Differential responses of respiration and photosynthesis to air temperature over a moist tundra ecosystem of Alaska and its impact on changing carbon cycle

This study analyzed the sensitivities of carbon cycle to surface air temperature using the CO2 flux data collected from June to September for six years (2014–2019) over a moist tundra site in Council, Alaska. The tundra ecosystem was a strong sink of carbon in June and July, a weak sink in August with rapidly decreasing photosynthesis, and a carbon source in September. The ecosystem respiration (Re) and gross primary production (GPP) were obtained from the net ecosystem exchange (NEE) of eddy-covariance system. Both the Re and GPP increased with temperature, enhancing carbon emission and uptake during observation period. Notably, Re showed higher sensitivity to temperature than GPP did. This result means that as global warming continues, the increase in carbon release is greater than the increase in carbon uptake. In other words, the tundra ecosystem is expected to become a weaker carbon sink in June and July and a stronger source of carbon in September. Possible mechanism of different temperature sensitivities of Re and GPP as well as temporal variations of temperature sensitivities are suggested. Present results highlight the importance of understanding the temperature sensitivities of Re and GPP in various tundra ecosystems to accurately understand changes in the carbon cycle in the Arctic region.


Introduction
The Arctic is experiencing much faster warming than the global average (Overland et al 2019, Rantanen et al 2022).If this rapid warming continues, the thawing of permafrost is expected to escalate (Camill 2005), leading to decreased area of permafrost by 24% (RCP 4.5) to 54% (RCP 8.5) by 2100 (Schuur et al 2013).Permafrost contains approximately 50% (1672 Pg) of the global soil organic carbon (SOC) (Tarnocai et al 2009) and approximately 62% (1035 Pg) of the carbon is stored in the surface layer (0-3m depth) which is susceptible to thawing (Hugelius et al 2014).Consequently, the thawing of permafrost can release a substantial amount of carbon stored in the soil into the atmosphere, thereby accelerating global warming (Schaefer et al 2011, Schuur et al 2015).Due to these concerns, there has been a growing interest in understanding permafrost carbon feedback (Sjöberg et al 2020).
The carbon exchange at the surface of an ecosystem is represented by Net Ecosystem Exchange (NEE) consisting of carbon emission by ecosystem respiration (Re) and uptake by photosynthesis (Gross Primary Production, GPP) by relation of NEE = Re -GPP.This means the role of an ecosystem in carbon cycle is determined by relative magnitudes of Re and GPP.For example, Schuur et al (2022) indicated that Alaskan tundra region was a carbon sink, storing carbon in the soil before industrialization.However, some tundra regions were found to change to carbon source in recent decades (Belshe et al 2013).Schuur et al (2021) observed that a moist tundra region such as the Eight Mile Lake site, had been carbon source in most years based on 15 years of observations (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018).In a broader region, Commane et al (2017) showed that the Alaskan tundra region (58 °N to 72 °N, 140 °W to 170 °W) was simulated to be a carbon source during 2012-2014.
The transition of tundra regions from carbon sinks to carbon sources is attributed to the non-growing season's carbon emissions offsetting carbon uptake during the growing season (Belshe et al 2013, Natali et al 2014, Parazoo et al 2016).As shown in Natali et al (2019), CO 2 emission in winter is projected to increase by 17% until 2100 compared to 2003-2017, attributing 32% of the non-growing season carbon emission variability to temperature changes.Considering stronger warming trend in non-growing season than in growing season in the Arctic (Box et al 2019), together with above studies, responses of Re and GPP to warming in each season need to be known to make more accurate prediction on carbon cycle change.But while there are many studies on the response of Re to temperature (Yvon-Durocher et al 2012, Mundim et al 2020), it is hard to find a study on the response of GPP to temperature based on the data obtained under natural condition.
In this study, we aimed to gain more accurate understanding of the carbon cycle changes in the moist tundra ecosystem by analyzing the carbon flux data observed from June to September for six years in the moist tundra region of Council, Alaska.We used NEE to calculate Re and GPP components and analyzed the monthly variations and temperature sensitivities of GPP and Re.Then, we projected how the Re and GPP of the tundra ecosystem in this region might change in response to warming, and consequently, how the carbon cycle might be affected.

Study site and instrumentation
This study analyzed observational data from six years (2014-2019) obtained from the Council site in western Alaska (64.844°N, 163.711°W).The Council site is approximately 120 km northeast of Nome, Alaska, and is a subarctic transition area from forest to tundra showing a greening trend (figure S1).The vegetation of the site consists of vascular plants (e.g., Erophorum scheuchzeri and Betula nana), moss (e.g., Sphagnum lenese pohle and Sphagnum russowii Warnst), and lichen (e.g., Cladonia stellaris) (Chae et al 2016).The monthly mean surface air temperature (SAT) measured at a nearby city of Nome (Station ID: USW00026617) ranged from −17.4 to 13.2 °C during the period of 2014-2019, according to the Alaska Climate Research Center, University of Alaska Fairbanks (ACRC 2023).
The Council site is a moist tundra region and has been used for the study of permafrost change and the water cycle (Yoshikawa and Hinzman 2003), variation of surface energy balance depending on vegetation type (Beringer et al 2005) and the impact of environmental parameters on soil respiration (Kim et al 2014).The site was registered in the Arctic-Boreal CO 2 flux (ABCflux) database (Site name: 'US-KOC, Council') (Virkkala et al 2022).Our time-lapse images showed that the Council site was usually covered by snow from late October to mid-May (not shown).
An eddy covariance system was used to measure the NEE at the study site.The eddy covariance system consisted of a sonic anemometer (CSAT3A, Campbell Scientific Inc., Logan, UT, USA) and an open-path gas analyzer (EC150, Campbell Scientific Inc., Logan, UT, USA) placed 3 m above the ground.High frequency (10-Hz) data of CO 2 , H 2 O, and 3-D wind components were recorded by a CR3000 data logger (Campbell Scientific Inc., USA) and Compact Flash card (16 GB).Calibration of the gas analyzer for CO 2 and H 2 O was performed annually, at the beginning of the summer measurements.Solar radiation was measured using a radiometer (CNR4, Kipp&Zonen, Netherlands) located 3 m above the ground level, where the SAT was also obtained.We used the moderate resolution imaging spectroradiometer (MODIS) Aqua Normalized Differential Vegetation Index (NDVI) on a selected 250 m × 250 m NDVI pixel plot closest to the council site at 16-day intervals to show the seasonal variation in vegetation phenology.A higher NDVI, close to 1, indicates more active vegetation activity (Pettorelli 2013, Didan 2021).
Precipitation was obtained from the Global Precipitation Climatology Project (GPCP) Monthly Precipitation Climatology Data Record (CDR) data.This data is an integration of satellite data sets and rain gauge data, at a 2.5-degree global grid on a monthly interval (Adler et al 2018).
General corrections for spectral loss and density were applied in the EddyPro.Then, additional quality control was applied to the NEE by excluding outlying values beyond ±3σ from the mean.
Data gaps of NEE caused by missing raw data or filtering were filled following the method of Reichstein et al (2005) using relevant variables such as air temperature, humidity, and solar radiation within ±7 days of interest.
Gap-filled NEE was then used for flux partitioning into Re and GPP, following the method of Reichstein et al (2005), i.e. so-called Nighttime partitioning method.Initially, the nighttime (when global solar radiation <20 W m −2 ) NEE was considered as the Re, assuming the absence of photosynthesis.Subsequently, the shortterm temperature dependency of Re was determined and extrapolated into daytime data (Lloyd and Taylor (1994)).GPP was estimated from NEE and Re as follows: NEE = Re -GPP.The 30-min interval NEE, Re, and GPP were summed to provide daily accumulated CO 2 fluxes for the Council site.We also applied Daytime partitioning method to partition NEE into Re and GPP following the method of Lasslop et al (2010).For the gap-filling and flux partitioning process, we used R-package REddyProc v.1 (Wutzler et al 2018).

Sensitivity analysis
Atmospheric and ground conditions that affect photosynthesis and respiration include temperature, precipitation, radiation, vegetation, and soil properties (Virkkala et al 2021).Among these factors, we focused on the impact of temperature on CO 2 flux, i.e. the intrinsic temperature sensitivity of Zhang et al (2017) by (1) filtering out low light conditions when solar radiation was below 25% of the clear sky radiation of the time, and (2) analyzing sensitivities every month to have similar thaw depths in each month.The soil moisture at the Council site did not show a significant impact on the variation in photosynthesis, at least for the months analyzed in our study (figure S2).Thaw depth is also a major variable affecting ecosystem respiration, but the thaw depth within each month can be considered to maintain within 15% from the monthly average (Kim et al 2016).
We then performed linear regression of air temperature for NEE, Re, and GPP to determine their sensitivity to variations of temperature using SigmaPlot for Graphing and Data Visualization v10.

Environmental conditions
We examined the SAT and solar radiation values recorded in the Council site from June to September for six years (2014-2019) and NDVI of MODIS, and monthly precipitation provided by the GPCP to provide an overview of the meteorological conditions of the moist tundra ecosystem at the Council site(figure 1).
First, data for the same date in each of the six years were averaged.The SAT values increased gradually from 9.6 ± 2.4 °C on June 3 (Day of year: DOY = 154) to a maximum temperature of 15.7 ± 3.2 °C on July 6 (DOY = 187), and then steadily decreased to −0.8 °C, on September 30 (DOY = 273) (figure 1(a)).Here the uncertainty was calculated as the standard deviation among the same date during the six years unless otherwise noted.The average monthly SATs were 12.5 ± 1.6 °C, 13.7 ± 1.2 °C, 11.4 ± 1.8 °C, and 5.8 ± 3.3 °C in June, July, August, and September, respectively, with the highest and lowest temperatures observed in July and September, respectively.
The vegetation condition represented by MODIS Aqua NDVI was 0.60 ± 0.05 on June 2 (Day of the year (DOY) = 153), which increased progressively until reaching the maximum value of 0.75 ± 0.07, recorded on August 5 (DOY = 217), after then it gradually decreased to the lowest value of 0.59 ± 0.03 on September 22 The highest daily mean downward solar radiation was 344 Wm −2 on June 1 (DOY = 152), after which it gradually decreased.The monthly mean solar radiation was 291.8 ± 54.0 Wm −2 in June, 163.2 ± 47.0 Wm −2 in July, 115.7 ± 30.8 Wm −2 in August, and 112.5 ± 23.9 Wm −2 in September.It was the highest in June and decreased steadily thereafter (figure 1(c)).
The monthly total precipitations were 28.5 ± 6.5 mm in June, 76.7 ± 25.5 mm in July, 96.4 ± 23.7 mm in August, and 89.9 ± 30.3 mm in September.So the largest precipitation was recorded in August, and June was the driest month during the observation period (figure 1(d)).

Carbon flux
The carbon flux observed during the study period at the Council site in Alaska is shown in figure 2. GPP and Re are expressed as positive values, whereas NEE can be negative or positive.A negative NEE value indicated the uptake of carbon from the atmosphere to the surface (i.e.carbon sink), whereas a positive NEE value indicated the release of carbon from the surface to the atmosphere (i.e.carbon source).
The NEE on the first day of observation, June 1 (DOY = 152), was −2.6 gC m −2 day −1 indicating carbon uptake.The carbon uptake progressively increased until July 5 (DOY = 186), when the maximum uptake of −3.0 ± 0.6 gC m −2 day −1 was observed, after which it gradually decreased (figure 2(a)).The first positive NEE value, thus carbon release, of 0.0 ± 0.4 gC m −2 day −1 was observed at the end of August(Aug.31, DOY = 243), after which positive NEE was consistently observed during the remainder of the observation period.In other words, carbon uptake gradually increased in June and then steadily weakened from July until the carbon cycle changed to carbon release on September 2nd.The monthly integrated NEE analysis indicated that the largest carbon uptake occurred in July (−59.7 ± 15.3 gC m −2 ) and the largest carbon release occurred in September (15.9 ± 6.9 gC m −2 ), respectively.
The Re and GPP values were estimated as 3.9 ± 2.6 gC m −2 day −1 and 4.4 ± 5.6 gC m −2 day −1 , respectively, on June 1 (DOY = 152).These values then gradually increased until they reached their maxima of 5.5 ± 2.0 gC m −2 day −1 (Re) on August 5(DOY = 217) and 6.9 ± 2.6 gC m −2 day −1 (GPP) on August 4 (DOY = 216), respectively.Afterward, the values steadily decreased until they reached near zero on September 30 (Re = 0.32 gC m −2 day −1 and GPP = 0.07 gC m −2 day −1 ) (figures 2(b) and (c)).The monthly integrated Re and GPP in figure 3 both showed the highest in July (Re = 132.9± 16.4,GPP = 195.3± 22.3) and the lowest in September (Re = 69.3± 27.6, GPP = 53.4± 29.1).Moreover, although GPP was larger than Re from June to August, the opposite was true in September.In other words, the tundra ecosystem of the Council site acted as a carbon sink from June to August but as a carbon source in September.These results showed the same trend even when partitioned using the daytime method, although numerically different (figures S3 and S4).

Responses of carbon fluxes to surface air temperature
As temperature is of the major factors controlling the carbon cycle, we focused on the response of tundra ecosystems to temperature increases based on observation data from the Council site (figure 4).The sensitivities of NEE, Re, and GPP to temperature variations were determined using the daily values of NEE, Re, GPP, and SAT for each month and were compared to identify changes in sensitivity during the observed months.
The sensitivity of NEE to temperature was analyzed as a positive value in July, August, and September.In other words, as the temperature increases, net carbon absorption becomes weaker during the carbon uptake period (i.e., June and July) and net carbon emission becomes stronger during the carbon release period (i.e., September), revealing that this tundra ecosystem tends to be a weaker net sink and stronger net source in warmer conditions.
Our results highlight that the sensitivity of Re to temperature was 1.4 times larger than that of GPP in June and July, and 1.1 times in September.In August, the temperature sensitivity of GPP was greater than that of Re, but the difference was statistically insignificant.These mean that Re responded more strongly to temperature change than GPP did, i.e.Re increased more rapidly with temperature than GPP did.

Discussion
To investigate how the Re and GPP that constitute the carbon cycle in the tundra ecosystem respond to warming, observational data from the Council site were analyzed.The results showed that in June and July, both Re and GPP exhibited high values due to the growth of vegetation and increasing temperatures, and the GPP value was even greater than the Re value, indicating a carbon sink (NEE < 0).In August and September, with the decline in vegetation activity and decreasing temperatures, both GPP and Re values decreased.In August, GPP, which is determined solely by vegetation, decreased more rapidly than Re, weakening the carbon sink in the region.In September, GPP became smaller than Re, resulting in a transition of the ecosystem from carbon sink to carbon source.The carbon flux by Schuur et al (2021) using the data from the moist tundra site located at a similar latitude to the Council site, Eight Mile Lake site (63°52′42.1″N,149°15′12′W) also showed that in July, the region acted as a strong carbon sink, turning into a weak carbon sink in August, and eventually becoming a carbon source in September.The relative magnitude changes in the monthly average NEE values from June to September were similar in both regions.
The analysis of carbon flux sensitivity to temperature in the Council site revealed that as the temperature increases, both Re and GPP increase.This result indicates that with rising temperatures, carbon emission increases due to ecosystem respiration, but at the same time, carbon uptake increases through photosynthesis.This implies that there is a greater movement of carbon between the soil and the atmosphere with temperature rise.Similar findings were mentioned by Jeong et al (2018) in the wet tundra region of Barrow, Alaska (71.29°N, 156.79°W).They observed that as temperature increases, carbon emissions from thawing permafrost increase, leading to a significant rise in atmospheric CO 2 concentration.Concurrently, the increased productivity of vegetation enhances carbon uptake, resulting in larger fluctuations in atmospheric carbon concentration and a decrease in carbon residence time.
We found that during the observation period from June to September, the temperature sensitivity of Re is higher than that of GPP, except for August, when is statistically insignificant.This result suggests that if warming continues in the moist tundra region, the carbon emission may increase more strongly than the carbon uptake.We found that the Re and GPP partitioned by Daytime method also shows consistent results with those of Nighttime method.In short, GPP is larger than Re except September and the temperature sensitivity of Re is larger than those of GPP in each month (table S1 and figure S5).As the relations of Re and GPP with temperature are not linear in the temperature range of natural conditions, we can speculate that the temperature sensitivity can vary depending on the temperature regime.In other words, sensitivity may be different in periods of low temperature and high temperature.Additionally, because the biomass of vegetation changes depending on the season, it is possible that the temperature sensitivity of GPP may also vary depending on the season.Since the temperature level and vegetation mass are different each month, sensitivity may also vary each month.
Among previous studies analyzing the temperature sensitivity of carbon flux of Alaska,, Ueyama et al (2013) analyzed the CO 2 flux variations and temperature sensitivity in several regions of Alaska, including moist tundra regions similar to the Council site (e.g., Atqasuk (70.47°N, 157.41°W),Ivotuk (68.49°N, 155.75°W), and Imnavait Creek (68.606°N, 149.304°W)).However, their results showed that from June to September, the temperature sensitivity of GPP was higher than that of Re.This different result from the current study may be attributed to different environmental factors (such as soil temperature, soil moisture, vegetation type, and soil microbial composition) that can influence the carbon cycle, even in regions classified as moist tundra.Even though analysis of the discrepancy with above tundra sites is not feasible without full environmental data, we suggest one possibility that since ecosystems with different temperature regimes might have different optimal temperature of carbon flux (Niu et al 2012), tundra ecosystems from different temperature regime might have different sensitivities depending on relative stage compared to thermal optimality.We think that those discrepancies emphasize the importance of reporting present results for better understanding carbon cycles in various regions.
Furthermore, many climate models (e.g., Earth system models) that simulate the terrestrial carbon cycle, have not yet fully captured CO 2 emissions from high-latitude regions (Natali et al 2019).Therefore, to accurately analyze the current and future carbon cycle in high-latitude regions, it is crucial to understand the carbon cycle characteristics of various tundra ecosystems by analyzing each tundra region's specific carbon flux dynamics, like this study.
We noted that in September GPP(Re) showed the second(third) -highest sensitivity to temperature among the four months.Given the average GPP and Re in September are lowest, when applying current temperature sensitivities, September exhibited the largest rate of change in Re and GPP compared to the present average value among the four months.This result suggests that in the tundra region, where the non-growing season is longer than the growing season with a strong winter warming trend, the transition speed of the carbon cycle due to increased carbon emissions may be faster than expected.The changes in the carbon cycle during the nongrowing season in tundra regions can have a significant impact, as shown in other studies (Belshe et al 2013, Commane et al 2017, Jeong et al 2018, Natali et al 2019, Virkkala et al 2022).The temperature increase during the non-growing season in the tundra region can lead to substantial changes in the carbon cycle, implying that the analysis of carbon dynamics for this period is critical in predicting the future carbon cycle.Our study highlights the importance of understanding the temperature sensitivities of Re and GPP and carbon dynamics during both the growing and non-growing seasons, respectively, in tundra ecosystems.As the climate is expected to warm continually, our findings provide valuable insights for predicting the future carbon cycle in the tundra region.
Although this study included 6-year data, it was not possible to analyze Re and GPP sensitivities throughout the year because the observation was only performed from June to September.Therefore, it is necessary to have extended period of flux measurement to understand changes of the annual carbon cycle.Furthermore, it's important to note that the Council data used in this study does not represent the whole moist tundra ecosystem.To gain a more comprehensive understanding of the carbon cycle across diverse tundra ecosystems, it would be beneficial to conduct additional research for various tundra ecosystems with distinct vegetation and soil environments.By gathering data from multiple sites and diverse ecological environments, we can better understand the response of the entire tundra ecosystem to climate change and provide more robust predictions for the future.

Conclusion
In this study, we analyzed how the carbon cycle components, Re, GPP, and NEE, in a moist tundra ecosystem would respond to temperature variability.We examined data collected over six years from a moist tundra region of western Alaska, Council, to observe the variations in carbon cycle components, from June to September.Additionally, we determined the temperature sensitivity of these components.Based on these findings, the conclusions are presented as follows.
1.The moist tundra region in Alaska acted as a carbon sink during June, July, and August but as a carbon source in September.
2. From June to September, both GPP and Re increased with rising temperatures.
3. Due to the higher sensitivity of Re to temperature than GPP in this region, future warming is expected to weaken carbon uptake during June and July and enhanced carbon release in September.
4. To better predict changes in the carbon cycle for the entire Arctic tundra ecosystem, it is essential to investigate the temperature sensitivities of Re and GPP across various tundra ecosystems and to understand underlying mechanism driving temperature sensitivities of carbon fluxes.

Figure 1 .
Figure 1.Time series of multi-year average of (a) daily mean surface air temperature (SAT), (b) 16-day normalized differential vegetation index (NDVI), (c) daily mean solar radiation, and (d) monthly total precipitation at the Council site observed from June-September during 2014-2019 ( : 1 year, : 2 years, : 3 years, : 4 years, : 5 years, : 6 years ).The NDVI was acquired from the Aqua satellite of MODIS, and the precipitation from the global precipitation climatology project (GPCP) monthly precipitation climate data record (CDR).The error bars show the standard deviation among the same dates during the six years.

Figure 2 .
Figure 2. Time series of multi-year average of daily averaged (a) net ecosystem exchange (NEE), (b) ecosystem respiration (Re), and (c) gross primary production (GPP) at the Council site between June-September 2014-2019 ( : 1 year, : 2 years, : 3 years, : 4 years, : 5 years, : 6 years).The error bars show the standard deviation among the same dates during the six years.

Figure 3 .
Figure 3. Monthly accumulated net ecosystem exchange (NEE), ecosystem respiration (Re), and gross primary production (GPP) at the Council site from June to September during 2014-2019.The error bars show the standard deviations among the same months during six years.

Figure 4 .
Figure 4. Sensitivity of net ecosystem exchange (NEE), ecosystem respiration (Re), and gross primary production (GPP) based on their response to surface air temperature (SAT) at the Council site from June to September during 2014-2019.n denotes total number of daily data in each month.

Table 1 .
The monthly average value and sensitivity of carbon fluxes from June to September during 2014-2019 for the Council site.Uncertainties are standard deviations among the same months during six years.