Observed Methane Uptake and Emissions at the Ecosystem Scale and Environmental Controls in a Subtropical Forest

: Methane (CH 4 ) is one of the three most important greenhouse gases. To date, observations of ecosystem-scale methane (CH 4 ) ﬂuxes in forests are currently lacking in the global CH 4 budget. The environmental factors controlling CH 4 ﬂux dynamics remain poorly understood at the ecosystem scale. In this study, we used a state-of-the-art eddy covariance technique to continuously measure the CH 4 ﬂux from 2016 to 2018 in a subtropical forest of Zhejiang Province in China, quantify the annual CH 4 budget and investigate its control factors. We found that the total annual CH 4 budget was 1.15 ± 0.28~4.79 ± 0.49 g CH 4 m − 2 year − 1 for 2017–2018. The daily CH 4 ﬂux reached an emission peak of 0.145 g m − 2 d − 1 during winter and an uptake peak of − 0.142 g m − 2 d − 1 in summer. During the whole study period, the studied forest region acted as a CH 4 source (78.65%) during winter and a sink (21.35%) in summer. Soil temperature had a negative relationship ( p < 0.01; R 2 = 0.344) with CH 4 ﬂux but had a positive relationship with soil moisture ( p < 0.01; R 2 = 0.348). Our results showed that soil temperature and moisture were the most important factors controlling the ecosystem-scale CH 4 ﬂux dynamics of subtropical forests in the Tianmu Mountain Nature Reserve in Zhejiang Province, China. Subtropical forest ecosystems in China acted as a net source of methane emissions from 2016 to 2018, providing positive feedback to global climate warming.


Introduction
Methane (CH 4 ) is an important greenhouse gas and accounts for approximately 32% of the global radiative forcing. It has 28-32 times higher global warming potential over a 100-year time horizon than does carbon dioxide (CO 2 ) [1,2]. The atmospheric CH 4 concentration has been increasing and has more than doubled since preindustrial times, showing a rapid increase until 1999, after which it remained nearly constant until 2006. From 2007, the atmospheric CH 4 concentration again began increasing, likely due to a combination of anomalously high temperatures in the Arctic region and more precipitation in tropical regions [3]. Although major CH 4 sources (e.g., wetlands, rice paddies, biomass burning, and fossil fuels) have been identified [4], we still lack a complete understanding of ecosystem-specific information on CH 4 sinks and sources that could be significant factors contributing to global variations in CH 4 sinks and sources [2,5,6].

Site Description
The studied subtropical forest was located in the Tianmu Mountain Nature Reserve, northwest of Lin'an District, Zhejiang Province, China (30 • 20 34.951" N, 119 • 26 08.671" E, Figure 1). This area has a subtropical monsoon climate, with an elevation of 1152 m. The annual mean temperature and the annual total precipitation were 11.9 • C and 1715 mm from 2016 to 2018, respectively. The region is covered by 140-year-old natural evergreen and deciduous broad-leaved forests. The dominant plants are Cyclobalanopsismyrsinifolia, Daphniphyllummacropodum and Pterostyraxcorymbosus. The canopy height, forest density and forest crown closure were 15-20 m, 3125 hm −2 and 0.7, respectively.

Site Description
The studied subtropical forest was located in the Tianmu Mountain Nature Reserve, northwest of Lin'an District, Zhejiang Province, China (30°20′34.951′′ N, 119°26′08.671′′ E, Figure 1). This area has a subtropical monsoon climate, with an elevation of 1152 m. The annual mean temperature and the annual total precipitation were 11.9 °C and 1715 mm from 2016 to 2018, respectively. The region is covered by 140-year-old natural evergreen and deciduous broad-leaved forests. The dominant plants are Cyclobalanopsismyrsinifolia, Daphniphyllummacropodum and Pterostyraxcorymbosus. The canopy height, forest density and forest crown closure were 15-20 m, 3125 hm −2 and 0.7, respectively.

Eddy Covariance System
The eddy covariance (EC) technique was used to measure and quantify CH4 flux from 2016 to 2018. The EC system was installed in a relatively flat region. Sensors were mounted 38 m above the soil surface. Sensor height was determined to ensure that the EC system was mounted at least twice the height of the plant canopy (15-20 m) during the peak growing season. The EC system included an open-path CH4 infrared gas analyzer (LI-7700, LI-COR Biosciences, Lincoln, NE, USA), an open-path CO2/H2O infrared gas analyzer (LI-7500A, LI-COR) and a sonic anemometer (WindMaster, Gill instruments, Lymington, UK). All raw data were collected at a frequency of 10 Hz and stored by a data logger (LI-7550, LI-COR). Data between 27 February and 30 June in 2016 were missing due to a lack of electrical power supply.

Data Processing
The CH4 flux of 30-min block averages was calculated by EddyPro software (version 6.2.0, LI-COR Biosciences, Lincoln, NE, USA). We used several advanced settings during processing. The angle of attack corrections for Gill WindMaster Pro firmware were used [53] and the block average method was used for detrending raw data. The time lag detection method we used was covariance maximization with default. The double coordinate rotation method was used to ensure that the mean vertical wind speed was zero, averaged over 30 min. The compensation of density fluctuations (WPL terms) was implemented according to Webb et al. [54]. The steady state test and the well-developed turbulence test provided a quality flag (1~9) [55]. We applied spike detection of raw data after Vickers and Mahrt [56].
After data processing by EddyPro, we further filtered the dataset to ensure data quality. We discarded the data when rainfall occurred. CH4 flux was used only when the

Eddy Covariance System
The eddy covariance (EC) technique was used to measure and quantify CH 4 flux from 2016 to 2018. The EC system was installed in a relatively flat region. Sensors were mounted 38 m above the soil surface. Sensor height was determined to ensure that the EC system was mounted at least twice the height of the plant canopy (15-20 m) during the peak growing season. The EC system included an open-path CH 4 infrared gas analyzer (LI-7700, LI-COR Biosciences, Lincoln, NE, USA), an open-path CO 2 /H 2 O infrared gas analyzer (LI-7500A, LI-COR) and a sonic anemometer (WindMaster, Gill instruments, Lymington, UK). All raw data were collected at a frequency of 10 Hz and stored by a data logger (LI-7550, LI-COR). Data between 27 February and 30 June in 2016 were missing due to a lack of electrical power supply.

Data Processing
The CH 4 flux of 30-min block averages was calculated by EddyPro software (version 6.2.0, LI-COR Biosciences, Lincoln, NE, USA). We used several advanced settings during processing. The angle of attack corrections for Gill WindMaster Pro firmware were used [53] and the block average method was used for detrending raw data. The time lag detection method we used was covariance maximization with default. The double coordinate rotation method was used to ensure that the mean vertical wind speed was zero, averaged over 30 min. The compensation of density fluctuations (WPL terms) was implemented according to Webb et al. [54]. The steady state test and the well-developed turbulence test provided a quality flag (1~9) [55]. We applied spike detection of raw data after Vickers and Mahrt [56].
After data processing by EddyPro, we further filtered the dataset to ensure data quality. We discarded the data when rainfall occurred. CH 4 flux was used only when the relative signal strength indicator (RSSI) was >20%. In addition, we filtered the CH 4 flux using a threshold of u* > 0.3 m s −1 to ensure well-developed atmospheric mixing conditions [57]. According to the study of Foken et al. [55], the quality of fluxes was classified by the quality flags of "0", "1" and "2", which represent high-quality data, intermediate-quality fluxes and poor-quality fluxes, respectively. Data with quality flags of "2" were not used for further analysis. These quality criteria occasionally caused equipment failures, resulting in data intervals of different durations.
After the quality check, 31.8% of the raw data were left for analyses. We obtained the daily CH 4 flux by averaging the quality-controlled half-hourly CH 4 flux for each day. Because of quality control, the amount of data remaining varied greatly from day to day. For reliable daily averaged CH 4 flux, only the days with more than 6 data points were used for analyses.
To estimate the budget of the CH 4 flux, missing data needed to be interpolated. The random forest (RF) method was used to fill the gaps in the data. The RF algorithm introduced by Breiman [58] is an ensemble method of regression trees. Kim et al. [59] tested RF for eddy flux gap filling at several sites and found that it outperformed other techniques for all sites and all gap conditions. Thus, we used the RF exactly following Kim et al. [59].
The following variables were used as potential drivers of CH 4 to train the RF: sensible heat flux (H), net ecosystem exchange (NEE), latent heat flux (LE), gross ecosystem product (GEP), soil temperature at 10 cm deep (T soil 10 ), air temperature (Ta), relative humidity (RH), pressure (P), vapor pressure deficit (VPD), Ustar, soil moisture at 10 cm deep (M soil 10 ) and precipitation. The gap-filling performance of RF method in our study was also very good (R 2 = 0.85). Then, the gap-filled flux was used for the calculation of the budget. The annual flux was the sum of the daily average flux of the year.
The uncertainties of the CH 4 flux include the random uncertainty and uncertainty of gap-filling for the CH 4 flux. The random uncertainty for each half-hourly CH 4 flux was estimated through the empirical models described by Finkelstein and Sims [60]. The uncertainty of gap-filling for CH 4 flux was also estimated following Kim et al. [59].

Temporal Variations in Environmental Variables
Both soil temperature and soil moisture showed distinct seasonal variations (Figure 2a,b). Daily soil temperature also showed a distinct seasonal variation, with a minimum temperature of 0.9 • C in winter and a maximum temperature of 35.9 • C in summer. The annual mean soil temperature showed small interannual variability ranging from 18.5 to 18.9 • C. In contrast to soil temperature, soil moisture was low in summer and high in winter during the three-year measurement period. For example, daily soil moisture decreased from June and reached the lowest value of 19.5% in August 2016. Then, it increased until October to about 2.5% and maintained a relatively steady state. Most of the rainfall occurred in summer during the measurement period. Annual rainfall was highest in 2016 and lowest in 2017. The annual rainfall totals in 2016, 2017 and 2018 were 2088, 1381 and 1677 mm, respectively.

Diurnal Variationsin CH 4 Flux
Diel patterns of CH 4 flux varied among different seasons ( Figure 3). The diurnal patterns between winter and summer obviously showed a contrast, although there was no consistent diurnal pattern during spring and fall. In winter, the CH 4 flux started to increase after sunrise (7:30) and reached peaks (0.0492, 0.0907 and 0.0606 µmol m −2 s −1 in 2016-2018, respectively) at 10:00-12:00. The diurnal pattern with a noon peak in winter was generally consistent for each year. In contrast to the winter values, the CH 4

Diurnal Variationsin CH4 Flux
Diel patterns of CH4 flux varied among different seasons ( Figure 3). The diurnal patterns between winter and summer obviously showed a contrast, although there was no consistent diurnal pattern during spring and fall. In winter, the CH4 flux started to increase after sunrise (7:30) and reached peaks (0.0492, 0.0907 and 0.0606 μmol m −2 s −1 in 2016-2018, respectively) at 10:00-12:00. The diurnal pattern with a noon peak in winter was generally consistent for each year. In contrast to the winter values, the CH4 flux at noon in summer had the lowest daily values. After 8:00 am, the CH4 flux gradually decreased to negative and reached uptake peaks (−0.112, −0.0685 and −0.1053 μmol m −2 s −1 in 2016-2018, respectively) at noon (11:30-13:00).

Seasonal Variations in CH4 Flux
Large seasonal variations in daily CH4 flux were observed ( Figure 4). The daily averaged CH4 flux ranged from −0.142 to 0.145 g m −2 day −1 . The CH4 flux started to decrease in winter (emission) each year and reached the minimum (uptake) in summer. Then, the CH4 flux continuously increased from negative to positive and reached the maximum emissions in winter. The maximum emissions were 0.045 g m −2 day −1 on 16 January 2016,

Seasonal Variations in CH4 Flux
Large seasonal variations in daily CH4 flux were observed (Figure 4)

Annual Budget of CH 4 Flux
The study site acted as a net source of CH 4

Environmental Controls on CH 4 Flux
The linear relationship between daily CH 4 flux and all environmental factors (including: Ta, RH, precipitation, VPD, T soil 10 , M soil 10 ) we measured was not significant. The monthly averaged CH 4 flux exhibited a significant negative linear correlation with soil temperature (Figure 6a, p < 0.01, R 2 = 0.34) but increased exponentially with soil moisture (Figure 6b, p < 0.01, R 2 = 0.35). Regression models for the monthly averaged F CH4 and soil temperature (T soil 10 ) and soil moisture (M soil 10 ) are as follows:

Environmental Controls on CH4 Flux
The linear relationship between daily CH4 flux and all environmental factors (including: Ta, RH, precipitation, VPD, Tsoil 10, Msoil 10) we measured was not significant. The monthly averaged CH4 flux exhibited a significant negative linear correlation with soil temperature (Figure 6a, p < 0.01, R 2 = 0.34) but increased exponentially with soil moisture (Figure 6b, p < 0.01, R 2 = 0.35). Regression models for the monthly averaged FCH4 and soil temperature (Tsoil 10) and soil moisture (Msoil 10) are as follows:

Temporal Variations and Annual Budget of CH4 Flux
The diurnal patterns with a single uptake peak and emissions peak, which all appeared at noon, occurred in summer and winter, respectively, each year in this study. A diurnal pattern of CH4 flux with an uptake peak occurring around noon (10:00 am-14:00 pm) in summer has often been observed in upland forests [61][62][63]. Another diurnal pattern of emission peaks occurring in winter has rarely been reported for upland forests [64] but has been well reported for wetlands [44,65,66]. Additionally, a phenomenon similar to that in this study, where both of the two diurnal patterns occurred in summer and winter respectively each year, has not been reported in upland forests. Meanwhile, both diurnal variations can randomly or sporadically occur in an ecosystem [65,67].
The seasonal variation pattern of CH4 uptake in summer and emission in winter was also found for the first time in upland forests in this study. Even u* threshold filter-

Temporal Variations and Annual Budget of CH 4 Flux
The diurnal patterns with a single uptake peak and emissions peak, which all appeared at noon, occurred in summer and winter, respectively, each year in this study. A diurnal pattern of CH 4 flux with an uptake peak occurring around noon (10:00 am-14:00 pm) in summer has often been observed in upland forests [61][62][63]. Another diurnal pattern of emission peaks occurring in winter has rarely been reported for upland forests [64] but has been well reported for wetlands [44,65,66]. Additionally, a phenomenon similar to that in this study, where both of the two diurnal patterns occurred in summer and winter respectively each year, has not been reported in upland forests. Meanwhile, both diurnal variations can randomly or sporadically occur in an ecosystem [65,67].
The seasonal variation pattern of CH 4 uptake in summer and emission in winter was also found for the first time in upland forests in this study. Even u* threshold filtering was not used, the seasonal dynamic was similar to the original result. The range of daily CH 4 flux in this study was −142~145 mg m −2 day −1 . The uptake range was approximately 10 times higher than that measured in other forest ecosystems [68,69], mainly because those studies measured only CH 4 flux in the soil but not at the ecosystem scale. The pattern of CH 4 uptake in summer has often been observed. Wang et al. [63] measured CH 4 fluxes from June to October in a temperate forest that reached an uptake peak in September with a range from −0.002 to −0.006 g m −2 d −1 . In addition to the temperate forest system research, a similar pattern of CH 4 flux with uptake in summer has been reported in mixed deciduous forests [69], broad-leaved/Korean pine forests [61] and spruce-fir forests [70]. Unlike the uptake pattern in summer, an emissions pattern in winter has rarely been reported. Only Sakabe et al. [67] reported that CH 4 emissions occurred in the summer, fall and winter in 2009 in a coniferous forest, mainly due to the increase in precipitation. Except for this example, we did not find a similar pattern in other ecosystems.
The annual budgets of the CH 4 flux for 2017 and 2018 were estimated to be 1.15 ± 0.28 and 4.79 ± 0.49 g CH 4 m −2 year −1 in this study, respectively. When u* threshold filtering was not used, the annual budgets was 0.22~2.68 g CH 4 m −2 year −1 . The range of the annual budget was different from that of other subtropical or tropical ecosystems (Table 1) based on the eddy covariance technique [44,52,71]. Shoemaker et al. [52] reported that the budget in spruce-fir forests was an order of magnitude smaller than that in this study. However, in alpine grasslands and mangrove forests [44,71], the budgets were an order of magnitude higher than that in this study. Except for the eddy covariance technique, we compared the range (−142~145 mg m −2 d −1 ) of CH 4 daily average flux with other subtropical upland forests (Table 1) in China and found that the range was higher than that of other ecosystems [35,72,73]. Except for subtropical or tropical ecosystems, we compared the range (−0.218 to −142 mg m −2 day −1 ) of CH 4 uptake (sink) at our site with that in temperate forests [74,75] and found that it was higher than that in temperate forests. Smith et al. [75] reported that CH 4 uptake ranged from −0.05 to −3.6 mg m −2 day −1 in forests located in six countries of northern Europe. Morishita et al. [74] found that CH 4 uptake ranging from −0.05 to −4.3 mg C m −2 day −1 was observed across 26 forest sites in Japan. According to these comparisons, we found that the subtropical forest can act as a significant CH 4 source in upland forests.

Control Factors of CH 4 Flux
The annual CH 4 budget (net ecosystem exchange) depends on the balance of the CH 4 sink (uptake) and source (emissions). Methane uptake and emissions are a combination of biochemical and physical processes [77]. It is widely recognized that CH 4 flux dynamics in forest ecosystems are controlled by multiple environmental factors, including soil temperature, soil moisture, soil nutrients, natural disturbances such as droughts and fires, and forest management practices (such as thinning and understorey removal) [13].
Temperature, as a primary driver, plays an important role in affecting CH 4 production, oxidation and emissions in various forest ecosystems [47,[78][79][80]. Changes in soil temperature affect not only the activities of soil microbes [81], but also the transport of CH 4 flux from soil to the atmosphere [29,82]. Numerous studies have indicated that the temporal variation in the CH 4 flux is mainly determined by temperature, and the CH 4 flux increases with increasing temperature [83,84]. At our study site, there was a negative correlation ( Figure 6, p < 0.01, R 2 = 0.34) between T soil 10 and the CH 4 flux, which was different from the results of many previous studies [42,80,85,86]. That is because, in winter, despite the lower soil temperature, the soil moisture in the study was highest, and the relatively favorable wet conditions were suitable for enhancing CH 4 production. In summer, a higher soil temperature may result in more CH 4 consumption, forming an enhanced sink of CH 4 ( Figure 5).
There is general agreement among mainstream scientists that soil moisture plays an important role in ecosystem CH 4 exchange [13,84,87]. Soil moisture can directly affect oxygen availability, gas diffusion rate and microbial activity, and that significantly alters CH 4 oxidation and production [28,88,89]. In our study, M soil 10 had a significant positive effect on the CH 4 flux (Figure 6b, p < 0.01) and accounted for approximately 34.8% of the variation in the daily CH 4 flux (Figure 6b). During the winter, the higher soil moisture should have created more anaerobic conditions for methanogens and thus increased CH 4 emissions. However, in the summer, the lower soil moisture due to the higher temperature could form aerobic soil conditions to promote the uptake of CH 4 via CH 4 oxidation (Figure 7). Although both soil temperature and moisture are important factors influencing the sources and sinks of CH4, it was obvious in this study that the effect of soil moisture on the CH4 flux played a dominant role in the annual CH4 budget during the winter. Therefore, the CH4 emissions in winter were higher than the uptake in summer. It is likely that the reason for the higher soil moisture in winter may be due to snow cover [14,[93][94][95]. On the one hand, the melting of snow water led to higher soil moisture and lower oxygen content, resulting in the reduction was more than that in summer, thus  In addition, CH 4 flux dynamics are affected by many other factors [15,17,19,20,[90][91][92]. All of the abovementioned factors eventually combined and interacted to affect the processes and activities of methanotrophs and methanogens (Figure 7). In summer, the effect of soil temperature on methanotrophs may be more dominant than that of methanogens. Meanwhile, soil moisture decreased due to higher evapotranspiration. Adequate oxygen may weaken the activity of methanogens. However, in winter, due to high soil moisture, soil oxygen is relatively low, and anaerobic conditions may increase methane emissions (Figure 7).
Although both soil temperature and moisture are important factors influencing the sources and sinks of CH 4 , it was obvious in this study that the effect of soil moisture on the CH 4 flux played a dominant role in the annual CH 4 budget during the winter. Therefore, the CH 4 emissions in winter were higher than the uptake in summer. It is likely that the reason for the higher soil moisture in winter may be due to snow cover [14,[93][94][95]. On the one hand, the melting of snow water led to higher soil moisture and lower oxygen content, resulting in the reduction was more than that in summer, thus reducing the proportion of CH 4 oxidation. On the other hand, snow cover can also keep the soil warm, thereby maintaining the activity of soil methanogens, increasing the production of CH 4 . Consequently, this may result in a positive CH 4 annual budget (a net source of CH 4 ), which provides direct evidence to support the previous model simulation study by Tian et al. [5].
Obviously, our current understanding, measurement data and analysis are still limited. First, several data gaps in the CH 4 flux observations existed during the measurements because of instrument failure and a lack of electrical supply. For example, the data from March to June 2016 were missing due to a break in the electrical power supply. Although we used the random forest (RF) approach to gap-fill data, this gap-filling may have introduced some artificial bias and errors for annual budget estimation. Second, additional auxiliary measurements on soil microbial activities, soil oxygen, tree species, ages, tissue types and site characteristics are needed to improve our understanding of the mechanisms of CH 4 uptake and emissions [8].

Conclusions
This study provides a first attempt to use the eddy covariance technique to continuously measure and quantify CH 4 uptake, emissions and annual budget and to investigate its control factors in a subtropical forest of Zhejiang Province, China. Our results suggested that natural evergreen and deciduous broad-leaved forests in the study area acted as CH 4 sinks (uptake of −0.84 g m −2 year −1 ) in summer and CH 4 sources (emissions of 3.815 g CH 4 m −2 year −1 ) in winter. The net annual budget (net source) of CH 4 was approximately 1.15-4.79 g m −2 year −1 during 2017-2018, which provides positive feedback to global climate warming. We also observed a clear diurnal and seasonal pattern of CH 4 flux. At the daily scale, there was a significant emission peak in winter and a significant uptake peak in summer. The peaks of emissions and uptake both occurred at noon. At the seasonal scale, the studied forest region acted as a CH 4 source during winter and a sink in summer. Soil temperature and moisture are the most important and dominant factors affecting the CH 4 dynamics of subtropical forests in China. In addition, this study filled the research gap of CH 4 flux observations of upland forests at the ecosystem scale, providing unique field observation data for informing and validating simulations of process-based CH 4 dynamic models for global upland forest CH 4 budgets.

Data Availability Statement:
The data presented in this study are available on reasonable request from the corresponding author. The data are not publicly available due to privacy.