The carbon dynamics of northern natural ecosystems contribute significantly to the global carbon balance. Periodic disturbances to these dynamics include insect herbivory. Larvae of autumn and winter moths (Epirrita autumnata and Operophtera brumata) defoliate mountain birch (Betula pubescens) forests in northern Scandinavia cyclically every 9–10 years and occasionally (50–150 years) extreme population densities can threaten ecosystem stability. Here we report impacts on C balance following a 2004 outbreak where a widespread area of Lake Torneträsk catchment was severely defoliated. We show that in the growing season of 2004 the forest was a much smaller net sink of C than in a reference year, most likely due to lower gross photosynthesis. Ecosystem respiration in 2004 was smaller and less sensitive to air temperature at nighttime relative to 2006. The difference in growing season uptake between an insect affected and non-affected year over the 316 km2 area is in the order of 29 × 103 tonnes C equal to a reduction of the sink strength by 89%.
 The terrestrial ecosystems of the world are generally considered net sinks of carbon [House et al., 2005]. In an average year the world's vegetation absorbs ca 60 billion tonnes of C by net photosynthesis and releases slightly less by respiration. The average difference between photosynthesis and respiration, the net uptake, is today estimated at about 2.6 billion tones [Denman et al., 2007]. This sink strength is variable and in a warmer climate the C sink may not persist due to warming soils and as a consequence, increased release of carbon. Disturbance patterns are also likely to change in the future and this may have further negative impact on the sink strength and a weakened C sink will produce a positive feedback on the climate. When estimating future terrestrial C uptake with present-day process understanding and various coupled carbon-climate models, a large spread of estimates is evident [Friedlingstein et al., 2006]. This illustrates differences in the simulated climate change, the current low model complexity and lack of process understanding [Heimann and Reichstein, 2008]. Events that damage vegetation are only one example of a factor that is currently under-represented in the models. To model future C uptake it is critical to include extreme, episodic and cyclic events such as wind throw, fires and insect pest damage. Insect outbreaks are already extensive [Jepsen et al., 2009] and lead to lower uptake of carbon and the likelihood of a prolonged negative effect on carbon sink capacity. Although indirectly documented before [Kurz et al., 2008], such impacts have to our knowledge not been directly quantified earlier as we do in the following.
 The impact of insect outbreaks can have surprisingly persistent effects on the balance of northern forests. In northern Fennoscandia, mountain birch forests defoliated by insect outbreaks need more than 70 years for the leaf biomass to fully recover [Tenow and Bylund, 2000]. This paper presents data from the most recent Epirrita outbreak which severely defoliated the mountain birch forest in subarctic Sweden within the Lake Torneträsk catchment (68.0–68.5° N, 18.1–20.2° E). This outbreak site was only one of many areas affected in the northern region of Scandinavia in 2004, where as much as 3300 km2 was detected [Jepsen et al., 2009]. The population cycle peaked in 2004 after a moderate defoliation in the area in 2003, about 50 years after the last serious outbreak in 1954–1955 [Van Bogaert et al., 2009; Wielgolaski et al., 2005]. In 2004, the average larvae density was estimated to 800 ± 436 (±standard error (SE)) larvae m−2 ground area (see auxiliary material).
 Subarctic Sweden has experienced warming over the last decades [Callaghan et al., 2010]. For example, palsa mires in the area show a clear fingerprint of climate change since the 1970s with vegetation changes and permafrost disappearance [Johansson et al., 2006]. A warmer climate and especially a lower frequency of extremely low winter temperatures [Callaghan et al., 2010] enhance the survival of the overwintering eggs of the autumnal moth [Tenow, 1972]. However, the extent to which climate warming will influence the population dynamics of insects, and their ecological [Jepsen et al., 2008] and biogeochemical impacts, is largely unresolved.
 To quantify the impact of the defoliation on C balance, we used eddy covariance methodology (EC) [Aubinet et al., 1999], a non- destructive measuring technique. This method measures net ecosystem exchange (NEE) of, e.g., CO2, i.e., the integrated flux of gross primary production (GPP) and the combined respiration of autotrophs (plants) and heterotrophs (soil bacteria, fungi), here denoted ecosystem respiration (Re,night and Rd). NEE over the 2004 (outbreak year) and 2006 (reference year) growing seasons was used in the analysis. The data was divided into daytime and nighttime (see auxiliary material) and grouped in two week periods. The daytime NEE (μmol m−2 s−1) was used together with photosynthetic photon flux density (PPFD μmol photons m−2 s−1) to estimate net photosynthesis at light saturation level. Ecosystem respiration and quantum yield (α), which is the initial slope of the photosynthesis light response curve, were calculated by using the Misterlich light response model [Falge et al., 2001] (see auxiliary material). Nighttime fluxes, assumed to present ecosystem respiration, were fitted to an Arrhenius respiration model [Lloyd and Taylor, 1994] driven by air temperature. For further information see auxiliary material.
3. Results and Discussion
 The EC system was installed and data retrieval commenced in June–July (Julian days (JD) 184 and 171, 2004 and 2006 respectively). For the reference year (2006) the forest's growing season (here considered to start at birch bud burst) commenced more than 2 weeks earlier than the measurements started. The bud burst days had basically identical timing in the two years of interest; June 1 (JD 153) and May 30 (JD 150) 2004 (leap year) and 2006 respectively (see auxiliary material), and both years fall within the natural variability (JD 156 ± 8 (±standard deviation (SD)) [Karlsson et al., 2003].
 The larvae reached outbreak densities around June 15, peaked around June 22 and crashed a week later when the vegetation was totally defoliated. As the larvae outbreak was synchronized with the bud burst, most of the leaves were eaten at or just after budburst (Figure 1).
 The forest type in the area is of heath character [Sonesson and Lundberg, 1974] with a very high abundance of Empetrum hermaphroditum and Vaccinium vitis- idaea in the ground cover, estimated as 68 and 14 percent of the ground cover respectively (green dry weight). The ground cover in this type of forest has an average LAI (leaf area index) of one (m2 m−2) [Dahlberg et al., 2004] and the total maximal forest LAI including the ground vegetation estimated in 2006 is ca 2.6 (see auxiliary material). Two growth forms of mountain birch trees are found, monocormic (single stemmed) and polycormic (multi stemmed). Polycormic trees, the predominant form at the investigated site, are better adapted to the severe stress outbreaks impose on the trees due to its regeneration by sprouting whereas the monocormic form primarily rejuvenate by seeds. In 2004, after the full defoliation, leaves were yet again produced from late July–early August from long shoots (a characteristic trait for heterophyllous trees). The ratio total leaf weight/leaf area of the late leaves was estimated to be ca. 20% compared to measurements in 2005 (see auxiliary material). Many leaves produced after the defoliation were much larger than normal (data not presented, similar effect was seen by Hoogesteger and Karlsson ). The ground vegetation is affected by an outbreak in contrasting and complex ways. Specific shrubs are affected negatively by herbivory, e.g., Vaccinium myrtillus leaves are consumed whereas Vaccinium vitis- idaea are not, and Empetrum hermaphroditum is almost untouched by the larvae, most likely due to phenolic compounds found in the leaves as a defense system. Nevertheless the Empetrum leaves wither shortly after the herbivory event (Figure 1) probably due to water stress following larval damage to leaf cuticles.
 The two years (2004 and 2006) had highly different seasonal development. In 2004 the forest stayed a source of C through the larvae attack, until the secondary leaf burst in early August (around JD 208) whereas in 2006 it had acted as a consistent net C sink for a month prior to this date (Figure 3, before JD 185 data not presented). The C balance for the growing season until middle of October for both years (105 days) is −12 g C m−2 and −105 g C m−2 for 2004 and 2006 respectively (Figure 2), where close to zero value for growing season in 2004 denotes a not quantified but certain significant net source of carbon to the atmosphere in a given annual context. The burst of photosynthetic activity after the late flush can be seen in the flux data and associated modeled parameters (Figures 3 and 4). The maximum saturation flux (Fcsat) during the time of measuring in 2006 concurs with peak PPFD in mid July, whereas the highest Fcsat in 2004 is delayed until second half of August, due to the larvae attack. In the late season, the two signals follow each other closely (Figure 4a). The light use efficiency (α) is higher during the second flush of leaves in 2004 (Figure 4b) which partly compensate for a much lower leaf area compared with 2006. One possible explanation for high Fcsat in 2004 is that we observed a signal of compensatory fluxes and higher capacity of new leaves owing to higher concentration of nutrients in the leaves [Hoogesteger and Karlsson, 1992] due to lower ratio of leaf area/fine roots. This effect of elevated photosynthetic rates of mountain birch in years following an insect outbreak was also observed by Prudhomme . The respiration terms derived from the response curves of the light response (daytime) and ecosystem respiration (nighttime) models show that the seasonal patterns and respiration rates differ from each other and between years. The nighttime ecosystem respiration (Re,night) shows the same general seasonal pattern as the average nighttime air temperature but with a lower sensitivity to temperature in 2004 compared with 2006 (see auxiliary material). The daytime respiration rates (Rd) in 2004 are higher during the second flush of leaves than at the same time in 2006 and the respiration peaks during this time, with a maximal rate of 1.9 ± 0.7 (95% confidence limit) μmol m−2 s−1. The peak respiration rates in 2006, with a maximal corresponding Rd of 2.2 ± 0.6 μmol m−2 s−1, occur during the peak of the season in the middle of July. This pattern for day time respiration is mainly a combination of three drivers: temperature, soil moisture and photosynthetic rate. Temperature is considered the main driver of respiration [Lloyd and Taylor, 1994] although differences in precipitation between years (which should relate to different soil moisture) can also partly explain inter-annual variation in respiration [Reth et al., 2005].
 A first estimate was made of the regional C setback by scaling the flux to cover the whole defoliated area (see auxiliary material). In 2004, forest areas over 316 km2 acted as a net sink of 4 × 103 tonnes C during the time of measurements due to the defoliation which can be compared with a net sink function of 33 × 103 tonnes C in 2006.
 The forest was unfortunately not monitored for CO2 fluxes during the year following the extreme outbreak. However, at visual inspection during spring leaf flush 2005, no dead trees or stems were found at the site, but discoloration (red tone, a possible nitrogen deficiency effect) of the leaves was observed on single trees. In contrast, an 80–90% mortality rate of the leaf carrying shoots of birches the year after a severe defoliation has been reported earlier [Tenow and Bylund, 2000]. In our study the leaf weight measured after senescence and fall in 2005 implies a lower leaf area at about 66% of that in the reference year (2006) with a corresponding even greater reduction in 2004.
 It has been suggested [Tenow et al., 2004] that Epirrita outbreak dynamics is a prerequisite for rejuvenating the forest ecosystem. The C exchange found for the undisturbed year (2006) may in fact be a result of earlier insect outbreaks. Areas may still at this date be recovering and producing new growth following earlier outbreaks. In the light of this the representativeness of any single year as “reference” may be questioned. Preliminary data, however, from the same time period in other years suggests 2006 as well within an average span of undisturbed years. While 2006 had a growing season uptake of −105 g C m−2, 2007 had −126 g C m−2 and 2009 −93 g C m−2 (2008 had instrumental problems) (M. Heliasz et al., manuscript in preparation, 2010). Consequently, it appears that the 2006 flux data presented here are much closer to an average accumulated seasonal flux than what the corresponding −12 g C m−2 in 2004 is.
 In this study we have shown and quantified the important impact that ecosystem disturbance, in this case in the form of an insect outbreak, may have on the ability of natural ecosystems to be acting as sinks for atmospheric CO2. The globally important terrestrial ecosystem atmospheric sink functioning currently estimated as amounting to 3 Pg C yr−1 [Le Quere et al., 2009], or roughly 30% of the current total anthropogenic emissions, is fragile. This study shows how a possible changed frequency of disturbance events may have dramatic impacts on the capability of natural ecosystems to maintain their sink functioning. A changed frequency (for whatever reason) of disturbance events may well have a greater impact on natural ecosystem carbon exchange than that of climate change per se.
 This work has been supported by the Swedish Research Councils VR and FORMAS, the Crafoord Foundation and the EU MULTIARC project. The authors are grateful to the staff at the Abisko Scientific Research Station for practical and logistical help.