 exchange may be especially sensitive to the weather during specific intervals of the year.
A quantitative assessment of the effects of climate variability on global GEE, R, and NEE will require combining long-term flux observations in all of the major biomes (20) with spatially and temporally resolved weather data (21). We can make a rough estimate of the interannual variability in Northern Hemisphere CO2 exchange by assuming that all northern biomes respond with half the intensity observed at Harvard Forest. Mid-winter snow cover in the Northern Hemisphere has varied over the past 20 years by 7 X 106' km2 (22), potentially shifting hemispheric R by 0.1 to 0.8 Gt C year-1, depending on whether the phenomenon observed in 1992-1993 occurs elsewhere. Similarly, variations in fall and spring temperatures, inferred from the extent of fall and spring snow cover, could shift hemispheric R by 0.2 to 0.4 Gt C year and GEE by 0.2 to 0.4 Gt C year-Finally, variation in cloud cover over continents in summer could shift global GEE by at least 1 Gt C year' (23,24). These fluctuations in hemispheric CO2 exchange are of the same magnitude as those derived from analyses of atmospheric CO2 data (1).
The climate over northern continents has tended in recent decades toward warmer springs (25), warmer autumn nights (26), diminished snow pack (22), and increased cloud cover (23). We have shown that the annual CO2 exchange at Harvard Forest is sensitive to each of these aspects of climate. If the responses observed at Harvard Forest are indeed typical of northern biomes and persistent over decadal time scales, these trends in cli-1578 mate may have altered the carbon balance of the northern terrestrial biosphere (21). Longer growing seasons and reduced snow cover may therefore account for some of the net uptake of CO2 attributed to the terrestrial biosphere (1).