Divergent responses of subtropical evergreen and deciduous forest carbon cycles to the summer 2022 drought

Ongoing shifts in climate, especially extreme drought, is posing considerable threats to the forest carbon uptake worldwide. In China, summer 2022 was the warmest and driest since the beginning of meteorological measurements. This study synthesized the tower-based carbon fluxes and climate data from two subtropical evergreen and deciduous forest ecosystems to investigate the effects of such summer drought. Interestingly, the net ecosystem production (NEP = −NEE) only exhibited a slight decrease at the deciduous forest while it even enhanced at the evergreen forest during the summer 2022 drought. Further analysis revealed that although reductions in gross primary productivity (GPP) and ecosystem respiration (R e) were found at both sites, larger decrease in R e than GPP at the evergreen forest led to stronger NEP compared to the previous year. However, the NEP of two forest ecosystems sharply reduced in the following 2023, which can be ascribed to the legacy effects of the summer 2022 drought. The results of multiple linear regression revealed that soil water content (SWC) was recognized as the primary driver of GPP and R e, and downwelling shortwave radiation (R g) regulated the variability of NEP during the summer 2022. Across these forest carbon fluxes including GPP, R e and NEP, the deciduous forest exhibited larger resistance, whereas the evergreen forest showed stronger resilience. All analyses emphasized the diverse adaptive strategies among vegetation types, which acted an important role in assessing ecosystem carbon sequestration in face of future climate change.


Introduction
With the intensification of global warming in the future, the risk of compound high temperature and drought events may be further increased (Seneviratne et al 2021).Elevated temperatures, shifting precipitation patterns and prolonged droughts disrupt the delicate balance of terrestrial ecosystem carbon cycles (Wolf et al 2013, Lu andYan 2023).Extreme climate events, including droughts and heatwaves, pose severe threats to both society (e.g.affecting food security, human well-being, and energy resources) and the function of terrestrial ecosystems (Tang et al 2017, Wolf and Paul-Limoges 2023, Li et al 2023a, Chen et al 2023b, Wang et al 2023b).However, the impacts of droughts and heatwaves on carbon sink and consequently, the capacity of various terrestrial ecosystems to adapt climate change remained unclear (Oddi et al 2022).
The substantial increase in terrestrial ecosystem carbon sequestration over recent decades is predominantly attributable to forests (Rodda et al 2021, Tong et al 2023).Forests, and especially subtropical forests, known as 'oases on the Tropic of Cancer' (Tan et al 2012), play a crucial role in alleviating global warming (Yao et al 2018).Despite their critical potential in carbon sinks, more recent studies have underscored the vulnerability of forests to extreme drought events (Zheng et al 2018, Forzieri et al 2022).Ongoing global climate change is exerting considerable pressure on forests, potentially causing a transition in their role from carbon sink to carbon source at both local and regional scales (Oddi et al 2022).
Previous studies have significantly advanced our understanding of the impact of extreme drought events on diverse vegetation types over different periods using satellite remote sensing (Ciais et al 2005, Xu et al 2019, Ding et al 2022, Deng et al 2023, Liu et al 2023).The current understanding of drought effects suggests that drought can lower forest carbon uptake and productivity by limiting water loss through transpiration and reducing stomatal conductance (Yuan et al 2019, Dannenberg et al 2022).In addition, evidence indicates that extreme drought not only exerts an immediate impact on carbon sinks, but also triggers lagged responses leading to plant mortality (Reichstein et al 2013, Li et al 2020) and releasing of stored carbon into the atmosphere (Oudin Åström et al 2013, Wolf andPaul-Limoges 2023).By contrast, some studies indicate that not all droughts reduce biome productivity, as vegetation resistance may alleviate negative impacts on certain ecosystem types (Li et al 2023b).Furthermore, increasing atmospheric CO 2 and N deposition are important factors in driving long-term carbon sinks on terrestrial forests (Pan et al 2011, Piao et al 2022, Ruehr et al 2023).Consequently, it is critical to examine the responses of forest ecosystem carbon cycle to climate change.
The carbon sink over subtropical forests in southwest China accounts for about 32% of the carbon uptake across the Chinese mainland (Yao et al 2018, Wang et al 2020, 2023c), which plays a vital role in realizing the national strategy of carbon neutrality.However, this region has been plagued by severe and frequent droughts, particularly for the summer drought of 2022 (Wang et al 2023b).Record-breaking heat along with minimal rainfall resulted in prolonged, large-scale, and high-impact summer heat waves, with severe vegetation dieback and wildfires (Sun et al 2022, Wang et al 2023a, 2023d), and exerted serious impact on the structure and function of forest ecosystems.However, few attempts have been made to the effects of such event on subtropical forest carbon cycles as well as the forest feedback mechanisms.Therefore, this study aims to fill this gap by analyzing consecutive years of flux tower measurements at the two subtropical evergreen and deciduous forests, (1) to examine the seasonal and inter-annual dynamics of carbon fluxes and climatic variables from January 2020 to August 2023; (2) to reveal the effects of such drought event on evergreen and deciduous forest carbon cycles; and (3) to compare the response differences of two subtropical forest types to the 2022 summer drought including resistance and resilience indexes.These analyses provide a better understanding of the interactions between terrestrial ecosystem function and the climate system.The soil types at both sites are primarily calcareous, with small quantities of brownish-yellow loam.

Flux measurements and data processing
The EC-based micro-meteorological measurement equipment was comprised of a 3D sonic anemometer (CSAT3, Campbell Scientific Inc., Logan, UT, USA) to measure three-dimensional wind speed and an infrared gas analyzer (EC155, Campbell Scientific Inc., Logan, UT, USA) to quantify absolute carbondioxide and water-vapor densities, which are used to measure carbon fluxes at 23 m above the ground.Furthermore, an automatic weather station was installed on the tower to continuously measure local hydro-meteorological variables, such as downwelling shortwave radiation (R g , W m −2 ), near-surface air temperature (T a , • C), soil water content at 10 cm, 20 cm, 40 cm depths (SWC, %), relative air humidity (RH, %) and precipitation (Pre, mm).In this study, RH and T a were used to calculate the vapor pressure deficit (VPD, hPa) (Zhao et al 2021, Chen et al 2022, Zhao et al 2023b).
We initially collected raw flux data at 10 Hz and subsequently processed it to a half-hour time scale.A series of standardized procedures were used for flux correction and quality control (Lee et al 2004, Mauder and Foken 2006, Liu et al 2011, Tu et al 2024).Following Papale et al (2006), average fluxes were then adjusted for the periods of low friction velocity using the USTAR threshold method.All of the gaps were filled using the marginal distribution sampling, look-up table, and mean diurnal course algorithms from Reichstein et al (2005).In this study, we utilized a nighttime-based flux-partitioning algorithm to separate NEE into GPP and R e (Lloyd andTaylor 1994, Wutzler et al 2018).All the aforementioned processes including USTAR-filtering, gap filling and flux partitioning were performed using the R package created by the Max Planck Institute for Biogeochemistry, accessible at www.bgc-jena.mpg.de/bgi/index.php/Services/REddyProcWeb (Reichstein et al 2014).Finally, in our study, the carbon use efficiency (CUE) was defined as the ratio of NEP/GPP (NEP = −NEE).

Resistance and resilience indexes
The response of forest to extreme drought event can be assessed by its sensitivity to drought events (Resistance, Rt) and its post-drought recovery rate (Resilience, Rs) (Li et al 2020, Bellanthudawa and Chang 2022).Resistance (Rt) is defined as the ability of a system variable (GPP, R e and NEP) to resist changes in response to a drought perturbation (Thompson et al 2009, Nimmo et al 2015), whereas resilience (Rs) is the capacity of a forest to withstand external pressures and return to its pre-disturbance state (Chang and Wen 2017, Ingrisch and Bahn 2018, Gessler et al 2020).Therefore, Rt and Rs are calculated by the following equation (Isbell et al 2015): (1) where Y n , Y e and Y e+1 represent the carbon fluxes (GPP, R e and NEP) during the normal years, the drought year and the following year, respectively.Both resistance and resilience are dimensionless values, and the values calculated for the same variable can be compared between different ecosystems and between positive and negative disturbances.Larger Rt and Rs values represent stronger resistance and resilience.Lastly, we used the average carbon fluxes during summertime from June to August over three consecutive years to calculate the resistance and resilience at the deciduous forest site (CH-SWP) and evergreen forest site (CH-LBJ), respectively.

Statistical analysis
First, we initially conducted a comprehensive assessment of the correlation between carbon fluxes (GPP, R e and NEP) and environmental variables (T a , R g , VPD and SWC) throughout the summers from 2021 to 2023 using Pearson correlation analysis.We then examined the potential collinearity issues among all variables (all VIF < 10).Subsequently, in order to quantitatively understand individual contributions of environmental variables to carbon fluxes during the summers from 2021 to 2023, we employed a multiple regression analyses based on a stepwise Akaike Information Criterion (AIC) regression with the 'leaps' package in R. The algorithm systematically evaluates all potential models based on a predefined set of variables, selecting the best model according to AIC values.Once the optimal model was identified, we conducted the multivariate regression analysis using standardized variables for the environmental variables.Standardization facilitated a quantitative comparison of regression coefficients.Finally, we calculated the relative significance of each variable using the 'relaimpo' in R (Elizabeth and Arain 2024).

Climatic anomaly in summer 2022
Both sites experienced an unusually hot and dry summer in 2022, marked by notable deviations in all meteorological parameters.Rainfall substantially decreased by 30% in 2022, accompanied by an apparent rise in air temperature (+1.1 • C above average) in comparison to the adjacent years (table 1 and figure 1).July and August were particularly dry and precipitation at both sites was below average, varying from −49% to −54% in July and −81% to −90% in August.
The anomaly in precipitation in summer 2022 was also reflected in the temporal patterns of toplayer soil water content, as illustrated in figures 1(e) and (f).The soil water content at two forest sites exhibited the largest deviations from the long-term mean (CH-SWP: −26%; CH-LBJ: −33%), because of extremely low precipitation in August 2022.Notably, the patterns of soil moisture recovery post the summer drought of 2022 differ between the two forest sites.Following substantial precipitation events at the end of September 2022, the SWC at the evergreen forest site (CH-LBJ) swiftly rebounded, while the SWC at the deciduous forest site (CH-SWP) showed a modest increase.Until the spring of the following year, the SWC at the CH-LBJ site had essentially returned to normal, owing to a replenishment of precipitation, whereas soil moisture conditions at the CH-SWP site appeared consistent with those of the summer of 2022 (table 1).In the summer of 2022, both sites experienced larger amounts of R g compared to 2021, with an increase ranging from +19.9% (CH-SWP) to +33.6% (CH-LBJ).VPD exceeded the longterm average at both sites with a rise of +29.7% for the deciduous forest site (CH-SWP) and +23.6% for the evergreen forest site (CH-LBJ) (see table 1).

Variability in carbon fluxes of two forest ecosystems
The seasonal dynamics of GPP and R e at two subtropical forests from January 2020 to August 2023 were depicted in figure 2, revealing a distinct inverted V-like curve throughout these years.During the winter period, the low temperature hindered the photosynthesis and respiration of vegetation, causing the forests to act as weak carbon sources.However, as T a and R g increased in the springtime, plants began to grow, leading to a gradual rise in both GPP and R e .Consequently, GPP surpassed R e and the forest transitioned from carbon source to carbon sink.The NEP reached its peaks in summertime.Interestingly, in the end of August drought of 2022, the deciduous forest site (CH-SWP) changed into a net source of carbon, which was earlier to the adjacent years.
During the summer drought 2022, large differences in carbon fluxes were observed at the two forest sites (figure 2, table 1).In comparison to the evergreen forest site (CH-LBJ), the deciduous forest site (CH-SWP) exhibited lower NEP (1.89 ± 2.39 g C m −2 d −1 ) with larger GPP (8.63 ± 2.89 g C m −2 d −1 ) and R e (6.74 ± 2.04 g C m −2 d −1 ), respectively.Meanwhile, the sharp declines in GPP and R e was observed at both sites due to drought, with the evergreen forest site (CH-LBJ) experiencing the larger losses (GPP: −21.0%versus 2021, R e : −36.5% versus 2021).Interestingly, compared to 2021, the effect on NEP varied between the two forest sites, showing a slight decline (−1.5%) at the deciduous forest site (CH-SWP), and even an increase (+12.7%) at the evergreen forest site (CH-LBJ) owing to larger reduction in R e relatively to that in GPP.Subsequently, compared to the previous year, it is noteworthy that both forest ecosystems recorded a considerable reduction

Climate factors Carbon fluxes
Site Year  in NEP during the summer 2023 by approximately 61.8% and 35.4% at the CH-SWP site and CH-LBJ site, respectively (table 1).

Effects of climate anomaly on carbon fluxes
Our study revealed significant positive correlations between meteorological factors (T a , R g and VPD) Given that VPD is a reliable indicator of atmospheric dryness, this study investigated the connections between VPD and carbon fluxes, including GPP, R e and NEE at two forest sites.As showed in figure 4, the influence of VPD on carbon fluxes presents a nonlinear relationship in the drought events.Our findings suggest that carbon fluxes including GPP, R e and NEP in forest ecosystems exhibited an increase with the enhancement of VPD.However, when VPD exceeded a certain threshold, both forest ecosystems experienced a gradual decline in carbon fluxes.

Carbon use efficiency
During three consecutive years of observations, the highest CUE occurred in summer 2022 at both forest sites (CH-SWP: 0.59; CH-LBJ: 0.73) (figure 5).In comparison to the summer 2021, a combination of decreased GPP along with increased NEP caused the higher CUE (+49%) at evergreen CH-LBJ site, while the larger reduction in GPP relative to NEP led to a slight increase in CUE (+3.5%) at deciduous CH-SWP site (table 1 and figure 5).Furthermore, there were notable differences from the summer 2022-2023, with both sites experiencing apparent reductions in CUE by approximately 27.1% and 15.1% at the CH-SWP site and CH-LBJ site, respectively.

Responses of ecosystem carbon fluxes to drought
This study systematically explored the resistance (Rt) and resilience (Rs) of two forest ecosystems in    There is growing evidence that responses of ecosystem carbon cycling to extreme drought events often hinge on surpassing eco-physiological thresholds (Forzieri et al 2022).It is well-established that vegetation resistance is constrained under intense drought stress.When stress induced by drought or heat exceeds a critical threshold, the state of ecosystem carbon balance is disturbed, particularly in forests (Sippel et al 2018, Li et al 2023b).In our study, we observed that VPD, as a good proxy for atmospheric dryness, affected physiological and ecological processes of forests in a non-linear manner (figure 4).During the initial stage of summer drought, forest ecosystems maintained its biophysical process by accessing deep water sources through the developed root system (Wang et al 2021).Meanwhile, the relatively high temperature is beneficial for vegetation photosynthesis and respiration.However, this phenomenon is not sustainable, and the potential mechanism relies severely on adequate soil moisture levels and nutrient availability.As VPD surpassed a critical threshold, extreme soil dryness could prompt the reduction of leaf stomatal conductance at the cost of diminished gas exchange, impacting both GPP and R e (Zhang et al 2013, Fu et al 2022).
Contrary to the previous hypothesis of a constant CUE among forests, our study supports the idea that the CUE is a variable, with CUE changing significantly with biotic and abiotic factors such as stand structure, forest age, climate and altitude (Zhu 2013, Khalifa et al 2018).In contrast to the observations by Kwon and Larsen (2013) regarding CUE in Eastern American forests, our study revealed a distinctive pattern in subtropical evergreen broad-leaved forest.Notably, we found that such ecosystem exhibited a significantly higher CUE compared to deciduous forest.

The legacy effects of drought
In the summer of 2022, we observed reductions in GPP and R e at both forest sites associated with soil moisture deficits (table 2).Despite relief from soil drought following sufficient spring precipitation, the phenomenon of forest soil drought still existed and carbon fluxes at both forest sites did not return to the normal levels in the summer of 2023 (see table 1), suggesting possible legacy effects of extreme drought events on the ecosystem carbon cycle.Unlike grasslands, which typically recover to pre-drought disturbance levels within the first year after a drought, the legacy effects of forest drought tend to be more prolonged, persisting for several years (Müller and Bahn 2022, Zeng et al 2023).Due to the increasing risk of tree dieback following drought events (Anderegg et al 2015, Hoover et al 2021, Pohl et al 2023), trees adopted a dynamic C allocation strategy that prioritized biosynthesis of new growth tissue to mediate tree recovery from drought stress, which may enhance the autotrophic respiration and decrease long-term carbon sink (Kannenberg et al 2019b, 2020, Lee et al 2021).Besides, increased ecosystem respiration can indicate an expansion in the substrate pool.The largest amount of litter fall was expected in 2023, which may lead to the growth of total substrate amount as well as higher R e in 2023 at both sites (Palmroth et al 2005).Additionally, precipitation pulses following drought events can enhance soil respiration by affecting the size and/or activity of soil microbial communities (Unger et al 2012).

Drought response across two forest ecosystems
The responses of deciduous and evergreen forests to drought highlighted distinctions in stomatal regulation mechanisms and water use strategies.Stomatal behavior plays a pivotal role in trees' adaptive strategies to environmental stress such as drought (Dungan et al 2003, Qi et al 2021).Under drought conditions, evergreen forests closed stomata to prevent leaf water potential from dropping below the critical level at the expense of reduced photosynthesis.In contrast, deciduous forests exhibited little or no stomatal control, enabling them to maintain photosynthesis at higher rates compared to evergreen forests (van der Molen et al 2011, Ishida et al 2014).Besides, more evidence indicated that hydraulic characteristics and drought strategies can interact to affect a species' ability to recover from water scarcity (Kannenberg et al 2019a).Interestingly, the SWC of deciduous trees at different depths decreased parallel (figure 1(e)), while the SWC of evergreen trees mainly decreased at 10 cm depth (figure 1(f)).We concluded that deciduous forests demonstrated a greater interspecific root uptake depth.The utilization of deep soil water by deciduous trees during drought stress helped mitigate hydraulic damage, supporting our findings of their superior resistance compared to evergreen forests (Lee et al 2021).

Conclusions
On the basis of the EC-based carbon flux observations in two subtropical deciduous and evergreen forests from 2020 to 2023, our study examined the divergent effects of the extreme summer drought in 2022 on forest carbon cycles as well as their response differences.Specifically, GPP and R e exhibited obvious declines at both forest ecosystems, whereas the NEP showed different characteristics, with only a slight reduction for the deciduous forest but a large increase for the evergreen forest.Further analysis revealed that it can be ascribed to R e decreased more than GPP at this forest type.A combination of decreased GPP along with increased NEP also caused the higher CUE in evergreen forest during the summer 2022.Yet it is worth noting that such phenomenon is unsustainable.Because of the legacy effect, the NEP during the summer 2023 decreased sharply by approximately 61.8% and 35.4% at the CH-SWP site and CH-LBJ site, respectively.Generally, the deciduous forest exhibited larger resistance while the evergreen forest showed stronger resilience.Therefore, our work suggests that a deep understanding of the response of different forest types to drought extremes is highly relevant to predict impacts of future climate change on terrestrial ecosystem carbon cycles.

Figure 1 .
Figure 1.Temporal variation in 8-day running means of (a) air temperature (Ta), (b) soil temperature in upper 10 cm layer (Ts), (c) vapor pressure deficit (VPD), (d) downwelling shortwave radiation (Rg), soil water content at 10 cm (black line), 20 cm (red line), 40 cm (blue line) depths (SWC) and total precipitation (Pre) at the CH-SWP (e) and CH-LBJ (f) sites from January 2020 to August 2023, respectively.Light-gray shaded areas represented the months of June, July and August (JJA) in each year.Abbreviations in the headings indicate different forest types, including deciduous broadleaf forest (DBF) and evergreen broadleaf forest (EBF).

Figure 2 .
Figure 2. Temporal variations in 8-day running means of carbon fluxes including NEE, GPP and Re at the two forest ecosystems (CH-SWP: (a); CH-LBJ: (b)).Light-gray shaded areas represented the months of June, July and August in each year as shown in figure 1.

Figure 3 .
Figure 3. Relationships between carbon fluxes and the environmental variables during the summer 2021, 2022 and 2023 at two forest sites.The colors represent positive and negative correlation, with darker colors indicating larger r values and the width of the ellipse indicating the magnitude of correlation level.The asterisks (i.e.' * ') in the figure indicate the magnitude of significance level (p < 0.001).Abbreviations in the headings indicate different forest types, including deciduous broadleaf forest (DBF) and evergreen broadleaf forest (EBF).

Figure 4 .
Figure 4. Relationships between VPD and ecosystem carbon fluxes including GPP (a), Re (b) and NEP (c) at CH-SWP site and CH-LBJ site during summer of three consecutive years from 2021 to 2023 on a daily scale.The dashed lines indicate the corresponding VPD threshold.Abbreviations in the headings indicate different forest types, including deciduous broadleaf forest (DBF) and evergreen broadleaf forest (EBF).

Figure 5 .
Figure 5. Carbon use efficiency (CUE) as the ratio of net ecosystem productivity (NEP) and gross primary productivity (GPP) in summer from 2021 to 2023.Significant differences in CUE (slopes) were detected at two forest sites CH-SWP (a) and CH-LBJ (b), both p < 0.001.The gray shaded bands represent 95% confidence intervals.Abbreviations in the headings indicate different forest types, including deciduous broadleaf forest (DBF) and evergreen broadleaf forest (EBF).

Figure 6 .
Figure 6.Differences in resistance (a) and resilience (b) to the summer drought of 2022 between CH-SWP and CH-LBJ site.Abbreviations in the headings indicate different forest types, including deciduous broadleaf forest (DBF) and evergreen broadleaf forest (EBF).Rt-value indicates the ability of forest ecosystems to withstand an extreme event and Rs-value indicates the ability of forest ecosystems to return to their normal state after an extreme event.
, Sun Y, Wu C Y, Ding Z and Tang X G 2024 Long-term dynamics of peak photosynthesis timing and environmental controls in the Tibetan Plateau monitored by satellite solar-induced chlorophyll fluorescence Int.J. Digit.Earth 17 2300311 Unger S, Máguas C, Pereira J S, David T S and Werner C 2012 Interpreting post-drought rewetting effects on soil and ecosystem carbon dynamics in a Mediterranean oak savannah Agric.For.Meteorol.154 9-18 van der Molen M K et al 2011 Drought and ecosystem carbon cycling Agric.For.Meteorol.151 765-73 van der Woude A M et al 2023 Temperature extremes of 2022 reduced carbon uptake by forests in Europe Nat.Commun.14 6218 Wang J et al 2020 Large Chinese land carbon sink estimated from atmospheric carbon dioxide data Nature 586 720-3 Wang J et al 2023a Unprecedented decline in photosynthesis caused by summer 2022 record-breaking compound drought-heatwave over Yangtze River Basin Sci.Bull.68 2160-3 Wang J, Dong J, Yi Y, Lu G, Oyler J, Smith W K, Zhao M, Liu J and Running S 2017 Decreasing net primary production due to drought and slight decreases in solar radiation in China from 2000 to 2012 J. Geophys.Res.Biogeosci.122 261-78 Wang M, Ding Z, Wu C Y, Song L S, Ma M G, Yu P J, Lu B Q and Tang X G 2021 Divergent responses of ecosystem water-use efficiency to extreme seasonal droughts in Southwest China Sci.Total Environ.760 143427 Wang Y F, Hao Y B, Cui X Y, Zhao H T, Xu C Y, Zhou X Q and Xu Z H 2014 Responses of soil respiration and its components to drought stress J. Soils Sediments 14 99-109 Wang Y N, Sun Y, Chen Y N, Wu C Y, Huang C P, Li C and Tang X G 2023c Non-linear correlations exist between solar-induced chlorophyll fluorescence and canopy photosynthesis in a subtropical evergreen forest in Southwest China Ecol.Indic.157 111311 Wang Y, Wang Y M, Chen Y N, Chen H, Li X T, Ding Z, Han X J and Tang X G 2023b Spatial and temporal characteristics of The study encompasses two forest ecosystem sites with continuous eddy covariance-based flux measurements at the Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station, China.Both flux towers are installed in the subtropical humid monsoon zone.The Shanwangping flux tower (CH-SWP) is located at 29 • 5 ′ 32 ′′ N, 107 • 20 ′ 36 ′′ E, with an altitude of 1350 m.The annual average air temperature measured by meteorological stations is about 10.75 • C, with a mean yearly precipitation of about 803 mm.The vegetation composition in this site is deciduous broad-leaved forest including Daphniphyllum macropodum Miq., Carpinus viminea Lindley and Quercus aliena var.acutiserrata Maximowicz ex Wenzig.The Lanbajing flux tower (CH-LBJ) is located at 29 • 1 ′ 18 ′′ N, 107 • 9 ′ 3 ′′ E, with an altitude of 1450 m.The annual average air temperature is about 8.2 • C. Mean yearly precipitation is about 1434 mm, mainly concentrated in summer.The main dominant tree species in the evergreen broad-leaved forest are comprised of Neolitsea pulchella (Meisn.)Merr., Rhododendron longipes var.chienianum and Eurya japonica Thunb.

Table 1 .
The averaged climate factors including Ta, Rg, VPD, SWC, total precipitation and carbon fluxes including GPP, Re and NEP during summer (June-August) 2021, 2022 and 2023.Relative deviations are shown for summer 2022 compared to 2021 and 2023.

Table 2 .
Relative importance analysis for multivariate regression of standardized GPP, Re and NEP against standardized explanatory variables (Ta, Rg, VPD and SWC) across summer 2021-2023.
4.1.Effect of drought on carbon cyclesOur findings showed an obvious decrease in both GPP and R e across two forest ecosystems during summer 2022, aligning with previous researches(Ciais et al 2005, Reichstein et al 2007, Wang et al  2023a).Our analysis of daily GPP and R e values showed that SWC was the dominant control on GPP and R e while R g and T a had a secondary the stomatal conductance to enhance water use efficiency during drought events (Bastos et al 2020).However, forests may suffer damage or mortality when SWC decreases to a threshold level, reinforcing stomatal closure and impeding hydraulic transfer from soil to leaves (Sperry et al 2002).In contrast to studies reporting significant declines in carbon

L
Han et al Lloyd J and Taylor J A 1994 On the temperature dependence of soil respiration Funct.Ecol.8 315-23 Lu J and Yan F Q 2023 The divergent resistance and resilience of forest and grassland ecosystems to extreme summer drought in carbon sequestration Land 12 1672 Mauder M and Foken T 2006 Impact of post-field data processing on eddy covariance flux estimates and energy balance closure Meteorol.Z. 15 597-609 Müller L M and Bahn M 2022 Drought legacies and ecosystem responses to subsequent drought Glob.Change Biol.Reichstein M, Ma X L, Mahecha M D, Lange H, Flach M and Frank D 2018 Drought, heat, and the carbon cycle: a review Curr.Clim.Change Rep. 4 266-86 Sperry J S, Hacke U G, Oren R and Comstock J P 2002 Water deficits and hydraulic limits to leaf water supply Plant Cell Environ.25 251-63 Sun Z X, Zhang Q, Sun R and Deng B 2022 Characteristics of the extreme high temperature and drought and their main impacts in southwestern China of 2022 J. Meteorol.40 764-70 Tan Z H, Zhang Y P, Liang N S, Hsia Y J, Zhang Y J, Zhou G Y, Li Y L, Juang J Y, Chu H S and Yan J H 2012 An observational study of the carbon-sink strength of East Asian subtropical evergreen forests Environ.Res.Lett.7 044017 Tang X G, Li H P, Ma M G, Yao L, Peichl M, Arain A, Xu X B and Goulden M 2017 How do disturbances and climate effects on carbon and water fluxes differ between multi-aged and even-aged coniferous forests?Sci.Total Environ.599-600 1583-97 Tang Y, Jia C, Wang L, Wen X and Wang H 2021 Solar energy dominates and soil water modulates net ecosystem productivity and evapotranspiration across multiple timescales in a subtropical coniferous plantation Agric.For.Meteorol.300 108310 Thompson I, Mackey B, McNulty S and Mosseler A 2009 Forest Resilience, Biodiversity, and Climate Change.A Synthesis of the Biodiversity/resilience/stability Relationship in Forest Ecosystems (Ecosystems.Technical Series) (Convention on Biological Diversity) Tong X J, Xiao J F, Liu P R, Zhang J S, Zhang J R, Yu P Y, Meng P and Li J 2023 Carbon exchange of forest plantations: global patterns and biophysical drivers Agric.For.Meteorol.
Canadell J G, Prentice I C, Sitch S and Terrer C 2023 Evidence and attribution of the enhanced land carbon sink Nat.Rev. Earth Environ. 4 518-34 Seneviratne S et al 2021 Weather and climate extreme events in a changing climate IPCC Sixth Assessment Report (Chapter 11)