Atmospheric and Surface-Condition Effects on CO2 Exchange in the Liaohe Delta Wetland, China

Qingyu Jia 1,2 ID , Wenying Yu 2, Li Zhou 3 and Chenghua Liang 1,* 1 College of Land and Environment, Shenyang Agricultural University, Shenyang 110866, China; beyond.22@126.com 2 Institute of Atmospheric Environment, China Meteorological Administration, Shenyang 110166, China; ywyywy20000@163.com 3 Chinese Academy of Meteorological Sciences, China Meteorological Administration, 46 Zhongguancun South Street, Haidian District, Beijing 100081, China; zhouli@camscma.cn * Correspondence: liang110161@163.com


Introduction
The global wetland area is only 20.3 × 10 4 km 2 , which is equivalent to 5-8% of the terrestrial ecosystem [1]. However, its carbon sink capacity reaches 830 Tg·year −1 and the average carbon fixation rate is 118 g·C·m −2 ·year −1 [2]. Wetlands are the largest carbon pool in the world, and their carbon storage measures 400-500 GTC, which is equivalent to 20-30% [3][4][5] of the terrestrial ecosystem. Wetlands are particularly sensitive to global change [6]. Therefore, they play an important role in the balance of the carbon budget [7,8]. "The Global Climate 2011-2015" issued by the World Climate Organization (WMO) concluded that CO 2 concentration, air temperature, and sea surface temperature in the Earth's atmosphere have reached their highest values in the past 5 years and extreme weather events on all continents have reached an unprecedented high, especially over the past 3 years. The annual average temperature is 0.76 • C higher than the average from 1961 to 1990, and the CO 2 concentration is now at 400 ppm. Global climate change has caused changes in the carbon cycle of wetlands, and has also affected the carbon stock stability [9]. The lack of identification of

EC Measurements
The EC observation system was installed at 40°56′29″ N, 121°57′36″ E, and a 90% footprint of the contribution area of the EC observation system was reed. The system consisted of a precision 3-axis sonic anemometer, an open-path CO2/H2O analyzer , and data collector . The sampling frequency was 10 Hz and the installation height was 4 m. The original output data included horizontal wind speed (Ux, Uy), vertical wind speed (Uz), CO2 absolute density, water vapor absolute density, ultrasonic virtual temperature (Ts), atmospheric pressure (pressure), and CSAT3 diagnostic value (diag_csat). The system calculates the online flux (every 30 min), and stores the flux (every 30 min), and the time series (10 Hz; Figure 2) data.
The calibration of the open-path gas analyzer was checked at semiannual intervals, using a calibration cell provided by the manufacturer. A standard gas was used to set the CO2 span, and a dew point generator , Licor, Lincoln, NE, USA) was used for the H2O span, while zeros were checked for both gases by scrubbing the airstream with magnesium perchlorate and soda lime (LI-670 flow controller, Licor, Lincoln, NE, USA).

EC Measurements
The EC observation system was installed at 40 • 56 29 N, 121 • 57 36 E, and a 90% footprint of the contribution area of the EC observation system was reed. The system consisted of a precision 3-axis sonic anemometer, an open-path CO 2 /H 2 O analyzer , and data collector . The sampling frequency was 10 Hz and the installation height was 4 m. The original output data included horizontal wind speed (Ux, Uy), vertical wind speed (Uz), CO 2 absolute density, water vapor absolute density, ultrasonic virtual temperature (Ts), atmospheric pressure (pressure), and CSAT3 diagnostic value (diag_csat). The system calculates the online flux (every 30 min), and stores the flux (every 30 min), and the time series (10 Hz; Figure 2) data.
Water 2017, 9,806 3 of 16 species occur in the lower distribution, and the yield is up to 14-15 t·hm −2 . The soil is typical of the marine saline soil subclass (Soils of China). Soil texture is typical clay, with pH > 8. The soil organic matter content, ammonium nitrogen, alkali solution nitrogen, nitric nitrogen, instant phosphorus, instant potassium, and total nitrogen are shown in Table 1. The water in the wetland originates mainly from irrigation, and flooding from March to September. The irrigation water supply is from the Liaohe and Daling rivers [37], as well as groundwater, precipitation, and sea water supply. The soil surface contains a lot of water.

EC Measurements
The EC observation system was installed at 40°56′29″ N, 121°57′36″ E, and a 90% footprint of the contribution area of the EC observation system was reed. The system consisted of a precision 3-axis sonic anemometer, an open-path CO2/H2O analyzer , and data collector . The sampling frequency was 10 Hz and the installation height was 4 m. The original output data included horizontal wind speed (Ux, Uy), vertical wind speed (Uz), CO2 absolute density, water vapor absolute density, ultrasonic virtual temperature (Ts), atmospheric pressure (pressure), and CSAT3 diagnostic value (diag_csat). The system calculates the online flux (every 30 min), and stores the flux (every 30 min), and the time series (10 Hz; Figure 2) data.
The calibration of the open-path gas analyzer was checked at semiannual intervals, using a calibration cell provided by the manufacturer. A standard gas was used to set the CO2 span, and a dew point generator (Li-610, Licor, Lincoln, NE, USA) was used for the H2O span, while zeros were checked for both gases by scrubbing the airstream with magnesium perchlorate and soda lime (LI-670 flow controller, Licor, Lincoln, NE, USA).  The calibration of the open-path gas analyzer was checked at semiannual intervals, using a calibration cell provided by the manufacturer. A standard gas was used to set the CO 2 span, and a dew point  (Li-610, Licor, Lincoln, NE, USA) was used for the H 2 O span, while zeros were checked  for both gases by scrubbing the airstream with magnesium perchlorate and soda lime (LI-670 flow  controller, Licor, Lincoln, NE, USA).
The CO 2 flux is calculated as: where Fc is the CO 2 flux, is the instantaneous deviation of the vertical wind speed, and the mean value, i.e., the disturbance value; ρ c is the instantaneous disturbance value of the CO 2 density; and w ρ C is the covariance of the vertical wind speed and the CO 2 density. The raw data were manipulated using Eddypro 5.1.1 software for, e.g., the coordinate rotation, spiking removal, time-delay removal and WPL correction; and the throughput data for 30 min were obtained. QA/QC quality control was performed to remove outliers. The calculation of spectral correction factors is performed in different ways for "small" and "large" fluxes, where the threshold between small and large fluxes is set in the graphical user interface (GUI) taken to be the "Minimum, unstable" values under "Spectral and cospectra QA/QC" in Eddypro Advanced Settings. During the period of precipitation, the observation error of the eddy correlation observation system is relatively large and, therefore, the observational data during precipitation are excluded.
Gap-filling: to accurately calculate the annual values of net ecosystem exchange (NEE) at the sites, gap-filling is required to account for the missing data. The commonly used methods for filling missing data (1-3 point) include Excel's interpolation FORECAST function (x, known_y s, known_x s). For longer data, the mean diurnal variation (MDV) is an interpolation technique where the missing NEE value for a certain time period (half-hour) is replaced with the averaged value of the adjacent days at exactly that time of day. Windows of 7 days during daytime and 14 days during nighttime were chosen for averaging in the application. Missing half-hourly NEE data (due to, e.g., instrument and power failure) were filled with the mean diurnal variation approach using a 14-day window during the winter season [38]. Using the above analysis, half-hour NEE data and the CO 2 mole fraction were calculated.
Using QA/QC analysis and gap-filling process, the entire year's half-hour of NEE was calculated, and the half hours of NEE were then combined into the annual NEE.

Abiotic Measurements
The installation height of the automatic weather observation system (A753 WS-X, Adcon, Vienna, Austria) was 2.5 m, and the installation height of the soil temperature sensor was −5 cm. The installation position of the salinity observation system (A755 SM-X, Adcon, Vienna, Austria) was −0.1 m. The WTL meter is a WTL gauge (YSI600, YSI, Yellow Springs, OH, USA). The weather sensors, soil temperature, salt content, and WTL were continuously monitored. The Plant Phantom installation (Plantcam, Wingscapes, Shelby, AL, USA) was located at 1.5 m. The canopy height and dried biomass of the samples were observed once every 15 days from the beginning of the reeds' budding until the end of the growing season [39].

Meteorological Data
Meteorological data were downloaded from China Meteorological Data Service Center homepage (http://cdc.nmic.cn/home.do). The data were passed through quality control, including the extreme climate boundary value housings, stations, fixed duration (average) with extreme internal consistency, and time consistency. The mean value X ± S (n = 50) represents the multi-year average.

Partial Least Squares Regression Analysis
The prediction model was constructed using partial least squares (PLS) regression analysis [40], which is a statistical method similar to principal components regression. Instead of finding hyper planes of maximum variance between the response and independent variables, it finds a linear regression model by projecting the predicted variables and the observable variables to a new space. Because both the X and Y data are projected to new spaces, the PLS family of methods are known as bilinear factor models. Partial least squares discriminant analysis (PLS-DA) is a variant used when the Y is categorical. PLS regression is particularly suitable when the matrix of predictors has more variables than observations, and when there is multicollinearity among the X values [41]. The analysis was carried out using the SIMCA software (SIMCA v.13.0, UMETRICS, Umeå, Sweden). We used variable importance in projection (VIP) to measure the importance of x j in interpreting Y [42], explaining the role of the independent variable x j in the dependent variable set x j .
Among them, the x j pair is interpreted by t h . If Rd (Y; t h ) is large, w h takes a large value, and x j s interpretation of Y is important. We calculate VIP j as: When Rd(Y; t h ) is large, w hj 2 is very large, and VIP j 2 also takes a greater value. Conversely, When all VIP j are 1, the independent variable x j (j = 1,2, ..., p), indicates that they have the same effect when interpreting Y. Otherwise, for x j of VIP j > 1, it is more important in interpreting Y. If an independent variable regression coefficient and VIP j are small, this means that the variable contribution to the model is very small, and can be neglected. VIP less than 0.8 can be considered to make a small contribution to the dependent variable [43].
We simulated the model based on PLS regression: Suppose the multiple regression model is Standardizing the raw data with a z-score, where z i is the normalized variable value; x j is the actual variable value; x i is the mean; and s i is the standard deviation. The regression coefficients w 0 , w 1 ,... w j of the z i principal component were calculated using the PLS regression method, and were entered into the PLS regression normalized data mathematical model Y = w 0 + w 1 z 1 + w 1 z 2 + . . . w j z j (8)

Seasonal and Inter-Annual Variation of the CO 2 Exchange
The CO 2 concentration in the reed wetland ecosystem increased by 22.7 ppm from 2012 to 2015, which was below the global CO 2 concentration. However, the annual average rate of increase was 5.67 ppm higher than the global rate ( Table 2). The annual variation of the CO 2 budget was large during the 4-year study period. The NEE absorption was 1472.97 g·CO 2 ·m −2 in 2012, up to 2450.97 g·CO 2 ·m −2 in 2015, and 1928.67 and 2296.84 g·CO 2 ·m −2 in 2014 and 2013, respectively. The CO 2 exchange accumulation curve suggests that the emissions in winter had little effect on the CO 2 accumulation. The average annual CO 2 absorption of reed wetland was 2037.15 g·m −2 , which indicates a carbon sink ( Figure 3). The seasonal variation trend of the CO 2 exchange volume was similar in different years. The daily emission of CO 2 gradually increased before the growth of reeds in the middle of April. From late April to May, the individual reeds were smaller and the CO 2 absorption capacity was weaker. From June to July, the biomass of the reed was large, individual growth was at its highest, and the uptake of CO 2  Soil respiration increased in August and September, with an increase in temperature, while the vegetative growth of the reed transformed into reproductive growth, and the CO 2 net absorption of the reed community gradually decreased. From January to February, CO 2 emissions were not significant, soil organic matter decomposition was slow, and soil respiration caused weak CO 2 emissions from wetlands.

Seasonal and Inter-Annual Variation of the CO2 Exchange
The CO2 concentration in the reed wetland ecosystem increased by 22.7 ppm from 2012 to 2015, which was below the global CO2 concentration. However, the annual average rate of increase was 5.67 ppm higher than the global rate ( Table 2 Soil respiration increased in August and September, with an increase in temperature, while the vegetative growth of the reed transformed into reproductive growth, and the CO2 net absorption of the reed community gradually decreased. From January to February, CO2 emissions were not significant, soil organic matter decomposition was slow, and soil respiration caused weak CO2 emissions from wetlands.

Seasonal and Inter-Annual Variation of the Atmospheric Measurements
The average temperature (Ta) in 2012 (8.88 °C) was 0.04 °C lower than the Ta over the last 50 years (1961-2010; Table 3). In 2015, the annual Ta measured 10.26 °C, the highest in nearly 50 years. In 2012, the annual cumulative precipitation (PPT) anomalies measured 999 mm, and there were 98 Accumulated CO₂-C Daily CO₂-C Flux

Seasonal and Inter-Annual Variation of the Atmospheric Measurements
The average temperature (Ta) in 2012 (8.88 • C) was 0.04 • C lower than the Ta over the last 50 years (1961-2010; Table 3). In 2015, the annual Ta measured 10.26 • C, the highest in nearly 50 years. In 2012, the annual cumulative precipitation (PPT) anomalies measured 999 mm, and there were 98 PPT days. The PPT during 2013-2015 was 111.4 mm, 260.8 mm, and 164.1 mm, respectively, which were less than the long-term average of 611.2 mm. The relative humidity during 2012-2013 was higher than the long-term average of 65.97%. The average wind speed for the 4 years was less than the long-term average (3.93 m·s −1 ). The annual average sunshine duration (SD) from 2012 to 2015 was 2688.2 h, 2534.8 h, 2525.9 h, and 2326.8 h less than SD long-term average, respectively. The order of meteorological factors for the wetland during 2012-2015 was Ta rise, PPT, WS, RH, and SD decrease. The monthly variation of meteorological factors was clear, with the highest average monthly temperature in July and August, and the lowest in January (Figure 4). In the 4-year study period, temperatures in the winter and spring rose significantly, causing an increase in the annual average temperature. Rainfall was concentrated mainly from July to August. The precipitation and frequency during June, July, and August in 2012 were higher. The total annual precipitation in 2014 was the lowest, but the monthly precipitation in May and June was higher than the 50-year average. The precipitation from summer to winter during 2013-2015 was significantly lower than the long-term synchronization. The average wind speed had two peaks in spring and autumn, with the highest wind speed in April, the lowest in August, where the average wind speed of each month was less than the 50-year average. The average relative humidity was the highest in March and August, and lowest in December. In 2014, the relative humidity per month was less than the 50-year average. The longest SD months were May and September. In the autumn and winter, haze conditions reduced the number of sunshine hours, resulting in less ground-heat loss, and a smaller rise in the winter temperature. PPT days. The PPT during 2013-2015 was 111.4 mm, 260.8 mm, and 164.1 mm, respectively, which were less than the long-term average of 611.2 mm. The relative humidity during 2012-2013 was higher than the long-term average of 65.97%. The average wind speed for the 4 years was less than the long-term average (3.93 m·s −1 ). The annual average sunshine duration (SD) from 2012 to 2015 was 2688.2 h, 2534.8 h, 2525.9 h, and 2326.8 h less than SD long-term average, respectively. The order of meteorological factors for the wetland during 2012-2015 was Ta rise, PPT, WS, RH, and SD decrease. The monthly variation of meteorological factors was clear, with the highest average monthly temperature in July and August, and the lowest in January (Figure 4). In the 4-year study period, temperatures in the winter and spring rose significantly, causing an increase in the annual average temperature. Rainfall was concentrated mainly from July to August. The precipitation and frequency during June, July, and August in 2012 were higher. The total annual precipitation in 2014 was the lowest, but the monthly precipitation in May and June was higher than the 50-year average. The precipitation from summer to winter during 2013-2015 was significantly lower than the long-term synchronization. The average wind speed had two peaks in spring and autumn, with the highest wind speed in April, the lowest in August, where the average wind speed of each month was less than the 50-year average. The average relative humidity was the highest in March and August, and lowest in December. In 2014, the relative humidity per month was less than the 50-year average. The longest SD months were May and September. In the autumn and winter, haze conditions reduced the number of sunshine hours, resulting in less ground-heat loss, and a smaller rise in the winter temperature.

Seasonal and Inter-Annual Variation of Phragmites Communis
From 2012 to 2015, the height of the reed canopy increased ( Figure 5). The germination times of the reed occurred in mid-to-late April. The plant height and biomass increased linearly after germination, and the photosynthetic biomass of the reed increased a little until July. After August, the biomass again increased linearly. At this time the leaf area index decreased, and the weight of the stem increased. There was mainly vegetative growth, the dry matter in the leaves transferred to the stem, the biomass growth rate increased abruptly, and the plant height increased. The plant height and biomass reached a maximum in August or September. In September, the reed entered the flowering period, and the ground biomass decreased abruptly. After October, the reed entered the withering period, the average height of the stem was approximately 200 cm, and the average height of the reed stubble was approximately 15 cm in the following year.

Seasonal and Inter-Annual Variation of Soil Water and Salinity
The WTL is the distance from the surface of the mud to the surface of the water, which is one of the key parameters that control the carbon budget of the wetland ecosystem [45]. The thickness of the WTL is related to the annual quota of artificial irrigation and precipitation recharge. The mean WTL during the growing season was 15 ± 6.2 cm, 11.3 ± 7.3 cm, 4.4 ± 5.0 cm, and 8.1 ± 6.8 cm below the vegetation surface in the years 2012, 2013, 2014, and 2015, respectively. During the growing season, the WTL trend was downward, with the lowest WTL in 2014 at 0.044 cm ( Figure 6). The decrease in precipitation in summer and autumn were the cause of the WTL decrease in 2014 and 2015. The trend of salinity change had some correlation with the change in WTL thickness, which indicates that irrigation and drainage had a regulatory effect on salinity. The salinity was less than 0.1% before April. From May, the salinity increased rapidly, reaching 0.3%. The reed absorbed the dissolved salts as it grew, and the salinity then gradually decreased [46]. The process of rainwater top-down leaching also reduced the salinity. The variation curve of soil temperature was similar to

Seasonal and Inter-Annual Variation of Phragmites Communis
From 2012 to 2015, the height of the reed canopy increased ( Figure 5). The germination times of the reed occurred in mid-to-late April. The plant height and biomass increased linearly after germination, and the photosynthetic biomass of the reed increased a little until July. After August, the biomass again increased linearly. At this time the leaf area index decreased, and the weight of the stem increased. There was mainly vegetative growth, the dry matter in the leaves transferred to the stem, the biomass growth rate increased abruptly, and the plant height increased. The plant height and biomass reached a maximum in August or September. In September, the reed entered the flowering period, and the ground biomass decreased abruptly. After October, the reed entered the withering period, the average height of the stem was approximately 200 cm, and the average height of the reed stubble was approximately 15 cm in the following year.

Seasonal and Inter-Annual Variation of Phragmites Communis
From 2012 to 2015, the height of the reed canopy increased ( Figure 5). The germination times of the reed occurred in mid-to-late April. The plant height and biomass increased linearly after germination, and the photosynthetic biomass of the reed increased a little until July. After August, the biomass again increased linearly. At this time the leaf area index decreased, and the weight of the stem increased. There was mainly vegetative growth, the dry matter in the leaves transferred to the stem, the biomass growth rate increased abruptly, and the plant height increased. The plant height and biomass reached a maximum in August or September. In September, the reed entered the flowering period, and the ground biomass decreased abruptly. After October, the reed entered the withering period, the average height of the stem was approximately 200 cm, and the average height of the reed stubble was approximately 15 cm in the following year.

Seasonal and Inter-Annual Variation of Soil Water and Salinity
The WTL is the distance from the surface of the mud to the surface of the water, which is one of the key parameters that control the carbon budget of the wetland ecosystem [45]. The thickness of the WTL is related to the annual quota of artificial irrigation and precipitation recharge. The mean WTL during the growing season was 15 ± 6.2 cm, 11.3 ± 7.3 cm, 4.4 ± 5.0 cm, and 8.1 ± 6.8 cm below the vegetation surface in the years 2012, 2013, 2014, and 2015, respectively. During the growing season, the WTL trend was downward, with the lowest WTL in 2014 at 0.044 cm ( Figure 6). The

Seasonal and Inter-Annual Variation of Soil Water and Salinity
The WTL is the distance from the surface of the mud to the surface of the water, which is one of the key parameters that control the carbon budget of the wetland ecosystem [45]. The thickness of the WTL is related to the annual quota of artificial irrigation and precipitation recharge. The mean WTL during the growing season was 15 ± 6.2 cm, 11.3 ± 7.3 cm, 4.4 ± 5.0 cm, and 8.1 ± 6.8 cm below the vegetation surface in the years 2012, 2013, 2014, and 2015, respectively. During the growing season, the WTL trend was downward, with the lowest WTL in 2014 at 0.044 cm ( Figure 6). The decrease in precipitation in summer and autumn were the cause of the WTL decrease in 2014 and 2015. The trend of salinity change had some correlation with the change in WTL thickness, which indicates that irrigation and drainage had a regulatory effect on salinity. The salinity was less than 0.1% before April. From May, the salinity increased rapidly, reaching 0.3%. The reed absorbed the dissolved salts as it grew, and the salinity then gradually decreased [46]. The process of rainwater top-down leaching also reduced the salinity. The variation curve of soil temperature was similar to that of air temperature. In the 4-year study period, Soil temperatures in the spring rose significantly (Figure 7).

Relationship Between Environmental Factors and CO2
The relationship between the environmental factors and the monthly average CO2 (Figure 8) showed that the monthly CO2 exchange increased with increasing temperature when the temperature was below 10 °C. When the average temperature reached 10 °C, the absorption of CO2 began to increase linearly. The monthly precipitation ranged between 0 mm and 150 mm, and the CO2 exchange increased with increasing precipitation. Above 150 mm, the CO2 exchange decreased with increasing precipitation.

Relationship Between Environmental Factors and CO2
The relationship between the environmental factors and the monthly average CO2 (Figure 8) showed that the monthly CO2 exchange increased with increasing temperature when the temperature was below 10 °C. When the average temperature reached 10 °C, the absorption of CO2 began to increase linearly. The monthly precipitation ranged between 0 mm and 150 mm, and the CO2 exchange increased with increasing precipitation. Above 150 mm, the CO2 exchange decreased with increasing precipitation.

Relationship between Environmental Factors and CO 2
The relationship between the environmental factors and the monthly average CO 2 (Figure 8) showed that the monthly CO 2 exchange increased with increasing temperature when the temperature was below 10 • C. When the average temperature reached 10 • C, the absorption of CO 2 began to increase linearly. The monthly precipitation ranged between 0 mm and 150 mm, and the CO 2 exchange increased with increasing precipitation. Above 150 mm, the CO 2 exchange decreased with increasing precipitation. CO 2 exchange was relative to the relative humidity, but the relative trend of wind speed and sunshine were not clear. The CO 2 exchange was linearly related to the thickness of the water table layer. When the thickness was less than 0 cm, CO 2 was weakly discharged. When the thickness ranged between 0 cm and 20 cm, CO 2 was absorbed. Within a specific range, the increasing thickness of the water layer promoted the absorption of CO 2 . Within a salinity range of 0-0.2%, CO 2 emissions increased with increasing salinity. Between 0.2% and 0.3%, CO 2 absorption increased with increasing salinity.
The relationship between the environmental factors and the monthly average CO2 (Figure 8) showed that the monthly CO2 exchange increased with increasing temperature when the temperature was below 10 °C. When the average temperature reached 10 °C, the absorption of CO2 began to increase linearly. The monthly precipitation ranged between 0 mm and 150 mm, and the CO2 exchange increased with increasing precipitation. Above 150 mm, the CO2 exchange decreased with increasing precipitation. CO2 exchange was relative to the relative humidity, but the relative trend of wind speed and sunshine were not clear. The CO2 exchange was linearly related to the thickness of the water table layer. When the thickness was less than 0 cm, CO2 was weakly discharged. When the thickness ranged between 0 cm and 20 cm, CO2 was absorbed. Within a specific range, the increasing thickness of the water layer promoted the absorption of CO2. Within a salinity range of 0-0.2%, CO2 emissions increased with increasing salinity. Between 0.2% and 0.3%, CO2 absorption increased with increasing salinity.
On the basis of the monthly average, we further analyzed the relationship between the daily average and the environmental factors (Figure 9) in the growing season. Figures 9 and 10 show a On the basis of the monthly average, we further analyzed the relationship between the daily average and the environmental factors (Figure 9) in the growing season. Figures 9 and 10 show a similar trend. The daily average correlation of air temperature, WTL, salinity, soil temperature and CO 2 was more obvious. The maximum daily absorption of CO 2 reached 12 g·C·m −2 , when the temperature range was 20-25 • C, the water layer thickness was 0.1-0.2 m, the salinity was 0.3, and the soil temperature range was 20-25 • C.
of the water layer promoted the absorption of CO2. Within a salinity range of 0-0.2%, CO2 emissions increased with increasing salinity. Between 0.2% and 0.3%, CO2 absorption increased with increasing salinity.
On the basis of the monthly average, we further analyzed the relationship between the daily average and the environmental factors (Figure 9) in the growing season. Figures 9 and 10 show a similar trend. The daily average correlation of air temperature, WTL, salinity, soil temperature and CO2 was more obvious. The maximum daily absorption of CO2 reached 12 g·C·m −2 , when the temperature range was 20-25 °C, the water layer thickness was 0.1-0.2 m, the salinity was 0.3, and the soil temperature range was 20-25 °C.  is the WTL, t2 is soil temperature; t3 is air temperature; t4 is the salt content; t5 is relative humidity; t6 is precipitation; t7 is biomass; t8 is sunshine duration; t9 is height; and t10 is wind speed. Environmental factors Figure 10. Variable importance in projection (VIP) of the environmental factors for CO 2 exchange. t1 is the WTL, t2 is soil temperature; t3 is air temperature; t4 is the salt content; t5 is relative humidity; t6 is precipitation; t7 is biomass; t8 is sunshine duration; t9 is height; and t10 is wind speed.
The environmental factors attributed to CO 2 exchange were studied using the PLS method. The VIP value of the WTL, soil salinity, temperature, weather, air temperature, precipitation, relative humidity, and biomass were greater than 0.8 ( Figure 10), which had significant explanatory meaning for monthly CO 2 exchange. The VIP of the WTL was maximum at 1.42, followed by surface soil temperature and air temperature. The VIP of the sunshine duration and wind speed were smaller, suggesting that the monthly CO 2 balance did not have significant explanatory meaning. Environmental factors may have multiple collinearity and, therefore, we gathered collinearity statistcs for the environmental factors. When 0 < VIF < 10 and tolerance > 0.1, there is no multicollinearity; when 10 ≤ VIF < 100 and tolerance < 0.1, there is clear collinearity [47]. The results show that t2 and t3, t7 and t8 have clear collinearity (Table 4). Using the PLS method, the environmental factors were divided into five groups to simulate the ability of the different combinations to explain the balance of CO 2 .
The first group, which included 10 environmental factors using the independent variable information rate R2X (cum) was the lowest, with a prediction accuracy of 62.9% (Table 4). The fifth group contained the VIP > 0.8 environmental factors, and the sixth group further eliminated the collinearity factor soil temperature. The ability and prediction accuracy of the sixth group was the best. The accuracy of the model increased from 62.9% to 72.5% (Table 5). This indicated that the model had 71.1% to 78.1% predictive power of CO 2 exchange. The factors of the fifth group were used to predict the mathematical model of the CO 2 monthly exchange, with a formula of Y = −0.588 − 0.647t 1 − 0.356t 3 + 0.028 t 4 − 0.243t 5 + 0.29t 6 + 0.255t 7 . The predicted value of CO 2 exchange is less than the observed value (Figure 11), with a correlation coefficient, R 2 = 0.781.

Discussion
In this study, we used the eddy correlation method to analyze the CO2 exchange during 2012-2015 of the reed wetland in the Liaohe Delta, and observed the characteristics of several environmental factors simultaneously. The results showed that the atmospheric CO2 concentration increased annually. During the 4 years, the Liaohe Delta wetland was a large carbon sink. The average annual CO2 absorption was 2.04 kg·CO2·m −2 , and the absorption peak was 23.51 µmol·m −2 ·s −1 . The solid carbon capacity is higher than that of the Chongming Island wetland [48] and the Yellow River Delta [49]. The weak release period occurred from November to April and the net absorption period from May to October. The net absorption period was the longest in 2013, reaching 161 days. From June to July, the amount of CO2 absorption accounted for 62.11-78.54% of the year. The CO2 emissions were higher in April and September, reaching 7.35 g·CO2·m −2 ·d −1 , and emissions were the lowest in December and January.
The annual average temperature during the 4-year period increased each year, and was clearly higher than that of the past 50 years, mainly owing to increased temperatures in winter. During 2013-2015, the accumulated annual precipitation was less than that for a normal year, especially during July-September; however, this did not affect the natural growth of the reeds. The annual sunshine duration decreased during 2012-2015. The decrease in the sunshine duration occurred mainly in winter, which had no effect on the absorption of CO2 by photosynthesis during the growing season of the reed.
There was correlation between environmental factors and CO2 exchange. The temperature range of 20-25 °C promotes photosynthesis of the reed. The WTL is one of the key control parameters of CO2 exchange, where the optimum depth is 0.1-0.2 m. Within a range of 0-0.2%, the increase in salinity can promote microbial activity, and increase organic carbon mineralization and CO2 emission. The salinity range of 0.2% to 0.8% is suitable for reed growth. At levels less than 0.2%, rush growth is too high, affecting the growth of the reeds.
The correlation of air temperature, WTL, salinity, soil temperature and CO2 was clearer than that of other factors. Water level and salinity are mainly affected by manual management measures. To increase the yield of the reeds, there should be initial irrigation, spring irrigation, and autumn irrigation according to the scientific irrigation of reed wetlands, where normal should be the "three

Discussion
In this study, we used the eddy correlation method to analyze the CO 2 exchange during 2012-2015 of the reed wetland in the Liaohe Delta, and observed the characteristics of several environmental factors simultaneously. The results showed that the atmospheric CO 2 concentration increased annually. During the 4 years, the Liaohe Delta wetland was a large carbon sink. The average annual CO 2 absorption was 2.04 kg·CO 2 ·m −2 , and the absorption peak was 23.51 µmol·m −2 ·s −1 . The solid carbon capacity is higher than that of the Chongming Island wetland [48] and the Yellow River Delta [49]. The weak release period occurred from November to April and the net absorption period from May to October. The net absorption period was the longest in 2013, reaching 161 days. From June to July, the amount of CO 2 absorption accounted for 62.11-78.54% of the year. The CO 2 emissions were higher in April and September, reaching 7.35 g·CO 2 ·m −2 ·d −1 , and emissions were the lowest in December and January.
The annual average temperature during the 4-year period increased each year, and was clearly higher than that of the past 50 years, mainly owing to increased temperatures in winter. During 2013-2015, the accumulated annual precipitation was less than that for a normal year, especially during July-September; however, this did not affect the natural growth of the reeds. The annual sunshine duration decreased during 2012-2015. The decrease in the sunshine duration occurred mainly in winter, which had no effect on the absorption of CO 2 by photosynthesis during the growing season of the reed.
There was correlation between environmental factors and CO 2 exchange. The temperature range of 20-25 • C promotes photosynthesis of the reed. The WTL is one of the key control parameters of CO 2 exchange, where the optimum depth is 0.1-0.2 m. Within a range of 0-0.2%, the increase in salinity can promote microbial activity, and increase organic carbon mineralization and CO 2 emission. The salinity range of 0.2% to 0.8% is suitable for reed growth. At levels less than 0.2%, rush growth is too high, affecting the growth of the reeds.
The correlation of air temperature, WTL, salinity, soil temperature and CO 2 was clearer than that of other factors. Water level and salinity are mainly affected by manual management measures. To increase the yield of the reeds, there should be initial irrigation, spring irrigation, and autumn irrigation according to the scientific irrigation of reed wetlands, where normal should be the "three irrigation, three rows" irrigation-management system. The first irrigation generally occurred from 10 March to 10 April. The main role of this was to advance thawing, increase the spring bud rate, and ensure the normal growth of the reeds. The spring irrigation was generally from 10 May to 10 June, where the main aim was to promote growth. Autumn irrigation generally occurred from 10 to 20 September to increase the reed fiber content and improve dry matter, while cultivating autumn buds, to lay the foundation for the second year of reed growth. The current water resources in the Liaohe Delta are mainly used for living and agriculture, and the remaining water can sustain a single spring irrigation, causing the single peak in the WTL curve. Over the 4-year period, the water supply for this irrigation management of "one irrigation and one row" was sufficient for the growth of reeds. During the last 3 years, precipitation in the Liaohe Delta decreased, and the decrease of the water layer had a positive effect on the uptake of CO 2 in the reed wetland. Under irrigation management, the salinity of the growing season in the area was maintained within a range of 0.2-0.3%, which was favorable for the growth of reed and the promotion of CO 2 absorption. The salinity tolerance range of reed (1.2-6.6 ng/L) [46] was better adapted to the salt marsh than that of the other types of wetland plants, such as Spartina alterniflora and Suaeda salsa.
The PLS regression model overcomes the shortcomings of traditional regression analysis, and dynamically quantifies the relationship between CO 2 absorption and environmental factors in reed wetlands. Variable importance in projection reflects the importance of environmental factors for the CO 2 exchange. The thickness of the water layer had the strongest effect on the CO 2 exchange, followed by soil temperature, temperature, and salinity, whereas wind speed and sunshine had little effect. Therefore, irrigation is the best way to adjust the carbon balance of the reed. Independent variables such as soil water and salinity, and weather, can simulate and explain the CO 2 exchange. This indicates that WTL and temperature are the main environmental factors that determine the distribution of reeds, ecotypes, and growth conditions [21].

Conclusions
The Liaohe Delta wetland is a large carbon sink. It plays an important role in wetland carbon sequestration in China. The variability of the CO 2 exchange over the past four years was affected by vegetation and meteorological and hydrological factors. The changes in the environmental factors include increasing temperature, decreasing precipitation and sunshine hours, and increasing CO 2 concentration.
Owing to the aforementioned environmental changes and irrigation management, the ability of the reeds to absorb CO 2 and convert it to organic carbon was enhanced, reflecting the ability of the reeds to adapt to the environment. The WTL was the main contributor to the carbon dioxide expenditure among environmental factors. Satisfactory results were achieved using the PLS method to simulate carbon dioxide.
In recent years, under increasing temperatures and decreasing precipitation, artificial irrigation has controlled the key WTL, and can increase the CO 2 absorption in the growing season. However, the increased temperatures in winter lead to increased emissions. The effect of coupling between the water level and salinity, and meteorological factors on the carbon balance of the wetland, will be the focus of future research.