Main

Boreal forests cover 11% of the Earth’s land surface1 and store around one-third of the global terrestrial carbon (C) stock2. As such, they are recognized as an important element in climate change mitigation policies3. Yet, the circumpolar boreal region is subject to the fastest rate of climatic change4,5. This includes more frequent and severe extreme climatic events such as droughts6,7, which are expected to diminish the C sink strength of this biome during the twenty-first century8. Advancing our understanding of how the forest C cycle responds to droughts is hence crucial for improving the predictions of boreal forest C cycle–climate feedback.

Drought stress in forest ecosystems commonly arises from prolonged periods of reduced precipitation combined with warm air temperature, causing high evaporative demand coupled with decreasing soil water availability9. The impact of drought on the forest net ecosystem production (NEP; that is, the ecosystem C balance) therefore depends on the distinct responses of its two primary underlying components, that is, net primary production (NPP) and heterotrophic respiration (RH), to both warmer and drier conditions. In situ experiments10, remote sensing11,12 and ecological modelling13 studies suggest that droughts reduce forest NPP primarily through suppression of photosynthetic activity by either ecophysiological or structural changes in plants, plant hydraulic damage, and/or forest dieback through concurrent abiotic and biotic stress14,15. At the same time, the increase in temperature occurring during droughts may stimulate RH rates16. However, beyond a certain threshold, RH might also become constrained if drought conditions lead to soil water stress on microbial communities17. Despite this understanding from previous works, considerable uncertainty remains about the relative importance of these processes in mediating the drought response of the forest NEP across the boreal landscape.

The variation in the drought-induced reduction in forest NPP is often attributed to differences in tree age and/or size18,19,20,21,22,23. Nonetheless, while most of this knowledge has been developed in temperate and tropical regions, empirical evidence supporting this dependency remains limited for boreal forests. Furthermore, most previous studies ascribe the impact of drought on NPP to the response of the tree layer, while less attention has been paid to that of the understorey vegetation. However, relative to deep-rooted trees, the understorey vegetation growing on the forest floor has been reported to be particularly vulnerable to drought stress because of shallower rooting depths and smaller C reserves24,25. Given that the understorey may contribute up to 80% to NPP in the boreal forest26, there is a critical need to account for the role of this component in both empirical and modelling assessments of drought impacts on the boreal forest C sink.

Most recently, several studies using eddy covariance measurements have provided valuable evidence regarding the impact of drought on the boreal forest NEP9,27,28,29,30. Nevertheless, eddy covariance studies provide limited insights into how the response of NEP to drought is regulated by the sensitivity of its underlying production and respiration component fluxes. In addition, eddy covariance studies commonly suffer from a lack of spatial replication, therefore failing to capture the large heterogeneity across the boreal landscape. In comparison, networks of extensively distributed and repeatedly measured monitoring plots that use biometric- and chamber-based measurements may provide valuable information on the variability in the drought response of the separate forest NEP component fluxes across the heterogeneous landscape.

During the summer of 2018, a record-breaking, hot drought occurred over extensive regions of Central and Northern Europe31. This study explored its consequences for the boreal forest C balance across a 68 km2 managed boreal landscape located in northern Sweden (Extended Data Fig. 1). Compared with the nearly normal environmental conditions observed in this landscape during the baseline period 2016–2017, the year 2018 was characterized by an exceptionally warm and sunny, but normally wet spring. It was followed by an unusual summer marked by moderate drought conditions in May and June, which reached severe levels in July (Fig. 1). The progression of the summer drought was primarily caused by extremely high positive anomalies in air temperature and radiation rather than by precipitation deficits. The environmental conditions at the forest-floor interface further confirmed that temperature rather than moisture was the main driver of the summer drought development (Supplementary Fig. 1). To investigate the landscape-scale variation in the response of forest NEP and its component fluxes to this exceptional 2018 drought relative to the baseline period 2016–2017, we used an extensive dataset of biometric- and chamber-based annual CO2 flux measurements obtained in 50 forest stands (spanning 5 to 211 years old) located across the landscape. Specifically, we analysed the distinct drought responses of forest-floor understorey and tree NPP as well as RH to understand how these in concert regulate the magnitude of the drought impact on the forest ecosystem C balance.

Fig. 1: Environmental conditions during drought and reference periods.
figure 1

a–h, Air temperature (Ta, °C; a,b), global radiation (Rg, MJ m−2; c,d), precipitation (P, mm; e,f) and standardized precipitation evapotranspiration index at a three-month time scale (SPEI, dimensionless; g,h) observed during the long-term reference period 1991–2020 (LT91–20), baseline period 2016–2017 (BL16–17) and drought year 2018 (D18). The letters on the x axes represent the first letter of each month. Drought magnitude in g was classified into five types including extreme (SPEI ≤ −2.0), severe (−2.0 < SPEI ≤ −1.5), moderate (−1.5 < SPEI ≤ −1.0), mild (−1.0 < SPEI ≤ −0.5) and non-drought (SPEI > −0.5). Values in a, c, e and g show annual means (±95% confidence intervals), while the solid lines show the mean monthly values. Panels b, d, f and h show the annual and monthly standardized anomalies (ɀX), with values showing the annual means and bars depicting the monthly values. Extreme environmental and drought conditions are defined as values of ɀX above and below one standard deviation unit, respectively (depicted as dashed lines).

Full size image

Non-uniform drought response of NEP across the forest landscape

During the drought year 2018, NEP decreased on average by 80 ± 16 g C m−2 yr−1 (mean ± 95%CI; Fig. 2a,b), which corresponds to a reduction of 57 ± 13% (Extended Data Fig. 2a) across the studied landscape. Nevertheless, our findings reveal a non-uniform response across stand ages. Specifically, a twofold greater NEP reduction was observed in 20–50 year old stands, averaging at 95 ± 39 g C m−2 yr−1, compared with a reduction of 54 ± 57 g C m−2 yr−1 in stands older than 130 years (Fig. 2b). It is further noteworthy that the drought induced a transition from C sink to source in only a small number of forest stands, mainly in young and low productive mature ones (Fig. 2a). Our results also suggest that the response of NPP rather than that of RH modulated this stand-age-related divergence in the drought impact on NEP. In particular, NPP exhibited an average decrease of 51 ± 16 g C m−2 yr−1 (Fig. 2c,d) or 13 ± 4% (Extended Data Fig. 2b), following an age-related response pattern similar to that of NEP. In comparison, RH increased on average by 29 ± 4 g C m−2 yr−1 (Fig. 2e,f) or 14 ± 2% (Extended Data Fig. 2c), but without a clear age-dependent pattern. We further noted that neither soil type nor tree species significantly affected the drought response of NEP, NPP and RH across the landscape (Extended Data Fig. 3).

Fig. 2: Drought response of forest NEP and its component fluxes.
figure 2

a,c,e, Annual values of NEP (a), NPP (c) and total RH (e) for the baseline period 2016–2017 (BL16–17) and the drought year 2018 (D18). b,d,f, Absolute anomalies (ΔX) of NEP (b), NPP (d) and total RH (f) (∆X). Circular open symbols indicate the values for each forest stand, while circular filled symbols indicate the means for each of the stand age classes, including initiation (I), young (Y), middle-aged (Ma), mature (M) and old-growth (O). The horizontal and vertical bars represent the 95% confidence intervals, while the solid lines show the best-fit regressions. The equation form and coefficient of determination (R2) of the linear regressions are also presented. Equations and associated goodness-of-fit statistics are derived from the individual forest stand data (n = 49). Box plots of the annual values along with the absolute anomalies are also shown. The boxes represent the 25th (bottom) and 75th (top) percentiles, the central line the median, and the cross the mean. Whiskers above and below the boxes denote the data within 1.5 times the interquartile range, with outliers presented as individual points. The values show the means (±95% confidence intervals) of NEP, NPP and RH, as well as their absolute anomalies. The P values for the non-parametric Kruskal–Wallis rank sum test comparing the differences between the means of stand age classes and BL16–17–D18 periods are also shown. The horizontal dashed lines indicate the negative-to-positive transition for the absolute anomalies.

Full size image

The partitioning of NPP into its components (Fig. 3) showed a greater decrease in understorey production (NPPu) compared with tree production (NPPt), with an average reduction of 33 ± 8 and 17 ± 2 g C m−2 yr−1, respectively (or 26 ± 5% and 7 ± 5%, respectively; Extended Data Fig. 4). Our results further reveal that the drought response of NEP corresponded most closely to the NPPu anomaly in the younger forest stands up to the age of ⁓70 years, beyond which the NPPt anomaly largely explained the NEP reduction (Fig. 4 and Supplementary Table 1). The shift in the relative importance of NPPu to NPPt in regulating the response of NEP to drought coincided with a decrease in their ratio with stand age (Fig. 5). Specifically, NPPu was the dominant contributor to NPP in young forest stands (≤25 years), whereas NPPt became the more important component of NPP in older stands. Consequently, the age-related reduction in NEP (Fig. 2a,b) was the result of the greater drought sensitivity of understorey and its decreasing contribution to NPP relative to trees as stands aged. We also found that drought reduced aboveground production rather than belowground production in trees (Extended Data Fig. 5), whereas for understorey vegetation, both above- and belowground production were vulnerable to drought (Extended Data Fig. 6). In addition, no effect of soil type and tree species was observed on the drought response of NPPt and NPPu (Extended Data Fig. 7).

Fig. 3: Drought response of tree and understorey NPP.
figure 3

a,c, Annual values of NPP of trees (NPPt; a) and understorey (NPPu; c) for the baseline period 2016–2017 (BL16–17) and the drought year 2018 (D18). b,d, Absolute anomalies (ΔX) of trees (∆NPPt; b) and understorey (∆NPPu; d) (∆X). Circular open symbols indicate the values for each forest stand, while circular filled symbols indicate the means for each of the stand age classes, including initiation (I), young (Y), middle-aged (Ma), mature (M) and old-growth (O). The horizontal and vertical bars represent the 95% confidence intervals, while the solid lines show the best-fit regressions. The equation form and coefficient of determination (R2) of the linear regressions are also presented. Equations and associated goodness-of-fit statistics are derived from the individual forest stand data (n = 49). Box plots of the annual values along with the absolute anomalies are also shown. The boxes represent the 25th (bottom) and 75th (top) percentiles, the central line the median, and the cross the mean. Whiskers above and below the boxes denote the data within 1.5 times the interquartile range, with outliers presented as individual points. The values show the means (±95% confidence intervals) of NPPt and NPPu, as well as their absolute anomalies. The P values for the non-parametric Kruskal–Wallis rank sum test comparing the differences between the means of stand age classes and BL16–17–D18 periods are also shown. The horizontal dashed lines indicate the negative-to-positive transition for the absolute anomalies.

Full size image
Fig. 4: Age-related shift in understorey versus tree production regulates the drought response of forest NEP.
figure 4

a, Moving-window correlations between the absolute anomaly of NEP (∆NEP) and the absolute anomalies of NPP of trees and understorey (∆NPPt and ∆NPPu, respectively) over stand age. Analyses were conducted using a window of seven forest stands with a step of one forest stand. For each relationship, circular open symbols indicate the Pearson’s correlation coefficient (r) values for each stand age, while solid lines show the best-fit regressions (see Supplementary Table 1 for further details). b, Differences between the correlation coefficients of each relationship for each stand age (circular open symbols) and the fitted regression (solid line). On the x axis, stand age corresponds to the mean age of the seven-forest-stand window. The vertical dashed line indicates the stand age at which ∆NPPt becomes dominant over ∆NPPu in the response of ∆NEP, which occurs at ⁓61 years. The horizontal dashed line shows the negative-to-positive transition in the correlation coefficient difference.

Full size image
Fig. 5: Age-related shift in the ratio of understorey to tree production.
figure 5

The data show the ratio of understorey to tree NPP (NPPu:NPPt) estimated for the period 2016–2018. Circular open symbols indicate the values for each forest stand, while circular filled symbols indicate the means for each of the stand age classes, including initiation (I), young (Y), middle-aged (Ma), mature (M) and old-growth (O). The horizontal and vertical bars represent the 95% confidence intervals, while the solid line shows the best-fit regression. The equation form and coefficient of determination (R2) of the linear regression are also presented. Equation and associated goodness-of-fit statistics are derived from the individual forest stand data (n = 49). The P value for the non-parametric Kruskal–Wallis rank sum test comparing the differences between the means of stand age classes is also shown. The horizontal dashed line indicates the equal contribution of NPPu and NPPt to NPP.

Full size image

We further analysed a greenness index (that is, green chromatic coordinate, gcc) to describe the seasonal phenology of understorey (gccu) and trees (gcct) in a mature mixed-species forest stand located within the studied landscape (Methods and Extended Data Fig. 1). During the 2018 growing season, gccu decreased by 36%, which was threefold greater than the decrease in gcct (12%; Fig. 6). In addition, the timing and length of the peak growing season were affected differently for trees and understorey during the drought year, with larger changes observed in the latter component.

Fig. 6: Divergent drought impact on understorey and tree phenology.
figure 6

a,b, Normalized green chromatic coordinate (gcc) for understorey (gccu; a) and trees (gcct; b) during the baseline period 2016–2017 (BL16–17) and the drought year 2018 (D18) over the day of the year (DOY). Circular open symbols and lines correspond to the 3-day means of normalized gcc and locally estimated scatterplot smoothing (loess) fits, respectively. Mean values of the absolute and relative anomalies of gccu (∆gccu and δgccu, respectively) and gcct (∆gcct and δgcct, respectively) during the growing season are also shown. Furthermore, the transition dates between the phenological stages of green-up (G), maturity (M), senescence (S) and dormancy (D) are indicated by vertical lines. The length in days of both the total and peak growing season (GSL and GSpeakL, respectively) are also reported.

Full size image

Understorey–tree composition controls drought impact on NEP

Our empirical evidence for the shift in understorey–tree composition with increasing stand age as a key control adds an improved conceptual dimension for evaluating the landscape-scale sensitivity of the forest C balance to drought. This result suggests that the common exclusive focus on the tree layer may lead to considerable bias in estimating the drought response of the boreal forest C sink. Consequently, there is a need to alleviate the widespread deficiency in understorey sampling32 in future biometric-based studies, particularly in young boreal forest stands where understorey exerts a key control on regulating the drought-induced NEP change. Furthermore, the different responses of forest-floor understorey and tree NPP to drought, and how they in concert regulate the drought sensitivity of NEP, call for further efforts to partition ecosystem responses in order to improve model-based predictions of C cycle–climate feedbacks in boreal forests.

The greater decline in understorey production compared with that of trees might reflect the lower competitiveness of shallow-rooted understorey versus deep-rooted trees during periods with soil water scarcity24,25. Specifically, the greater drought sensitivity of NPPu in younger stands (⁓20–35 years) could be due to an increase in: (1) competition for resources due to high understorey plant density; (2) heat stress as a result of increased exposure to incoming solar radiation and radiative soil warming; and (3) soil water depletion attributable to increased evapotranspiration. The subsequent decrease in the negative impact of drought on NPPu in older stands may be related to the age-related decline in understorey biomass and production26,33. Moreover, the capacity of overstorey to act as thermal forest-floor insulator (Extended Data Fig. 8) through shading and transpirational cooling may mitigate drought stress for the understorey34. This effect may, however, be compensated by an increase in rainfall interception and tree transpiration, which consequently may lead to a reduction in soil water storage with the increase in tree biomass. In addition, the age-related changes in the drought response of NPP (and subsequently NEP) might also result from contrasting drought tolerances within the tree layer during stand development. In particular, tree age and/or size have been previously noted as key factors in controlling the drought response of tree growth18,19,20,21,22,23. Possible mechanisms causing such age- and/or size-related variations may include changes in leaf-to-sapwood area ratio35 and/or greater access to soil water provided by deeper, more extensive tree rooting systems36, which tend to counteract the effect of the increasing hydraulic challenge, water requirements and evaporative demand experienced by trees during stand development21,37. At present, however, it remains inconclusive whether forest stands consisting of young (small) or old (large) trees are more resistant to drought, particularly in the boreal region. Although we observed that the production of trees in young stands was somewhat less resistant to drought, overall our findings do not strongly support the hypothesis of a significant age-dependent drought sensitivity of NPPt. Nonetheless, a more detailed understanding of how trees at various growth stages respond to drought is warranted, as it may have important implications for regeneration failure and/or future forest composition and structure38.

During the 2018 drought, both understorey and tree production were reduced, but in general they allocated relatively more biomass to roots than to aboveground components (Extended Data Fig. 9), likely to improve access to soil water. While this strategy for coping with drought stress is well known39, our results suggest a divergent age-related pattern. Specifically, trees and understorey prioritized belowground C allocation in the younger and older stands, respectively, which may be related to a competitive gradient with the respective dominant layer during stand development. This finding illustrates that the role of competition for resources in the root system becomes more critical under drought stress40. This strategy could be adopted by plants to maximize survival and might be more prominent (that is, higher C root investment) in the understorey than in the trees. Our work therefore suggests that both understorey and overstorey components in boreal forests experience greater susceptibility to drought events when facing high inter-layer competitive stress.

Our phenological observations provide further evidence that the impact of drought on the seasonal development of understorey biomass was greater than that on trees. In particular, the trees exhibited a steady decrease in greenness throughout the 2018 growing season and maintained a phenological timing consistent with normal conditions. In contrast, the short peak growing season of the understorey and its anticipated senescing stage, triggered by the summer drought, was partly due to the legacy effects of the preceding warm and sunny spring. The latter promoted an early and rapid understorey development, which was not evident in the case of trees. This fact may have contributed to earlier topsoil water depletion through enhanced water consumption and forest-floor evapotranspiration, which ultimately may have induced a greater reduction in C assimilation in understorey than in trees later in the summer. Similar to our findings, previous studies have reported the effect of earlier spring greening on vegetation growth during a drought year41,42. Overall, this suggests that the peak growing season is the most sensitive phenological stage to drought, with carry-over effects from spring vegetation phenology, particularly for the understorey, strongly influencing the subsequent repercussions of summer droughts on the C sink of boreal forests.

Overall, NPP was more sensitive to drought stress than RH and acted as the primary factor contributing to the change in NEP. These observations align with previous reports across multiple biomes13,43,44, suggesting that climate-driven alterations in production, rather than respiration, may emerge as key drivers in regulating the C sink strength across the boreal forest landscape. In contrast, a recent study based on a network of eddy covariance observations in 11 Nordic forests9 identified RH as the driver of the forest NEP anomaly across Scandinavia and the Baltic region in 2018. This divergence may indicate that RH could be more important in regulating the drought response of NEP at larger latitudinal scales compared with our studied boreal landscape. However, in contrast to our empirical assessment of NPP and RH dynamics, it should be noted that RH was not directly measured in ref. 9, but rather derived from the partitioning of ecosystem respiration, assuming autotrophic respiration as a fixed fraction of gross primary production across all sites and environmental conditions. Nevertheless, although both approaches suffer from method-specific shortcomings, these contrasting findings emphasize the need for a better understanding of the individual component fluxes that regulate the impact of drought on the forest C balance.

Managing the boreal forest NEP in a future drier climate

Our results showed that the NEP was considerably reduced by drought (on average by 80 ± 16 g C m−2 yr−1 or 57 ± 13%) across the managed landscape. It is worth noting that the NEP reduction in our mature forest stands (58 ± 31 g C m−2 yr−1) was closely related to independent eddy covariance observations conducted in forest stands of similar age class, both within and in close vicinity of the studied landscape (Methods and Supplementary Table 2; ⁓41–59 g C m−2 yr−1). Furthermore, this decline in the forest C sink during the 2018 drought is consistent with that estimated (87 ± 87 g C m−2 yr−1) from observations reported in the recent eddy-covariance-based study9. Considering the broad range (10th–90th percentile) of 8–154 g C m−2 yr−1 or 3–111% observed in the drought-induced NEP reduction across the studied forest stands, our results suggest, similarly to previous eddy-covariance-based reports9,13, that the weakening of the forest C sink is highly variable across the heterogeneous landscape. This spatial variation needs to be taken into account when extrapolating the future drought response of the forest C balance across the boreal biome.

Our study further explored the importance of key spatial attributes in determining the impact of drought on forest NEP across the managed boreal landscape. It was somewhat surprising to find that soil type and dominant tree species did not have a significant impact on the drought response of NEP, despite previous studies demonstrating interaction effects between drought, soil and tree species on tree growth45,46,47. This result may be related to the counterbalancing drought responses of the NEP components (Extended Data Figs. 3 and 7) and/or the similar development of drought impact (Supplementary Fig. 2) between sediment–till soils and pine–spruce stands. It is also worth noting that our studied forest stands were not pure monocultures26. Therefore, it is possible that the mixture of tree species may have levelled out to a similar stand-level drought response45,48. Our results suggest that variations in forest age structure, including understorey-tree composition, may largely outweigh the effects of soil conditions and tree species distribution in controlling future drought impacts on the C sink strength of boreal forests.

The age-dependent drought sensitivity of NEP implies that the response of the forest C balance to more frequent and severe droughts projected for northern latitudes during the twenty-first century6,7 may be largely non-uniform across the boreal forest landscape. At present, approximately two-thirds of the boreal forest biome is actively managed4, resulting in a complex mosaic of forest stands at varying developmental stages. Consequently, large gradients in the drought response of NEP can be expected across managed landscapes, a fact that may be particularly important in Fennoscandia, where up to 90% of the forest area is managed4. Furthermore, these managed landscapes are dominated by a large proportion of young forest stands (Extended Data Fig. 10) and hence, according to our results, may experience a greater drought-induced NEP reduction. Thus, a shift towards young forest stands at the present8 or at even higher rates as a result of more frequent stand-replacing disturbances49,50 and/or shorter rotation periods, might critically weaken the C sink strength of the boreal landscape under future droughts. Therefore, forest management strategies that promote the transition to late-successional stands (for example, increasing the share of continuous cover forestry and/or extending the rotation period) may have the potential to increase the drought resistance of the C sink of the boreal forest landscape.

Methods

Study site characteristics

This study was carried out in the 68 km2 Krycklan Catchment Study (KCS), which is located in the boreal region of northern Sweden (64° 14′ N, 19° 46′ E), ca. 60 km west of the Baltic Sea coast (Extended Data Fig. 1; see ref. 51 for further details). The climate is cold temperate humid, with permanent snow cover persisting for ca. 167 days per year from mid-November to late April. The 30-year (1991−2020) total annual precipitation averages 638 ± 40 mm (±95% confidence interval), of which approximately 30% falls as snow. The mean annual, January and July temperatures are 2.4 ± 0.3, −8.3 ± 1.0 and 15.0 ± 0.6 °C, respectively. The terrain is gently undulating and spans elevations from 339 m above sea level (a.s.l.) in the northwest to 138 m a.s.l. at the outlet in the southeastern part of the KCS. Till soils (51% of total catchment area) dominate the higher elevations, whereas the lower elevations are characterized by sediment soils (that is, silt, sand and glaciofluvial deposits; 30%). The KCS is predominantly covered by forests (87%) consisting of Scots pine (Pinus sylvestris L., 63%), Norway spruce (Picea abies (L.) H. Karst., 26%) and deciduous trees (Betula spp., Alnus incana (L.) Moench. and Populus tremula L., 11%). The understorey vegetation at the forest floor is composed of ericaceous shrubs, mostly bilberry (Vaccinium myrtillus L.) and lingonberry (Vaccinium vitis-idaea L.), which grow on moss mats of Hylocomium splendens (Hedw.) Br. Eur. and Pleurozium schreberi (Brid.) Mitt51. Mires (9%), arable land (2%), built-up areas (1%) and lakes (1%) encompass the remaining catchment area. A network of small streams, both naturally formed and manmade, connect the 13 partially nested sub-catchments (ranging from an area of 0.12 to 19.13 km2) to the river network51.

Environmental conditions

Environmental measurements were available for the long-term reference period 1991–2020. Specifically, air temperature at 1.7 m above ground (Ta, °C; T107, Campbell Scientific Inc.), global radiation at 1.7 m above ground (Rg, MJ m−2; CM11, Kipp & Zonen) and precipitation (P, mm; ARG 100, Campbell Scientific Inc) were measured in the vicinity of the central part of the KCS at the Svartberget (SVB) reference climate station (64° 14′ N, 19° 46′ E, 225 m a.s.l.; Extended Data Fig. 1a). In addition, the standardized precipitation evapotranspiration index (SPEI) computed at the 3-month time scale was used to characterize the drought conditions over the KCS. This was retrieved from the 0.5° gridded dataset supplied in the Global SPEI database (SPEIbase v2.8, https://spei.csic.es/database.html). The estimation of the potential evapotranspiration in this database is based on the FAO-56 Penman–Monteith method.

Selected forest stands

A subset of 50 forest stands, with ages ranging from 5 to 211 years, was selected by a stratification from the KCS plot network (350 × 350 m square grid) established in autumn 2014 and comprising 556 permanent forest inventory plots (Extended Data Fig. 1a). The age (that is, number of years after stand establishment) of each plot was determined by the basal area-weighted average age, obtained by coring 8–10 dominant trees outside each plot in 2015. Forest stands were categorized into five age classes: initiation (n = 8), young (n = 9), middle-aged (n = 13), mature (n = 14) and old-growth (n = 6). These age classes spanned 5−27, 31−58, 61−78, 80−105 and 131−211 years old, respectively. Moreover, each forest stand was grouped by soil type (sediment and till, n = 15 and 35, respectively) and dominant tree species (pine and spruce, n = 28 and 22, respectively). It should be noted that the forest stands were not pure monocultures, but commonly included a mixture of tree species (see ref. 26 for further details). Within each forest stand, the inventory plot (10 m radius) belonging to the KCS network was used for biometric- and chamber-based CO2 flux measurements performed during the period 2016–2018. One forest stand was subjected to thinning operations in spring 2018 and was thus excluded from the analysis to prevent confounding effects (see Supplementary Fig. 3 for further details).

Biometric- and chamber-based annual CO2 flux estimates

Annual above- and belowground coarse root biomass stocks of living trees (diameter at breast height (DBH, 1.3 m) ≥ 3 cm; AGBt and BGBt−cr, respectively) were calculated for each inventory plot in a subset of 47 forest stands during 2016−2018 using successive forest inventories, tree increment cores (n = 150 trees, 2–4 representative trees per plot) and species-specific allometric equations52,53,54. Tree seedlings and saplings (DBH < 3 cm) were not included in the inventories and their total biomass was considered negligible. Using measurements from each inventory plot, above-mentioned allometric equations and/or volume estimates as well as species- and decay-class-specific wood densities55, we computed the annual above- and belowground coarse root biomass stocks of standing dead trees (snags, AGBdw–s and BGBdw–s, respectively) and the annual total biomass stock of downed dead wood (logs, Bdw–l). In addition, the annual litterfall production (L) was also measured over the 3-year study period using 3 systematically established circular funnel-shaped litter traps (area 0.25 m2) per inventory plot, which were placed 1 m above the ground level. Plant material was collected at the beginning of the snow-free period (that is, mid-May) and at monthly intervals thereafter from August to November, oven-dried (60 °C, 48 h), sorted into fractions (foliage, branches (≤1 cm), cones and miscellaneous) and weighed. Annual AGBt, BGBt−cr and L in 3 recent clear-cut forest stands (5–7 years old) were further estimated based on their relationships with stand age obtained from the remaining forest stands in the initiation age class as described in ref. 26. Finally, the annual aboveground NPP of trees (ANPPt) was computed for each of the 50 forest stands by adding the annual increment in both aboveground tree and dead wood biomass (∆AGBt and ∆AGBdw–s, respectively) and the annual L (equation (1)). Annual belowground NPP of coarse roots of trees (BNPPt–cr) was further computed for each forest stand as the sum of the annual increment in both belowground tree and dead wood biomass (∆BGBt–cr and ∆BGBdw–s, respectively; equation (2)).

$${{\rm{ANPP}}}_{{\rm{t}}}={\Delta {\rm{AGB}}}_{{\rm{t}}}+{\Delta {\rm{AGB}}}_{{\rm{dw}}-{\rm{s}}}+L$$
(1)
$${{\rm{BNPP}}}_{{\rm{t}}-{\rm{cr}}}={\Delta {\rm{BGB}}}_{{\rm{t}}-{\rm{cr}}}+{\Delta {\rm{BGB}}}_{{\rm{dw}}-{\rm{s}}}$$
(2)

Aboveground biomass of understorey vegetation growing on the forest floor, consisting of small-sized plants such as dwarf shrubs, mosses, herbs and/or lichens, was measured for each of the 50 forest stands in June and August 2017 using 3 systematically arranged quadrats (area 0.25 m2) per inventory plot. Understorey species were mainly those described above, but their contribution varied between forest stands. All understorey plants inside each quadrat were clipped, sorted by plant functional types (PFTs; lichens, herbs, mosses and dwarf shrubs), oven-dried (60 °C, 48 h) and weighed. The biomass production of each individual PFT was then added up to calculate the annual aboveground NPP of understorey (ANPPu) in 2017. Thus, herb production was evaluated from peak standing biomass in August, whereas the production of lichens, mosses and shrubs was determined by their respective biomass increments between June and August. Annual ANPPu in 2016 and 2018 was computed from the annual shoot length increments (ASL, mm) of dwarf shrubs. The ASL was measured in June 2019 from the 3 closest ramets of each shrub species present at 4 m intervals along a 20-m-length transect in each inventory plot. It was assumed that the proportion between the different PFTs remained constant over the 3-year study period.

Belowground fine-root production (BNPPfr) was calculated for each of the 50 forest stands using the ingrowth core method. At each inventory plot, 3 plastic mesh cores (diameter 10 cm, length 30 cm) containing a fixed volume of root-free native soil were installed in June 2017. One core per plot was removed at the end of August 2017, while the remaining two were collected at the end of September 2018. On each sampling, roots were washed free of soil over a 2 mm sieve, oven-dried (60 °C, 48 h) and weighed. Initially, we computed the daily fine-root production (FRP, g C m−2 d−1; fine-roots C content was assumed to be 50% of dry biomass) of each plot in 2017. FRP was calculated from the total root biomass at the end of August 2017 divided by number of days of incubation in the field. We then obtained the corrected FRP for each plot over the entire growing season by multiplying the daily FRP by the number of growing days (n = 122) between June and September 2017. Subsequently, the annual BNPPfr in 2018 was derived by subtracting the FRP of September 2017 from that of September 2018. Following this, the ratio of BNPPfr to aboveground NPP (ANPP, ANPPt + ANPPu) calculated in 2018 was used to compute the annual BNPPfr based on ANPP in 2016 and 2017. Annual BNPPfr during 2016−2018 was further partitioned into understorey (BNPPu) and tree (BNPPt–fr) components. As a first step, the annual BNPPu in the 3 recent clear-cut forest stands in 2018 was calculated by multiplying the ratio of the understorey fine-root to the total fine-root biomass C stock (see ref. 33 for further details about its estimation) with annual BNPPfr. From these values, we determined the average ratio of above- to belowground understorey production (that is, ANPPu:BNPPu). Assuming this ratio to be constant, we derived the annual BNPPu in 2018 for the remaining 47 forest stands based on their annual ANPPu. Next, a linear regression between the ratio of understorey to total fine-root production (Fu) and aboveground NPP of vascular plants (ANPPu(vp)) was defined on data obtained in 2018 across the 50 forest stands. Then, Fu in 2016 and 2017 was estimated for each forest stand using the previous linear regression along with their annual ANPPu(vp) values. Annual BNPPu in 2016 and 2017 was thereafter calculated by multiplying Fu with annual BNPPfr. Finally, the annual BNPPt–fr during 2016−2018 was computed as the difference between BNPPfr and BNPPu.

Annual belowground NPP of trees (BNPPt) was computed for each of the 50 forest stands as the sum of BNPPt–cr and BNPPt–fr (equation (3)). Then, the annual NPPt was calculated for each forest stand by summing ANPPt and BNPPt (equation (4)), while the annual NPPu was determined as the sum of ANPPu and BNPPu (equation (5)). Finally, the annual total NPP was calculated for each forest stand as the sum of NPPt and NPPu (equation (6)). Dry biomass of snags and logs was converted to C by using species-specific C contents for each decay class55,56, while the C content was considered 50% of dry biomass for the remaining NPP components. In our approach, we assumed absent or negligible C losses due to herbivory, biogenic volatile organic compounds and root exudation.

$${{\rm{BNPP}}}_{{\rm{t}}}={{\rm{BNPP}}}_{{\rm{t}}-{\rm{cr}}}+{{\rm{BNPP}}}_{{\rm{t}}-{\rm{fr}}}$$
(3)
$${{\rm{NPP}}}_{{\rm{t}}}={{\rm{ANPP}}}_{{\rm{t}}}+{{\rm{BNPP}}}_{{\rm{t}}}$$
(4)
$${{\rm{NPP}}}_{{\rm{u}}}={{\rm{ANPP}}}_{{\rm{u}}}+{{\rm{BNPP}}}_{{\rm{u}}}$$
(5)
$${\rm{NPP}}={{\rm{NPP}}}_{{\rm{t}}}+{{\rm{NPP}}}_{{\rm{u}}}$$
(6)

Heterotrophic soil respiration (RHs) was measured manually every 3–4 weeks during the snow-free period (that is, early May to late October) for each of the 50 forest stands during 2016–2018. CO2 measurements were conducted in an experimental plot (1 m2) located a few metres outside the inventory plot, where all living vegetation was clipped and roots were trenched along the plot border. CO2 fluxes were measured with an opaque custom-made closed steady-state chamber (45 × 45 × 20 cm) connected to a portable infrared gas analyser. Three different infrared gas analysers (MI70, Vaisala; LGR-GGA-24EP, Los Gatos Research Inc.; GasScouterTM G4301, Picarro) were used over the 3-year study period. No significant differences in CO2 fluxes among the 3 analysers were observed in a cross-comparison.

Annual RHs was estimated by defining stand-specific Arrhenius response functions that relate RHs to soil temperature for the snow-free and snow-covered periods (the latter based on measurements performed in October). These functions were then used to extrapolate RHs based on continuous half-hourly soil temperature records to annual sums. It is noteworthy that soil moisture was not included as an additional factor because we observed no significant effect of soil moisture on RHs. Further details about the RHs estimation are shown in ref. 33. Annual heterotrophic dead wood respiration (RHdw) was derived for each forest stand by multiplying the annual above- and belowground dead wood C stocks in 2016, 2017 and 2018 with their respective decomposition rate constants (kdw, yr−1). We used species-specific kdw rates for each component and decay class57,58. Then, RHdw was added to RHs to obtain the annual total RH (equation (7)). It is worth noting that RHs was the dominant component of RH in all forest stands, while the contribution from RHdw was limited to less than 5% on average26. Finally, the annual NEP was determined for each of the 50 forest stands over the 3-year study period as the difference between NPP and RH (equation (8)).

$${\rm{RH}}={{\rm{RH}}}_{{\rm{s}}}+{{\rm{RH}}}_{{\rm{dw}}}$$
(7)
$${\rm{NEP}}={\rm{NPP}}-{\rm{RH}}$$
(8)

In this study, all component CO2 fluxes (g C m−2 yr−1) were presented as positive terms, whereas positive and negative NEP refers to net C sink and source, respectively. In addition, forest-floor environment records were available for each of the 50 forest stands during 2016–2018 as described in detail in ref. 33. These records included below-canopy air temperature (Tabc, °C), soil temperature at 10 cm depth (Ts, °C) and soil volumetric water content at 5 cm depth (SWC, %).

Vegetation phenology

Seasonal changes in understorey and overstorey vegetation greenness phenology were assessed over the 3-year study period in the central part of the KCS using the SVB experimental forest, a ⁓110-year-old mixed-species forest stand (that is, spruce 64%, pine 35%, and birch 1%51). Greenness was derived from hourly images collected through digital repeat photography at the Integrated Carbon Observation System (ICOS) SVB ecosystem station (named SE-Svb in the ICOS network, 64° 15′ N, 19° 46′ E, 270 m a.s.l.; https://www.icos-sweden.se/svartberget; Extended Data Fig. 1a). Below- and above-tree canopy images were taken with web cameras (NetCam SC 1MP, StarDot Technologies), mounted on a vertical pole and the ICOS mast at 3.8 m and 36 m above the ground level, respectively, with a downward-looking angle of 12°. Specifically, a greenness index was quantified using the gcc (dimensionless, equation (9)), which uses the average digital numbers (DN, 0–255) of the red (RDN), green (GDN) and blue (BDN) image channels within a selected region of interest.

$${\rm{gcc}}={G}_{{\rm{DN}}}/\left({R}_{{\rm{DN}}}+{G}_{{\rm{DN}}}+{B}_{{\rm{DN}}}\right)$$
(9)

Three-day mean time series for understorey (gccu) and trees (gcct) were then derived for the 3-year study period. This was achieved by assigning the 90th percentile of all values within a 3-day window to the centre day of a discrete (non-overlapping) moving window as described in ref. 59. The values for gccu and gcct were then normalized (0–1) to describe the seasonal minimum and maximum of vegetation biomass development. We then fitted a locally estimated scatterplot smoothing (loess) curve using the normalized data points in order to improve the visualization. We further used the gcc to derive the four successive phases of vegetation phenology, namely green-up (G), maturity (M), senescence (S) and dormancy (D). Specifically, we determined the transition dates (that is, day of the year) for G and D by extracting the 10% and 25% thresholds of the total amplitude in gccu and gcct, respectively. Similarly, the transition dates for M and S were obtained from the 90% threshold of total amplitude in both gccu and gcct. Subsequently, we calculated the length in days of the total and peak growing season (GSL and GSpeakL, respectively) as the difference between D–G and S–M, respectively.

Eddy covariance measurements

Eddy covariance measurements at two local sites were used to support the drought response of NEP observed by our biometric- and chamber-based approach (see Supplementary Table 2). One site was the SVB station, whose location and surrounding forest have been described above. The other site was Rosinedalsheden (ROS; 64° 10′ N, 19° 45′ E, 145 m a.s.l.), a ⁓100-year-old naturally regenerated homogenous pine stand, located outside the KCS about 10 km south of SVB. Annual NEP values for SVB and ROS sites during the baseline period and the year 2018 were previously reported in ref. 9.

Drought anomalies of annual CO2 fluxes and vegetation phenology

The impact of drought on annual estimates of CO2 fluxes and gcc index was evaluated by calculating both the absolute and relative anomalies (∆ and δ, respectively) of the drought year 2018 (D18) relative to the baseline period 2016–2017 (BL16–17) according to equations (10) and (11):

$$\Delta {X}_{i}={{X}_{i}}_{{{\rm{D}}}_{18}}-{{X}_{i}}_{{{\rm{BL}}}_{16-17}}$$
(10)
$${{\updelta }}{X}_{i}=\frac{{{X}_{i}}_{{{\rm{D}}}_{18}}-{{X}_{i}}_{{{\rm{BL}}}_{16-17}}}{{{X}_{i}}_{{{\rm{BL}}}_{16-17}}}\times 100$$
(11)

where ∆Xi and δXi are the absolute and relative anomalies of each variable (X) expressed in absolute and relative units, respectively, \({{X}_{i}}_{{\text{D}}_{18}}\) and \({{X}_{i}}_{{\text{BL}}_{16-17}}\) are the variable values during D18 and BL16–17, respectively, and subscript i refers to the studied forest stand and/or understorey–tree component layer. Positive and negative anomalies represent the increase and decrease, respectively, of each variable in 2018. Note that the absolute anomalies were used to analyse the drought response of the variables.

Statistics

Linear, polynomial and nonlinear (Levenberg–Marquardt method) regression analyses were used to explore the relationships presented in Figs. 2–5. In all cases, several types of regression were evaluated and the final model was selected based on the highest R2 criterion. Equations and associated goodness-of-fit statistics were derived from the individual forest-stand data. Non-parametric Kruskal–Wallis rank sum tests were also performed in Figs. 2–5 to examine differences between the BL16–17 and D18 periods as well as among the initiation, young, middle-aged, mature and old-growth stand age classes. These previous statistical approaches were also used in Extended Data Figs. 2–9. A moving-window correlation was conducted between the absolute anomaly of NEP (∆NEP) and the absolute anomalies of understorey and tree NPP (∆NPPu and ∆NPPt, respectively) using a seven-forest-stand window with a one-forest-stand step to identify temporal shift of the dominant contributor to ∆NEP (Fig. 4). Furthermore, a locally estimated scatterplot smoothing (loess) fit was applied as an estimate of the seasonal trajectories for the normalized gccu and gcct (Fig. 6). All analyses were conducted using SPSS software (version 29.0; IBM corp.) and MATLAB software (MATLAB R2019b, MathWorks Inc.).