Intensified inundation shifts a freshwater wetland from a CO2 sink to a source

Climate change has altered global precipitation patterns and has led to greater variation in hydrological conditions. Wetlands are important globally for their soil carbon storage. Given that wetland carbon processes are primarily driven by hydrology, a comprehensive understanding of the effect of inundation is needed. In this study, we evaluated the effect of water level (WL) and inundation duration (ID) on carbon dioxide (CO2) fluxes by analysing a 10‐year (2008–2017) eddy covariance dataset from a seasonally inundated freshwater marl prairie in the Everglades National Park. Both gross primary production (GPP) and ecosystem respiration (ER) rates showed declines under inundation. While GPP rates decreased almost linearly as WL and ID increased, ER rates were less responsive to WL increase beyond 30 cm and extended inundation periods. The unequal responses between GPP and ER caused a weaker net ecosystem CO2 sink strength as inundation intensity increased. Eventually, the ecosystem tended to become a net CO2 source on a daily basis when either WL exceeded 46 cm or inundation lasted longer than 7 months. Particularly, with an extended period of high‐WLs in 2016 (i.e., WL remained >40 cm for >9 months), the ecosystem became a CO2 source, as opposed to being a sink or neutral for CO2 in other years. Furthermore, the extreme inundation in 2016 was followed by a 4‐month postinundation period with lower net ecosystem CO2 uptake compared to other years. Given that inundation plays a key role in controlling ecosystem CO2 balance, we suggest that a future with more intensive inundation caused by climate change or water management activities can weaken the CO2 sink strength of the Everglades freshwater marl prairies and similar wetlands globally, creating a positive feedback to climate change.


| INTRODUC TI ON
Climate change has altered global precipitation patterns, resulting in significant changes in regional hydrology and increasing the frequency and intensity of both seasonal and episodic drought or flooding (IPCC, 2013). Ecosystem processes, especially those associated with greenhouse gas emissions (e.g., carbon dioxide), are highly sensitive to these alterations in hydrology and, in turn, may provide feedbacks to ongoing climate change. While many studies have focused on the effect of drought on ecosystem carbon dioxide (CO 2 ) exchange (e.g., Fenner & Freeman, 2011;Malone, Starr, Staudhammer, & Ryan, 2013;Rocha & Goulden, 2010;Taylor, Ripley, Woodward, & Osborne, 2011), less effort has been made to understand the effect of inundation (Han et al., 2015;Larmola et al., 2004;Morison et al., 2000).
Inundation is especially crucial in wetlands ecosystems, where CO 2 exchange and carbon sink potential are mainly driven by hydrology (Davis & Ogden, 1994). Wetlands have significant soil carbon pools (Kayranli, Scholz, Mustafa, & Hedmark, 2010). As climate change continues to shift precipitation patterns and, therefore, hydrology, the fate of wetland carbon becomes more uncertain (Burkett & Kusler, 2000;Erwin, 2009). Thus, there is an important need to understand how the ecosystem CO 2 balance, and its underlying processes, respond to changes in hydrological conditions, such as inundation regimes.
At the ecosystem level, the effect of inundation on net ecosystem CO 2 exchange (NEE) is determined by the difference between gross primary production (GPP; CO 2 uptake) and ecosystem respiration (ER; CO 2 emission). Previous wetland studies reported stronger reductions in GPP than that of ER during the inundation period, weakening the ecosystem CO 2 sink strength (Han et al., 2015;Schedlbauer, Oberbauer, Starr, & Jimenez, 2010). In contrast, enhanced CO 2 sink strength was found under higher water levels (WL) in an Amazon floodplain due to greater reductions in ER than GPP (Morison et al., 2000). Based on data collected from different wetland sites, Larmola et al. (2004) suggested that the effect of inundation on NEE can vary based on the dominant plant species. To date, the effect of inundation has mostly been studied as a binary or discrete factor (e.g., inundated vs. noninundated or high vs. low WL). To better understand the effect of inundation, a comprehensive picture of how continuous variation in hydrology, in terms of WL depth and inundation duration (ID), affects NEE and its components (i.e., GPP and ER) is still needed.
Furthermore, inundation may have legacy effects on CO 2 fluxes, such that effects extend into the season after inundation. For example, studies found that plant photosynthesis and growth can be suppressed for a period of time after inundation as vegetation recovers from the physiological stresses (Chen, Zamorano, & Ivanoff, 2010;Hu, Wu, Yao, & Xu, 2015;Kozlowski & Pallardy, 1979). Accordingly, lower net ecosystem CO 2 uptake was often noted following extensive inundation (Dušek et al., 2009;Hu et al., 2015). Therefore, to fully understand the influences of inundation on the ecosystem CO 2 budget, it is also important to evaluate these carryover effects of inundation and determine the length of their impact.
In the Florida Everglades, hydrology (i.e., inundation) plays a critical role in determining wetland types by affecting the plant species composition (Davis & Ogden, 1994;Todd et al., 2010). Of particular interest is the short-hydroperiod freshwater wetland ecosystem (sometimes referred to as marl prairies), which has seasonally alternating dry periods and inundation that fluctuates substantially with precipitation patterns (Davis & Ogden, 1994;Schedlbauer et al., 2010). This ecosystem has been shown to be a sink or neutral for annual CO 2 budgets (Jimenez et al., 2012;Schedlbauer et al., 2010). The water depth and ID are the two variables that characterize the inundation intensity (Childers et al., 2006;Todd et al., 2010). Although previous studies have recognized that carbon dynamics of the ecosystem are driven by the seasonal inundation, how changes in water depth and ID interactively affect the underlying CO 2 fluxes is largely unknown.
In 2015-2017, due to a combination of extremely high precipitation at the end of 2015 and the beginning of 2016 (Figure S1), associated with El Niño Southern Oscillation (ENSO), and water management activities (i.e., greatly increased freshwater flow in an adjacent canal), the short-hydroperiod freshwater wetland (with an average inundation of ~6 months) in the eastern Everglades experienced an extended inundation period lasting for 17 months. This extreme event allowed us to test ecosystem response to inundation scenarios which were out of the normal range for this system but may occur more frequently in the future. In this study, we used eddy covariance (EC) data and the normalized difference vegetation index (NDVI;2008-2017 from a site in the Everglades short-hydroperiod freshwater wetland to answer the following questions: (a) How do changes in water depth and ID affect the components of ecosystem CO 2 flux (i.e., NEE, GPP and ER)? (b) How does extreme inundation change the sink/source capacity of the ecosystem? (c) Does extreme inundation has lag effects on ecosystem CO 2 dynamics during the following seasons? Answering these questions can improve our current understanding of the influences of hydrologic shifts on the carbon dynamics of freshwater wetland ecosystems and inform the processes needed for accurate representation of wetlands in ecosystem and Earth-system models.

| Site description
This study was carried out in a short-hydroperiod oligotrophic freshwater marl prairie (25°26ʹ16.5ʺ N, 80°35ʹ40.68ʺ W, elevation: ~1 m above sea level) near the headwaters of Taylor Slough on the eastern edge of Everglades National Park (Figure 1). The mean annual temperature is 23.9°C, with the average monthly temperature being lowest in January and highest in August (18.1°C and 29.4°C, respectively) (Jimenez et al., 2012). The site experiences a significant seasonal variation in irradiance and daily net radiation ranges from 6 to ~20 MJ m −2 day −1 (Malone, Staudhammer, Loescher, et al., 2014). The photosynthetically active radiation (PAR) reaches up to 2,350 μmol m −2 s −1 in April/May and up to 1,500 μmol m −2 s −1 in December/January. Annual precipitation averages 1,380 mm, ~70% of which occurs during June-November (Davis & Ogden, 1994). The hydrology of the site is also influenced by the comprehensive everglades restoration plan (Perry, 2004), which aims to improve the freshwater availability in the Everglades.
As a result of seasonal precipitation and water management activities affecting WL in an adjacent canal, the study site is inundated (i.e., WL exceeds soil surface level) for 4-6 months each year with WLs reaching 20-50 cm above the soil surface ( Figure 2c). The surface water flows through the site from the north at a slow rate (0.5-0.8 cm/s) (Schaffranek & Ball, 2000) and the overall lateral carbon balance within the water is likely close to zero (Schedlbauer et al., 2010). The site is not impacted by salinity and has an ~14 cm layer of marl soils on top of limestone bedrock (Schedlbauer et al., 2010). The vegetation has a height of ~0.73 m and is codominated by the macrophytes sawgrass (Cladium jamaicense Crantz) and muhly grass (Muhlenbergia filipes M.A. Curtis) (Schedlbauer et al., 2010). Periphyton, which consists of algae, fungi, and bacteria, also develops progressively at the site during the inundated period as floating mats, benthic carpets, and "sweaters" around submerged macrophyte stems/leaves.

| Eddy covariance and meteorological instrumentation
An EC tower, which is a part of the AmeriFlux network, was estab- that measures wind speed and air temperature. The instruments were installed 3.3 m above the ground. A CR1000 data logger (Campbell Scientific Inc.) was used to record the EC data at a frequency of 10 Hz.
The IRGA was calibrated monthly according to manufacturer instructions against standard calibration gases (i.e., nitrogen gas with soda lime and Drierite, 492.6 ppm CO 2 gas, and gas from a dewpoint generator; LI-610, LI-COR Inc.). The EC system was estimated to have a 90% fetch distance within 200 m of the tower (Schedlbauer et al., 2010;, an area that represents typical vegetation in the marl prairie with minimum impact from the adjacent canal (~400 m away, Figure 1).
For the meteorological variables used in our analysis, PAR was

| Data processing
Eddy covariance data from January 1, 2008 to September 5, 2017 were used in this study. We calculated 30 min NEE from 10 Hz EC data with the program EdiRe (University of Edinburgh, http://www.geos. ed.ac.uk/abs/resea rch/micro met/EdiRe ) following protocols including coordinate rotation, despiking and air density corrections (Aubinet et al., 2000;Baldocchi, Hicks, & Meyers, 1988;Webb, Pearman, & Leuning, 1980). In the process, canopy height was adjusted against the WL change during the inundation period. The CO 2 storage term was estimated according to Hollinger et al. (1994) and included in the 30 min NEE values. The sampling rates were checked by implementing spectral analysis on the raw data. To ensure quality, data satisfying the following criteria were removed: (a) evidence of rainfall, contamination (e.g., condensation, spider webs or bird fouling) in the sampling path of IRGA or sonic anemometer, (b) incomplete 30 min data during calibration or maintenance, (c) friction velocity (u*) <0.15 m/s (Goulden, Munger, Fan, Daube, & Wofsy, 1996) or (d) implausible values found in our data sets. Overall, 27% of daytime and 70% of nighttime NEE were removed over the study period.
To calculate CO 2 budgets, missing NEE data were gap-filled on a monthly basis using separate equations for day-and nighttime. For daytime (PAR ≥ 10 μmol m −2 s −1 ), data were gap-filled using a hyperbolic equation as a function of PAR: where parameters to be estimated are: α (apparent quantum efficiency), ER (ecosystem respiration rate, μmol m −2 s −1 ), and P max (maximum ecosystem CO 2 uptake rate, μmol m −2 s −1 ). For nighttime (PAR < 10 μmol m −2 s −1 ), NEE is comprised entirely of ER and an exponential equation was used for the gap-filling as a function of air temperature (T a ): Seasonal variations of the daily mean air temperature (a), photosynthetically active radiation (PAR) sum (b), water level (WL) (c), and cumulative net ecosystem CO 2 exchange (NEE) (d) during 2008-2017. In (c), a negative value indicates a WL that is below the soil surface while a positive value indicates the WL is above the soil surface where the parameters to be estimated are: R 0 (base respiration rate when air temperature is 0°C) and b (an empirical coefficient indicating the temperature sensitivity of the fluxes). Gap-filled data were only used for calculating CO 2 flux budgets. Because data from September 5 to November 29, 2017 were missing due to system outages caused by Hurricane Irma, the annual CO 2 budget of 2017 was excluded from the analyses.
Following gap-filling of missing data, 30 min ER was estimated via Equation (1), for daytime and via Equation (2)  The EC data are available through AmeriFlux (https ://ameri flux.lbl. gov/).

| Vegetation indices
To explain the responses of CO 2 fluxes to inundation, values of the NDVI at the site were obtained from Moderate Resolution Imaging Spectroradiometer (MODIS), which represent the vegetation dynamics over the study period (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). The index is a part of the Level 3 product "MOD13Q1" available at 250 m resolution over 16-day compositing periods (Didan, 2015). The data were accessed from https ://modis.ornl.gov on June 6, 2018 and cover a 250 × 250 m area centered on the EC tower. NDVI is sensitive to chlorophyll greenness (Huete et al., 2002) and represents a normalized ratio of the reflected near-infrared and red radiation from the land surface: where NIR and RED are the surface reflectance factors for near-infrared and red bands, respectively, which correspond to wavelengths of 841-876 and 620-670 nm, respectively.

| Data analysis
In this study, inundation is indicated by WL >0 cm and inundation intensity refers to the WL and ID.
We used the nongap-filled 30 min data over the study period to fit the light response curve for GPP adapted from Equation (1) to derive the parameters α and P max as following: At the same time, the temperature response curve (Equation 2) was used to derive the parameters R 0 and b for ER. Based on the light and temperature response curves, we examined the interactive effect of WL and ID on the parameters of the curves by using a parameter prediction approach. Specifically, the light and temperature response curves were fit with the parameters (i.e., α, P max , R 0 , and b) replaced by linear functions of WL, ID, and the interaction term as follows: where a 0 -a 3 , b 0 -b 3 , c 0 -c 3 , and d 0 -d 3 are coefficients to be estimated, water level and inundation duration are indicated by WL and ID. Where interaction terms between WL and ID were not significant (i.e., the ER model, p > 0.05), we further examined the individual effect of WL and ID on the parameters (i.e., R 0 and b) by excluding the interaction terms. Data with WL <0 cm as well as gap-filled data were excluded in the models. To visualize the relationships of the models we plotted the least square mean predicted parameters (i.e., α, P max , R 0 , and b) against WL under different IDs.
To determine the effect of WL and ID on NEE, we fit a linear mixed model to daily NEE as a function of WL, ID, and their interaction term. A first-order autoregressive variance-covariance structure, AR (1), was also included in the model to account for the temporal autocorrelation among measurements taken in adjacent time periods.
Based on this model, the compensation points (i.e., the values of WL and inundation that correspond to NEE = 0 g C m −2 day −1 ) were calculated and their 95% confidence intervals were estimated by block bootstrapping (300 random bootstrap replicates).
To determine the importance of inundation on annual CO 2 budgets, we performed linear regressions, using hydrological factors (e.g., mean WL during inundation period, length of inundation period, etc.) as predictors of the annual budgets of flux components (i.e., GPP, ER, and NEE).
To reveal the vegetation response to inundation, we fit a linear mixed model to NDVI as a function of WL, ID, and their interaction term. A first-order autoregressive variance-covariance structure, AR (1), was also included in the model to account for the temporal autocorrelation among measurements in adjacent time periods. Individual effects of WL and inundation were further assessed by plotting binaverage of NDVI as a function of WL (from 0 to 70 cm at 5 cm intervals) and IDs (from 0 to 360 days at 30 day intervals), respectively. As aids for trend visualizations, smooth lines were added based on the nonparametric Local Polynomial Regression (LOESS) in the boxplots.
Lastly, to examine whether inundation effect on the CO 2 fluxes persists into the season following the inundation period, we compared the temporal changes of the parameters (i.e., α, P max , R 0 , and b) from the light and temperature response curves after inundation among the studied years. For this purpose, light (Equation 5) and temperature (Equation 2) response curves were fit to the nongap-filled data within 90 days after inundation with parameters replaced by a linear function of days after inundation (similar to Equations 6 and 7). The models were fit separately for each year (Table S2). To account for nonlinear temporal variations, LOESS curves were used to represent the variation of daily NEE and NDVI during the postinundation periods and compared among years.
The light and temperature response models were fit using the nonlinear least squares function "nls" in the program r 3.5.0 (R Development Core Team, 2018). The linear mixed models accounting for temporal autocorrelation was carried out using the package "nlme" (Pinheiro, Bates, DebRoy, Sarkar, & R Development Core Team, 2013). Bootstrapping was performed using the "boot" package (Canty & Ripley, 2017). Graphs were made using the r package "ggplot2" (Wickham, 2016). All reported average values in this study are shown as mean ± SE, unless otherwise specified.

| Seasonal variation in environmental conditions and annual NEE
Mean daily air temperature at the study site peaked in June-August (29.5 ± 0.2°C) while it was lowest from November to February  Table 1).
However, in 2016, annual NEE was 32 g C m −2 year −1 , becoming a net CO 2 source; NEE switched from net CO 2 uptake in the early months of the year to net emissions from June to the end of the year. In 2017, cumulative NEE was positive before September; however, due to the missing data after September, no conclusions about the annual CO 2 budget can be drawn.

| Effects of WL and ID on GPP and ER
In the GPP models, the effect of WL on α and P max depended on the length of inundation, as evidenced by the significant interaction terms (p < 0.01, Table 2). Specifically, a higher WL was associated with a higher apparent quantum efficiency (α) while extension of inundation period greatly restrained the magnitude of α by decreasing both the slopes and intercepts with changing WL (Figure 3a). In contrast, lower maximum ecosystem CO 2 uptake rates (P max ) were present at higher WLs (Figure 3b). Like α, the slopes and intercepts of the relationships between P max and WL also decreased as inundation extended. Overall, the effect of WL on P max was more notable as the ID was short. Once the ID approached 1 year, P max became almost irresponsive to the change of WL and its magnitude was already rather low (<2.7 μmol m −2 s −1 ).
For ER, WL and ID did not exhibit a significant interactive effect on the parameters b and R 0 (p > 0.05, Table 2). When excluding the interactive effect, WL showed significant effects on the parameter R 0 (i.e., base respiration rate when air temperature is 0°C) (p < 0.01) but not on b (i.e., temperature sensitivity of ER) (p > 0.05) (Figure 3c,d, Table S1).
In contrast, ID showed no significant effect on either R 0 or b (p > 0.05).

| Effects of WL and ID on daily NEE
The interaction between WL and inundation was also significant in the model for NEE (Table 2). Based on the model, the daily NEE was ~−0.51 g C m −2 day −1 as net CO 2 uptake when the site was not inundated (i.e., WL <0 cm and inundation = 0 day). As the WL increased to 45.6 cm (95% confidence interval: 41.5-51.6 cm), the ecosystem changed from a net CO 2 sink to source even at the start of inundation (i.e., ID = 1 day) ( Figure 4). As the inundation period extended, the WL corresponding to the carbon compensation point (i.e., NEE = 0 g C m −2 day −1 ) declined exponentially. As the inundation exceeded 214 days (95% confidence interval: 201-235 days), the ecosystem was a net CO 2 source regardless of the water depth.
regressions was lower than those with the length of the entire inundation period as the independent variable (Figure 5d,e).

| Effects of WL and ID on NDVI
Normalized difference vegetation index showed a significant relationship with WL (p < 0.01) and this relationship was independent of the ID (Table 2). Moreover, this relationship tended to be nonlinear, where NDVI value remained at ~0.41 under a WL of 0-40 cm and declined to ~0.29 only when WL increased from 40 to 70 cm (Figure 6a). In contrast, the NDVI showed no consistent relationship with ID ( Figure 6b).  (Table S2).

| Postinundation CO 2 fluxes and NDVI
Overall, compared to the other years (2009)(2010)(2011)(2012)(2013)(2014)(2015), daily NEE in 2017 showed greater net carbon emissions during the first 4 months following the inundation (Figure 8a). Its value became negative (net CO 2 uptake) only at the end of the third month after inundation, which was ~2 months later than other years. After the fourth month following inundation, NEE tended to be more negative in 2017 than the other years. Similarly, NDVI was lower in 2017 than other years (2009)(2010)(2011)(2012)(2013)(2014)(2015) within the first four postinundation months, and its value increased to a similar magnitude of other years only at the beginning of the fifth month (Figure 8b).

| D ISCUSS I ON
This study has shown links between inundation and ecosystem CO 2 exchange rates in a freshwater wetland. To the best of our knowledge, this is the first study that has analyzed the effects of inundation Abbreviations: ID, inundation duration (days); NDVI, normalized difference vegetation index; NEE, net ecosystem CO 2 exchange; WL, water level (cm). "WL-ID" indicates the interaction between WL and ID. * and ** indicate the coefficient is significant at p < 0.05 and p < 0.01, respectively.
TA B L E 2 Summary of the models relating parameters from light (α, P max ), temperature (R 0 , b) response curves, NEE, and NDVI to WL and ID F I G U R E 3 Parameters (a: α, b: P max , c: R 0 , and d: b) from the light response curve for gross primary production (GPP) and the temperature response curve for ecosystem respiration (ER) as a function of water level (WL) at different inundation durations (IDs). The relationships were determined based on least square mean predictions of the model with interaction terms between WL and ID for GPP in Table 2 and of the model without interaction terms for ER in Table S1 F I G U R E 4 Net ecosystem CO 2 exchange (NEE) as a function of the interaction between WL and ID based on the least square mean predictions from the model in Table 2. The solid curve indicates the compensation points (i.e., NEE = 0 g C m −2 day −1 ) under the corresponding conditions. The dashed lines represent the 95% confidence interval of the compensation points as continuous variables (i.e., WL and ID), rather than as binary or discrete variables (i.e., inundated vs. noninundated or high vs. low WL), on NEE and its components (i.e., GPP and ER). Our study highlights that the combination of WL and ID is the main driver controlling the annual CO 2 budget at the site. Importantly, with the influences of ENSO and water management activities, the extremely long ID and high WL that occurred in 2016 greatly restrained ecosystem CO 2 uptake, which shifted the wetland from what is typically CO 2 neutral or F I G U R E 5 The sum of annual gross primary production (GPP) (a, d, g), ecosystem respiration (ER) (b, e, h) and net ecosystem CO 2 exchange (NEE) (c, f, i) as functions of average WL during the inundation period (a-c), length of inundation period (d-f), and length of the period with WL >40 cm each year (g-i). Solid lines indicate significant linear regressions (p < 0.05) and dashed lines indicate regressions are not significant (p > 0.05). Gray areas denote the 95% confidence intervals

| Effects of inundation on GPP
Our results indicate that an increase in WL caused a decline in maximum ecosystem CO 2 uptake rate (P max ) (Figure 3b; Table 2). Theoretically, ecosystem CO 2 uptake is determined by the amount of photosynthetic tissue in the system and the photosynthetic rate of that tissue. At this site, macrophyte photosynthesis plays an important role in primary production (Schedlbauer, Munyon, Oberbauer, Gaiser, & Starr, 2012). An increase in WL submerges the macrophyte leaves, which reduces the total amount of exposed leaf area in the ecosystem (Schedlbauer et al., 2010). This decline in photosynthetic leaf area was also supported by the associated decrease in NDVI (Figure 6a), which partially explained the relationship between P max and WLs.  also found a substantial reduction in photosynthetic rates for the dominant plants (especially muhly grass, M. filipes) under inundation at the site.
Therefore, we suggest that both the declines in photosynthetic leaf area and photosynthetic rates contributed to the decrease in ecosystem level P max as WL increased.
At the same time, the effect of WL on P max strongly depended on the ID (Table 2, Figure 3b), suggesting that the effect of WL is more noticeable under a short inundation period. An increase in ID can enhance the physiological stresses that are imposed by inundation and further decrease plant photosynthesis (Bragina et al., 2002;Chen et al., 2010). However, unlike GPP, NDVI was unaffected by ID changes (Table 2, Figure 6b), which may be due to the poor responses of the vegetation index to plant physiological changes (Huete et al., 2002;Running & Nemani, 1988), especially under inundation conditions. As inundation period exceeds 1 year, the effect of WL changes generally became minor because the plants were already severely stressed as indicated by the low P max .
Overall, we suggest that enhanced inundation intensity (i.e., increase in WL and ID) decreases ecosystem CO 2 uptake by restraining the macrophyte photosynthesis and reducing the nonsubmerged photosynthetic leaf area. Nevertheless, this relationship can vary among ecosystems that are dominated by plant species with different inundation tolerance. For example, Jones, Stagg, Krauss, and Hester (2018) observed enhanced photosynthesis in Spartina alterniflora, a more inundation-tolerant plant, when exposed to a WL of ~10 cm. This increase of plant-level production may potentially enhance the ecosystem level net CO 2 uptake under a mild inundation. Therefore, more studies in different wetland ecosystems are still needed to develop a full understanding of the relationships between GPP and WL or ID.

| Effects of inundation on ER
The R 0 showed a decline as WL increased (Figure 3c, Table S1).
Particularly, we found that this relationship tended to be nonlinear, where no obvious reduction in R 0 was present when WL increased beyond 30 cm ( Figure S2, Table S3). While a decrease in soil heterotrophic respiration under anoxic conditions has been well-documented in previous studies (Anderson & Smith, 2002;Conner & Day, 1991;Fenner & Freeman, 2011;Happelll & Chanton, 1993), this nonlinear relationship between WL above the ground and ER has never been reported freshwater wetlands and these processes generally cause the water body in the ecosystem to be net heterotrophic (i.e., net CO 2 emission) (Hagerthey, Cole, & Kilbane, 2010). Increase in WL may potentially enhance the heterotrophic contribution from aquatic metabolism, which may dampen the overall decline of heterotrophic respiration at some points. Nevertheless, such nonlinear processes need to be considered in the ecosystem modeling.
Plant root respiration usually decreases with inundation due to oxygen limitation in the soil (Gleason & Dunn, 1982;Islam & Macdonald, 2004;Schlüter, Furch, & Joly, 1993), which may have also contributed to the decline of R 0 in this study. Unlike root respiration, leaf respiration usually decreases only when severe physiological stress develops and impedes the metabolism of the plants (Bradford, 1983;Gleason & Dunn, 1982;Liao & Lin, 1996). Some even observed a slight increase in leaf respiration under inundation stress (Chen et al., 2010). In this study, the duration of inundation showed no effect on ER, which may indicate that stress-induced changes in leaf respiration have only minor impacts on CO 2 emission at the ecosystem level.
Overall, our results indicate that inundation did reduce the rate of ER; however, a more intensive inundation (i.e., deeper and longer inundation) did not differ in its effect on the rate of ER from a mild inundation.

| Effects of inundation on NEE
Due to the unequal responses of GPP and ER, inundation weakened the net CO 2 sink strength of the ecosystem in our study, which is in line with research carried out in an episodically flooded wetland dominated by common reed (Phragmites australis) in China (Han et al., 2015). However, another study reported opposite results indicating that inundation enhanced the ecosystem CO 2 sink strength of a wetland dominated by Aleman grass (Echinochloa polystachya) in the Amazon floodplain (Morison et al., 2000). Based on our results, the changes in NEE under inundation were driven more by changes in GPP rather than ER, as ER was less responsive to high WLs (i.e., >30 cm) ( Figure S2, Table S3) and prolonged inundation ( Figure 3)

| Postinundation effects
As expected, the extreme inundation in 2016 was indeed followed by lower P max (Figure 7) and a longer recovery period to resume the average net CO 2 uptake (i.e., as indicated by negative NEE) in 2017 compared to other years ( Figure 8). Inevitably, this longer period of low P max enhanced the annual CO 2 source capacity in 2017 ( Figure 2d). Similar effect of extreme inundation has been reported for NEE in the following season in other wetland studies (Dušek et al., 2009;Hu et al., 2015). This reduction in P max was related to slow plant physiological recovery from severe inundation stress that limited photosynthetic rates and new leaf development (Chen et al., 2010;Hu et al., 2015;Kozlowski & Pallardy, 1979). This explanation was also supported by the relatively low postinundation NDVI in 2017 (Figure 8). Even though this extreme inundation event influenced the CO 2 uptake rates of the following season, no significant correlations were present between inundation intensity and postinundation CO 2 fluxes across the studied years (p > 0.05, data not shown). Therefore, we suggest that inundation reduces the net ecosystem CO 2 uptake in the following dry season only when it leads to extraordinary physiological stress to or even death of the dominant plants.
Interestingly, a greater ecosystem net CO 2 uptake was present after the recovering period (i.e., 4 months after the inundation)

| Implications for future hydrological changes
Everglades short-hydroperiod freshwater marl prairies are representative of wetlands with a regular seasonal inundation period of less than a year. Increases in WL and ID facilitated stronger declines in GPP than ER, and, consequently, a reduction in CO 2 sink strength.
An extremely intensive inundation event eventually turned the ecosystem into a CO 2 source. Other freshwater wetlands that have regular inundation periods may exhibit similar responses of CO 2 fluxes to inundation depending on the inundation tolerance of the dominant species. Given that extreme flooding occasions tend to keep increasing in many regions due to climate change (IPCC, 2013) or water management (e.g., Perry, 2004), we suggest that the carbon sink strength of the wetlands in these regions may be weakened, which would positively feedback to ongoing climate change. On the other hand, if inundation intensity remains high for longer periods, the vegetation in these systems can shift toward more inundation tolerant species within years (Armentano et al., 2006), which could further alter the relationship between CO 2 fluxes and inundation. Moreover, development of submerged photosynthetic capacity (e.g., from periphyton and submerged aquatic plants) over the long-term inundation may also partially compensate for reduced above water photosynthetic capacity under inundation. Our study reveals the relationships between the components of ecosystem CO 2 flux and inundation intensity, which are important for improving ecosystem and Earth-system models. However, the processes underlying the relationships and the long-term changes are largely unexplored. More studies that focus on macrophyte physiology, species composition, and aquatic carbon processes in response to inundation are still needed.

ACK N OWLED G EM ENTS
This research is based on support from the Department of Hernandez.